Clinical Research Oriented Workshop (CROW) Meeting: May 27, 2021

 

Present:   Levi Bonnell, Justine Dee, Nancy Gell, Emily Houston, Ben Littenberg, Jen Oshita, Liliane Savard, Adam Sprouse-Blum, Connie van Eeghen (9)

1.                   Warm Up: Masks are starting to come off!

2.                   Justine Dee’s RCT Introduction and Methods: Review the latest draft of intro and methods, as well as initial results. I am looking at ways to clearly communicate the findings. I have a lot of outcome measures. I am hoping for feedback on the tables formatting and style and any other suggestions you might have.

a.       How best to share univariate and multivariable regressions, both unadjusted and adjusted models: Levi provided an example; Ben provided an alternate table of presenting Justine’s results to include coefficients, confidence intervals, and p values.  Discussion: why include the unadjusted results?

                                                   i.      Unadjusted coef: 1.48 with a CE from -0.1 to 3.9,  p of 0.22

1.       For every one unit increase in DSEN,  Physical Function increased of about 1.5

                                                 ii.      Adjusted coef: 2.16, CI -.03 to 4.6, p of 0.08 (so far)

1.       The DSEN group came out 2.16 higher than the comparison group

a.       Years of chronic Pain CI -0.09: older people have worse physical function while controlling for sex, age, and which group they were assigned to (the association between the covariate and the outcome)

                                               iii.      This model does not explain relationships, and it shows effect on DSENS group

                                               iv.      For the secondary outcome, Pain Interference, use the same rationale for use of covariates: the 10% rule.  Report on all covariates used; reviewers often insist on seeing everything.

1.       Include telehealth as a logical potential confounder

2.       Consider a sensitivity analysis for telehealth, which was 10% of the group

3.       Be careful about too many covariates: they may be interdependent and if there’s too much to analyze, it becomes incomprehensible. Go with the five or six strongest, or at least test them out.

                                                 v.      Proposed language: “All models adjusted for potential confounders that influenced the main outcome… by …”

                                               vi.      Randomized, groups matched, we tested for confounders, no effect, our unadjusted effects are about right (or adjusted effects)

                                              vii.      Adjust for covid burden: all in VT/Chittenden County, during pandemic, so no adjustment needed, but may need to explain this

b.       Discussion section: there was a reduction in a secondary outcome; everyone got better in both function and pain (both groups).  Why did everyone get better?  Placebo, common elements in both treatment arms, engagement with therapist, a self-healing condition (not likely), regression to mean – they enrolled on a bad day/week/month of a fluctuating condition.

                                                   i.      Protection against regression to the mean: enroll, wait six months (depending on how long it takes for poor status to regress), retake the baseline, and then run the study.  

c.       Reporting on race/nationality: create a category of Asian or Other, depending on the population included, rather than listing out nationalities (for which the data are incomplete) or including these peoples under “Non-Hispanic/white). Be aware of how the reporting of data can be used to counter discrimination, rather than reinforce it.

d.       Recruitment was higher than planned; is this OK to recruit 107 rather than 100?

                                                   i.      Powered for 84 analyzable records.  Planned for 100 to account for drop offs. 

                                                 ii.      Limits are determined by IRB protocol – get an amendment if enrolling more than 100

                                               iii.      Higher samples provide power for smaller effect size – but does that effect size matter?  If not, then it’s still an intervention that doesn’t matter.

3.                   Next week:  TBA