Data and Analysis for Clinical and Health Research Clinic

Archive

Current Notes

2013 Dec 19

Bennett Spetalnick, Tara A.Nielsen, OB/GYN

  • Retrospective study, about 2 year period. 4500 deliveries per year. About 400 vs 400 for each treatment group per year.
  • The primary endpoint is composite adverse outcome (maternal and neonatal).

DIa Beachboard, PMI

  • Electronic images of host cells invaded by virus. Compare wild type virus and mutant virus. The endpoint is binary pheonotype of vesicle. Another type is vesicle size.
  • 30% aberrant phenotype in wild type vs 60% in mutant virus. 200nm vesicle for wild type vs 280nm for mutant.

2013 Nov 21

Dia Beachboard, Department of Pathology, Microbiology, and Immunology and Pediatrics

  • I am trying to determine statistical power and what tests to use for electron microscopy quantification I am doing.

    I have two types of analysis that I will be preforming on 3 data sets. The first analysis is a nominal categorical analysis of normal vs. abnormal vesicles within my EM sections. The second analysis will be to determine if there are differences in the diameter of these vesicles (quantitatiive/continuous).

    For my three data sets, I will be answering different questions. For data set 1, I have 5 samples and want to compare 4 mutant to wild type. For data set 2, I will have 6 samples but I will be comparing two samples at a time. I am testing 3 mutations in two backgrounds and will be determining if there are differences in the vesicle between background but not across mutations. For data set 3, I have wild type and mutant virus in the presence or absence of a drug (4 samples total) and want to compare each sample in the presence or absence of drug then compare whether the mutant has a different response to the drug than wild type.

    My specific questions about statistically power are: 1) how many cell sections do I need to image to get statistical power? 2) how many vesicles do need to analyze for statistical power?

Allie Greenplate

  • I'm working on a VICTR grant application and need help determining a minimum sample size. The hypothesis of my proposal is that melanoma infiltrating T cells will have a different phenotypic and functional profile compared to healthy T cells. We have primary melanoma tumors from a clinical collaborator that have been disaggregated to single cells. From there I will perform mass cytometry on each sample to determine phenotype and function. Mass cytometry is similar to traditional fluorescent flow cytometry, however the antibodies are coupled to metal isotopes, allowing us to measure 30+ parameters as opposed to 4-8 parameters measured in flourescent flow. This allows for greater resolution of small subpopulations

2013 Nov 7

Imani Brown, MPH student

  • REDCap branching logic question. Whether missing field when not applicable will affect statistical analysis.
  • Suggest attend REDCap clinic about how to start new section.

2013 Oct 31

Sharidan Parr, Nephrology Fellow, Department of Nephrology and Hypertension

  • Retrospective study on hospitalized patients.
  • Propose to identify pts hospitalized with AKI and matched hospitalized pts without AKI, measure urine protein at baseline and at least one after baseline
  • Will do propensity score matching on age, ethnicity, diabetes (severity of DM by A1C), BP, medication exposure
  • Some pts do not have protein at baseline, for those pts, change from no to yes is also important
  • Can analyze the whole cohort on the post protein measures using linear mixed-efects model: post protein ~ baseline protein + AKI + covar
  • Can do subgroup analysis on pts with protein at baseline (continuous outcome) and without protein at baseline (binary outcome).
  • If post protein is measured at different times between patients, need to adjust for time length as well

2013 Oct 10

Pratik Pandharipande, Jennifer Guiseffi, Cardiovascular Medicine

  • REDCap database. N=234 total pts. Retrospective study.
  • There are two different surgical procedures.
  • Primary outcome: post-op day 0, 1, or 2 delirium development.
  • Covariates: age, charlson index, pre-op ef, intra-op midazolam
  • Second outcomes: 30 days and 6 months mortality, related to delirium.
  • Will apply for VICTR voucher for biostat support. Estimate ~$4000. Need $1000 cost share.

2013 Oct 03

Michael O’Connor pediatric pulmonary fellow

  • Inpatient admission for cystic fibrosis exacerbations with increased airway clearance and intravenous antibiotics will result in improvement in serum fatty acid profiles as well as a decrease in pro-inflammatory metabolites

2013 Sep 26

Lynne Caples Vanderbilt School of Medicine

  • ongoing cardiovascular RCT; want to stop early
  • Stopping rules are usually specified before the trial starts. Use of particular method depends on individual study. (type of disease, type of treatment, question of interest...)
  • Early stop due to efficacy.
  • The important thing is to provide proper operating characteristics

Heather Maune, PGY4 Obstetrics & Gynecology

  • A new policy on Preventing the Primary Cesarean Delivery
  • Interested in the some outcomes before and after the implementation of new policy. Time is confounder.
  • Power analysis shows patients collected from four months maybe enough.

2013 Sep 19

Alex de Feria

  • Questions about what type of analysis and also about sample size for the database. The project is centered around patients seen at the Vanderbilt Center for Inherited Heart Disease.

Brooke Weaver, VPH

  • Have REDCap data base, want to generate some graphics.

2013 Sep 5

David Francis, ENT

  • Need biostatistics assistance with a surgical study. N=56.
  • Evaluate grade change over time for each patient up to 20 years
  • Grade has 4 levels: invasive, severe, mild/moderate, none.
  • Want to relate the grade change over time with the procedure. The patients started at difference state and had difference # of procedures at different time.
  • Longitudinal data analysis. Goal to publish paper.
  • Suggest apply for voucher in amount of $5000.

Jill Danford, ob/gyn

  • Retrospective study on the procedure of taking out the mesh. Outcome: after surgery pain (improved, unchanged, or worse)
  • N=226. Measurement of pain came from chart by description. About 70% of patients will have improved outcome. N=18 worse, N=44 unchanged.
  • Apply for VICTR voucher, suggest $2000.

Cesar Molina, Ortho

  • Retrospective study on development of minor complication, major complication, or death after procedure (2005-2011)
  • Problem: huge sample size, all risk factors are significant
  • Suggestion: report odds ratio with confidence interval as well as the p-value. Consider forest plot.

2013 Aug 29

Rachel Thakore, Vanderbilt Orthopaedic Institute Center for Health Policy

  • Compare length of stay and costs for tibia fracture patients who have received flap coverage (N=45) to those that have not (N=106)
  • Original data has been analyzed and manuscript has been written
  • Association between treatment and length of stay adjusting for age, gender, race, and ASA score; also want to look at the difference of cost between two groups
  • Retrospective chart review from 1/1/2010 to 12/31/2011, should consider time for cost analysis
  • The distribution of length of stay is probably skewed, can consider survival analysis

Pampee Young and Fred Oakley, Pathology

  • Look at transfusion reaction comparing peds with adult population, can use chi-square test
  • Only had age for pts with reaction. There several different reaction. Want to see relationship between each reaction and age.
    • Logistic regression of certain reaction on age. Consider nonlinear effect of age.

2013 Aug 22

Heather Kistka, PGY-5, Department of Neurological Surgery

  • a survival analysis investigating the predictive value of pre-op KPS score compared to post op KPS score.
  • KPS score takes the value of 0, 10, 20, 30, ..., 90.
  • Usually pre-op KPS is predictive
  • N = about 140.
  • Question: which one is a better predictor for patients' survival, pre-op KPS or post-op KPS
  • Apply for a VICTR Voucher. $4000

2013 Aug 8

Shaoying Li, Dept. of Pathology, Microbiology and Immunology

  • T0 is time of diagnosis. Event is death. Censored at today. Many patients have very short follow-up time.
  • Would like to examine the association between some risk factors and prognosis (survival).

2013 July 25

Sharmin Basher, Cardiovascular Medicine - Fellow

  • Sample size/power: investigate the effectiveness of supplementary written information to women during cardiovascular disease prevention consulting compared to consulting alone.
  • Sample size/power using PS
  • need a method to calculate the total score

2013 June 13

Lara Hershcovitch, Emergency Med-Housestaff

  • Study of patients with Parkinson disease, patients will have medication or medication with deep brain stimulation

2013 June 6 - Canceled

2013 May 23

Erin Fortenberry, Department of Obstetrics and Gynecology

  • Cohort: all the moms that gave second birth during 2007-2013 and also had c-section before
  • Are interested in the effect of gestational age (37-42weeks, could be in days) on the composite morbidities for unsuccessful TOLAC
  • Sample size estimation using PS

Shwin Krishna, Peds

  • Need biostat support for analysis only. Suggest applying for $2000 Voucher

2013 May 9

Wissam Abdallah, Ben Shoemaker, Fellow, Division of Cardiovascular Medicine

Correlation of left ventricular fibrosis by cardiac magnetic resonance imaging with success of radio-frequency ablation in atrial fibrillation
  • Check the estimate on the median time
  • Need use simulation to calculate sample size based on Cox model; Might dichotomize the continuous predictor and then use PS
  • For Aim 2, could use the marginal error of correlation for sample size justification
  • Could apply for $2,000 for biostat support

Olivier Boutaud, Department of Pharmacology

  • Sample size calculation based on t test using PS

2013 April 25

Jenna Faircloth, Pharmacy Resident, VCH

IRB approved research project with two pharmacists and two physicians reviewing antithrombin III supplementation. I have completed data collection and summary but have questions about which statistical tests I should use for data analysis.

Natasha Rushing, OB/GYN Resident

Retrospective chart review focusing on outcomes associated with diagnosis of chorioamnionitis.

2013 April 04

Connie Lewis, Cardiology

  • Hypothesis: More African American males with hypertension have heart failure.
  • What are the risk factors associated with young people having heart failure.
  • n=360 patients in database all with heart failure (not case-control)
  • Descriptive summary table to describe risk factors for heart failure by age (20-30, 30-40,...etc.)
  • This is a pilot study, risk factors for heart failure in young patients have not yet been described in the literature
  • Create awareness of heart failure in a young population
  • About 20 hours of support is recommended

Summer Wirth, Erin Rebele, OB/GYN

  • The Role of Transvaginal Ultrasound in the Diagnosis of Retained Products of Conception.
  • Is radiology report consistent with the pathology findings.
  • Pathology is the gold standard, don't want to perform unneeded procedures
  • Can the radiology report be depended on to prevent unnecessary procedures.
  • 650 total DNC (final number will be a subset DNC of detained product, about n=200), most of them will have had an ultrasound beforehand
  • Needs for residency presentation in May
  • About 20 hours of support is recommended

2013 March 28

Carla Sevin, Pulmonary Critical Care

  • After critical care, pts have follow up to determine if they are still having problems.
  • Studying if ICU recovery program is beneficial on outcomes (readmission, mortality)
  • Secondary outcomes: health care utilization, physical impairments, cognitive effects, quality of life.
  • Has intervention and control group., plus follow up over 6mos
  • Intervention gets a hospital visit.
  • Suggested getting an option for determining mortality.
  • Consider doing a time-to-event.
  • Suggested finding the confounders associated with readmission and mortality
  • Determine risk of the outcome in relation between the two groups in order to calculate sample size.
  • Using PS with a Detectable Alternative, Power = .8, m1 = 12, n = 220/group, accrual = 27mos, Follow-up =6, and 1:1 for the two groups: Results were .725 or 1.421.

2013 March 21.

Ruki Odiete, Cardiology

  • Give rats diabetes and then give heart attack
  • There are 5 groups: control for DM, DM, control for MI (SM), MI, DM+MI; have 5-7 rats per group
  • Outcome variable is strain. There are differences in strain among groups. Adjusting for other variables is difficult due to the limited sample size
  • Try to plot and calculate correlation between LARSS (strain) and EDV (volume) for all together, then separate into 5 groups
  • Regress strain on volume and group: strain ~ volume + group
  • Suggestions: 1. plot raw data and check distribution; 2. fit regression model of strain on volume for all together; 3. fit regression model for DM and control only: strain ~ volume + group

Verena Wyvill Brown, Department of Pediatrics

  • Project to understand the patterns of child maltreatment reporting as it relates to cultural and societal variables in Latino, white, and black children in Middle Tennessee.

2013 March 14

Benjamin Shoemaker, Cardiovascular Medicine

  • Outcome: atrial fibrillation burden (continuous) after bariatric surgery. It will be repeatedly measured (up to every month) for 2-3 years.
  • Question1: association between weight loss and AF burden
  • Question2: association between repeated measures of AF burden and weight change over time as well as other comorbities (6-10).
  • Sample size: 30 patients.
  • AF burden is expected to decrease over time. Comorbidities might also change over time
  • Can power the study based on the first primary aim, and put other as secondary analysis
  • VICTR voucher for biostat support in grant preparation in amount of $2000

Brandon Perry,

  • Evaluation of an Early Hemoglobin A1c-Based Screening Strategy for Gestational Diabetes N=1850 with 150 GDM cases
  • Q1: Develop a risk assessment tool for early diagnosis of Gestational Diabetes
  • Q2: Calculate incidence and prevalence of GDM at vanderbilt
  • Q3: Is there a statistically significant difference between GDM and controls regarding C-section rates, post-op complications, birth trauma, preeclampsia, and weight gain
  • Q4: Is there a statistically significant difference between GDM babies and control babies in terms of gestational age at delivery, baby location, birth weight, presence of shoulder dystocia, feeding issues, and baby bilirubin levels?
  • Q5: Calculate the time to event (% of patients that are being picked up early)?
  • Q6: Is there a difference in cost between GDM and control patients
  • Apply for VICTR voucher and $2000 is suggested

Kaushik Mukherjee, Surgery

  • Apply for VICTR voucher for obtaining data from national data base

2013 Feb 21

Fred Oakley, Pathology

  • Study on comparison of transfusion reactions between adult and pediatric populations
  • Total 108 events for peds and 277 events for adult during 2 years observed at Vanderbilt Medical Center
  • Need to know total number of transfusions (peds and adults) and report incidence rate for each adverse events
  • Since it is not a random sample, statistical test is not needed

J Newton, OB/GYN

  • Retrospective cohort study from two hospitals (academic medical center and community hospitals)
  • Compare exposed (borderline fluid level in pregnant women from ultrasound) with unexposed (normal fluid level) for a number of outcomes
  • There are total 175 exposed and 562 unexposed
  • 8 categorical outcomes and 3 continuous outcomes (age of mom, avg days in hospital, gestational age of baby)

2013 Feb 14

Romina Sosa, Clincal Pharmacology

  • The effect of aspirin on lung cancer is mediated by COX2
  • Treat the lung cancer patients with aspirin for 7 days and measure PGE2 (pilot study)
  • Applying for VICTR funding, requires sample size justification
  • Preliminary data on PGE2 level in the normal patients. 50 normal patients. Mean +/- SD before and after aspirin. 135 +/- 100 vs 75 +/- 60 (raw data is available to calculate within-subject correlation)
  • In lung cancer patients, would expect PGE2 level reduce by 40% with a baseline level of 135 or higher.
  • $500 biostat support

David Johnson, Aaron Fritts, Radiology

  • The goal of the study is to optimize the use(?) of Port cather
  • data set: 96 patients, during the past two years, who had port cather check.
  • The response variable: whether the cather needs intervention (yes/no), about half ascertained; the presence of problem with the cather (multiple situations).
  • The independent variables: tip position (yes/no), left/right side, vein (2), operator (2 level plus few NAs)
  • Could apply for a Voucher of $4000. The department copay $1000.

2013 Feb 7

Bennett Landman, EE, Pathology

  • Analysis of survey results. There are three surveies. Most are descriptive statistics and comparisons between two groups.
  • Need expert on survey analysis to oversee the method used. PI can do the actual data analysis.
  • Suggest involve Biostatistician in the design of the survey
  • $500 VICTR voucher is appropriate

J Newton, Ob/Gyn

  • Analysis of two cohort studies, possibly combined or birth outcomes. Investigator will be teamed with Li Wang for some funding arrangement.

2013 Jan 17

Rachel Wolf

Third year medical student applying for funding for a clinical research project. Currently working on proposal and need some statistics guidance

2013 Jan 3

Kathy Niu

Outcome: Catatonia subtype are categorical. Multiple visits overtime. DSM IV or V to establish criteria over 2009-2012.N=610 with repeats ~279 patients

--Assuming two levels of Catatonia - superous vs. excited can also be ordinal.

Predictor or etiology variable interest is whether it is medical or psychiatric with many different ways to assess.

Additional variables: Demographic variables medical history and drugs that they are taking.

Time effect may be associated with treatment variables. The medical era or practice is going to affect the outcome measurement or assessment.

Need to take into consideration the missingness of covariates.

Analysis considerations for multivariable regression. Correlated data analysis. Complex analysis example: Mixed Effect Logistic Regression with patient id as random effect.

May want to pick the first occurrence of the outcome as there are only very few with multiple outcomes and in that case the logistic regression analysis will be appropriate.

Additional notes: CTSA vouchers for biostatistical input. Check the excel spreadsheet from heaven vs. hell in biostatistics website.

2012 Dec 20

Paul Morphy

  • Study background: Special education, total of 38 subjects. Determine if two interventions are equivalent.
  • To test equivalence, two one-sided tests, or operationally 90% confidence interval (CI) is used to see if it would lie within the limit a priori decided (Reference: Schuirmann (Biometrics, 37: 617, 1981) proposed in bioequivalence testing).
  • The limit should be decided a priori: convention in education, a fifth of standard deviation considered for difference boundary.
  • For this study, the confidence interval approach is a better approach since it may be hard to set the limit that can be acceptable for everyone. Also calculate 95% CI.

2012 Dec 13

Mike Baker, Cardiology; Stewart Benton, Medicine

  • Compare traditional treatment (put the valve in through operation) and a new procedure (via leg), primary outcome is measured bleeding parameters
  • Every patients will undergo the new procedure and will compare their outcomes with historical findings
  • If the outcome is binary, 10% will have normal values before procedure and 70% will have normal values after, use McNemar 's test
  • If the outcome is continuous, use paired-t test to compare values before and after. Need to get estimate of the distribution and variation of the outcome from a pilot study (need about 6 or 7 subjects)
  • Better use the raw readings. Need to find the standard deviation of the pre post difference, also try to search literature for the distribution
  • Another approach is to sequentially estimate margin of error and stop when it reaches some threshold which is defined in front
  • Can apply for a VICTR voucher and estimate $2000 is suggested

2012 Dec 6

Edem Binka

  • Three groups of patients: specific renal support therapy, traditional renal support therapy, without any renal support therapy
  • Outcome: sodium and potassium level
  • Also want to look at association between age, weight, gender and outcome, should not be any other disease associated
  • Suggest descriptive statistics due to limited sample size, nonparametric test to compare between groups
  • If have ~50 patients, linear model is possible
  • Will apply for VICTR voucher support and $3000 is suggested

2012 Nov 29

Robyn Tamboli, Dept. of Surgery

  • Advised the investigator that sample size re-estimation needs careful thinking and planning; References on this topic were provided in clinic follow-up; A VICTR study design voucher was recommended.
  • Email: I need some guidance on a power calculation. I was awarded VICTR funds to look at how ghrelin affects insulin sensitivity before and after RYGB (VR178.3). I have complete pre and post data on four subjects. Would you advise that I re-determine my sample size based on the data from these subjects? Also, I would like to submit an amendment to add a lean comparator group and need to determine sample size. There is one paper on the effect of ghrelin infusion in lean subjects; however the ghrelin dose and methods for measuring insulin sensitivity are different from my approach. How do I determine sample size with the less than ideal available data?

2012 Nov 15

J Michael Newton, OB/GYN

  • pilot project to evaluate the association between BMI and pregnancy outcomes in obese women;
  • recommend a descriptive study to collect preliminary data;
  • n=360 with over-sample in extreme BMI range;
  • recommend a VICTR voucher of $3500.

William Dresen, medicine resident

  • Investigator email note:
I am a third year medicine resident working with Ben Shoemaker, one of the cardiology fellows, on a retrospective multi-variate model analyzing a-fib ablations
  • Retrospective review of sample size n=1800 with about 600 events, a-fib ablations. There is a list of about 20 variables;
  • recommend a tree-based model to understand the data and possibly logistic models to analyze the data;
  • Investigators requested a VICTR voucher of $2000.

2012 Nov 8

Peter R. Martin, Psychiatry and Pharmacology

  • Investigator email note:
I would like to request help with determining the sample size of a BioVu study to replicate preliminary genetic data we have previously obtained suggesting a protective human mu opioid receptor variant with respect to addiction. The only day I and my genetics collaborator Al George are available is Thurs Nov 8th. Please let us know how to proceed as this is my first time attending the Biostatistics Clinic.
  • Have identified two groups of subjects using the Synthetic Derivative (opioid and non-opioid dependent)
  • Look at association between addiction and phenotype
  • Try to match case and control by some important factors
  • Comparing 1.5% to 7%, need 169 cases and 169*2 controls

Romina Sosa

  • Investigator email note:
Question is on sample numbers I am writing a grant and need to figure out how many patients I will need to analyze to look at a biomarker. Briefly, I am looking at a marker on platelets that may be able to give good predictive information of platelet activation in disease processes. Our lab has preliminary data in a patient population with metabolic syndrome, that this is the case. I am trying to look at the same marker in a population with hematologic disease. How do I come up with numbers to write on the grant? Does the preliminary data we have in metabolic syndrome help me predict the sample numbers I will need?
  • Pts with metabolic syndrome, mean 9.6 (SD 3.7); normal pts, mean 3.7 (SD 1.1)
  • Plan study on pts with hematologic disease, need to calculate sample size.
  • If can get measures before and after treatment, then use paired test which needs less pts
  • Standardise treatment period (take second measurement after 3 months treatment)

2012 Nov 1

David Hak Kim, Cardiovascular Medicine Division

  • Investigator email note:
I would like to attend the biostatistics clinic regarding two related studies.

First, I have a database of 125 pts that have presenting with acute coronary syndrome under the age of 35 yrs old, hypothesizing that an increased BMI is related to worse atherosclerotic disease and outcomes. I would like to perform descriptive statistics of the cohort as well as stratify outcomes based on BMI. I have de-identified the data and it is attached.

The second study will be utilizing the synthetic derivative, identifying all patients under the age of 45 that has suffered an MI. The goal of the study is to stratify the group into different age groups (ie <20 yrs, 20-30, 30-40, >40) and note differences in the prevalence of traditional risk factors as well as BMI. Our hypothesis is that early presentation of MI is due to risk factors independent of traditional risk factors.

Would these be appropriate to present to the biostats clinic, and if so, Thursday?
  • Look at young patients with Acute Coronary Syndrome (ACS)
  • Study 1: retrospective cohort study, age between 18 and 35, ACS, year 2000 to present, N=124
    • A paper published based on 10 clinical trials about ACS (age < 45)
    • Want to compare between this study and the results published - not appropriate
    • Association between obesity and other risk factors in ACS
    • Suggest include all the pts with and without ACS and design as a case-control study
  • Study 2: use synthetic derivative to replicate the results published
  • Qualify for instant biostat voucher

2012 Oct 25

John Schneider, Department of Otolaryngology

  • Interested in conducting a survey looking at patient expectations regarding potential therapies for chronic sinusitis.
  • Now in the stage of developing the survey but need to conduct a power analysis to determine the sample size.
  • Patients have two choices for treatment: medical management or surgical management
  • Patients available for the study: already had multiple medications; already had multiple surgeries; or newly diagnosed patients without any treatment
  • Interested in identifying patients' characteristics regarding their choices of medical or surgical
  • Suggest do a pilot study first and the analysis will be mainly descriptive, then design a hypothesis generating study

Marissa Blanco, Kathryn Carlson, Department of Pediatrics

  • Study of developing appropriate discharge instructions for non-English speaking patients
  • Two arms: control and Spanish speakers
  • Will compare proportions for the two groups after intervention (control will remain same as 10%, Spanish speaker group will increase from 10% to 35%)
  • Can calculate power comparing two binomial distributions
  • Suggest work with Ben and Kelly through PEDS collaboration

2012 Oct 18

No clinical investigators.

  • Discussed Prof. Pena's seminar, and the paper on FDR-controlling procedure by Benjamin and Hochberg, JRSSB, 1995 with a group of students.

2012 Oct 11

Vincent Agboto, Meharry Medical College CTSA

  • Vincent could not attend clinic on 10/11. Rescheduled for another date.

  • Investigator email note:
I am sending you this message because I want to pay a visit to the clinic this week with an investigator that I am helping with an R01 on vaccines to new born in Bangldesh. We had studio a few days ago and we are working on revising the draft based on the suggested changes at the studio. I revised the statistical concepts and I would like to stop by just to discuss them with you.
Please let me know if we can stop by on Thursday to discuss the statistical help that I provided to her so we can get her moving.

2012 Oct 4

Jeremy D. Moretz, PGY1 Pharmacy Practice Resident

  • Attended clinic on 10/4:
The research question was discussed, suggesting the possibility of treating bleeding as a continuous endpoint instead of binary; analyzing the association between blood thinning index and bleeding outcome might also be a possible research question; propensity score approach is needed in the logistic regression or GLM.

  • Investigator email note:
Briefly, we will be attempting to retrospectively analyze the bleeding risk associated with either argatroban or IV UFH drug use for patients listed for cardiac transplantation.
Project Title: Use of DTI Anticoagulation for HIT Risk Reduction during Heart Transplantation Listing and Evaluation
Questions to be Answered: Is there a difference in bleeding events among inpatients awaiting cardiac transplantation who are anticoagulated with argatroban as compared to intravenous UFH?
Expected Outcomes of the Study: We hypothesize that patients anticoagulated with argatroban will experience more bleeding events that those anticoagulated with IV UFH. The identification of bleeding risks and potential reduction of risk will enlighten current practice of anticoagulation and contribute to the understanding of the practice of inpatient anticoagulation in patients awaiting cardiac transplantation.

2012 Sep 27

Rajshri Mainthia, Cancer Center; Alexander Parikh, Surgery

My name is Rajshri Mainthia. Dr. Parikh (PI) and I are applying for a VICTR grant (voucher) to be used for biostats support for a project we are working on. In order to complete our application, we have been told that we should attend a biostats clinic to get an idea of specifically what type and how much biostats support we need.

If possible, we would like to come to the biostats clinic on Thursday Sept 27th.

Attached is a description of the project, as well as the data and results we have thus far. Also attached is the database as a Stata file.

The analysis we would be requesting help with includes:
1) Looking over our univariate analysis/demographics data (Tables 1-6)

2) Multivariate regression analysis.
Outcomes:
-Radiotherapy use in Stage 2A and higher rectal cancer (neoadjuvant, adjuvant, neither or both)
-Chemotherapy use in Stage 3A or higher (all sites)
-Sphincter preservation (LAR/coloanal vs APR, extent, etc) for rectal cancers only
-Overall survival

Primary exposure variable: Insurance type (4 types)

Covariates: age, gender, race, rural, metro, primary site, tumor grade, tumor stage, margins, #lymph nodes examined, time to first treatment, tumor site, and type of resection

  • Attended clinic on 9/27.
  • Study the influence of health insurance in Tennessee on the treatment of colorectal cancer
  • There were 4 insurance types: private, government, TN care/medicare, and uninsured
  • Primary hypothesis: patients with certain type of cancer at certain stage should be receiving certain treatment (like chemotherapy) regardless of what insurance he/she has
  • Need help with going through the univariate analysis which have been done and the multivariable model
  • Possible apply for a VICTR voucher. Suggested $6000

Shanna Arnold, PUI

  • Study looking at whether biomarker can predict recurrence in bladder cancer patients within two years
  • Will have N=100 patients per year. About 50% will have recurrence within 2 years.
  • Asked about what study design can be used if want to stop the study earlier when observed a clinically meaningful hazard ratio
  • Can use Bayesian design, but require lots of work in front
  • Flexible frequentist method will also be possible, as long as the stopping rule is clearly defined in front (for example, after observing 50 events, we will stop and look at the data)

2012 Sep 20

Eric Millica, Dermatology

I am a dermatology resident looking to put together a clinical research project on diagnostic drift in the grading of atypia by dermatopathologists.  I am a little lost on the best way to set up the analysis looking at three classes of atypia and how they change over 3+ time periods for each pathologist (something along the lines of a kappa statistic, but with multiple time periods).

Some of the faculty members here just told me about your clinics and I see that you have one tomorrow.  I know it is less than two days notice, but I was wondering if it would be possible for me to come to the clinic tomorrow.  If not, do you know when your next clinical research clinic will be held?

Thanks,

Eric Millica

  • Has attended the clinic and may apply a voucher for assistance of sample size calculation and the analysis. $2000 is appropriate.

2012 Sep 13

Kendell Sowards, Surgery Trauma

  • Study of association between cpk and the outcome N=200 (renal failure)
  • Many patients had missing cpk values which are probably not random (need to justify)
  • cpk above 210 vs. below 210; dichotomization loses power and is not suggested
  • Try logistic regression, use original continuous cpk value and adjust for other factors
  • Some outcome only has 11 events. Is it possible to define the outcome as continuous?
  • Suggest apply for a design voucher $2000

Thais Plama, Urology

  • Study of urinary incontinence in relation to age, BMI, number of children, type of delivery
  • N=~700. Urinary incontinence as outcome, use logistic regression, age, BMI continuous
  • Score as outcome, use linear regression

2012 Aug 23

Claudia Ramirez, School of Medicine

  • Controlled clinical trial on bp-control drug.
  • 45 subjects in each arm. Within each arm, 3x3 crossover design.
  • Endpoints: MAP, plasma NE and other five secondary endpoints
  • substudy: 7 subjects in each arm
  • Applying for Voucher, estimate $7500

Phil Lammers, School of Medicine

  • Pilot study. The effect of Aspirin on PGE-M production derived from cox-2 activity in smokers
  • No previous information on effect size and variability available
  • Plan to enroll 20 male smokers at first, and then probably enroll more after effect size obtained
  • Applying for Voucher, estimate $6000 for sample size justificaiton, study design, data analysis, manuscript preparation
  • Pi is Cancer Center member and can use collaboration instead of VICTR?

Heather Kistka, Neurosurgery

  • Retrospective study. The effect of anti-depression on the progression-free survival and overall survival
  • Large brain tumor dataset, 141 subjects dead
  • Covariates: age, diabetes, smoking, chemotherapy, gender
  • May want to also include patients who are still alive. Should have representitive sample of the population. Don't subset based on the endpoint of interest.

2012 Aug 16

Romina Sosa, Clinical Pharmacology

  • Question about power calculation
  • Patient with metabolic syndrome, two groups (intervention and placebo)
  • Outcome: Lysyk-MDA-crosslink
  • Mean (SE) for the two groups: 3.7 (0.44) N=6 vs. 9.58 (1.15) N=10
  • Use SD and the difference to calculate the power with PS software

S.Nicole Chadha, Allergy Immunology

  • Retrospective chart review of 40 patients; Need descriptive statistics
  • Will apply VICTR voucher, estimate of ~20 hours ($2000)

2012 Aug 9

Victor Nwazue, Clinical Pharmacology

  • Sample size for a study of long term outcome
  • POTS patients who came to the clinic ~10 years ago and will bring them back (around 16 patients if all can come)
  • Pilot study with fixed sample size (N=16 maximum)
  • Outcome is continuous, have before and after measures, use Wilcoxon signed rank test (use nQuery to calculate power based on nonparametric test); or to use the precision approach (report margin of error)
  • Have to think about other confounders, like life styles

2012 July 5

VICTR statistician: Li Wang

Ryan Hollenbeck, Cardiology; Jeremy Pollock, internal medicine

  • Find optimal blood pressure in survivors of cardiac arrest treated with hypothermia
  • First 24 hours blood pressure. Survivors will stay in hospital ~ 10-12 days.
  • Outcome: in hospital mortality. Secondary outcome: neuological function.
  • Usually blood pressure is measured every hour, but will be more frequent for sicker patients.
  • Dose of medication (to adjust the blood pressure) should be considered since it might relate to the outcome
  • Possible approach: AUC for time spent above 80.
  • Plot of blood pressure over time for each patient
  • N=200, 120 died, can have around ~5 covariates
  • Apply for VICTR voucher, suggested $6000

Amlan Bhattacharjee, Anesthesiology

  • 60 cases, ~200 matched controls (matched on age, CPT code, surgeon)
  • Prediction of violations

2012 June 28

VICTR statistician: Chang Yu, Li Wang

James Johnson, Tom Talbot (Department of medicine)

  • Interrupted time series analysis *Total 3000-4000 observational opportunities, had convinient sample every month and watch how many times they wash hands
  • Had data from 2006 to 2012; Enhanced program started in 2009; Percent of hand hygiene adherence over time *Adherence is defined as # correctly performed / # total opportunities *Sampling method: same observer goes to the same clinic every month *Primary question: compare adherence rate before and after intervention *Contact Frank, apply for VICTR voucher requesting working with Amy Graves, estimated ~40 hours ($4000)

2012 June 06

Diego Hijano, Pediatrics

  • Study about breast milk vs. formula (N=700) within 28 days of birth
  • Outcome: disease yes/no (abour 10% prevalence rate); secondary: severity
  • Collect information: mother, birth weight,
  • Plan to apply for VICTR voucher for data cleaning/checking, analysis, manuscript preparation, suggest $6000

2012 May 31

VICTR statistician:

Drs. Courtney Horton and Candice Mcnaughton

  • Retrospective chart review pediatric trauma patients Level I and Level 2, start 2008-N= 4000 but will limit to those that have
  • an INR value-To have data for future study.
  • Using INR values as predictor of pediatric trauma patients outcome
  • Examine whether INR has added predictive ability to routinely collected clinical data
  • Outcomes considered: transfusion (Yes, No), among those transfused amount transfused, length of stay, in hospital mortality
  • Time to event analysis if looking at time to mortality and account for censoring.
  • Use INR as continuous variable - make recommendations to physicians using all available data
  • Other factors that are routinely collected: for adjustment age, gender, race, insurance status mechanism of injury vital signs mechanism of arrival, ESI, injury severity index. (Other measure GCS, not as helpful among children)
  • Suggest apply for a VICTR voucher for pulling data. Peds. collaboration for analysis (Dr. Ben Saville)

2012 May 10

VICTR statistician: Chang Yu, Li Wang

Peter Bream, Everett Gu, Radiology

  • Retrospective review of the use of stent graft. ~100 grafts. If there is problem, will put stent
  • Follow total 20-30 stents
  • Interest in primary and secondary patency rate (the survival time for the graft)
  • The patients have more than one interventions, Record the data in long format. Each patient will have multiple rows for each observation.
    • column names: id, age, gender, physician, visit date, outcome
  • Suggest apply for a VICTR voucher in amount of $6000. First $2000 is free, then 50% cost share.

2012 Apr 19

Jennie Esbenshade, Peds, Hospital Med, ID, Adam Esbenshade MSCI

  • Consultants: Yuwei Zhu, Frank Harrell
  • Detecting flu virus in health care workers in pt care area 2009
  • If sx, asked to come back and get nasal swab
  • Serology also; no flu detected
  • Multiplex PCR to detect other viruses
  • Questionnaire data to capture symptoms
  • Random 200 regular non-sick swabs selected as controls
  • 2394 swabs, 42+ (35 sx, 7 asymptomatic)
  • 119 "sick" swabs
  • Also interested in variation with type of health care personnel
  • +/- specimen vs. cough, runny nose, aches, fatigue, fever >24h, sex, age, child @home, MD/RN
  • Did not use P-values in deciding which variables to put in multivariable model
  • Second model predicting the probability of + and sick from non-sx variables sex, age, MD/RN
  • Suggest variable clustering and redundancy analysis
  • Estimate standard $2000 biostat voucher will be adequate

Margot Lazow, working with Dr Kim in ophthalmology

  • Uveitis - autoimmune - immunosuppressants for a year - topical and systemic
  • Taken off treatment, look at recurrences of inflammation compared with those who did not have recurrence
  • What are the risk factors?
  • Age, sex, type of med, underlying med condition
  • Watch for treatment by indication bias in original selection of immunosuppressants; capture original reasons
  • Variable clustering may be helpful in dealing to lots of symptoms/risk factors
  • Ask at least 3 clinical experts to list factors thought to be related to recurrence and those thought to be related to drug prescribing
  • Duration of disease, severity of disease

2012 Apr 12

VICTR Biostatistician attendees:

Meredith Pugh (Pulmonary and Critical Care)

  • 10 year retrospective cohort study for a rare disease called PAH which is related to pulmonary hypertension
    • Have approximately 246 (min age 65 yr) patients meet inclusion criteria and of those, 37 had PAH
  • Looking for the association between PAH and following potential risk factors
    • age
    • gender
    • tissue disease
    • other clinical measurements
  • Thoughts on proposed analysis:
    • Logistic regression - report adjusted odds ratio and 95%CI for selected 3-4 important risk factors, may have model over-fitting problem if including too many covariates, ORs along with CIs can be displayed by figure
    • Descriptive analysis
  • Suggested possibly applying for a VICTR Biostat Resource Request
  • Because some analyses were done, recommend requesting $2,000.

2012 Mar 29

VICTR Biostatistician attendees: Chang Yu

John Grave, Peter ?? (Health Service Research)

  • Finite mixture models and methods to differentiate primary care provider from speciality visits based on medical records.

2012 Mar 8

VICTR Biostatistician attendees: Chang Yu, Li Wang, Hui Nian

Chris Anderson (Urology)

  • Using SEER data (national cancer registry) to look at bladder cancer patients
    • Have approximately 300 meet inclusion criteria
  • Looking at the management of these patients and their survival
    • Have good info on the patients' follow-up
      • Approx half died
    • Some received chemo after surgery --- about 70 patients.
    • Because treatment was not randomized, considering a propensity score analysis
    • Wanted to know if this is correct approach or if some other analysis is better
  • Thoughts on proposed analysis:
    • Logistic regression - what variables are associated with getting more chemo post-surgery --> where he wants to use the Propensity Score analysis
    • Cox regression - outcome is death and primary predictor is whether patient received chemo after surgery; use propensity score outcome as covariate

2012 Feb 23

VICTR Biostatistician attendees: Chang Yu

Nursing project on evidence-based practice survey and electronic medical record review.

Meta-analysis of respiratory infection in developed and developing countries.

2012 Feb 16

VICTR Biostatistician attendees: Chang Yu, Li Wang

Josh Smith, DBMI

  • Study drug adverse effects or indication
  • Have table listing findings for all the drugs, 164,000 pairs, want to know method of taking sample for the reviewers to review
  • Need to standardize the procedure, from the 3 sources to evaluate the drug symptoms as: correctly identified as AE, indications, or undefined
  • Can take the trained reviewers as the gold standard and compare our tool to the reviewers
  • website "sider" for side effects, can it be used as the gold standard?
    • compre the tool to "sider", pay attention to those which don't agree and let reviewers to review
    • also take sample from those not in "sider"

Diane Andens, CRC

  • Look at the correlation between the score from questionaire and the urine residual in bladder
  • 25 subjects is a good number for pilot study
  • Primarily descriptive: describe the distribution of the scores and compre to ultrasound results

2012 Jan 19

VICTR Biostat attendees: Li Wang, Chang Yu

Carrie Geisberg (Cardiology)

  • Has death proportions from cardiology patients from two different groups across time
    • Do not have raw data, have only aggregate --- eg, proportion of patients in each group who were dead at 1 year, 3 years, 5 years.
  • Issue with data: not necessarily same follow-up on all patients; data is broken down into time frames when patients received their surgery
  • Would like to know if proportion of dead patients at each of the time points is significantly different between the two groups.
  • Discussed trying to get a better "cohort" of patients --- ask for those patients who had surgery during a certain time frame and are followed for a specific period of time; then tally the proportion dead at 1, 3, and 5 years.
    • Also discussed trying to get "time to death/follow-up" (in days) for each patient, instead of aggregate level data.
  • Suggested possibly applying for a VICTR Biostat Resource Request
    • Because eventual goal is publication, recommend requesting $4,000.

Beatrice Stefanescu (Neonatology/Peds)

  • Randomized control trial regarding ventilator associated pneumonia (VAP)
    • Want to look at time to off of ventilator --- may or may not have had VAP while on the ventilator
    • Recommend applying for a VICTR Voucher --- $2,000 should cover it
  • Have 2nd study looking at neurological impairment of babies (at 18 months) put on one of two different breathing machines

2011 Dec 8

VICTR Biostat attendees: Li Wang, Chang Yu

Rubin Baskir (Cardiology)

  • Brought in data in electronic form; was also able to get a raw death rate for each county.
  • Also have raw data from California
  • Generated a two-way table of Lithium Category vs Death Rate Category as well as Chi-Square test
  • Also generated a sunflower plot as discussed last week
  • Discussed performing a Poisson Regression to estimate the effect Lithium Category using the raw Death Rate as the outcome.
  • Suggested moving forward with the VICTR request for $2,000 Voucher in order to get "pretty graphs" and Poisson Regression analysis.

Dr. Zhaoliang Li (Cell Biology)

  • Has submitted a manuscript and needs some help on how to address the reviewers' comments.
  • Reviewers asked for some "additional" statistical analysis for a graph
    • Graph represents expression of over time
    • Graph has three lines depicting three different cell lines --- normal expression, under expression, and over expression
    • Each line represents 3 samples (ie, a mean +/- SD is shown each time point on each line)
    • Chang's suggestions:
      • Calculate slope for each line between time 0 and time 12hrs; report slopes only in response to reviewers.
      • Perform non-parametric Wilcoxon rank-sum (aka, Mann-Whitney U test) test to compare (1) over expression to normal expression and (2) under expression to normal expression at time point 12 hrs only.
        • So, will report two p-values.
      • Make sure you state limitations of data (ie, only 3 samples per cell line).

2011 Dec 1

VICTR Biostat attendees: Li Wang

Rubin Baskir (Cardiology)

  • Has county level data from Texas for a given year --- ion level (in ground water) and death rate.
    • Both variables are categorical.
  • Would like to know if there is any correlation between ion level and death rate.
  • Recommend creating a 2-way table (ion level bvs death rate).
  • Discussed creating a sunflower plot.
  • Discussed running a Chi-Square test, Spearman correlation, and Chi-Square trend test.
  • Recommend returning Thurs Dec 8 with data in electronic format.

2011 Nov 10

VICTR Biostat attendees: Li Wang, Chang Yu

Jason Williams (VUIIS - Imaging Science)

  • Regarding a submitted VICTR request - have already submitted a protocol; initially requested $9,000 --- Dan Ayers helped write stat analysis plan and sample size.
  • In the past, Dan Ayers has supported PIs in related studies
  • Support requested is actually for meetings/consultation over the data accrual time period (2 years) and eventual analysis at end of study.
    • Dan Ayers would fulfill the meetings/consultation over the two years and mentoring of the MS statistician at the end of the 2 years
    • MS statistician would fulfill the analysis at the end of the 2 years
  • Li Wang to send an email to Frank Harrell (CC Chang Yu, Dan Byrne, and Jason Williams) to ensure this structure of work over time is "okay" under VICTR; also to ask Frank if this study may fall under the new Imaging Collaboration.
  • Update: Chang and Li met with Dr. Williams on Nov.16, 2011. We felt that $5000 is reasonable to apply for support of the statistical analysis and preparation for the manuscript, which will be accomplished by VICTR biostatisticians (master biostatistician doing the analysis under supervision of PhD biostatistician).

2011 Nov 3

VICTR Biostat attendees: Chang Yu, Terri Scott

Sabina Gesell (Pediatrics)

  • Completed study looking at children's physical activity (measured by accelerometers) during an after school program (either at a school or a community center)
  • Main objective: Compare the community center program to the school program --- community center program was designed to get the kids active.
  • Roughly 50 kids in each group
  • All African American children
  • Measured the children's activity multiple times in each child.
    • Have number of minutes spent in each level of activity (rest, sedentary, lowintensity, moderateintensity, and vigorousintensity).
    • Sum of these numbers = total time spent in the program (on that day) --> this total is different across the children (ie, some children stay longer than others).
  • How much time (specifically, proportion of time) is spent in (1) moderate+vigorous intensity activities, and (2) sedentary+rest
  • Possible regression: Poisson or Negative Binomial regression
  • Final goal: manuscript
  • Cost estimate: $6,000 (60 hours) --- will need to submit a protocol.

2011 Oct 20

VICTR Biostat attendees: Chang Yu, Dan Byrne, Terri Scott

Laurie Cutting (Pediatrics)

  • Sent email regarding "Neurobiology and Treatment of Reading Disability in NF1" NIH grant application.
  • Hypothesis: drug magnifies affect of tutoring
  • Discussed how many arms to move forward with -- 4 (placebo, only tutoring, only drug, or tutoring and drug) or 3 (only drug, only tutoring, or tutoring and drug)
  • Need revised sample size calc -- based on hypothesis, study should be powered based on the interaction effect.

Emily Reinke (Sports Medicine)

  • Take X-rays of individual's knees and measure the space (ie, distance in mms) between bones
  • Have two raters that will be making measurements from the same images using the same method
  • Need to determine how many times each rater needs to examine the same image and how many times the two rates need to examine the same image
  • Major aim of project: to measure and describe the distance --- so, don't need both raters to examine all images each (that is, each raters can examine a subset of the images)
  • Recommend a Bland-Altman plot to further examine agreement between raters
  • nQuery Advisor -- sample size for a precision (ie, width of a confidence intervals) around the desired intra- and inter-rater ICC values.
    • Will determine how many images the raters will have to examine in common
    • Would be nice if the images chosen for the two raters to examine each are also the set of images each rater will examine twice

2011 Oct 13

VICTR Biostat attendees: Chang Yu, Dan Byrne, Terri Scott, Li Wang

James (Jim) Powers (Medicine)

  • Discussed "Exploring the Utility of Ultra-Brief Delirium Assessments in Non-Intensive Care Geriatric Population: the GEM study" research study that was explained in an email sent to the biostat-clinic email address on Oct 6, 2011.
  • Dr. Powers has conducted some initial analyses, but would like support for additional analyses.
  • Goal: manuscript.
  • Desired additional analyses, include examining the variability and confidence intervals (possible bootstrap CIs) in more detail. Also would like some possible sub-group analyses (may be difficult because have only 7 patients with delirium; possibly can use a mixed effect models). Possibly calculate kappa.
    • Would be interesting to look at time from admission to study involvement (and time from admission to unit to study involvement), where "study involvement" is defined as the first day the CAM/DSM-IV are assessed.
    • Would also be interesting to compare the data from each rater (for each patient) to each other.
  • IMPORTANT: because this is a VA project, can only have someone who has WOC clearance (eg, Ayumi, Jennifer, Svetlana, Sam, Shirley, David) work with the data --- none of the current VICTR folks have VA clearance.
    • Ayumi is happy to provide oversight; she suggests Svetlana or Jennifer for the bulk of the work.
  • Suggest requesting $4,000 Biostat support from VICTR.
    • Already has support from Chair for additional 50-50 cost sharing plan.

2011 Sept 29

VICTR Biostat attendees: Daniel Byrne, Chang Yu, Hui Nian, Li Wang

Mary Sundell, Nephrology and hypertension

  • Assess the agreement between two measurement methods
  • Each patient has observations at three time points
  • Suggested plotting the data at baseline as a start, can use Bland-Altman plot

Ryan Hollenbeck, Cardiovascular Medicine

  • Effect of early catheterization on clinical outcome
  • The patients were not randomized
  • Suggested use propensity score to account for the group differences, matching, and Cox proportional hazard model. The outcome is time to discharge from hospital, and the patients can be censored at that time. Also good to have complete record of patients' current status and analyze the overall survival.
  • Will be applying for VICTR --- suggest applying for $2,000.

2011 Sept 22

VICTR Biostat attendees: Terri Scott, Li Wang

Shubhada Jagasia, Endocrinology

  • Reference to email sent by Brandon Perry (9/21/2011).
  • Related to "Hemoglobin A1C as a screening tool for gestational diabetes" study (VR2144)
  • Would like to look at sub-groups of women who have different HA1C "profiles" across their pregnancy --- in terms of result of early screening, screening at usual time spot (24-28 weeks), etc.
  • Would like to also compare these subgroups of women to "control" women (ie, those whose H1Ac and glucose testings were normal throughout their pregnancy) --- would like to see if their are "predictors" of the various profiles.
    • Questions of interest "Is there an H1AC in the first trimester that correlates with GDM in the second trimester?"; "Is it possible to intervene with certain groups of women to reduce the incidence of GDM?"
  • Goal is a manuscript.
  • Will be applying for VICTR --- suggest applying for $4,000.

2011 Sept 15

VICTR Biostat attendees: Terri Scott

Pampee Young, Pathology

  • Human stem cells --- not understood if/how stem cells are different across people
    • Looking at possible potency assays; and wondering "are stems cells different?" (across people)
    • 10 patients; bone marrow from each; made stems cells from each persons sample
    • How do the "test" results compare?
  • Need support for analyzing the data --- descriptives (including, mean +/- SD; 25th, 50th, and 75th percentile; and range) and graphs (ie, boxplots and stripcharts).
    • Also would like to see how distributions differ across gender and age.
  • Need support for manuscript writing, etc.
  • Suggest requesting $4,000.

2011 Sept 8

VICTR Biostat attendees: Terri Scott

Oscar Gomez, Peds Infectious Disease

  • Wish to submit two Biostat requests to VICTR --- one for support for an IRB proposal on a grant that's already been submitted; and one for a grant to be submitted in Jan 2012.
  • For IRB proposal --- need confirmation of needed sample size as well as statistical analysis plan. Would be good to have a general review of study design and other statistically related issues. Lastly, discuss using REDCap to collect and manage data; include sentences in IRB proposal.
    • Feel $2,000 (~ 20 hours) would be sufficient support for work required.
  • Grant to be submitted in Jan 2012 will be an RO1 --- Will need support regarding specific aims, study design, choice of measures (ie, outcomes and predictors), sample size calculation & justification, statistical analysis plan, estimate of biostats support needed for grant budget, and data collection/management.
    • Because it's an RO1, would recommend asking for $4,000 support --- will need to submit a letter of support for the 50% cost-sharing plan (over $2,000).

2011 June 2

VICTR Biostat attendees: Terri Scott

Sunil Kripalani, Medicine

  • Submitted a Voucher request for sample size calc for R01 (Sept)
  • Will need: sample size calc, stat analysis, and study design
  • Will be looking at a subset of patients from the ISCHEMIA worldwide clinical trial (David Maron)
  • Subset who speak English and are in the US or Canada - want to engage them in an ancillary study -- either traditional clinical management or a telephone based intervention that would improve their medical adherence
  • Hypothesis: intervention would improve their medical adherence as well as cardio outcomes
  • Was suggested (from ISCHEMIA PI) to also have a life style intervention --- 2 by 2 factorial study
  • Was suggested (also from ISHEMIA PI) to not randomize at the patient level because of another ancillary study that will be randomizing two different BP interventions --- so, have cluster randomization instead
  • Thoughts from Dr Kripalani: ask each patients at enrollment whether they are adherent then enroll those folks who are non fully adherent
  • Thoughts of a three arm study --- adherence intervention, lifestyle intervention, and no intervention
  • Will have ~150 sites in US/Canada, but only ~800 patients across all sites --- so small number of patients per site (will affect sample size calc for cluster randomization)
  • Will be following the patients longitudinally for a year or so --- analysis will involve some repeated measures modeling
    • Will need to consider sample size in order to perform proposed analysis (yet to be determined) and dropout/missing data
  • Thoughts for revision of VICTR request:
    • Request Frank's involvement since he's been involved in the ISCHEMIA design
    • Increase requested amount to $4,000 (because of many nuances of study design)
    • Approach the sample size calc from the point of you of "we'll have X patients enrolled in the US/Canada, what effect size can we find with 80% power?"

2011 May 26

VICTR Biostat attendees: Frank Harrell, Dan Byrne, Terri Scott

David Tabb (Biomedical Informatics)

  • Wanting to submit an RO1 in October --- involving proteomics
  • Needs a statistician's time for writing the grant --- preferably someone with proteomics experience (eg, Ming Li)
  • Question: should Dr. Tabb use VICTR funds to get help? Or should he utilize Ming (and the Cancer group)?

Crystal Rice (CRC Nurse)

  • Comparing two types of saline flushes --- used to flush IVs
    • (1) pre-filled syringe; and (2) bag flush
  • Patients have complained about taste and/or smell of the different flushes
    • Would like to validate that the flushes actually cause a smell/taste
    • At this point, goal is not to determine if there is a difference (ie, hypothesis test) but to describe their reaction to the flush (if any)
  • Would like to blind patients to which flush they are receiving
    • Have a cross-over randomized trial
    • Would like to piggy back on an existing study that involves flushes
  • Smell - Yes/No; Taste - Yes/No; Degree of taste/smell; Type of taste (metallic, bitter, sweet, etc)?
  • Also collect simple descriptives of each patient --- age, gender, race, smoking, pregnant, concomitant meds
  • Plan to exclude chemo patients

Amy Dreischerf (Endocrinology) & Charles Keil (Human Nutrition & Gastro)

  • Have data from a joint PI from a previous from a genotype and nutrition study
  • Would like to explore some additional hypotheses with "extra" data that were collected in original study
  • Goal: poster (for conference) & brief summary of findings (for summer internship)

2011 May 5

VICTR Biostat attendees: Terri Scott, Frank Harrell

Karen Chen & Samit Patrawala (Dermatology)

  • Follow-up from 2011 April 21
  • Prioritized analysis:
    • Plaque & patch stage vs tumor stage patients (at time of diagnosis) -- looking at outcomes (ie, number of patients, outcome, survival, etc)
    • Medical therapy vs radiation therapy --- looking at outcomes
    • Whether received Bexaritene (Targretin capsules) or not -- looking at outcomes
    • Focus will be descriptive --- should calculate some confidence intervals
  • Feel $6,000 estimate is accurate

Christi Parker (Pharmacy)

  • W/ Anesthesiology; studying cardiac surgery patients
  • Adrenal insufficiency
  • Etomidate - studying incidence of adrenal insufficiency in patients who receive Etomidate
  • Initial StarBRITE request: $2,000 for "Data analysis"
  • Desires:
    • Primary predictor: Etomidate yes/no
    • Primary outcome: Adrenal sufficiency yes/no
    • Cortisol levels w/in 72 hours of being induced with Etomidate for surgery
    • N approx 250 patients (1 record per)
    • Basic desired analysis: 2x2 of Etomidate vs Adrenal insufficiency
    • 2ndary outcomes: vaso-presser hours, mechanical vent hours, hosp LOS, ICU los, receipt of stress dose steroids
      • Does Etomidate have an effect on the secondary outcomes
    • Possible adjusted analysis - will need to determine possible confounders
    • Suggest collecting continuous adrenal outcome
    • Want a manuscript --- way in the future!
  • Suggestions: continue with $2,000 request; draft detailed ranked analysis plan.

2011 April 21

VICTR Biostat attendees: Chang Yu, Terri Scott

Karen Chen (4th year medical student working with Dept of Dermatology)

  • 40 patients with Tumor stage T-cell lymphoma
    • Either initially diagnosed with tumor stage or progressed to tumor stage
  • Have collected various demographic and clinical data on each patient
    • Have REDCap database --- have a lot of questions that want to answer (what's possible will depend on the data once we get a look at it)
      • Data: clinical characteristics collected at presentation; data collected for each treatment period (possible multiple treatment perdios; treatment periods can overlap)
    • Want to look at "survival" (time to death; time to progression; disease specific survival; time to "treatment failure")
    • Final goal: abstract/poster leading to a manuscript
  • Estimate of hours: 60 --> $6,000 --> will have to cost share on $4,000 (Need to talk with mentor to identify key question(s) and thus $ may vary.)

2011 April 14

VICTR Biostat attendees: Frank Harrell, Chang Yu, Terri Scott, Li Wang

James (Jim) Powers & Mac Buchowski (Medicine)

  • Elderly patients -- two groups (inpatient (much sicker) and "more healthy")
  • Two techniques for measuring amt of water in patient --- want to compare the two techniques (for clinical utility); one more invasive/time consuming than the other (other based on bedside measures)
  • Question: can we develop a "simple" measure (based on bedside measures) to predict stuff for care of these patients.
  • Statistical suggestions:
    1. Calculate rank correlation b/w the two measures
    2. Perform regression analysis -- try to predict one measure using the other(may adjust for other covariates)
      • Can also calculate mean abs error from predicting one w/ the other (good measure of clinical utility)

Ken Monahan & Evan Brittain (Cardiology)

  • Submitting grant (due May 15) to examine how right side of heart works using an ultrasound method.
    • Need help with statistical portions of grant -- sample size & stat analysis plan.
      • Feel $2,000 would be enough for this.
  • Bland-Altman comparison of MRI vs ultrasound method
  • Transit time --- using MRI and new technique
    • Standard deviations of transit time (using both methods) would be good to have for sample size calcs
  • Establish reproducibility in "normative" population; also, reproducibility in popn w/ known pulmonary hypertension
    • Interested in how transit time differs b/w groups
  • Sample size will be determined (at this point) based on funding
    • Consider calculating sample size needed to estimate a correlation coefficient w/in a specific "margin of error" -- need SD estimates
    • Would like to calculate sample size needed in the future for further studies
  • Recommendation for amt to specify in grant budget for stat support (if grant awarded): 100 hours of time (max)

2011 February 17

James (Jim) Powers, Medicine

  • Also with Jim, Bill Gregg (Informatics)
  • Attending VICTR junior biostatisticians: Terri Scott, Hui Nian, Li Wang
    • Frank Harrell and Cindy Chen also in attendance.
  • Discussed data to be pulled from StarPanel:
    • For each patient, at each clinic visit -- measures like # high risk meds, total # meds, # htn meds, # diabetes meds, # dementia medications, BP, GFR, Pneumovax, and Flu vaccine.
    • For each patient, for each day 1 or more "contacts" was made with Clinic --- total # contacts made on that day.
    • For each patient --- dates of any "visit" (hospital admission, ED visit, PCP visit) and type of visit.
      • Similarly, no shows/cancellations.
      • For hospital admissions -- length of stay, primary reason for admission.
        • Similar for ED visit, PCP visit.
  • Discussed possible analysis:

2011 January 27

James (Jim) Powers, Medicine

  • Also with Jim, Renee Porier
  • Attending VICTR junior biostatisticians: Terri Scott, Hui Nian
    • Also, Cindy Chen
  • Email Biostat Clinic email address with data set, protocol, and data questions on 1/21/2011.
    • NOTE: not all of the needed variables are included in the emailed data set, including BP and Study ID.
    • Also discussed getting longitudinal data on each patients --- emailed data set has one row per patients.
    • Also, before & after group only includes those patients are only those who came to the clinic throughout the study period --- that is, those that died (for instance) are not included.
  • Has submitted Biostat services request in StarBRIITE -- perform statistical analyses and assist with publication.
    • Suggest to revise request to include discussion of what data needs to be collected (ie, pulled from StarPanel and StarTracker dashboard) for the analysis.
  • Interventional study in geriatric population --- have before & after measures for core group of ~600 patients.
  • Recommended revising request to $6,000 --- will need letter of support from Dept/Div.

2011 January 20

Michael Osgood, Research Fellow, Surgery

  • Goal of study: sample size calculation for a clinical trial
  • Compare two groups of patients with binary outcome
  • Estimated proportions in each group and use PS software to calculate the required sample size

Amy Pyle, Fellow, Pathology; Dan Anderson, RAII, Pathology

  • How to get Biostatistical help for a funded VICTR study

2010 June 17

Cyndya Shibao, Asst Prof, Clinical Pharmacology

  • Goal of study: differences in energy expenditure (total and resting) between autonomic failure patients (N=10) and matched controls (N=15)
  • Resting energy expenditure correlates highly with body fat free mass
    • Want to adjust for this and gender in a model
  • Interested in how to graph results
  • Linear model: REE ~ grp + FFM + grp*FFM
  • Could do a logistic regression with group as the outcome, adjusting for other factors
  • Could graph FFM by REE for both groups, adding a fitted line with confidence interval
  • Also recommend "Forest Plots"

Jorge Gamboa, Fellow, Clinical Pharmacology

  • 3 x 3 cross over design, patients (N=15) receive all three treatments with 7 day wash-out period in between each.
    • Each patient is measured 5 times per treatment
  • Problems with missing data - none are missing all time points
  • What would you like to compare among three treatments? Ending value? Peak? Time to peak? AUC?
  • Suggest making plot to start
    • x-axis - time; y-axis - outcome measure; color - treatment
    • Use graph to determine how to compare across groups
  • To deal with missing data, use mixed effects model.
    • Could also use data around it to impute

2010 June 10

Michelle Griffith and Jeff Boord, Endocrinology

  • Need help estimating statistical support needed for project
  • Working with Bioinformatics to get data concerning patients and blood sugar levels
    • Interested in building model predicting hypo- and hyper-glycemic events
    • About 35,000 admissions in dataset
    • Estimate what percent will have hypo- and hyper-glycemic events.
    • Consider separate models for each disease
  • Recommend getting initial voucher of 20 hours to get a better picture of what will be needed.
  • New system beginning July 1: After first $2,000, department will have to pay half of biostat request

2010 June 3

Jeff Kantor, Pediatric GI

  • Studying association between isoprostanes (measure of oxidative stress) and percent body fat
  • Retrospective cohort of 158 patients between the age of 8 and 17, currently healthy, many obese
    • Combination of patients previously studied in CRC
  • Chang suggested to make a list of factors that could effect isoprostane levels
    • diet, stresses to body, undiagnosed, dm, etc
  • Can include around 10 variables in the model.
    • would need to include age interaction with many variables.
  • Consider redundancy of some variables.
    • Ex: Weight in lbs and kgs are perfectly correlated and it would not make sense to put both in a model
    • Are there some variables which could be considered highly correlated?

Current Notes
Topic revision: r1 - 11 May 2015, DalePlummer
 

This site is powered by FoswikiCopyright © 2013-2020 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