CEASAR Aim 1 with 3-year follow-up data

This will address the same aim 1 objective comparing treatments with respect to function outcomes that we did for the one-year data, but we will include the new data from the 3-year surveys.
  • May stratified by risk. One for low-risk patients. One for intermediate- and high-risk patients.

Cohort definition

Same as for the 12-month aim 1 paper, except now will include patients who had only a 3-year survey.
  • Age 80 years or younger
  • Clinical tumor (cT) 1 or cT2
  • Prostate specific antigen (PSA) < 50
  • Prostate adenocarcinoma diagnosed within six months of enrollment
  • Treated with RP, EBRT, or AS (as their one-year treatment assignment)
  • Completed a 6-month, 12-month and/or 3-year survey in addition to the baseline assessment
  • Why are there 8 patients who had AS for txThusFar3yrA1.F06 but have missing .F12 and .F36 versions? The answer is that after their 6 month survey, they received treatments such that their treatment category would then be hormone or ablation. Since we are excluding those groups of patients from this analysis, they have the .F12 and .F36 tx variables blank, and only their 6 month data are used.

Analysis plans

  • Treatment is the independent variable
  • EPIC domain scores (including hormone function) = dependent
  • Longitudinal GEE (use earliest instrument (baseline?) results as covariates)

  • For any subset analyses distinguishing rads patients who did and didn't get ADT, per email communication with DB and DP c. 2016-04-13:
    • If adt was started more than 6 months after ebrt it should be salvage.
    • classify patients in the EBRT+ADT id they received ADT within 6 months of the conclusion of EBRT. If there is no ADT either before, during or within 6 months after EBRT, I would classify them as ADT only.

Output / Results planning

  • For showing the results, we will focus on the predicted function/differences in function at 3-years
  • For the figures, we can probably show continuous time
  • Figure 1. Consort diagram
  • Figure 2. Functional trajectories (unadjusted)
  • [Figure 3. Functional trajectories (unadjusted) by baseline function. Consider also stratifying by risk group (low vs. intermediate/high) or putting in separate publications.] Ė maybe
  • Table 1. Baseline characteristics and EPIC function scores
  • Table 2/3: Comparisons of primary interest.
    • Instead of format from Karenís paper letís have a visual format.
    • VISUAL FORMAT AND ALSO 3-YEAR DATA IN A TABLE.
    • PRESENT ONLY BASELINE AND 3-YEAR, ELIMINATE 6MO AND 12MO FROM TABLE (but the 6 and 12 month data could be shown in figures)
    • Mean score difference or mean score? DB would like the mean scores, but have advised that the difference is the measure of interest. Can present differences in text and predicted probability in tables/figures, but have also advised that this may confuse readers.
  • Sensitivity analyses. Functional trajectory and multivariable model restricted to nerve-sparing RP and non ADT IMRT. Consider also stratified by risk group or putting in separate publications.
  • Number excluded and why
  • Number in each group that experienced recurrence, second treatment, prostate cancer mortality, any mortality

Meeting notes

2017 February 1

  • New temporary appointment
    • Upper limit of 4 hours
    • Probably not analyses, just questions.
  • DUE BY END OF THURSDAY (TOMORROW)
  • TK needs all the knot locations. Can access from model object? Or just use the quantiles.
  • Add tick mark at end of x-axis, with no label.
  • Change the cutoff for high baseline from 80 to 90 and rerun figures.
  • Add the number of nonmissing, by tx, on the figure 1 panels. Numbers directly from the etable 1 (Supplement). Can incorporate this into my existing functions. AND for the high and low baseline function. 552/2550?,
  • Editor wants all the tables to the supplement, and put figures for all the individual items UNADJUSTED figures.
  • Incorporate the individual items by making plots of trajectories of UNADJUSTED predicted probabilities with CIs
    • Make a couple of prototype figures to show Dan before doing all and assembling.
  • Dan's wife's grandparents used to own a linen cleaning service.

2017 January 16

  • Alex and Li-Ching came to learn about the project

2017 January 4

  • Discussed three-year paper resubmission
  • Next thing is 3 year outcomes stratified by risk

Merged in the capsure provider data, now that Ali has answered questions. For radiation patients, at least in Dan's cohort, there were 42 patients, and about 40 of them had a provider id. 7 of them had a secondary provider. In Svet's surgery cohort, there were 109 capsure patients, and 94 had primary npi.

2016 November 28

  • Look at Svetlana's project and check whether NS was significant at all in the full model. Done
  • Appropriateness of use of AS. Dan is thinking about how to quantify life expectancy.

2016 November 21

  • Send Max and Mark updated table with fixed margins.

  • Dan Lee
  • Refit race model with 3 levels and get comparisons between different groups.

2016 November 7

  • A lot of what we've been studying is really pertinent to cancer survivorship.
  • Aim 1 year 3 paper has been in review for 40 days, and when it comes back, if they have revisions or questions, we will drop everything and take care of them.
Ask about providers with inconsistent specialties. "RO" in urologist file for patients that did have surgery and for patients who didn't have XRT, One urologist in the rad onc file. This was in atlanta.

  • Dan Lee's project: The areas we found as different in white/black were dose and unnecessary pelvic radiation. Focus on racial differences. Black men may tend to go to providers that tend to provide less guideline-compliant care. Could show that the black men tend to go to the "one-star" providers.

  • AUA/SUO: 3 year-aim 1, Dan (1 or 2), Svetlana, Max.

  • Just need to send Svetlana some pictures. Make sure the anova plot has the right variables.
Focus on what Svetlana and Dan need for AUA abstract, and then work on their papers.

  • Appropriateness criteriaof use of AS

2016 October 31

2016 October 24

  • Go ahead and link in the provider data.

2016 October 10

  • Discussed the amount of 5-year ceasar surveys that we've received.
  • Get the provider ids linked up with the data and get distribution of volume (number of records per provider).

2016 September 26

  • Do an analysis comparing EBRT + ADT to NS and NNS, and get comparisons at 1 and 3 years.

2016 August 15

  • Aim 1 year 3 is the priority
  • Dave's last minute changes to Mark's Aim 2: Risk abstract: Now that we've shown that the diff in tx effects is smaller among high risk patients, Dave wants to see if the difference is explained by the difference in the actual procedure that different risk groups get. This is because for most high risk surgery patients, they must (and do) have non nerve sparing surgery.
  • To answer this, we think we will need to have 5 treatment groups.
  • Check the output that Dan Lee made and why it's different than what I got...(Or between what I got and Sharon got?).

2016 August 8

  • Eden wants me to copy her/keep her posted on anything with Cornell.
  • Dan (the research fellow) is working on Karen's radiation paper that Sharon started a couple of years ago.
  • Dan and Dave have a meeting with Shelly. He thinks we should have done the analysis completely differently, and they got completely different results. They used instrumental variable analysis. In this, you conceptualize one patient feature as the 'driver' of who got which tx.
  • Consideration of moderators: risk, bl, etc.
    • We also need this for high and low baseline. **
    • Could give just estimates of the model covariates for these terms. (They are additive )
    • The 'importance' plots.
  • Figures for aim 1 year 3: consort, main event with 3 treatments trajectories and for 5 domains using unadjusted curves, the 5-tx group curves.
  • MAKE from BOTH unadjusted and adjusted models

2016 July 25

  • Discussed the first batch of the 5-year surveys. I will need to examine these.
  • Svetlana's new project is about function after prostatectomy
  • MMS medical marketing survey is

2016 June 27

  • Short tern paper goals: radiation patterns of care and quality, 3 year outcomes stratified fy D'amico risk, and anxiety as an outcome.
  • Anxiety: Questions under letter N. They are planning on using this as an outcome in the future.

  • Aim 1, 3 year data: Dan B is thinking about what to displace and how to display it.
  • Add new model/ add 5th group to the 4-group model
    • Output a trajectory figure from this model
  • Dan B. wants the group of "baseline" figures to be the main figures in the paper. These are the ones I made that use baseline function as an outcome along with the other assessments (6m, 12m, and 36m), and I included more knots than in the other mods (at 0.25, 0.50, 0.75, 1.25, 3.00).
    • Figure out why I didn't include AS patients in these curves. _I didn't use only a subset of the patients. In the model fit, for the tx variable I used a new variable that was missing if AS. I think this was because it could not estimate the shape of the curve between 0 and 6 months. However, it appears that the times are already jittered for the AS group, so can I run it including the AS group? _ Yes, I got it to run.
    • Also figure out what the interpretation of this model is, as the baseline function is both controlled for and included as an outcome.
    • On the figure, why do the function values at baseline line up perfectly between the tx groups? This is because I did control for baseline. I also included the time 0 values in the outcome.
    • I used 5 knots for time in this model, and they are specifically placed. However, I also used 5 knots for time in the main models. What are those knot locations?

2016 June 20

  • DB wants to have the primary analysis to be the 3-group comparison/model
  • Tasks are to make (program and run?) tables (1-5) for manuscript according to the table shells that DB selected
  • Five-group analysis? To replace the 4-group analysis
  • Survival data and table
  • Consort diagram

2016 June 6

  • Cornell is sending us a test batch of 5-year surveys. I will need to go through carefully and check that the entry is correct and double check the database set up.
  • Mentioned that we still have another paper in the pipeline which is ceasar aim 2(?) on 3-year data, looking at variation in treatment effect by risk.

2016 May 23

  • Aim 1 Yr 1 paper got rejected from JAMA Oncology. They are going to submit as rapid review.

2016 April 28

  • Per 2016-04-13 emails from DP and DB, we will consider a patient in EBRT + ADT if they started ADT within 6 months of the conclusion of EBRT. "If there is no ADT either before, during or within 6 months after EBRT, I would classify them as EBRT only." "We probably didn't worry much about the timing in the one year data because salvage wouldn't have been much of an issue. Now it is more important. If adt was started more than 6 months after ebrt it should be salvage."

2016 April 11

  • For timing of ADT and EBRT, if it's more than 6 months after rad start. See the documentation from Sharon's tenure. I looked through the code and several doc files. The doc files didn't have anything about timing of hormones and rads, except that if they had both, they should be in the rads group. "anyhormonetx" was FIRST just literally defined as yes if any of the sources said yes. Sometime later, I recoded it to be yes if the 31-level tx variable included hormone tx. So whenever we used the anyhormonetx variable, we were not taking timing into account. Also note that anyHormoneTherapy.Derived only includes data through the 12 month survey, and befor the 3 year survey, because it is based on thr original treatment assignment code.
  • Talk to Mark offline about this.

2016 April 4

  • Going over Dan's manuscript spreadsheet.
  • Discussed Dan's ideas for aim 1 year 3.
  • Table 3
  • Discussed whether to only include the ebrt vs. rp comparison in the body and tables of the text. Then we would include table output for comparisons with AS in supplemental material.

2016 March 28

  • Dan B is interested in doing risk-stratified (and adjusted) analysis. They aren't sure whether that would be included in this or another paper.

2016 March 7

  • Question 65 on the 3-year survey asked about any other things the patient wanted to share. The capture data includes these responses. Lots of people wrote something here. Curiously, I don't see this question in the data we got from Cornell. This is because Tara put these in to a redcap database separately.
  • Adjusting for the use of hormome tx is also tricky. I prefer handling this using sensitivity analyses. Would it still make sense to control for hormone tx?
  • For now, just use the 3 tx groups without breaking EBRT into ADT yes/no.
  • Will still have aggregate

2016 February 8

  • Talked about definition of tx group
  • DP wants it to be ITT. People will remain in their initial tx bins for the 3-year, aim 1 analysis.

-- JoAnnAlvarez - 04 Jan 2016
Topic attachments
I Attachment Action Size Date Who Comment
radHormTimes.pdfpdf radHormTimes.pdf manage 4.4 K 06 Apr 2016 - 16:13 JoAnnAlvarez Histogram of times from hormones to radiation
Topic revision: r43 - 01 Feb 2017, JoAnnAlvarez
 

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