Recommendations, Analyses, and Data for Health Services Research, Diagnosis, and Prognosis Clinic

click here for older notes

Current Notes


Petrice Cogswell, Radiology (resident)

Survey on radiology program directors toward MD PhD residents and resident research. The responses were likert scale, how do you view PhD residents vs non PhD score -2, -1, 0, 1, 2 representing much worse, worse, similar, better, much better in multiple areas. Question: is there a statistical test to evaluate this type of data?


  • First step: see how comparable the responders are to the non-responders (or the whole set of programs) in terms of # residents, # MD/PhD residents, NIH funding amount
  • From there we can talk about statistical testing. We will probably want to use a finite-population correction since the whole population = 63 programs
  • Regardless of comparability, descriptive statistics (10 out of 23... etc.) will still be interesting to report

Yolanda McDonald, Human & Organizational Development/Peabody College (Faculty)

Project Title: An environmental justice review of drinking water quality in the United States, 2011-2015

Abstract: Despite the need for potable water for human life and EPA regulation of U.S. public water systems, there has not been a comprehensive study to quantify disparities in residential drinking water. This research systematically reviews results of the National Primary Drinking Water Regulations (2011-2015) by community water systems at the county-level. This study utilizes an environmental justice framework to (1) elucidate if legally enforceable drinking water quality standards differ based on community race/ethnicity, socioeconomic status, and rural-urban classification and (2) determine if communities with predominantly underrepresented groups are disproportionately burdened with repeat violations of drinking water violations.

Data Sources and Variables:

Dependent Variable: Drinking water violations for arsenic, atrazine, chlorine, coliform (Pre -TCR), coliform (TCR), combined uranium, di(2-ethylhexyl) adipate, di(2-ethylhexyl) phthalate, nitrates, nitrate-nitrite, lead and copper rule, radium, TTHM, TCE, haloacetic Acids, and Trichlorethane were downloaded from the Safe Drinking Water Information System (SDWIS) federal reporting services for the years 2011-2015 (N = 58,018). Of the violations, there were N = 30,981 repeat violations. Violations and repeat violations were operationalized as dichotomous variables (0 = no violation, 1 = violation).

Explanatory Variables: Race/ethnicity and socioeconomic status variables were obtained from the U.S. Census, American Community Survey, 5-year estimate (2011-2015) and were operationalized as continuous variables measured as proportions. The rural-urban classification is based on the USDA’s Rural Utilities Service (USDA RU) definition of rural. Rural-urban classification was operationalized as dichotomous variables (0 = urban, 1 = rural).

Data structure: The database structure format is one violation per row, Public Water System ID (PWSID) is the unique identifier. A PWSID may appear more than once in the database. For example, a PWSID could have multiple violations during the study year. The column data points are violations, race/ethnicity, SES, and rural-urban classification.

Proposed Data Analysis Strategy: The unit of analysis is county-level. Descriptive statistics were run to characterize the data. Correlation matrix measured the magnitude and direction of association between water violations and explanatory variables. To determine the relationship between water violations and the explanatory variables univariable and multivariable logistic regression analyses were used. The Variance Inflation (VIF) diagnostic was used to detect multicollinearity in the multivariable and interactions analyses. To detect confounding, all explanatory variables unadjusted odds ratio were compared to adjusted odds ratio to determine if there was a change of ≥ or ≤ 10% in the odds ratio (Szklo and Nieto 2014). And, multivariable logistic regression was used to adjust for confounding (Pourhoseingholi, Baghestani and Vahedi 2012). The Pearson goodness-of-fit statistic was used to compare the observed values to the expected. Covariates that had a likelihood ratio P value of <0.050 (two-tailed) and an odds ratio that did not cross 1.00 with a 95% confidence interval were considered to be statistically significant in the univariable and multivariable analyses.


Do we need to adjust for water systems, i.e. counties vary in the number of community water systems that service the area? If so, which of these options are recommended? a. Do we need to adjust the variance estimators of the estimated coefficients to account for the variance within the county, i.e. robust standard errors using clusters? b. Weight counties by population served by the community water systems? c. Stratify by community water system size (i.e. number of people served): Small Level 1 ≤ 3,300; Small Level II 3,301 ≤10,000; Medium 10,001 ≤ 50,000; and Large ≥ 50,001.

Do you recommend that we use Pearson goodness-of-fit statistic to compare the observed values to the expected?

Do you recommend post-hoc analysis for logistic regression? If yes, are there different post-hoc test for interaction terms?


  • We are concerned because we don't know the number of times each system was tested. If it's not possible to get this information, one possibility might be to simulate data to try to get a sense of the possible scope of the impact of frequency-of-testing
  • The overall project seems like a good fit for a VICTR voucher or short-term biostatistics support (its scope is too large for clinic). To inquire about short-term biostats support, email Yu Shyr, Chair. Another possibility might be working with a student (email Jeffrey Blume, director of graduate studies).
  • Will need to keep in mind: some systems get swallowed up into other systems.
  • Longitudinal data analysis won't be feasible without the complete testing data (we would need the non-violations in addition to the violations).
  • Next level (after other issues resolved): geospatial correlation (tricky, though, because of the upstream/downstream issue)


Kelly Schuering, Internal Medicine/Vanderbilt Familiar Faces (medical student), with Ed Vasilevskis (mentor; Department of Medicine, Division of General Internal Medicine and Public Health)

This study is looking at the prevalence of housing instability, risk factors for instability, and utilization of community resources among patients working with the Vanderbilt Familiar Faces program. Our research questions are as follows: Primary: Among patients with high health care utilization working with the Vanderbilt Familiar Faces staff, what is the prevalence of housing instability, potential future housing instability, and secure housing and what factors predict this? Secondary: What community resources are people using to help address housing instability and how would individuals describe their relationship with those resources? What predicts whether patients are connected to resources to assist with finding housing?

Data will be collected through a self-administered redcap survey on an ipad while patients are in the hospital.

Our analysis plan is as below:

Housing stability (ordinal):
  • Pearson’s chi-squared or Fisher’s exact test, depending on n in each category
  • Ordinal regression (vs. multinomial?)
We plan to include the following variables in the regression based on literature review and experiences with similar populations: consistent income (binary), employment (binary), current substance abuse (binary), legal history (binary), and current/recent intimate partner violence (binary)

Resource usage (binary):
  • Pearson’s chi-squared or Fisher’s exact test, depending on n in each category
  • Logistical regression
Based primarily on our experiences as there is not literature in this area, we plan to include having an outside case manager/social worker (binary), having a regular monthly income (binary), history of drug use (binary), and current housing stability status (categorical) in this logistical regression.

Since our multivariable analysis will not be able to account for every potential confounder, we will also conduct a sensitivity analysis to determine how big of an effect an additional cofounder would have to be to change the observed relationship.

Finally, we will also do a subanalysis of the pre-identified VFF patients compared with those who were assigned to the VFF team due to risk and bed space. This binary variable could also be included in the multivariable analyses.

I was hoping to get feedback on the above analysis plan and input on how many variables can realistically be included in the regressions if the estimated sample size is 200. I am also hoping to get an estimate for how much time we would need to purchase from biostats in order to get the above analysis completed.


For the binary outcome, the best-case scenario would involve 100 patients per outcome group, in which case it would be reasonable to adjust for 5 covariates in the regression model. Pre-specifying the covariates without looking at the data would preserve the Type I error rate, but with an exploratory analysis like this, that might not be your highest priority. If you are planning to present the analysis as exploratory and don't need to prespecify the model, a good starting place would be to look at plots and descriptive statistics for all variables by outcome group, and then to look at a scatterplot/correlation matrix and also to make a variable-clustering plot to get a sense of whether some variables can be used to "represent" others.

An ordinal logistic regression would be appropriate for the ordinal outcome, but depending on the sizes in the groups, you may need to collapse two of the outcome groups.

Possibilities for longer-term help: VICTR voucher and/or contacting Dr. Yu Shyr, Chair, to see whether short-term help is available.


Christian Okitondo, Psychiatry (Staff)

Topic: Increased tendency for proximal proprioceptive errors in limb bisection for individuals with autism spectrum disorder is not mitigated by too use.

Previous studies involving tool use tasks have shown that typically developing (TD) individuals commit distal errors in limb bisection after using tools, presumably due to perceptual extension of the peri-hand space. Given that individuals with ASD are less susceptible to visual override of veridical proprioceptive information in other proprioceptive paradigms, we hypothesized that individuals with ASD would not demonstrate these distal errors after tool use.

Questions I would to address: How to incorporate repeated measure on my ANOVA? For each subject, I have a pre training means and post training means. How to explain the repeated measure ANOVA to the world with no statistical background?

Ricky Shinall, Surgery (Assistant Professor)

I have a dataset consisting of about 350 responses to a quality of life instrument that has not been previously validated. I would like to get an estimate on the biostatistical effort needed to analyze the data for consistency and validity in order to obtain a VICTR voucher.


Devika Nair, Nephrology (Postdoc)

Attending this clinic is part of a requirement for my Biostatistics I class that I am attending for my MSCI, but I do have a question related to one of my projects.

I'm interested in exploring the coping behaviors of African American patients with advanced, non-dialysis dependent CKD. Based on what is available in the literature (which is limited), minority patients in general use religious coping to deal with the stresses of chronic illness. African American patients in particular seem to use denial/avoidant coping mechanisms. I believe that these coping mechanisms could in part explain why many of these patients disengage and disappear when the need for dialysis is mentioned. I believe that these behaviors are more related to cultural differences, rather than socioeconomic status or educational level.

If I am trying to illustrate a causal mechanism for why AA patients with adv CKD disappear when dialysis is mentioned (independent of their SES/educational status), would the best study design be to compare AA patients of both low and high SES, or would it be to compare AA patients with low SES with patients of other races with low SES?

  • You are welcome to come back to clinic, but as a member of the Nephrology division you are also welcome to work directly with Thomas Stewart
  • Recruit patients across a range of SES's; will probably want to limit to patients who are either AA or white, due to likely low numbers in other groups

Baldeep Pabla, GI (Fellow)

Also attending clinic as part of a requirement for Biostatistics I class for MSCI; particular project involves looking at a predefined set of SNPs in patients with and without GI cancer or metaplasia. Current literature suggests that environmental factors may play a greater role than genetics in the development of these conditions.

  • You are welcome to come back to clinic, but as a member of the Gastroenterology division you may be able to work directly with Chris Slaughter
  • Identifying appropriate controls for this study will be tricky


Shaina Willen, Clinical Fellow (Pediatric Hematology/Oncology)

I am preparing a VICTR proposal to study the impact of biomarkers of lung injury on complications in children and adults with sickle cell disease. I would like some assistance with my statistical analysis plan and how to determine power and sample size calculations.

  • for enrolled patients, look at previous 3 years then follow forward 2 years
  • 250 children & 300 adults seen at clinic
  • plasma & DNA samples
  • outcome is pain and acute chest syndrome
  • 3 genotypes and plasma biomarkers
  • believe 2,2 genotype will have increased pain and chest syndrome (1,1 (26%) 1,2 (55%) 2,2 (19%))
  • look at incident rate at 1 yr and 2 yr; poisson model; need to know what difference would be expected between the groups
  • sample size - graph of incidence rate that can be detected vs sample size needed
  • Simplest approach: find confidence interval formula for a Poisson rate; assume lowest true rate and solve for n such that multiplicative margin of error is 1.5 with 0.95 confidence
    • Simplest confidence interval is lambda +- 1.96 * sqrt(lambda / n); once you have an upper limit on lambda can solve for n to give acceptable margin of error for lambda
    • Would be better to get the multiplicative margin of error for the ratio of two Poisson rates (to simplify we may assume the sample size in each group is the lowest of the three genotype group sizes)
    • See

Alice Hoyt, faculty (Medicine/Allergy)

The aims of this R21 are to determine the preparedness and knowledge of K-12 schools on the topics of asthma and food allergy, then to pilot an asthma telemedicine program.


Ricky Shinall (Surgery)

Palliative care consultation has been shown to reduce utilization in end of life care, but this hasn’t been rigorously studied in trauma patients. I have access to Vanderbilt’s trauma registry which can be cross referenced against the palliative care registry to identify patients with palliative care consultation. I’d like to discuss the procedures for creating a propensity matched comparison of patients with and without palliative care consultation to compare resource utilization between the two groups. I’d also like to get a sense of the complexity of this analysis and the amount of effort from a biostatistician it would take to complete it.

Propensity score model resource:

Katie Tippey (Anesthesiology)

We did a card sorting projects where each participant sorted 4 decks of index cards. We recorded the time it took them to sort and notes based on asking them to think aloud while they sorted the cards. We created annotated files of photos of their final sorting arrangement and have created a raw data set based on these photos. We think we may want to use factor analysis on this raw data set but are unsure both if this is the appropriate analysis and, if so, how to perform this analysis.

Resources: section 16.1


Nicolas Forget (Emergency Department)

  • Perception of collaboration between doctors and nurses in Guyana
  • Pre- and post-team-building exercise, then 4-6w later
  • 27 participants with 2 dropouts by the end; 15 nurses, 10 doctors at the end
  • Issue of using means vs. proportions for Likert scales; want to look at disagreements of perception before and after training
  • Nurses used more spread of answers than doctors
  • 2 demographic variables, 15 Likert questions; need to combine into a single global scale for graphical individual profiles and for stat analysis
  • Can do a formal analysis of variability of responses within subject, e.g. compute the SD over 15 questions within subject and see if nurses have more variation than doctors
  • Main analysis on mean
    • Form within-person difference from baseline (paired data)
    • Do 2-sample (unpaired) t-test comparing these differences - nurses vs doctors
  • Be sure to graph all measurements (on summary score)
  • Pre-post design often provides an upper limit to an intervention effect


Maureen Saint Georges Chaumet (fellow)

Project description: I am starting a project that compares the cosmetic outcomes of 3 different laceration closure methods in kids: sutures, tape and glue. I will also be looking at several secondary outcomes.

Overall, study design seems reasonable. Recommended that parents rate the cosmetic outcome of the laceration in addition to 3 reference pictures. If parent's responses can be captured via an online Redcap survey, consider the possibility of a sliding scale response.

Recommend 90 hours of biostat work. (Does not involve writing more than one manuscript)

Samuel Younger (Nurse Practitioner)

Interested in determining sample size and best statistical approach. HLM-SEM vs Path analysis?

Research Abstract

Many organizations are looking to their staff to creatively engage in improving the safety of patients. Further, within the Magnet health care environment, transformational leadership is the theory that has been promoted as core to the achievement of patient outcomes, thus is the core focus of this study. The purpose of this research is to examine the role that leaders play in bringing together elements of a safety culture and a climate of innovation that support and enable staff to engage creatively in improving the quality and safety of patient care. There is little empirical evidence in the nursing literature related to patient safety in an innovative climate, and none could be found that study the leadership behaviors of nursing managers that are conducive to an innovation climate and impact on patient safety outcomes in a Magnet designated, Academic Medical Center. Therefore, this study seeks to fill that gap in knowledge and expand the leadership and innovation literature to include patient safety within a Magnet work environment.

This research uses a multi-level, cross-sectional, descriptive correlational design aimed at examining the relationship between nurse manager transformational leadership and front line nurse rated patient safety score, and to further investigate how, if any, does communication and feedback about error and the innovation climate influence the relationship. The independent variables in this study are transformational and transactional leadership. The dependent variable is front line nurse rated patient safety score. The innovation climate is proposed to be a mediating variable. Feedback and communication is proposed to be a moderator variable between transformational leadership and patient safety score. The variables will be measured through an online survey based the three validated and reliable survey instruments (54 questions): the MLQ-5x short (MLQ-5x), the Team Climate Inventory-short (TCI), and Feedback and Communication About Error and Patient Safety Grade (subscales of the AHRQ Hospital Survey on Patient Safety Culture) which are all appropriate for collecting data about the perceptions of front line nurses.

If findings confirm these relationships, then in order to impact outcomes, nursing managers may need to be adept at navigating and promoting the complex nature of innovation through communication and establishing an innovation climate. In this context, leadership facilitates communication and an understanding of the innovation climate, which supports creative solutions to patient outcomes and improved quality, in this case, patient safety. On a practical level, this study will contribute to a greater understanding of how to prepare future nursing leaders for the challenges of a changing healthcare landscape through an understanding of what behaviors are necessary to generate innovative and safe care delivery models.

H1a: There is a significant, positive relationship between nurse managers’ transformational leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and nurses’ perception of patient safety as measured by patient safety grade (AHRQ HSOPSC).

H1b: There is a significant, positive relationship between nurse managers’ transactional leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and nurses’ perception of patient safety as measured by patient safety grade (AHRQ HSOPSC).

H1c: There is a significant relationship between nurse managers transactional leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and nurses’ perception of patient safety as measured by patient safety grade (AHRQ HSOPSC), but to a lesser degree than transformational leadership. Included per our discussion on transformational leadership predicting quality above and beyond that of transactional leadership.

H2a: There is a significant relationship between nurse managers’ transformational leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and innovation climate as measured by the Team Climate Inventory (TCI-short).

H2b: There is a significant negative relationship between nurse manager’s transactional leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and innovation climate as measured by the Team Climate Inventory (TCI-short).

H3: The relationship between nurse manager transformational leadership as measured by the Multifactor Leadership Questionnaire (MLQ-5X) and nurses’ perception of patient safety as measured by patient safety grade (AHRQ HSOPSC), will be mediated by innovation climate as measured by the Team Climate Inventory (TCI-short).

H4: The relationship between transformational leadership and patient safety grade will be moderated by feedback and communication about error. In terms of this relationship, transformational leadership will have a stronger, positive relationship with patient safety scores when feedback and communication about error is high.


Ryan Skeens (fellow)

This is a patient activation measure survey conducted on parents/caregivers of NICU patients. Survey will be conducted at NICU enrollment, NICU discharge, and 30 day after discharge. The hypothesis is that patient activation measure will decrease at NICU discharge but increase over time (30 day after discharge). In addition, characters such as social economic status that links to high patient activation measure will be identified.

The measure has been validated and used by mentor team. This is a fellowship project, and Ryan will apply an internal grant for the 6-9 months project. Further, CTSA support will be explored.

  • Sample size is fixed based on fellowship time. Power and sample size should be calculated accordingly.
  • Keep the measure in the continuous form (0-100) instead of dichonimization.
  • Consider to have CTSA statistician's early involvement at the design stage. Given this involves design, grant writing, data collection, data analysis, and manuscript preparation, a 90 hour work maybe needed.
  • As prediction is involved (identify characters that are related to high measures), model validation should be considered.



Danxia Yu, Epidemiology (faculty)

We will examine the associations of diet quality scores (assessed at baseline) with body weight change (from baseline to following visits) in a prospective cohort study. Generalized estimating equation model has been used in other studies, which we are not familiar with. We need statistical inputs on this model and the power estimation. We also would like to find a statistician whom we may work with on this project. Thank you.

  • If dropout is not random, either GLS with a serial correlation structure or a linear mixed-effects model would be more appropriate than GEE.
  • Do not collapse the diet variables into quintiles; leave them as continuous variables
  • For the power calculation, it may be possible to ask for conditional approval to have access to a subset of the data to get estimates of the quantities needed for a power calculation.
  • You can do a simplified power calculation with just one wave of data, and argue that the power will be higher when there are more data points per person.
  • Possibly useful R packages: longpower (thank you for bringing this to our attention!), pwr (in particular, the pwr.f2.test function).
  • Simulation could also be a useful approach, but it would also require some background information about the standard deviations of the variables

Joshua Cohn, Urologic Surgery (clinical fellow)

I have two questionnaire-based databases on overactive bladder that I have merged. I would like to use this data to develop a model that predicts bother based on symptoms and comorbidities and prioritizes necessary treatments. I am not sure if cluster analysis is the best way to do this.



Paul Yoder, Special Education (faculty)

I'd like evaluation of area under the curve (AUC) as a way to quantify the magnitude of the between treatment-group-difference and its confidence interval for RCT with repeated measures of the dependent variable. A reference for an example is Gallop, R. J., Dimidjian, S., Atkins, D. C., & Muggeo, V. (2011). Quantifying treatment effects when flexibly modeling individual change in a nonlinear mixed effects model. J Data Sci, 9, 221-241.

  • Email Hakmook Kang to talk about the possibility of working through the KC biostatistics core to get an estimate of how many children and timepoints you would need to do the flexible-breakpoint approach discussed in the article
  • We also discussed an approach using restricted cubic splines. It's possible that this approach would let you use fewer subjects; it may be useful even though you are expecting a linear relationship

Bryan Hill, OB/GYN (fellow)

This is a follow up from recommendations from 5/15/2017 regarding a logistic regression model of post operative complications as the output variable and clinical and demographic variables as the independent variables. The recommendations, in summary were:

1) Treating the outcome as an ordinal, rather than binary, variable if there are enough people in the additional groups

2) Look at the cross-tabulation between physician and sling type to see whether it is feasible to include both

3) Leave the continuous variables as is (do not categorize them). May want to consider log-transforming age.

4) Try variable clustering to see which variables may be collinear/redundant

5) Consider combining less important (less interesting) variables into a score

Goal for the session: to discuss results of the model.



No clinic---Memorial Day



Bryan Hill, Fellow, Gynecology

Reporting complications after surgery are important for quality improvement. Two methods of finding complications are: 1) administrative data from diagnosis codes and 2) key-word search from a manual chart review. We suspect the administrative reporting method under-reports complications. The primary aim of the study is to determine sensitivity and specificity of the administrative method compared to the manual reporting method. The secondary aim is to determine which risk factors are associated with having a complication.

We think that creating a logistic regression model would help address our secondary aim. Our plan is the following: setting the output as "complication present (1)" and using the variables: asa class, age, body-mass index, setting (outpatient or inpatient), sling type, attending, if a concomitant procedure was done, anesthesia time, operation time, smoking history, diabetes, and prior surgery.

Question #1: We need guidance on how many variables we can include in our model. Some have high numbers, and some are quite low.

#2 Some variables may influence each other. For example, sling type is heavily dependent on attending (they like to chose a particular brand or type). How do we adjust our model for that?

#3 It is known that older patients are more likely to experience complications. How do we determine if age is independently associated with "complication presence" versus just being a confounder influencing other variables?

Files we plan to append: data dictionary, STATA file, table of variables with total numbers of responses.

  • In deciding which categories to collapse, look at the sample overall (not by complication status)
  • To increase power, consider treating the outcome as an ordinal, rather than binary, variable if there are enough people in the additional groups
  • Look at the cross-tabulation between physician and sling type to see whether it is feasible to include both
  • Leave the continuous variables as is (do not categorize them). May want to consider log-transforming age.
  • Try variable clustering to see which variables may be collinear/redundant
  • Consider combining less important (less interesting) variables into a score
  • For binary logistic regression, we generally want to have 10--20 people in the smaller outcome group for every degree of freedom (continuous variable or single category) in the model
  • If you apply for VICTR funding, we recommend the larger time amount if you are interested in a publication or presentation. In your application, you can cite these notes as evidence that you have been to a biostatistics clinic.

Mike Temple, Biomedical Informatics, faculty

I am comparing the results of 2 surveys and need help calculating p-values and odds ratios to determine significance between the 2 surveys. I am using R

  • Get more information about the survey design (especially number of people surveyed) so that you can compare the response rates in 2012 and 2016. If they are not close to each other, it will be harder to justify comparing the results of the two surveys
  • If possible, get info about demographic makeup of the people surveyed in 2012 and 2016 from the organization's records. If, for example, the mean age of respondents is very different from the known mean age of the people surveyed, you will know that in at least that one aspect, the respondents are not representative of the people surveyed.
  • Chi-squared tests should be fine if the categories are exhaustive (but this is secondary to the nonresponse issue)
  • If possible, get more info about the outcomes and model specifications used for the regressions in Table 3.


Chirayu Patel, resident physician, radiation oncology

The project is VEEP-C - Visually Enhanced Education for Prostate Cancer, a randomized, controlled trial to assess the impact of a visual presentation on prostate cancer treatment decision-regret, anxiety, satisfaction, and patient-reported symptoms, in the radiation oncology department. The expected accrual for patients was 112 patients based on 120 prostate cancer patient consultations seen within a 6-month timeframe. Unfortunately, due to a drop in consultations, only ~30 patients have been accrued, and only 1 patient has completed external beam radiation therapy over a 6 month timeframe (other have undergone brachytherapy, surgery, active surveillance, or are still deciding).

1. The sample size is based on an instrument which only 1 patient has completed. As originally written, the study is not feasible. Determination of new outcome and sample size?

2. Role for interim analysis on secondary outcomes?

3. Thoughts on closing the trial due to poor accrual?


Cara Singer, PhD Student, Speech and Hearing

  • This project investigates speech-language imbalances in children. We are interested in the best way to measure imbalances using five standardized tests. Simple range scatter and standard deviation have been discussed. We are also interested in the best way to analyze whether increased synchrony between the five tests is associated with a decrease in stuttering frequency based on two years of development.

Hatun Zengin-Bolatkale, Faculty, Hearing and Speech

The purpose of the present study was to longitudinally assess sympathetic arousal (i.e., physiological correlate of emotional reactivity) of preschool-age children with persisting stuttering (CWPS), those who recover from stuttering (CWRS), and their normally fluent peers (CWNS) during a stressful picture-naming task. The apriori research questions/ hypotheses are as following:

The first question addressed whether change in SCL in response to stress at initial testing - close to the onset of stuttering - is associated with stuttering chronicity (i.e., persistence vs. recovery). We hypothesized that children whose stuttering persists, compared to those who recover and those who do not stutter, would exhibit increased skin conductance reactivity to a stressful picture naming task at their initial testing (i.e., prior to stuttering resolution for children who recover).

The second question addressed whether change in SCL in response to stress - approximately 18 months after their first testing – is associated with stuttering chronicity (persistent vs. recovered patterns). We hypothesized that children whose stuttering persists, compared to those who recovered and those who do not stutter, would exhibit increased skin conductance reactivity to a stressful picture naming task at 18 months-post-initial testing (i.e., after stuttering resolution for children who recover).

The third question addressed whether changes in SCL in response to stress are associated with changes in stuttering frequency. We hypothesized that for children who persist, compared to children who recover and children who do not stutter, increased skin conductance reactivity would be associated with increases in stuttering frequency.

We would like help from the clinic with the analyses of the hypotheses above, especially for #3.





Sarah Diehl, Hearing and Speech Sciences , PhD student

* Questions for the clinic:

1. After removing the ratings that have a mean score of 2 or below, there will be ratings that will highly correlate. Should we first do something like a multi-dimensional scaling approach to identify dimensions and then a cluster analysis to see how these dimensions cluster? Or do we throw all ratings (potentially 38 if none receive a mean score of 2 or below – realistically perhaps something like 20 to 25) into a cluster analysis.

2. If we expect at least 2 or 3 clusters, what is a reasonable sample size given the number of items we have on the rating scale?

3. What do we need to put into a proposal that is going to use cluster analysis? What kind of information is critical?

4. Is there another approach that would work better than cluster analysis?

Gurjeet Birdee, Health Services Research, Faculty.

  • The objective of this study was to measure the energy expenditure (oxygen consumption O2/kg/min) of adults practicing common yoga movements. For each individual, participants were asked to do movements in a standing position, lying position, and seated position (body orientation). In addition, each movement was done with different variations serially. In addition, participants were asked to walk at low and moderate intensities to compare energy expenditure of a comparative aerobic exercise to yoga.

The main questions we would like addressed:

What is the best approach to measure if there was significant variation between individuals for mean energy expenditure by body orientation?

What is the best approach to measure if there was significant variation between individuals for each movement?

When considering if variation exists above, should we take into account resting energy expenditure for each individual?


Cara Singer, Hearing and Speech Sciences, PhD student

  • This project investigates differences in skin conductance levels in children who stutter and are persisting, children who stuttered and recovered, and children who do not stutter. All children were followed 3-4 times across a two year period. At each visit, skin conductance levels were measured during a neutral video and speaking task, a positive emotion-inducing video and speaking task, and a negative emotion-inducing video and speaking task. We would like to discuss the best statistical models for our hypotheses.

  • Note that at each timepoint, there are 7 skin conductance measures (a "baseline" and 6 other measures)

  • Recommendations:
    • Keep all possible timepoints from all possible subjects. Do not exclude subjects based on their trajectories or baseline characteristics
    • Use continuous versions of the stuttering outcomes if possible; at a minimum, collapse the outcomes into 5 ordinal categories
    • Use a longitudinal mixed-effects model. Each subject will contribute 1, 2, or 3 rows depending on how many of the timepoints they have. You can model severity as a function of time-1 severity, age, sex, the seven time-1 conductance measures (or a reduction thereof; try a redundancy analysis first), time in days, and squared time in days, with random effects for subject (and possibly time and squared time). We recommend a continuous-time correlation structure, but this might be tricky with the mixed-effects model; generalized least squares might work better.
    • If we can get a clear, simple plan and the analysis is not a multi-step analysis and the dataset is clean (and tall and thin, with the relevant time-1 variables and non-identifying subject ID on each row), we may be able to conduct the analysis during a clinic.
    • Starting next month, we will be able to take on longer short-term projects for a charge.
    • The Kennedy Center statistics core may also be able to do this. If you come back to a clinic, please remind us to invite Hakmook.


Kristy Broman, Surgery Resident

Method to compare standardized incidence ratios using SEER data


Katie McGinnis (MPH candidate)

(followup from last two weeks)

  • For each overall question category, try a scatterplot of a) the means and b) the standard deviations for each item, with staff values on the x-axis and parent values on the y-axis (or vice-versa). Label each point with the question number or a short phrase to identify it
  • Do variable clustering within the staff items and the parent items, to see which items tend to be answered similarly by the same person (hcavar in stata)
  • Rather than doing several univariate analyses comparing the relationship between the demographic items and each survey item, do a single regression analysis for each survey item, with all the demographic items included in the model at once. Collapse the categorical items into 2 or at most 3 categories, and just assign numeric values (e.g. 1--5) to the levels in the binned continuous items like distance and treat those as continuous variables (so they will have just one term in the model). Actually, though, drop distance altogether and just use travel time. The overall F-statistic from the regression will tell you whether anything in the model matters. The best approach would be a proportional odds model, but ordinary regression will be next best.
  • It's ok to take the means of means (across items in a particular category) and talk about those, but there aren't enough data points to warrant a statistical test.


Katie McGinnis (MPH candidate)

(followup from last week)

  • Instead of doing t-tests, do wilcoxon rank-sum test (only 5 response options)
  • Rather than overlaying the parent and staff histograms, show the parent mean as a dot on the staff histograms
  • Do the "dot-histograms" by hospital because the hospitals are so different, even if tests comparing hospitals are not significant
  • Don't put too much weight on the p-values; this is exploratory research with relatively small sample sizes
  • For the two similar staff questions, run a correlation on the responses to help justify using only one of the questions. Use a Spearman rank correlation.
  • We don't think it would make sense to take the mean of the responses for the parent "how often" questions
  • For any set of questions, it could be interesting to order the means to see which questions had the highest or lowest means, but it wouldn't make sense to do a statistical test comparing the means of the different items.


Antje Mefferd, Hearing and Speech Sciences

I’m an assistant professor in the Hearing and Speech Science department and I’m currently preparing a manuscript. I would like to have someone take a look at the analysis that I completed to make sure they are correct. I’m a bit unsure about some things (assign fixed and random effects, reporting of degrees of freedom). I have my data in excel spreadsheets and can share it ahead of time.

The topic is how the tongue and the jaw change in their range of motion during various speech tasks (speaking typical, loud, slow, clear). These speech tasks are used in speech therapy to help people with brain diseases (Parkinson’s disease) to be better understood. In this data set I look at this in just one group of speakers (healthy speakers).

Participants complete 5 repetitions for each task (5 reps x 4 tasks = 20 data points from each participant). There are 11 females and 10 males in this study (sex has a significant main effect due to anatomical differences between males and females, but it is typically not statistically controlled for in our field in repeated measures). There are three measures – tongue movement, jaw movement , and the acoustics. For all three I need to analyze task effects in separate analyses. I also need to look at how changes in tongue movements predict changes in acoustics and how well changes in jaw movements predict change sin acoustics using data of typical to loud speech, typical to clear speech, typical to slow speech -- this time regressions within females and within males.

In the meeting I would like to make sure that I ran these analyses correctly and also would like to verify that I used to correct degrees of freedom in my write-up.

Recommendations: 1. For primary analysis, either ANOVA using each subject's mean or mixed-effects model with fixed effect for task and random effects for subject would be fine. 2. For secondary analysis, it would be best to use the same approach (either one mean data point per person per task, or a mixed-effects model). If doing mixed effects model for secondary analysis, be careful with the interpretation of R-squared.

Katie McGinnis (MPH candidate)

I have questions about my MPH Thesis project, specifically related to the best options for comparing some of my variables and running a few other statistical tests

Practicum in Kenya; originally a needs assessment, not designed for research. 16-page staff surveys (n= 94) & parent surveys (n= 69) from 2 children's hospitals, plus demographic data. Hoping to compare parent responses to staff responses in some way. Challenges: 1. parents are responding about 1 child but staff are responding about all children, and 2. for some items, the response scales for parents and staff are slightly or very different. She is comfortable treating the response options as numeric (taking the mean would be meaningful to her). The thesis does not have to contain a formal statistical analysis.

Recommendation for next steps: For survey items where the response scales are the same, continue the exploratory data analysis by plotting histograms for the staff responses, and then marking the mean of the parent responses on the x-axis.


Frances Anderson, MPH Global Health

I am an MPH Global Health track student and I need some assistance with ANOVA analysis on my thesis project. My project is an evaluation of Minnesota's TB screening of refugees and immigrants across four counties in the state. The data I am looking at for ANOVA includes mean days to initiation (TB testing) and mean days to disposition. There are some outliers in the data that I need to consider dropping. I seek advisement in this, completing the test, and if ANOVA is not appropriate for this dataset finding a new test.


Joshua Cockroft, MD student

We are looking to design and validate a new psychometric scale that measures a patient/client's trust in new providers. Though psychometric scales currently exist that measure trust in healthcare systems, trust in existing personal providers, and measures of global trust, there is currently no scale described in the literature that specifically measures trust in new providers. The hope is that such a scale would be of use in many underserved populations, particularly those populations with histories of either substance use disorder or severe mental illness, who are not regularly active participants within the healthcare system. We would hope to be able to use such a survey to measure the effect of this specific type of trust on outcomes such as healthcare service utilization. Like other healthcare trust-related scales, this scale would likely be a Likert-scale with questions that would span multiple domains of trust (i.e. competence, dependability). As there is no current gold standard for this type of measurement, advice on important considerations for internal validation would be greatly appreciated. We may consider the validation of this scale in multiple sub-populations if able. Conceptualization of this scale will be derived from the literature and our own qualitative research.


Angela Maxwell-Horn, MD, Assistant Professor of Developmental Pediatrics, Monroe Carell Jr. Children’s Hospital at Vanderbilt

I am a pediatrician wanting to do a study about the effectiveness of a medication to treat ADHD symptoms in children with autism. I would like to come to a biostats clinic to help me figure out what type of analysis that I should do and how many subjects I need to effectively power my study. I have attached a copy of my study proposal.

  • Recommend a randomized cross-over study design with double blinding if possible
  • Select a side-effect measurement tool
  • Clearly state inclusion/exclusion criteria

Heather Limper, Center for Clinical Quality and Implementation Research

"I would like to get some help with execution of times series analysis using STATA (ideally)."


Katie McGinnis, MPH Candidate, Global Health

Perform surveys in three children hospitals on parents and staff. 69 respondents from parents and 97 from staff. Parents survey: demographics about parents and children, how the experiences in hospital impact parents and children, patient satisfaction Staff survey: demographics, education, child's hospitalization needs

Research questions: what do you think caused the child's illness? The language barrier in receiving proper care? The correlation between child's experience in hospital and staff's education and experience.

Survey matrices are similar in parents' survey and in staff's survey (a dozen of likert-scale questions). Want to check the correspondence between parents' responses and staff's. First check if parents agree with each other. Code the answer to each question as 1,2,3,4,5. Summarize the score of each question across all the patients. Small SD is an indication of better agreement between parents. Second check the consensus of staff. Third, to evaluate the staff's characteristics, compare staff's responses to parents' consensus; to evaluate the parents' characteristics, compare parents' responses to staff's consensus. Take the difference between staff's response and parents' consensus as outcome, fit a regression model on providers' characteristics.

Could generate a summary score over multiple questions in one category (Rockwood's index).


Samantha Gustafson, Hearing and Speech Sciences

VICTR application for dissertation research. EEG measures for speech sound processing in quiet and in noise. Looking for age effects. How does the effect of noise change with age? Proposed analysis based on linear regression. Expects one EEG measure to be more sensitive than the other. Second question is to look for mediator with EEG response and how well they do behaviorally depending on age. Particularly tricky how to size a study for an exploratory mediation analysis. Have replaced repeated measures ANOVA with a linear model. Each EEG task takes 10 minutes. Two listening conditions, same task. Quiet vs noise order is randomized. Half of participants hear "da" and the other half receive "ga" (randomized). Model: EEG = intercept + age effect + noise/quiet + age x noise/quiet interaction. Can use generalized least squares (correlation structure irrelevant except don't assume the correlation is zero, since only 2 times per subject) or repeated measures ANOVA if very careful to use the correction for correlation (if can handle interaction between group (noise/quiet) and age). But GLS is ideal. Need to check normality assumption of residuals.

Power of a test of interaction is much lower than a test of main effect (difference in slopes vs. slope not being flat). Data not available for making initial guess of sample size required to achieve a given precision or power. Only thought is related to a minimum possible sample size - the size needed to estimate a difference in mean EEG for an adult with very good precision. The SD of the noise-quiet difference is used here. Once the acceptable margin of error (half-width of 0.95 confidence interval for the mean difference) is determined can plug in formulas related to precision - see e.g. . Beware: sample size needed for interaction is easily 4 times as large.

Alec Pawlukiewicz, Neuroscience and Psychiatry

Effect of exercise on neuro cognitive testing. Database of 20,000 participants - 9,000 after exclusions. Control for covariates sex, age, education level, # prior concussions. Interested in matched analysis. Not having enough controls. Suggested using full qualifying sample without matching, to maximize power and avoid any arbitrariness in how matches are determined. Non-matched analysis requires careful specification of the statistical model.

Several neuro scores are given by the test. If scales are continuous enough can use the standard multiple regression linear model if analyze one score at a time. May need to model age as a smooth nonlinear effect and perhaps likewise for education. Age and education may be co-linear. Variable of major interest is exercise (binary). Need to consider whether exercise may interact with age, sex, etc. What about type of exercise? For variables such as # prior concussions a quadratic effect often suffices.

Dillon Pruett, Hearing and Speech Sciences

Respiratory sinus arrhythmia. Comparing in children who do not stutter, stutter and persist, stutter and stop. They watch a video followed by a task, and this is repeated with different videos/tasks. Baseline re-measured at the end. Question about whether to form groups or to have a continuous-time longitudinal model with stuttering measures as the response variables (without categorization). Answer questions by estimating difference in means over time. Need to interpret the result in a clinically meaningful way. Need to adjust for baseline stuttering measure as a covariate. This might possibly be interacted with the intervention effect. Need to carefully formulate the linear model and account for within-subject correlation using something like GLS or mixed effects models (the latter is mainly used if there are more than 2 or 3 measurements over time within subject).


Omair Khan, Center for Research on Men's Health

  • "I would like to request some time to talk to another statistician about exploratory factor analysis I am doing in R with the psych package. This procedure is fairly new to me and I have some questions that I would like help with."


Mary Lauren Neel, Neonatal

  • Association between ITSP and illness severity score
  • Association between parenting style (PSDQ) and infant adoptation.

Mark Tyson, Urology

  • Bladder neck size on incontinence, controlling for BMI, age, preop score, disease status, and stitch.
  • Restricted cubic spline examples: MSCI Biostat II STATA


Dillon Pruett, PhD student in the dept of hearing and speech sciences working with Dr. Robin Jones

  • I'm working on a project involving longitudinal data with children who stutter and persist, children who stutter and recover, and children who do not stutter.


Scott Karpowicz

  • Matched design, 1:1, 1:many, BOOM
    • match on socio-economic, clinical factors, etc.
  • Change point analysis
    • see if readmission rates change at time of policy implementation
  • REQUEST FOR VICTR SUPPORT: Clinic statisticians recommend a 90 hour voucher.


Sam Gannon


  • Developing a randomized controlled clinical trial in mental literacy. Working notion, to increase mental literacy, communications which in turn increase mental health outcomes.
  • Submit concept paper to NIMental Health. Questions to address and want to get statistical expertise.
  • Questions: 4 educational arms and a control group for a total of groups. Setting community mental health clinics
Metric for outcome measure clinician reports notes - self management, behavioral adherence to protocol and rate of compliance These response measures are known to be correlated. Intervention: different educational programs. Control will have standard of care.
  • Consider cluster randomization. Figure out how many clinics that you will have access to. Five arms note one clinic receive one arm.
  • How to assess "fidelity"? Recording data consistently. Approach with assessment for some of inter-rater reliability.
How do you capture your outcome? If survey or standard form then it will be much easier to make results consistent. For example if reporting is done through RedCap, you will have the opportunity to formalize or standardize process.
  • Mediation analysis (Baron & Kenny, structural equation modeling). First you need to show that your intervention has an association with response variable. Mediator will be communication for example
. What factors mediate the intervention?
  • * (Y~X) Education is associated with improved mental health.
  • * (X~M) Education works through health literacy and/or communication(Mediators) to improve mental health.
  • Will I benefit from cross-over design? We believe that once knowledge is gained it will be difficult to have a "wash out". Cross over design will be more appropriate to a set up such the development of new drug with clear wash out.
  • Question from biostatisticians: do you need 4 arms? Can you combine some of these educational programs.
  • Transient effect: Is it common in the literacy literature and look into other clinical studies such as in diabetes which require behavioral changes. There are issues of relapse and maintaining adherence.
  • Timeline: Extend two years follow up time to address the "transient effect" although most studies have short follow up. Can you follow up subjects on StarPanel to show that you can address long term effects. Need to sit down with statitiscians to address realistically the multiple issues. How many clinics do you think that you could have access to? Recruitment time? How many subjects are needed?
  • Consider short term effects and long term outcomes. Can you design you study pragmatically without too much effort to collect data? Using the real set up Dr. entries for follow up assessment.
  • Recommendation: Follow up with VICTR voucher and statistician for help with proposal.


Heather Lillimoe

General Surgery Resident

I am currently in the process of designing a research study pertaining to resident feedback within the department of surgery. My hope is to utilize REDCap for my primary mode of obtaining data. I was hoping to meet with a biostatistician as I apply for VICTR funding for the study. It involves an educational timeout before an operation. This is a 3rd year rotation in plastic surgery. There is an iphone app to do a competency rating.
  • Survey - baseline assessment - residents and attendings - 85 questions
  • Additional survey after rotation

Cara Singer

I am a PhD student in the Department of Hearing and Speech Sciences. I would like to attend the biostat clinic today (if possible) to discuss appropriate analyses for a study I am conducting under the mentorship of Robin Jones (Developmental Stuttering Lab). The study is investigating whether a risk factor assessment (a mix of categorical and continuous variables) can predict stuttering persistence. 70-80% spontaneously recover. Would like to identify those likely to persist, in advance, for focusing therapy. Multiple risk factors have been identified. Empirical evidence for supporting predictive ability of the risk factors is sought.
  • Children previously seen - diagnostic visit; 4y ago; stuttering up to 18m; English is primary language
  • New follow-up for status at one point in time
  • Baseline variables that originate from continuous measurements (e.g., age at onset) need to be analyzed as continuous variables
  • Include baseline stuttering severity as a predictor
  • With a maximum of 150 children the maximum number of candidate predictors might be around 10 if the outcome variable is almost continuous (it's worse if outcome is almost binary)
  • Stuttering is multi-dimensional, e.g., some children may reduce amount of speaking because of the problem, so they seem to stutter less
  • May consider a compound summary of all the outcome measures, e.g., average rank across children; clinical ranking of scenarios can also be used
  • Dependent variable needs to have at least 5 frequently levels and be ordered or continuous
  • If there is one standout, popular scale, that one could be used by itself
  • Empirical variable selection requires an enormous sample size to reliably find the "right variables" so it's best not to use selection procedures; can find various approximations to the model for clinical non-computerized application
  • Data reduction methods (variable clustering, principle components, redundancy analysis) can be useful for effectively reducing the number of predictors to use in the multivariable model


Chris Brown, Internal Medicine Resident

  • To go over analysis produced by VICTR biostatisticians


Kazeem Oshkoya, Division of Clinical Pharmacology, Dr. Dan Roden's Lab

Data analysis on blood sample storage and drug concentration - look at whether a gel absorbs too much of a drug in the blood to make drug assessment accurate enough. Measured at baseline and 4h. Need to know how to describe the base value. Triplicate measurements available. More interested in relative comparison.
  • Best to present all the raw data
  • Might use 3 quartiles (25th and 75th percentiles and median) as descriptive stats and use Wilcoxon signed rank test for testing for a difference between baseline and 4h
  • There's also two types of samples - same study repeated with different samples, sample drug concentration
  • Only have 2 patients; plan to have 5 later
  • Better to not average over the 3 replicates - may hide variability
  • Bland-Altman plot (mean-difference plot) is a good way to show agreement and whether variation is stable over base levels. If band of variability expands going from left to right, this is an indication that perhaps the analysis should be done on the log concentration scale.
  • Other useful ways to summarize data: mean absolute difference between estimated and true concentrations - separately by no gel and gel
  • Can also show mean absolute differences between replicates ignoring the true concentrations
  • There are problems with lower limit of detection, representing missing values that are not randomly missing; ordinary analysis may be problematic

Jessica Dennis, Lea Davis, Genetic Medicine

Modeling lab values to look for genetic variation; data from the synthetic derivative
  • Interested in variation over time within patient
  • Variants are summarized into polygenetic risk scores
  • Difficulty in interpreting results if patients are being treated for the lab abnormality being studies
  • How to define time zero?
  • May want to ignore records corresponding to post-Rx periods
  • Started with HDL
  • Side study: confirm that med initiation that is supposed to modify HDL really does
  • Simplest longitudinal analyses:
    • Compute within-patient Gini's mean difference to correlation with gen. risk score; asks whether gen. risk is correlated with variability
    • Similar but summarize with the median to correlate gen. risk with overall height of the longitudinal records
    • Summarize entire longitudinal record with slope and intercept; AUC and relate summary measures to gen. risk score
  • Would be useful to summarize the data using representative patients after clustering on mean HDL, shape, number of observations, maximum time gap between any two measurements
  • Another type of analysis: summarize each patient using the 9 deciles of HDL; use these deciles to predict polygen. risk score
    • Does not take time ordering into account
    • Might add a slope or shape summary to the deciles


Amanda Peltier, Department of Neurology

Discuss Aims and power analysis for R01


Jake Landes, PT, DPT Vanderbilt Sports Medicine, Rehabilitation Services

  • I am a physical therapist in the Sports Medicine outpatient department and we are planning two studies that we would like to discuss. Primarily, though, we would like to discuss a prospective observational study we will be performing this coming school year with overhead athletes – we will be looking at the relationship of core strength to the likelihood of shoulder injury in overhead athletes. We plan to test the athletes’ core strength at start of their season and then collect data on injuries and time lost from playing their sport during the season. Specifically, we have questions about what our number of subjects should be in order to determine a difference and what we will need to do statistically in order to analyze the data.
  • Outcomes: number of days (or proportion) lost during the season due to shoulder injuries
  • Need information on the proportion of athletes who would get shoulder injury during a season. Sample size needed would be large if the proportion is very low.
  • Could use logistic regression to examine association between core strength and incidence of injury
  • Consider other factors that could affect shoulder injury such as the type of sport, number of years practicing, etc. These factors can be adjusted for in the regression model.
  • To calculate the sample size, need to specify the outcome, type of analysis used, the meaningful difference (effect size: odds ratio of injury upon one unit change in core strength) you want to detect, and some preliminary data on the outcome measurements (rate or variation). A rule of thumb: 20 cases of injury are needed for each factor you'd like to analyze.
  • Consider choosing a type of sports with the greatest association between core strength and shoulder injury.
  • how to quantify core strength, a single summary score?
  • A second study I am wondering about is an Anterior Cruciate Ligament Reconstruction study where we are going to compare a group of patients in a home based program versus standard care (control). We are wanting to do a feasibility study this year in our clinic, and I think it will be a prospective case-control study, or maybe prospective cohort—we also want to know about N size and analysis after ward.
  • Enroll 7 patients in one month. Feasibility study.


Katherine McDonell, Neurology

  • Parkinson's disease - norepinephrine; VICTR application
  • Original intention peripheral blood pressure support
  • Interested in a combined medication regiment
  • Goal to get nor. into CNS
  • Propose to study n=16 patients
  • Need dose titration 100mg bid -> 600mg 3/day
  • Which dose do patients tend to end up with?
  • Is a safety & tolerability study, partly dose-finding
  • Patient response that is monitored is blood pressure - minimizing orthostatic symptoms without side effects; target supine BP plus headaches, dizzyness, mania; symptoms are of primary emphasis
  • Is there an accepted symptom summary scale? If not may need to just count the number of symptoms present
  • But dose adjustments are clinical adjustments based on a symptom "gestalt"
  • Target for analysis is final dose
  • Need SD of dose; best available data will probably come from what doses are used long-term in clinical practice; we'll assume this is a stand-in for the final tolerable dose
  • Once a useful SD estimate is found, it can be used to compute the likely margin of error in estimating the population mean required dose when n=16, with say 0.95 confidence. The margin of error is the half-width of the confidence interval.
  • Would be good to know what evidence exists for the usefulness of plasma drug concentrations in estimating the final required dose


Reagan Leverett, MD, MS, Assistant Professor, Department of Radiology, Women's Imaging

  • PQI project. Two types of images (new vs. old method) were performed for each patient.
  • Examine the agreement between the two methods based on the paired data (kappa stat). Readings are ordinal values.
  • Let a few radiologists read the two sets of images in random order to study the agreement.
  • May need a couple of hundreds of patients, and a few (2 to 6) radiologists. (also want to have good agreement between radiologists, that is, readings of a certain method do not heavily depend on the experiences of radiologists).


Akshitkumar Mistry

Reserved spot for consulting with Chris F. about meta-analysis


Stephen Patrick, Assistant Professor of Pediatrics and Health Policy, Division of Neonatology

  • Mary-Margaret Fill, TDH EIS
  • Neonatal abstinence syndrome and long term outcomes
  • Merge TennCare data with educational data
  • Suggest regression model with traditional covariate adjustment unless need to do special matching (family, neighborhood)
  • Biggest assumptions: children move away from TN for reasons unrelated to potential educational achievement
  • Confounding: women giving birth to infant with NAS may tend to be different from those not having an NAS child; need to adjust for all factors related to this that might be associated with educational outcome
  • Also what is the effect of school on test scores?
  • Birth records have mother's educational level, zip code, tobacco use
  • Matching records may be challenged by mother changing last name
  • Might also look at infant and mother utilization of services, diagnosis of ADHD, etc.; cross-correlate with educational achievement


Lindsey McKernan

Here is the feedback I received on my application: Power analysis never should involve having a power of detecting a previously observed (and probably measured with bias) effect. Power should always be defined as the probability of detecting a minimal clinically meaningful effect. Also, this type of study is more suited for justifying sample size on the basis of precision of an effect of interest (usually a difference or a correlation). Precision is stated as a margin of error e.g. half-width of a confidence interval. Please revise Section E of the proposal and feel free to attend a clinic to discuss.

What was initially written: Power Analyses: Previous researchers have found moderate relationships between trauma severity and pain symptoms (r = .29; Poundja, Fikretoglu, & Brunet, 2006). Power analyses using unadjusted effect size from this study based on their sample size of 130 suggest a necessary sample size of at least 97 for the present study to reveal similar effects. Power analyses of the results of studies of the relationships between trauma severity, pain severity, experiential avoidance, and anxiety sensitivity (Gootzeit, 2014; Ruiz-Párraga & López-Martínez, 2015) suggest that a sample size of 144-158 is necessary to find these associations. The hypotheses outlined above will be tested through bivariate correlation and linear regression analyses. Specifically, relationships among variables of interest (Hypotheses 1A, 1B, 2A, 2B) will be assessed through Pearson product-moment correlation analyses to determine the strength of the association among these constructs in our sample. Tests of moderation (Hypotheses 2C, 3) will be tested using multiple linear regression with cross-products of the variables of interest to assess the interaction between predictors. All analyses will be carried out on either SPSS 22 (IBM, 2013) or the R statistical package (R Development Core Team, 2010)
  • See Chapter 8, P. 8-12 of - suggest using the r=0 curve. This approach is using the margin of error based on 0.95 confidence limits. E.g.: "With a sample size of N subjects we can estimate the correlation coefficient between two variables to within a margin of +/- xx with 0.95 confidence (see graph)."
  • Important to prioritize the comparisons and to report them in this pre-specified order so that no multiplicity corrections will be needed
  • A regression model that allows for interaction between time since trauma and amount of trauma would allow for estimation of the time-decay or enhancement of memories-effect. The time interaction effect may be nonlinear.

2015 Dec 14

Sachin Patel, Psychiatry

  • Animal model for exposure to stress, long at differential response to stress
  • Interested in susceptibility to stress
  • Measure of anxiety is a key measure (high = more anxious)
  • Each animal has a baseline measure
  • Would be good to do a Tukey mean-difference plot (Bland-Altman plot) to be sure that the delta is an adequate summary of the two measures
    • Also watch for floor and ceiling effects
  • Using the delta as a continuous stress response measure will optimize power and minimize arbitrariness
  • Discussed regression to the mean
  • Problem with choice of anxiety measure out of many
  • A composite measure may help, e.g., average z-score or average rank; can do Spearman rho rank correlation on the result, against another variable; can describe variability in ranks across anxiety measures
  • Otherwise analyses of disparate measures can be hard to reconcile

2015 Dec 7th

Pierce Trumbo

  • Shade tree clinic, where patients do not have insurance or do not have enough insurance can get medical service.
  • Primary outcomes: number of ER visits, length of hospital length of stay. Will compare before and after pts visited the clinic.
  • N=680 patients and estimate to have ~300 meet inclusion (time span between first visit and last visit greater or equal to 1 year).
  • Need estimate for VICTR application. Suggest for $5000.

2015 Nov 30th

Andrew T. Hale, Medical Scientist Training Program

  • Need a quote for biostatistical support for a VICTR grant submission
  • Want to assess the association between brain tumor grade (total 128, 93 1s and 35 2s) and gender, age at diagnosis, Edema (0-3), draining vein, necrosis, location (8 different location).
  • Will apply for VICTR voucher in amount of $2000.

Christopher John Prendergast, Tracy McGregor

  • We will specifically be seeking some guidance regarding graphical representation of data related to statin doses in children and adolescents.

Christopher Lee Brown

  • Discussed analysis for reviewer's comments

2015 Nov 23rd

Mark A. Clay, Divisions of Cardiology and Critical Care

  • The purpose of the study was to evaluate whether patients with single ventricle physiology undergoing the second stage of surgical palliation, who’s length to weight ratio was >90% were at higher risk for increased ICU length of stay, ventilator times, and increased non-invasive ventilation when compared to those whose length for weight was <90%. Analyzing the data with the Mann-Whitney U Test there was a statistically significant difference between ICU length of stay and ventilator hours for those with weight for length >90% compared to those <90%. However, I attempted to analyze the data again with Spearman’s to see if there was a correlation between increasing z-score percentile and there was no statistically significant correlation.
  • Clinic question: Has the data been analyzed appropriately to answer the question? Should I be concerned that Spearman’s correlation did not show a statistically significant correlation between the variables even though there was a statistically significant difference between the groups? Should I use and how might I best demonstrate association or risk related to weight for length z-score >90% with linear regression?

Rebekah Griesenauer (Conley), Biomedical Engineering

  • I am designing a study for a small group of human subjects to test the feasibility of a new tool that I designed for breast cancer assessment using medical images. I would like some guidance on effective study designs for a small number of patients and for determining the accuracy of a new tool when there is no current clinical equivalent to compare to.
  • Need a measureble outcome to calculate the required sample size

2015 Nov 16th

Aaron C. Shaver, M.D., Ph.D. Assistant Professor of Pathology, Microbiology, and Immunology

  • The csv consists of sample ID, the covariates I want to test (age as an integer and categorical variable; poor.risk through transcription, which are all categorical variables; and num.muts, which is an integer) and the OS and PFS data (for censoring rows, 0=censored and 1=dead). I would like to include the interaction between age and poor.risk, because I have biological reason to believe that that interaction is relevant. My questions concern: measuring goodness of fit of the model; how to interpret the interaction term; how to estimate power, given the large number of covariates and small sample size

2015 Nov 9th

Fernanda Maruri

  • "If possible I would like some help interpreting results of 2 Wilcoxon Rank Sum tests in which one is significant and the other is not."
  • Compare

Jessica Kaitlin Campbell

  • The goal of the project is to examine the impact that the palliative care unit has had on the medical intensive care unit in terms of patient length of stay and mortality. I have collected data regarding some parameters per and post opening of the palliative care unit. I am interested in the best approach in analyzing the data.
  • Have data a year before and a year after the unit opened. Want to compare LOS and mortality in MICU. Both groups had palliative consult, only some patients after went to the palliative care unit.
  • Wil apply for VICTR biostat support. Suggest for $5000 study.

2015 Nov 2nd

Gabriella D. Cozzi

  • GDM project analysis. Associaion between hoursehold income and education with the five primary endpoints.

Michael Chomat

  • To discuss experimental study design and data analysis for a project within REDCaps
  • Two REDCap data base can be merged based on common identifier.

Jamie Robine

  • R questions about fitting logistic regression model and plotting the figure.

2015 Oct 19th

Rebecca Cox, Psychology

  • I am working with the data from the National Comorbidity Survey Replication, a nationally representative sample used to estimate prevalence rates of psychological disorders. I have questions about what types of analyses to use with a complex sampling design that includes strata, clusters, and weights.

*Suggest review survey document to specify correct strata, clusters and weightings variables *Set up complex survey design effects in SPSS complex module.
  • Subpopulation command in stead of subset anaysis.

2015 Oct 5th

Stevenson, David, Health Policy

  • Stratified cluster randomized trial.
  • Intervention group: predicted mortality risk score obtained for all the patients, based on which "top patients" will be provided with hospice and will be expected to get better life quality. Control group: standard care. Individual agencies (50 in total) will be randomized to intervention/control group.
  • Cutoff of risk score may vary within and across the sites. Information obtained from prediction model: median life expectancy, probabilities of death during certain time periods.
  • If the primary outcome is continuous, would need SD to calculate sample size. If it's the time to event, we will need expected median time in each group.
  • Consider the flexible/sequential design, having pilot sites included in the final analysis.
  • Biostat resources: VICTR Voucher (35 hours). Dr. Matt Shotwell

Conor McWade, ED, PhD student

  • Apply for Voucher (90 hours)
  • Have collected car collision/victims data, demographics of the passengers, road characteristics.
  • Define collision as fatal vs serious.
  • Aim to develop a prediction model to predict the severity of collision based on location, time, etc.

2015 Sep 28

Katrina, Electrical Engineering

  • Our study is on incidence of eye disease seen at Vanderbilt. We have data on 33,000 patients looking at incidence of disease and I would like to discuss how to best analyze this data.
  • Whether incidence at Vanderbilt can represent incidence in Nashville


  • I was hoping to come to biostats clinic today to get some help with sample size calculations for my project.
  • Cross over design. Within subject correlation is 0.7. Need power calculation.

Wes Clord

  • Power analysis of survival analysis

2015 Sep 21

Jose A Arriola, PGY 3 - Psychiatry

  • "I plan to implement a different type of interview in the first episode psychosis outpatient clinic at Vanderbilt Psychiatric Hospital and investigate how it contributes to improve adherence and management. The type of interview is called Shared decision making approach which is a little bit different to what we are used to. I am planning to train the MDs and providers on this technique and then compare measurable outcomes before and after the training. The outcomes would be no-show clinic rates, hospitalizations, etc. (things that are recorded automatically on the patient's chart). "
  • Statistical tests: Wilcoxon signed rank test for continuous outcomes, and McNemar's test for binary outcome
  • Primary outcome: number of times that pt did not show up within 3 months raning 0-4. Proportional odds logistic model to analyze. num of no show after intervention ~ number of no show before intervention + age + gender
  • Sample size calculation: use PS for paried binary outcome

Tamara Moyo, Hem/Onc

  • Want to correlate the resistanze to therapy based on imaging with cell signal
  • Wilcoxon signed rank test (paired t-test); Wilcoxon rank sum test (two group t-test)
  • Mixed-effects model to adjust for other covariates.

2015 Sep 14

Mhd Wael Alrifai, Neonatology

  • Name of project: Paretneral Protein Calculator (PPC)
  • Type: Randomized controlled clinical trial, un-blinded
  • Help needed: Discussing the primary and secondary outcomes, designing the database
  • Study status: IRB approved, enrollment starting next week
  • Research question: the effect of intervention on the accuracy of protein prescription. The primary endpoint is the ratio of target days to total days (target days are the days when prescriptions are given with correct amount).

Sudipa Sarkar

  • My research topic is on the effect of statins on non-alcoholic fatty liver disease
  • retrospective cohort study.

Michael C. Dewan, Department of Neurological Surgery

  • I am interested in discussing sample size calculations. We are conducting a clinical trial evaluating the effectiveness of a prophylactic antiepileptic drug (levetiracetam) in brain tumor patients. For 14 days following surgery, patients will be randomized to either drug or no drug. The primary outcome is the development of a clinical seizure and the follow-up time to primary endpoint is 14 days.


  • I would like to address a few questions regarding sample size calculation for a translational study on the role of alternate complement activation in sickle cell lung disease

2015 Aug 24

Christopher Brown

  • Retrospective cross-sectional study on heart failure patients.
  • Outcome is the low potassium, related to urine output per hour.

Maya Yiadom, Emergence Medicine

  • Criteria for giving EKG to diagose STEMI.
  • Trigger criteria: typical symptom, atypical S

Megan Pask, Tricia Russ, BME

  • Compare CT values between four groups.
  • Use non parametric test: Kruskal Wallis test (ANOVA), Wilcoxon Rank sum test (two sample -t-test)

2015 Aug 17

Karl Zelik, Assistant Professor of Mechanical Engineering, Assistant Professor of Physical Medicine & Rehabilitation

  • sample size calculations for a grant proposal

2015 Aug 3

Lan Wu, PMI

  • Had questions about VICTR proposal review. Suggest use Wilcoxon Rank Sum test or Wilcoxon Signed Rank test to compare between and within subjects
  • Try to identify subset of b-cell in this set of subjects - will be able to provide descriptive statistics
  • Consent 60 subjects will estimate to have 30 subjects. Will quantify b-cell and compare b-cell among two different locations. First get a percentage of b-cell of the mixture then calculate the absolute number of b-cell per gram tissue.
  • Will find SD from preliminary data and calculate required sample size based on that.

2015 June 29

Aaron Noll, VMS IV

  • retrospective chart review of 1750 patients. correlation between screening exam results with 15 diseases. 923 patients had actual visits within two years (gold standard of disease).
  • Analysis data set: two-by-two tables based on 923 patients. Compare demographics between 923 patients with (1750-923) patients.
  • I am currently finishing a research project that is regarding various diagnoses that are able to be picked up on a screening exam (for diabetic retinopathy). To this point, I have calculated the following values for the 16 diagnoses of relevance: true positives/negatives, false positives/negatives, positive/negative predictive values, and sensitivities/specificities. However, I am unsure what the best test is to determine statistical significance or importance of these numbers--eg, do I use a 95% CI, odds ratio, etc. One issue with these results is that although I have a very large sample size for the initial screened population (over 900), many of the diagnoses have less than 5-10 true positive results.
  • Zero or close to zero number in certain cells. Wilson confidence interval. binom.confint() of binom package.

2015 June 22

Nelleke van Wouwe, Department of Neurology

  • We are working on a grant and we have some questions about a power calculation for a Repeated Measures ANOVA (based on effect size from a previous study).

2015 June 15

Daniel J. Miller, Department of Psychology, Psychological Sciences

  • Discussion about microstimulation data to develop a test of the hypothesis that stimulating two areas in the brain from which evoked movements differ produces a blend of those movements (endpoint neuronal encoding)
  • Need help understanding how to organize the data in order to build a model to explain physiological results (e.g., how the dual stimulation sites interact)
  • Suggest apply for a $5000 VICTR voucher.

Kendall Anne Ulbrich, Pediatrics

  • I am requesting assistance in figuring out statistical significance. We see a trend in the data with the diagnosis of chronic lung disease leading to increased risk of death after trach placement vs other diagnosis.
  • Babies in NICU, outcome is alive/died, want to compare chronic lung disease to other diagnosis.
  • There were ~15 diagnosis, among whom 12 had chronic lung disease.
  • Total 115 babies (25 died in NICU). Primary outcome is the death in NICU. 8 (or 11) babies who had lung disease and died.
  • Plot Kaplan-Meier curve first for description, use log-rank test.
  • Can use Cox proportional hazard model to analyze the association between lung disease and survival in NICU.
  • Could also apply for a $2000 VICTR voucher.

2015 June 1

Robert Lentz, cardiology

  • My project is looking at radiation-induced atrial fibrillation, specifically in patients with breast and lung cancers. I have raw data extracted from the Synthetic Derivative and am hoping for some guidance regarding my data analysis plan and how I might be able to best display my data.
  • There are ~3000 breast cancer pts (125 had AF), ~2000 lung cancer pts.
  • To test the association between radiation and AF, include all pts (y=AF, x=radiation y/n, cancer side); then take subset of pts who had radiation, fit a model of radiation dose/side with AF.
  • Length of follow up is different for all patients. Can use survival analysis. If certain proportion of pts died before developing AF, should treat those pts as competing risks events.
  • If apply for VICTR voucher, suggest $5000.

Robert K. Tunney, Jr., Cardiology Resident

  • Email: My research is investigating statin dose intensification according to the ACC/AHA 2013 Cholesterol Guidelines in post-ACS patients. I am interested in performing logistic regression analysis on ~300 patients and potentially Spearman rank r correlation coefficient.
  • Two groups: historic control and intervention group. Binary outcome. Primary aim is to assess the outcome difference between groups.
  • Chi-sq test and multivariable logistic regression can be used to test the primary hypothesis.
  • Suggest propensity score adjustment.
  • Will apply VICTR voucher in amount of $2000.

2015 May 11

Zac Cox, PharmD, BCPS

  • Email: I would like to request a biostats clinic reservation on Monday 5/11 from 12-1 for a comparative effectiveness research project. The main question is selection of the best primary outcome to maximize power for a population size that will be fixed (secondary to funding and patient enrollment). Our second question is the best statistical analysis method for 3 independent continuous variables (ANOVA vs the 2 experimental groups independently compared to the standard of care comparison arm). Please let me know if you would like me to send anything in advance.
  • Use continuous outcome to maximize power
  • Wilcoxon rank sum test to compare two new treatment groups to the standard care group
  • Multivariable linear model adjusting for baseline weight and treatment regimen

Michelle K. Roach, Obstetrics and Gynecology

  • Email: We will be completing a retrospective chart review looking at pregnancy and delivery outcomes in women with gestataional diabetes. We plan to use RedCap database for data entry.

2015 Apr 27

Aaron Noll, Medical Student

  • I am a third year medical student working on a research project that is evaluating the teleretinal imaging program at the Nashville VA Hospital. I attended one of your clinics about a month and a half ago and greatly appreciate the help I received at that time. I have now completed my data collection and am moving on to the data analysis portion of the research, and would like to discuss my revised project with you to see what the best way is for me to proceed.
  • As an overview, I am looking into the teleretinal screening program to evaluate its efficiency and its accuracy at diagnosing abnormalities other than diabetic retinopathy (the true purpose). I have recorded the data on the following topics:
    • Demographics (Age, sex, ethnicity)
    • Months from consult entry to screening
    • Days from screening until note loaded to chart
    • Screening diagnoses, diagnoses found at subsequent visits, and diagnoses found at previous visits
    • No-show rate for the screenings
    • Consult timing
    • Months since prior screenings and clinic visits
  • Had imaging readings and clinic diagnosis on ~1700 subjects. There were 18 diagnosis categories, looking at their agreement.
  • Will apply for VICTR voucher. Suggest $2000 for up to 35 hours

2015 Apr 20

Lexy Morvant, Pediatric

  • NICU data analysis
  • time trend of gestational age when receiving ECMO (Y2004-2014) for C-section babies. To evaluate the effect of policy change (increase gestational age for C-section baby in 2007) on ECMO.
  • Only have the information on birth year available. Fit a linear regression model
  • Also have the information on the total number of all ECMO babies. With an assumption that the proportion of C-section babies remains the same, could fit a poisson linear regression model.

2015 Apr 13


  • I have a retrospective dataset of patients who underwent a new cochlear implant programming procedure. The data contain pre- and post-intervention objective performance data, demographic data, and information about the cochlear implant type and location. I am trying to develop model(s) that can answer the following questions: 1) How can we predict whether a patient will be a responder to re-programming? 2) Which variables are most predictive of change in performance from baseline?
  • 177 patients.
  • Endpoint: measurement performance (0-100)
  • Predictors: 15 ~ 20
  • Fit a multivariable linear regression model. Predictor importance can be measured based on the model.

2015 Mar 9

Taylor Leath

  • We attended a biostats clinic on February 23rd to develop a statistical plan. Now that we have a dataset completed, we are having difficultly with our regression models and would appreciate your input.

2015 Feb 23

Katie Rizzone, M.D., Clinical Instructor, Orthopaedics and Rehabilitation

  • I would like to request a methods clinic (to review my methods) for a retrospective chart review study on female college athletes and stress fractures I am writing an IRB for.

Taylor Leath

I would like to reserve a time on Monday, February 23rd to develop an appropriate statistical plan for our study and dataset. I've attached the study protocol which details our specific aims and hypotheses. Our primary questions: 1) Is linear regression the appropriate model to use? Predictors would be sex, age, years of education, participant's current health, trauma exposure and religiosity (all continuous except for sex), and the outcome variable would be each of the individual health states (GOSE 2-8). If so, this would mean six different regression models for the six health states? 2) Alternatively, would it be more appropriate to develop one regression model that includes the health state (GOSE 2-8) as an additional predictor? 3) Do we have sufficient sample size to answer our study questions? Current n=2156 after exclusions. 4) We would also like to show whether the utility values for each of the six health states are significantly different from one another-- would that simply be a within-subjects ANOVA with pairwise comparisons? 5) Should we consider transforming the worse-than-death values?

2015 Jan 12

Dr. Heidi J. Silver, Ph.D., R.D Research Associate Professor of Medicine

  • Study of diet intervention, body composition, insulin resistance, lipo.
  • Could apply for a VICTR voucher of $4000.

2014 Dec 15

Amos M. Sakwe, Department of Biochemistry and Cancer Biology, Meharry Medical College

2014 Nov 24

Paula DeWitt, Center for Biomedical Ethics and Society; Madhu Murphy, pediatric cardiac intensive care unit

Email: We are wanting to test the effectiveness of a “journey board” (see attachment) designed to better prepare parents for their child’s stay and eventual discharge from Vanderbilt’s pediatric cardiac intensive care unit. This will entail giving self-administered surveys with preparedness and satisfaction items to parents of children hospitalized in the pediatric ICU immediately before and immediately after the parent has been exposed to a 15-minute educational intervention using the journey board. The intervention will take place in the child’s hospital room (or another room in the unit) and will consist of a clinician walking the parent through the journey board, and answering any questions the parent may have. Immediately before this, a researcher (not the clinician) will approach the parent, explain the study, and ask if the parent would like to participate. If yes, the parent will be given two short (5-10 minute) self-administered surveys to complete. He/she will be asked to complete one prior to the intervention and one immediately after the intervention. The data will be used to assess the effectiveness of the journey board in preparing parents, and we would like your advice concerning numbers of parents we will need to interview to obtain statistical significance and statistical techniques to be used.
  • Frank's note: Design is confounded with time/fatigue/learning. Also there is little precedent for doing a pre-post study with such little time between pre and post. I think you will need to do a randomized study to attribute any effect to the intervention. Randomize 1/2 of families to get the intervention, 1/2 to get the prevailing treatment, and give survey at the "after" time point for both groups.

Amma N. Bosompem, MPH graduate student

I'm a graduate student in the Masters of Public Health program who will be looking at treatment completion rates in a specific population for my thesis. I need some guidance on study design and what analyses are appropriate given my sample size.
  • TB clinic in Nashville, 203 cases (information on case only) in year 2013.
  • Treatment: completed vs not completed (refused, lost to follow up, etc.)
  • Research question: 1. treatment completion rate. 2. the association between patient's characteristics and treatment completion.
  • Prepare data set as
  • Is there an association between country of origin and acceptance of treatment (accepted vs. no accepted)
  • Apply logistic regression analysis (dichotomous or binary response variable) and include the variables of interest. *General rule of thumb the smaller sample size/10 will help you assess your regression power or how many variables you can include in regression model *N=48 that refused and will be the limiting sample size in regression analysis
  • Country of origin main factor and will have to think of best way of grouping *The covariates of interest: Age as continuous non-linear; gender, marital (married vs. non-married) and country of origin

2014 Nov 17

Mary Van Meter, medical student

My project is looking at the cost of packaging surgical instrument trays (possiblly using regression) and calculating the percentage of instruments used in trays across various Gyn specialties.
  • Total number of instruments used per tray (25-100), usually less than 50% is used.
  • Will compare unnecessary cost between specialty
  • Will apply for VICTR voucher. A standard $2000 is appropriate.

Tomas DaVee, GI

  • Patients underwent liver transplant who had plastic stent to treat leak, about 20-30% needed mental stent later
  • Want to predict early whether patient needs mental or not so pt does not need to surfer pain
  • The current data only gives conditional needs to mental if had plastic already
  • Suggest do descriptive statistics and plan bigger study to develop prediction model
  • Use R for internal validation and calibration using bootstrapping method (rms package)

2014 Nov 3

Monica Ledoux

I am an adjunct at Vanderbilt's Dermatology department, working with Zhengzheng Tang from biostats on microbiome and skin and would like to know the biostat budget for VICTR application(s)
  • Want to know the relationship between Cortisone treatment and bacterial change.
  • Each subject will be his own control: cortisone on one arm and no cortisone on the other. Each arm will be tested at two sites, one normal skin and one tape stripping skin. Observe bacterial change. Therefore, each subject will have 4 tested samples and each sample measured twice (total 8 per person)
  • Look at treatment effect on normal skin. Suggest amount of $2000.

2014 Sept 29

Bryan I. Hartley, MD Department of Radiology

Review an abstract.
  • There are limitations of pre post design. Many factors will affect the outcome besides the intervention like time.
  • Box plot with raw data to explore the distribution
  • Can use Wilcoxon signed rank test to compare continuous outcomes before and after
  • Consider ANCOVA (Analysis of Covariance) to analyze post while adjust for pre specified covariates like previous experience

Adam P. Bregman, MD, MBA, Annette Ilg, Vanderbilt Internal Medicine Resident

I am starting a project with Dr. Anthony Langone in the renal transplant division and I have come across some questions with my redcap database that I would like to ask
  • Retro spective study of post renal transplant patients. Follow those patients for two years to observe a rare event.
  • Describe users characteristics. Some pts took medication for entire 6 months, some stopped prior 6 months for certain reasons, some retook it later. Can consider using certain amount of time to define user.
  • Binary/categorical variables can be described as frequency and percentage

2014 Sept 22

Kasim Ortiz, Sociology, Doctoral Student

I’d like to obtain assistance with a project I completed this summer while attending the Fenway Summer Institute on LGBT Population Health. I used a dataset that permitted me to analyze the effects of states having restrictions on same-sex marriage on smoking, extending previous work by examining these impacts among racial minorities that are sexual minorities. My findings were counterintuitive to previously published work, insofar that I did not find where states restricting same-sex marriage policies had a negative impact on smoking among racial sexual minorities. The dataset is the Social Justice Sexuality Project.

Most previously published papers examining the effects of same-sex marriage policies on the health of LGBT populations have utilized GEE in their statistical analyses. However, I was having problems getting GEE models to converge with the cross sectional data I am using and thus I utilized GLM for the binomial family with adjustments for state clustering (vce) using STATA. I’m apprehensive about wrapping up the manuscript out of fear that I might get a ton of push back because of the methodology I’ve chosen. Also I was having problems with conducting post-hoc analyses on some of the interactions that I included in the analysis (sample sizes kept changing across models preventing me from appropriately conducting likelihood ratio tests). Thus, I wanted to obtain statistical advice on the differences between the two and consultation on how I should proceed. It is my understanding that R is utilized predominantly and since I am learning R in my statistical sequence this year, I thought this would be a great opportunity to compare STATA with R on a question that I’m invested in and a dataset that I’m familiar with.

  • Logistic regression model with robust standard error is appropriate.

Robin Jones, Assistant professor

We have two questions regarding our project on respiratory sinus arrhythmia (RSA) and stuttered disfluencies.

We are interested in how that RSA (respiratory sinus arrhythmia, parasympathetic activity on the heart) is associated with speech disfluencies. RSA during various tasks is influenced by it’s baseline value, thus, in our model we need to account for baseline RSA in some way. When running a model predicting stuttered disfluencies with RSA during an emotional speaking task, should we use a) RSA during the task and RSA at baseline as covariates, or b) create a residualized difference score of RSA during task relative to baseline as a covariate? With option “a” we would be using two covariates in the model (RSA during task & RSA at baseline), whereas with option “b” we would be using only one (RSA difference score).
  • Could include baseline RSA if colinearity is not an issue.

I would like to verify that our analytic plan is the most appropriate. We are using a Generalized Linear Regression with Negative Binomial Distribution. We are using this distribution because stuttered disfluencies are not normally distributed.
  • Generalized Linear Regression with Negative Binomial Distribution is good.

The model for our 3rd research question: Is the relationship between RSA and stuttered disfluencies different for children who stutter versus children who do not stutter during various emotion conditions. We used a 3-way interaction (as well as the various 2-way permutations) to assess this: RSA*Group*Condition. However, when running this model, I am not getting an intercept term. Is the model flawed in some way?
  • Probably.

Lastly, we have some outliers, and I am curious on what is the preferable way to handle this: 1) next case imputation, 2) log transformation, 3) square root transformation?
  • Not a real outlier

  • Take into account the correlation within each subject.
  • Might have carry-over effects between different periods. Could test on equivalent carry-over effects.

2014 Sept 15

Stacy S. Klein-Gardner, Ph.D. Biomedical Engineering

My data that I would like to analyze is survey data from a Likert scale that I believe to generate interval data. I am attaching the raw data to this message. We can focus only on the first spreadsheet as the analysis will be the same for all. The data that is shown includes pre and post data from each of three summers of educational intervention. On the spreadsheet, I have entered the data first under the summer in which the intervention and testing took place. I then moved the data around to indicate which summer it was for the student. For example a student may have come first the first time in 2012, 2013, or 2014. She may have come only one summer or she may have repeated the program. My question is: Does repeated participation in the intervention improve the impact that the program has on the measures indicated by this survey? Is one summer sufficient to increase this measure?

  • Outreach for engineering education
  • looking at data from engineering camp for girls (looking for changes in self- efficacy)
  • self- efficacy- feeling that you can accomplish something in your life (this scale has been validated) * some girls have participated for one year, some have participated for about three years
Research Question: * interested to see if there is an effect of attending camp for more than one year
  • also interested in differences in pre-post scores for the one year attended by student
  • descriptive statistics: consider summary statistics across different categories (i.e. pre and post scores by different school types, year of study, grades)
  • Consider repeated measures type analysis (longitudinal data analysis) for assessment of slope over time (year) of self-efficacy variable
  • Per question of interest - may need to reformat data to “long style or vertical format” (i.e. have row 1 id=1: 2012 post self-efficacy score, row 2 id=1: 2013 post, row 3 id=1 2014 post self-efficacy score, etc. for each of the girls)
  • adjust for age , school type (consider the role of additional potential confounders)
  • Consider applying for VICTR funding– for assistance in repeated measures type of analysis.
  • Need to account for the correlated nature of data and verification of assumptions (such as in Mixed effects modeling or generalized least squares)
  • Account for the missing data

  • Limitation: lack of control group (there is no way to conclude that the program is the only thing that is improving self efficacy)

  • pre vs. post score for any given year
  • consider doing boxplots for each of the pre and post scores for each year (these can serve as your summary statistics)
  • Univariate analysis: Wilcoxon Signed rank test to see if there is a difference between the distributions of pre and post scores (data in horizontal format works)
  • cautioned combining the pre-scores over the three years, and post scores over the three years (year may be a confounder and impact trend of the data)
  • pre vs. post study (may see a difference, however no guarantee that improvement is from the program- not an randomized controlled trial)
  • Motivated and selected group of girls and may have higher self-efficacy baseline score (pre) - consider comparing self-efficacy scores with those reported in other studies among girls.

2014 Sep 8

Jason Winnick, MPB

I perform biomedical research where the sample sizes are generally approximately 6 to 10 per group. The primary outcome variables are usually something like counterregulatory hormones levels, glucose in fusion rates, or muscle glucose uptake. I will be submitting an R01 in October and I would like to receive your advice regarding the estimation of sample size. In addition there may be the potential to receive statistical assistance over the life of the grant if it's funded.
  • Two primary endpoints; one with greatest variability (glucose infusion rate needed to maintain desired blood glucose level) has most variability and hence will be conservative to plan for
  • 10 with type I DM 10 without
  • Other covariates: age, insulin required to maintain blood glucose, HbA1c
  • Baseline liver glycogen assessment
  • Start with 3 hour fructose infusion to stimulate liver glucose update vs. saline infusion (randomized), then insulin infusion then 2 hour period where become hypoglycemic (using clamp)
  • Need a good estimate of the standard deviation across patients for infusion rate - use the dog data taking all relevant time periods and stratify by liver glycogen to compute 12 SDs; then we can compute an averaging by averaging the variances and taking the square root
  • Need clinically relevant difference (in mean infusion rates) not to miss: estimate 1 mg/Kg/min
  • Language for grant application something like: The power calculation was based on a 2-sample t-test without covariate adjustment for HbA1c, age, etc. The actual statistical test will be ANCOVA adjusting for these factors, which will increase the actual power a bit (increase would be more had the sample size been larger; the sample size chosen has a penalty for estimating the effects of the baseline covariates).
  • Last aim: most general way to assess to to fit a smooth function of time to the longitudinal (serial) measurements, separately for each of two groups, and test for differences in shape of the two curves. A convenient choice is to fit a quadratic function of time to each curve. This increases power over individual time point tests. Suggested statistical method: generalized least squares or mixed effects linear model.
  • Suggested contacting Li Wang to tell her that a VICTR voucher is in the works

2014 Aug 18

Cecilia Di Pentima, Zachary Willis, Pediatric

  • Want to assess the impact of the implementation of an ASP on antibiotics use.
  • Monthly antimicrobials (AMs) use in days from 2009-2012 April. Data from many hospitals including Vanderbilt. Want to compare Vanderbilt to ALLCHA.
  • ASP intervention started 2012 March at Vanderbilt. Can see less use of AMs after intervention.
  • The comparison of pre and post might be biased by other factors like time not just by intervention. Institution effect is hard to assess since all institutions started intervention at different times.
  • Also needs to adjust for other factors like date for seasonal effect.
  • Linear model of VCG ~ intervention + rcs(time)
  • Better to have individual data for all the hospitals which had both pre and after data to assess intervention effect using mixed-effects model, or just compare between hospitals using data after intervention to see whether Vanderbilt does better than others
  • Consider get Vanderbilt rank among all CHA

2014 July 21

Kelvin Moses, Urologic Surgery

  • wants to do a pilot study to get preliminary results for a grant submission.
  • requesting data from Southern Community Cohort Study (SCCS). Needs power analysis and statistical plan for the data request.
  • applying for VICTR biostats support for funding for this prelim project. Needs estimate.
  • about 3200 men enrolled. max follow-up 10 years. about half finished the whole study period.
  • Prediction of screening frequency by baseline characteristics. Association between prostate cancer stage and frequency of screening.
  • all patient self-reported data, at 5 year and 10 year. (have you had screening within the last year?)
  • GEE model of screening frequency (recent screening yes/no at 5 year, 10 year) on age, race, interaction between age and race, ...
  • Ordinal logistic regression model of prostate cancer stage/grade on screening frequency (need be carefully defined) prior to diagnosis. Need consider different follow-up of the patients.
  • Contact Li Wang( for budget estimate.

Austin Beason, summer medical student, MOON Group

  • How can I determine the required sample size (i.e. number of subjects or raters) for interval estimation of the Kappa statistic for an intraobserver and interobserver study with multiple raters? Our number of subjects is currently 20 (N=20) and our current number of raters is 27 (n=27). Further, we are hoping the given sample size will give at least 80% power at the 0.05 level of significance (two-sided).
  • >library(kappaSize)


April Barnado, Leslie Crofford, Division of Rheumatology

  • Email: I am submitting an early career grant for a starter type project due August 1 and needed help with performing and writing up power/sample size calculations.
  • Specific Aim #1: identify group of lupus patients of about 1135. Lupus nephritis patients of about 400. Nephritis is severity indicator.
  • Specific Aim #2: Determine the association between ED use and meeting standards of quality of care in management of SLE and in the treatment of SLE nephritis, as defined by the Quality Indicator Set for SLE. For aims #2, I would likely be performing Chi squared tests comparing 3 groups (non, occasional, and frequent ER users) for most of those sub-aims.
  • Specific Aim #3: Determine the association between ED use and corticosteroid use in SLE and SLE nephritis. For aim #3, I would likely be using multiple linear regression.
  • For binary outcomes, use logistic regression with adjustment of other confounders.
  • Ratio will be treated as continuous variable and will be analyzed using general linear model.
  • Hypothesis: more ED use will have higher steroid dose. Will analyze current steroid dose and #ED visits in the past 12 months. Steroid dose will be a ordered categorical variable with 4 levels. Can use Chi-square test. Proportional odds model can be used to adjust for other confounders.
  • Grant due Aug 1st, need to be done July 21st.

Taylor Leath, Trauma in Surgical Sciences

  • Survey on quality of life (N=1000). There are 7 GOSE questions about health states (0-100). Can describe the distribution for each GOSE. Predictors include gender, age, and years of education.
  • Want to compare between GOSE scores. Multiple comparison issues (21 comparisons).
  • Can use mixed-effects model taking into account of within subject correlation.

Ola Oluwole, Medicine

  • N=36 patients who had CLL transplant with two types (8 vs. 27). Want to compare survival between two groups.
  • Time from transplant to death or relapse. Sample size is limited. Mainly descriptive. Want to write manuscript.
  • Can apply for voucher of $4000.


Neelam Patel, Medical Student

  • I am fourth year medical student doing a project for dermatology. We are doing a meta-analysis of pediatric vitiligo patients to assess which populations need thyroid studies performed. I have a spreadsheet of the data. I need help analyzing it.
  • Research question: the percentage of thyroid abnormalities in pediatric vitiligo patients.
  • Only have aggregated data. Could have an overall estimate of percentage. Also could explore the variability between studies.
  • Apply for a $2000 Voucher.

Tyler Kendrick, Anesthesiology, Medical Student

  • One-year prospective study. Will record the numbers of surgeries in Ethiopia (an African country) and the number of perioperative mortalities.
  • Sample size calculation to reach a desirable precision of mortality rate estimate.


Wei Xie, Computer Science, Brad Malin, DBMI

  • we want to find out if the IRLS estimation algorithm is reversible -- e.g., given only the Fisher information matrix and scoring function (and \beta coefficients), can we go back to the original Y or X matrices
  • Context is confidentiality with data coming from multiple sites, with each site's data maintained independently, and controlled
  • How to do model diagnostics without residuals?
  • Does the distributed computing model lead to good statistical modeling practice? E.g.: covariate transformations, Y transformation, normality of residuals [could compute residual vector separately by center and share an ECDF of the residuals)
  • How often are practitioners of distributed statistical analysis assuming linearity of covariate effects? Being careful about transforming Y or modeling Y robustly?
  • Can't reverse the process to solve for an individual's datum if model is full rank, n > p, no parameter is devoted to only one subject, residual vector is secret
    • If a single parameter is devoted to 5 subjects at one site, may possibly be able to solve for a summary statistic for the 5 (e.g., race has 4 levels and one of the levels only applies to 5 subjects at a site)
  • May be able to discern that one site has an overall better level of Y than another site
  • Not able to get a robust sandwich covariance matrix estimator if residual vector is not provided; sandwich estimation requires U matrix not just U vector
  • Even if residuals are available, it may not be possible to work backwards to an individual from a given site because estimates come from a global beta vector over all sites
  • We seldom use OLS with health care data; the need for weighted X'X (X'VX) instead of X'X as used in OLS makes the identification problem more difficult in general, because V is a function of the current beta estimate (for all sites combined)
  • Worthwhile working out the special case where Y is binary and there is a single X that is binary or polytomous, and there is no special knowledge (e.g., k subjects are of type x and all have the same Y)
  • Worth taking another look at data squashing

Neil Templeton, Engineering, CHBE

  • Metabolic flux analysis
  • Rate of metabolite turnover
  • Which metabolic phenotypes are produced in high titre-achieving production processes
  • Protein therapeutics; cost of production
  • 14 conditions (cell lines); correlations between fluxes (80 reactions- flux, mass spec); looking for up-regulation
  • 80 Spearman rank correlations x 14; each correlation 10 observations (clones)
  • Two controls; secondary controls
  • Independent experimental units: clones, manipulations of cell lines
  • See if a unified model would be a better approach than pairwise analysis
  • Must be able to precisely estimate a quantity such as a correlation coefficient in order to be reliable in picking "winners" across reactions
  • Low precision (low number of independent experimental units) implies low probability of selecting the optimum reaction/condition
  • Dimensionality is high enough that an "omics" method may be needed
  • Recommend contining discussion at a Tuesday or Friday clinic


Elizabeth Morse, RN, MSN, FNP-BC, MPH Vanderbilt University School of Nursing

  • My project involves survey data of 220 Spanish and Arabic-speaking patients in the Center for Women's Health. I've completed all of the descriptive statistics but need help with the correlations. For example, I know from having surveyed patients myself that those patients who reported speaking "Arabic only" at home were more likely to self-report speaking English "not very well", but I don't know how to express this statistically.
  • To test association between two variables A and B,
    • If A is a continuous variable and B is categorical variable, use Kruskal Wallis test (or Wilcoxon rank-sum test)
    • If A and B are both categorical variables, use chi-square test
    • If A is ordinal variable and B is binary, use chi-square trend test
    • If A and B are both continuous variables, use spearman's correlation coefficient.


Brett Byram, BME

* Clinical image degradation with ultrasound
  • What are major factors of degradation? Pulling apart mechanisms.
  • Clinical target: liver tumors/biopsy; visualize needle
  • What is the best study design?
  • Ask trained readers to assess utility of image
  • Discusssed hypothesis testing vs estimation study
  • One estimand could be the mean absolute number of levels different
  • Can relate an ordinal measure to quantitative measures of image quality
  • Can estimate # patients needed if have a reliable estimate of the standard deviation of an absolute difference of interest
  • May consider progressively ruining an image to see when it becomes uninterpretable
  • One goal is to develop a model to predict expert's quality rating from multiple quantitative physics-based measures
    • May consider an ordinal response model / multinomial model


Steve Kahn, General Surgery Fellow

  • can't arrive before 1pm on Wednesdays, so attending Monday clinic
  • "I am going to perform an email survey of surgical residents (approx 5500 in the US) and wanted to know what you think an appropriate response rate would be and the best method to do statistical analysis (rough draft of survey attached). Or should the questions be revised to facilitate a better statistical analyisis?"
  • make the variable as continuous as possible using sliding bar

Philip Budge, Fellow, Division of Infectious Diseases

  • grant proposal relating to the development of new diagnostic technologies for neglected tropical diseases

LIsaMarit Wands, nursing

  • Survey on two cohorts, VA-based cohort and university-based cohort.
  • Outcome: global physical and mental health score. Pain is part of global score, and also a barrier to level of reintegration success. Could calculate a global score without pain. Could examine how pain correlates with reintegration and outcome.
  • A specific question (meaning of life) in two standardized questionnaire. Could include both in the model predicting outcome.


Stephanie Fecteau, Psychiatry Post-Doc

  • Cortisol measures 3 per day
  • % of increase because times not noted accurately
  • Need Bland-Altman plot to check proper transformation: post - pre vs. (post + pre)/2 or log(post) - log(pre) vs. geometric mean of pre and post
    • want the transformation that makes the graph flat and random
  • 1/2 of families received a service dog after 3 weeks
  • Suggest longitudinal analysis using 3 daily x 15 weeks, allowing for correlation; only one day per week
  • Correlation structure based on approximate time of measurements in days + fraction of day
  • Model smooth time trend, allowing for separate trend in those randomized to service dog; check for shape change between two groups
  • Easiest-to-interpret method generalized least squares with AR1 continuous-time correlation structure

David Dantzler and Donald Lynch, Cardiovascular Medicine

  • ECMO: what predicts survival to hospital discharge; initiated by cardiac surgeons
  • Collecting patients from last 2 years (N=60 so far)
  • Discussed margin of error of 0.1 in estimating a single probability with n=96
  • Alternate endpoints: LOS, censor on death, i.e. Y=time to successful discharge
  • Or: ordinal outcome Y=1, 2, 3, ... longest LOS, dead = longest LOS + 1; effective sample size almost equal to # subjects
  • Also have Glasgow coma scale at discharge; could factor into ordinal outcome
  • May be possible to use a complex high-information scale to derive a severity of illness-based score that is then used to predict mortality
    • Has reduced many variables to one
  • What to do with patients who died before ECMO was available?


Mitchell Odom, VUMC 3rd year medical student, Department of Neurosurgery.

I am currently helping with a project that requires a survey be employed, and we are creating an original one to send out. I would like to get some expert opinions on the questions that we ask, and to make sure we are honing in on what we're really looking for.
  • CTE - Chronic Traumatic Encephalitis caused by multiple concussions. Survey is designed to ask questions about awareness of CTE among parents of young athletes (junior high and high school). The plan is to distribute the survey using Vanderbilt connections with local high schools.
  • Recommendations:
    • Maximize response rate (by giving parents incentives of some sort)
    • Ensure that the survey is brief
    • Make sure the responses are anonymous
    • Use numbers instead of categories
    • Simplify the language
    • Branch questions
    • Incorporate visual analog scale (instead of categories)
    • Order questions in a logical way

Older Notes

Topic attachments
I Attachment Action Size Date Who Comment
BoxPlotR.RR BoxPlotR.R manage 5.7 K 17 Apr 2006 - 11:44 QingxiaChen  
InforegardingwhatmySPSSfilesays.docdoc InforegardingwhatmySPSSfilesays.doc manage 24.5 K 17 Apr 2006 - 11:44 QingxiaChen  
LOA_condensed_data.sxcsxc LOA_condensed_data.sxc manage 22.1 K 04 Dec 2006 - 09:17 PatrickArbogast Data from Edward Butterworth
Oluwole_Biostat_Clinic.xlsxls Oluwole_Biostat_Clinic.xls manage 46.5 K 25 Aug 2014 - 11:30 SharonPhillips data file for Olalekan Oluwole
StatisticalAnalysisRequest.docdoc StatisticalAnalysisRequest.doc manage 22.5 K 17 Apr 2006 - 10:26 QingxiaChen  
WellsIschemicCollat.pngpng WellsIschemicCollat.png manage 37.0 K 31 Jan 2011 - 13:58 MattShotwell  
WellsIschemicEF.pngpng WellsIschemicEF.png manage 37.4 K 31 Jan 2011 - 13:55 MattShotwell  
analysisEXT analysis manage 3.9 K 11 Feb 2006 - 20:30 QingxiaChen  
biost_clinic_stephanie_vaughn.csvcsv biost_clinic_stephanie_vaughn.csv manage 4.3 K 23 Apr 2007 - 11:37 PatrickArbogast  
biost_clinic_stephanie_vaughn.dtadta biost_clinic_stephanie_vaughn.dta manage 1.7 K 01 May 2007 - 11:12 PatrickArbogast Stata datafile for Stephanie Vaughn
biost_clinic_stephanie_vaughn.loglog biost_clinic_stephanie_vaughn.log manage 8.1 K 01 May 2007 - 11:13 PatrickArbogast Analysis results for Stephanie Vaughn from April 30th clinic
biost_clinic_stephanie_vaughn.xlsxls biost_clinic_stephanie_vaughn.xls manage 25.0 K 23 Apr 2007 - 11:37 PatrickArbogast  
boxplotdata.csvcsv boxplotdata.csv manage 2.7 K 17 Apr 2006 - 10:27 QingxiaChen  
clintCarroll.sxcsxc clintCarroll.sxc manage 40.4 K 26 Feb 2006 - 21:30 FrankHarrell Clint Carroll Langerhans Data
clintCarrollabstract.sxwsxw clintCarrollabstract.sxw manage 8.7 K 26 Feb 2006 - 21:27 FrankHarrell Clint Carroll Langerhans Abstract
specificaims.docdoc specificaims.doc manage 25.5 K 13 Feb 2006 - 10:11 ChuanZhou Specific Aims
tang.rdarda tang.rda manage 13.4 K 19 Dec 2009 - 08:42 FrankHarrell Data from Yi Wei Tang processed using R code above
Topic revision: r455 - 16 Oct 2017, LaurieSamuels

This site is powered by FoswikiCopyright © 2013-2017 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Vanderbilt Biostatistics Wiki? Send feedback