Clinical and Health Research Clinic

Notes (2022)

2022 December 1

Jonah Fox (Bassel W. Abou-Khalil), Neurology - Epilepsy

Using data from the Human Epilepsy Project (http://www.humanepilepsyproject.org/) we will attempt to compare the dose response relationship between levetiracetam and lamotrigine for the treatment of focal epilepsy. Our hypothesis is that lamotrigine does have a strong dose-response relationship and that levetiracetam dose not.

- Prospectively collected epilepsy data: demographic, MRI imaging, seizures, medications, etc.

- Lamotrigine has been shown to have a pretty clear dose response relationship (full dose cannot be started right away, risk for rash; 100mg twice/day)

- Levetiracetam trial results have been inconsistent

- Intervals on medication could differ

- For analysis, patient should be on mono-therapy

- Inter-patient comparisons?

- Seeking one-month period of stability for a patient

- Frequency of seizures is outcome

- Time interval may vary widely across subjects

- Ordinary CI formulas are invalid here

- Nonlinear function of dose relating to frequency of seizures can be assessed using ordinal proportional odds model

- Sample size of ~ 50 in each group; 100 total

- Inclusion/Exclusion criteria

- Study done exclusively among newly diagnosed focal patients

- Measurement of pre-treatment seizure should be available (important!)

Haseeb Tariq (Wesley Thayer), Plastic Surgery

To determine if PEG can successfully be applied to spinal cord injury in ex-vivo samples. The second aspect of this aim will provide ancillary data to confirm that our developed DTI assessment parameters are effective in visualizing structural repair in the spinal cord. Mentor confirmed.

- Sample size question to detect particular level of improvement

- Spinal cord sections come from fresh cadavers, not fixed

- DTI: diffusion tensor imaging; motion of water in an area to study nerves

- One segment per cadaver? Been using pig spines previously

- 3 or 4 cadavers => ~12 specimens

- Population for cadavers is heterogenous

- Problems with non-representative cadavers

- Previous trial demonstrated 50% improvement with 15% SD deviation

- Collaborators' calculations: 10 subjects in each group should allow for power of 0.95

- Applicable in more studies: estimate something; calculate precision of some key thing you want to estimate

- Property of spinal cord under given compound; calculate property under different compound

- Comparative: difference in characteristics after treating with compound A - treating with compound B

- Calculate margin of error for estimating that mean difference (if small, you've nailed something down to a good precision)

http://hbiostat.org/bbr/htest.html#sample-size-for-a-given-precision

2022 November 17

Jenine Stone (Shelagh Mulvaney, Marianne LaNoue), School of Nursing

We will be assessing insulin habit automaticity at mealtimes using ecological momentary assessment over a 7 day period. We will be investigating the relationship between the variability of this measure with an outcome measure of mealtime insulin adherence. Mentor confirmed.

Insulin adherence in people with Type I diabetes with specialized pens

Hypothesis: for people who say "automatic", how does this relate to adherence?

"Automatic" definition: lack of cognitive decision, self-reported

- Is self-reporting of automaticity reliable?

Data collected via app "MyDay"

Question: "taking insulin before meals is something..."

- Challenges in "observer study" format

Scale validated in physical activity contexts, ...

Analysis of "completers" will lead to serious bias

- Need "analyze all available visits" approach with longitudinal analysis that works under MAR assumption

Need to strive for representative patients (representative before the fact; don't do any "after the fact" exclusions)

Possible that you did not measure the thing that predicts missingness

Less biased approach if you do not do any after the fact exclusions

Prospectively defined cohort such that you can say who results apply to

- Cannot exclude non-completers. Use all available data

Outcome is binary (were they adherent or not?) => random effects binary logistic model

Important to not remove permanent drop-outs from sample

Ordinal longitudinal model is something to think about

Analyze all available data; don't assume normality, use non-parameteric tests

2022 November 10

Marjorie Butler (Yaa Kumah-Crystal), School of Medicine

Previous clinic session September 29, 2022.

We are planning to do a study to evaluate efficacy & efficiency of email-based interventions to increase voice recognition adoption by VUMC physicians. This study is relevant because of the growing need to educate providers about how to leverage EHR efficiency tools such as use voice recognition technology (VRT) in a way that can best accommodate provider time constraints and operational resource limitations and identifying ways to promote incorporation of new workflows in the EHR& other evolving technologies within the EHR without compromising patient safety.

Our proposed study design is a prospective observational study, with 3 intervention groups. Our hypothesis is that emails with in-line GIF instructions will result in higher VRT adoption rates that a static email, narrative videos, or email outreach for hands on intervention. We believe animated gifs will be most beneficial because animated gifts provide easy to follow visualization in line with email that does not require users click and load a clip that requires the additional focus for audio narration. 3 Groups: 1. email about VRT onboarding with static instructions & pictographs 2. email about VRT onboarding with animated gif instructions 3. email about VRT onboarding with narrated instructions.

We plan to explore: 1. which modality is associated with the greatest VRT adoption rate 2. how the interventions impact EHR and documentation efficiency (ie. Time to close encounters) 3. comparison of individuals who recieve intervention group and either a) dont engage b) set up VRT software but don't become an adopter vs c) become fully adopted We hope to randomize providers into intervention groups.

We anticipate a population of ~1500 providers for randomization. Relevant potential variables to control include: - MDM access Other variables to evaluate via ordinal logistic regression - provider role (house staff vs faculty role) -department -patient volume Analysis plan: We hope to compare outcomes in VRT adoption between intervention groups using one-way ANOVA OR ordinal logistic regression (seeking advice around which of these is best). We plan to use paired t-testing to evaluate pre and post intervention data for provider documentation. We are seeking guidance about the use of ANOVA vs ordinal logistic regression for statistical analysis with consideration that ordinal logistic regression has greater ability to preserve degrees of freedom. Mentor confirmed.

Questions for today: project design

Initial primary outcome: adoption rate

- Change from ordinal to continuous outcome? Make the most out of your sample size and increase power. Thresholding can't handle close calls

- Do we want to differentiate between people who use resource "a lot" from people who use it "a bit"? (make secondary analysis instead)

Pre-post comparison considered secondary outcome (not as interesting)

Emphasize parallel group design; compare frequencies

Randomization important! Want it to be reproducible (set random number seed) and where assignments are difficult to mess up

- Blocked randomization?

Jessica Plager (Bill Cutrer), Allergy and Immunology

My project asks the question whether a team-based learning exercise for medical students during the Internal Medicine clerkship increase their proficiency in taking an accurate and thorough drug allergy history, and improves their knowledge of penicillin allergy? Mentor confirmed.

Medical education project on penicillin allergy (population: 10%), worse antibiotics and worse health outcomes

Design: students arrive, take pre-test (quantitative and maybe a little qualitiative), session, post-test, maybe follow-up post-test six weeks later

Not possible to randomize, same group pre/post

40-60 participants, comparison against own pre/post scores (make sure responses are linked by individual)

- Likert style questions

- Slider button (careful with default so know person answered)

Other confounders when it comes to pre/post design that could explain results other than the thing you're measuring

How many questions to result in a meaningful score? Depends on correlations between question responses, difficult to answer without pilot data

The key is getting people to answer the survey the right way (easy to read/understand and well-tested)

- Don't let survey get too long to combat participant fatigue

- Put most important questions at beginning

2022 October 20

Andrew Wooldridge, Medicine/DGIM

Music Therapy Pilot in Palliative Care Unit: Writing up the experience from our nonrandomized pilot and need help with compiling/visualization of data extraction from clinical patient records (music therapy consult notes).

Relatively small sample size

Interested in data visualization of ~450 binary variables

How to measure benefit from music therapy?

Descriptive visualization: dot chart, ordered by frequency of condition

Two-dimensional: clustering diagram (dendogram) -> determine which conditions are frequently co-exhibited

Four data categories: recreative, compositional, receptive, ...

Could define a severity hierarchy for reason for referral (get rank-order)

Ashika Kuchhangi (Clare Melanie Schuele), Arts and Sciences

Assessment of language and literacy skills in children with persistent speech sound disorders. Recommended by VICTR review of application for funding support. We intend to classify children based on typical cut offs for disability and test against means of norm-referenced assessments. Questions have been raised by reviewers that these analyses are not appropriate. Mentor confirmed.

After several years, children with persistent speech sound disorders normalize, indistinguishable from peers

Evaluate speech, language, & literacy of children aged 9-17 (clinically referred, not population-based)

- Disability (Y/N, determined by arbitrary cut-off of 1 SD < mean)

- EEG task, speech perception of children compared to control group

- Raw EEG data = very high-dimensional

What interventions can target speech/language issues of these children? Understand prevalence in general population

Problems with comparisons to normative mean for cut-off (sometimes norms change) -> norms redone ~ 7 years, national norm comparison = "most fair"; reading assessments done for 3rd and 4th grade

Genetic factors to consider

Remaining issue: categorization of variables (thresholding for what is acceptable eliminates close calls => perfect success/perfect failure)

Correlation analysis

Sample size ~ 30

2022 October 13

NO SHOW: Srishti Nayak

2022 October 6

Kaleb Wolfe (Romney Humphries, Milner Staub), Infectious Diseases

To assess the prevalence of adopting up-to-date antimicrobial breakpoints amongst micro labs across the state followed by a targeted intervention in a subset of these labs to improve adoption. Current outcome is rates of adoption as well as differences in resistance data at each site after the intervention is completed. Goal of the clinic is to ensure that the survey we have developed will give us the data needed to answer our outcomes. Mentor confirmed.

Concern: antimicrobial resistance defined by breakpoints based on Monte Carlo sims (few treatment options)

Goals of project:

- improve rates of adoption, particularly in TN

- determine what proportion of labs have adopted different types of breakpoints

- assess barriers to adoption

Want to look at labs currently, design intervention, then re-evaluate

Example of breakpoints:

Create lawn of bacteria on petri dish, drop disk with antibiotic on it

Look at multiple isolates and determine at what point is this sensitive?

Survey design:

contact data

descriptive data, sentinel sites in TN

laboratory data (role of the lab, staffing info, lab accreditation)

questions about adoption of breakpoints, barriers to adoption

Q: Who will you contact? A: sentinel sites

Q: Response rate? A: ~ 90%, labs required to make updates

- introduce bias if labs lagging behind are more likely to not respond

Comparing sites' date + time of response

66-68 total sites, compare adoption by site location?

Findings would be generalizable to the state of TN, but not USA

Labs have permanent identifiers; will help with tracking

2022 September 29

Marjorie Butler (Yaa Kumah-Crystal), School of Medicine

We are planning to do a study to evaluate efficacy & efficiency of email-based interventions to increase voice recognition adoption by VUMC physicians.

This study is relevant because of the growing need to educate providers about how to leverage EHR efficiency tools such as use voice recognition technology (VRT) in a way that can best accommodate provider time constraints and operational resource limitations and identifying ways to promote incorporation of new workflows in the EHR& other evolving technologies within the EHR without compromising patient safety.

Our proposed study design is a prospective observational study, with 4 intervention groups and 1 control group. Our hypothesis is that emails with in-line GIF instructions will result in higher VRT adoption rates that a static email, narrative videos, or email outreach for hands on intervention. We believe animated gifs will be most beneficial because of the ability to provide a scalable form of social interaction.

5 Groups: 1) no email contact about VRT onboarding (control) 2) email about VRT onboarding with static instructions 3) email about VRT onboarding with animated gif instructions 4) email about VRT onboarding with narrated instructions 5) email about VRT onboarding with static instructions and opportunity to sign up for 1:1 instructions.

We plan to explore: 1) which modality is associated with the greatest adoption of VRT use 2) percent of users able to complete self-onboarding with given instructions vs requesting additional support 3) attrition rate for each intervention modality 4) how the interventions impact EHR and documentation efficiency (ie. Time to close encounters).

We hope to randomize providers into intervention groups. We anticipate a population of ~1500 providers for randomization. Relevant potential variables for randomization include: -provider role (house staff vs faculty role) -department -Vanderbilt app store already setup on their phone -inpatient vs outpatient -patient volume.

Analysis plan: We hope to compare outcomes in VRT adoption between intervention groups using one-way ANOVA. We hope to use two-way ANOVA to compare retention/attrition between and within groups during certain stages of the adoption process. We plan to use paired t-testing to evaluate pre and post intervention data for provider documentation. We are seeking guidance about the study design, power, and statistical analysis plan to ensure we have a robust approach to answer the questions we propose to study. Mentor confirmed.

2022 September 22

Lindsay Panah (Soha Patel), Cardiology

We are doing a survey based project on contraceptive counseling in patients with cardiovascular disease. We had a question about the number of responses we should target to have a meaningful study. Mentor confirmed.
Sample size question. Currently have ~100 participants. There are significant concerns with the methods, but if the team proceeds with analysis, would reccomend CIs only.
This is a patient survey regarding contraception counseling. Outcome is Y/N, was counseling provided. Multiple recruitment methods used, Patients self select into the survey. Concern is representativeness of this population (generalizability). We don't know the population of people who could have taken the survey. Non response bias is the concern.

Jennifer Connell (Tanya Marvi), GI- Hepatology

We are doing a retrospective study. Aims below: Aim 1: Describe the current acute management of GI bleeding in patients with cirrhosis. (a): We hypothesize that the management of acute upper GI bleeding (UGIB) in patients with cirrhosis will exhibit variability in patterns of laboratory test ordering, medication administration, blood product administration, and performance of procedures. To test this hypothesis, we will identify the target population using the EHR order of octreotide infusion, an order ubiquitous in and unique to the target population, and subsequently complete a retrospective review of their management in their acute hospitalization. (b) We secondarily hypothesize that specific management strategies of GI bleed in patients with cirrhosis is associated with the unit to which the patient was admitted, specialty of the treating provider, and between providers of the same specialty. To test this secondary hypothesis, we will compare patterns of ordering and prescribing between these groups. Aim 2: Describe clinical outcomes associated with different strategies in management of cirrhotic patients presenting with upper GI bleed. We hypothesize that certain management strategies of acute UGIB in patients with cirrhosis are associated with better clinical outcomes. Management strategies under investigation include, but are not limited to, intensive care unit (ICU) admission, early endoscopy, placement of thromboelastography (TEG) order, decreased transfusion volume (when controlled for weighted average of INR/PTT). To test this hypothesis, we will compare clinical outcomes including but not limited to 5 day treatment failure, 6 week mortality, time to mortality, transfusion requirement and subsequent development of transfusion related complications, subsequent cirrhotic decompensations, the need for salvage therapy, and days of ICU/hospital stay. VICTR Voucher # VR60351.
Requesting assistance with statistical analysis planning. Mentor confirmed.
This will provide background for a planned Pilot study that is the planning stages. Projected N is ~2,000
Keep aim 1 as mostly descriptive.
Aim 2 not as appropiate for a pilot study. Comments in application are mstly about aim 2.
This is a VICTR voucher that has been in the line for a while.

2022 September 8

Yash Vaishnav (Louise Mawn), Ophthalmology

My goal is to study the response to treatment of thyroid eye disease (TED) with teprotumumab, a novel biologic that represents the newest horizon of therapy for TED. While teprotumumab has been demonstrated to be an effective treatment for TED, its use for dysthyroid optic neuropathy has not yet been well characterized. Our goal is to not only examine the clinical response to teprotumumab on imaging, but to also explore the response of dysthyroid optic neuropathy patients to teprotumumab. The research that this fellowship would fund would utilize software to improve care for TED patients by providing tangible, imaging-based evidence that teprotumumab works to improve extraocular motility and decrease the risk of vision loss in TED patients. Mentor confirmed.

Condition is part of Graves disease.This medication blocks one of the mechanisms of the disease. Goal of treatment ultimately is to protect optic nerve. Measurements include measures of the optic structure. The measures should be precise, we need to understand the variation in the measure. Technical replicates.

Have ~50 patient treated and untreated. Two images per patient, on average. Could do a longitudinal trajectory analysis, if we know an estimated time zero. Or cross correlation. Use all available scans/data.

Optho does have a biostat collaboration. A VICTR voucher is an option. Research studios are also an option for big design questions. This may fit into a voucher framework, but would suggest a return visit before this step.

2022 September 1

Rei Ukita, Cardiac Surgery

I am drafting a research proposal that aims to establish clinical relevance of an animal disease model with actual data from human patients. To prove this, I want to collect blood/tissue samples from patients with the actual disease and compare to data from animals. Gene expression analysis to compare what pathways are enriched in humans vs animals

For animal study, these animal subjects serve as their own control because I can sample data over time, starting from their healthy state until they achieve disease state. I was wondering what would be an appropriate control for humans. I wanted to know the specific methods researchers use to create a control group, how characteristics are matched, etc.

Human patients are s/p Fontan who develop liver disease. Mechanism is thought to be changes in blood flow. We observe a pathway in animals (sheep models), question is will we see this in humans.

Question is how similar are the human/animal model? This may be problematic due to sample size. A blunt approach could be to predict the pathway using the type of sample (human/animal). But sample size is prohibitive. Sheep are expensive, budget would limit to about 20 sheep, which would allow the study of one attribute (roughly speaking). A challenge to this study is the number of measures, and measures over time. Bootstrapping is a possible method. Pre-establishing ranks.

Krista Suojanen, Internal Medicine

I will be using QuizTime (mobile app) to provide spaced education to residents and faculty on limiting daily labs. I will then see if the intervention led to any decrease in daily lab ordering. The project is in its planning stages and has not yet been implemented. Before I begin, I would like to think about the best way to collect and analyze the data that I will be gathering.

Intervention could improve knowledge and may change behavior. Intervention would be targeted to different services.

Goal would be publication, demonstrate that this intervention can change behaviors. This would need some type of randomization most likely. At least 20 teams would be needed. Teams change every two weeks.Possible that there will be some overlap, but not a lot.

Learning health system group may be of interest. https://www.vumc.org/implementation/learning-health-system-scholars-program. VICTR studios. https://victr.vumc.org/overview-of-resources/

2022 August 25

Megan Wright (Jessica Gillon), Pharmacy

We are doing an interventional, retrospective chart review related to the impact of requiring durations of therapy for antibiotic orders at time of order entry within VCH. Mentor confirmed.

Duration of therapy is measured in days of therapy. Ties in the data are a concern since data are likely in a few buckets (7 days, 10 days, 14 days). Could display as a histogram with descriptive statistics.Will stratify by primary service. LOS is a concern, ICU and hem/onc will have LOS amenable to the intervention. Other services may not. Time frame full encompasses COVID. May 2020 was start date of intervention. Acuity was high during COVID, though census was low. So antibiotic days were high. Could adjust for calendar time as a non linear adjustment.

Question: Did length of therapy decrease after implementing this requirement?

Could apply for VICTR voucher--adjusting for the COVID impact is complicated. This study would fit into a voucher framework.

Sriya Nemani (Wesley Thayer), Plastic Surgery

We will be investigating the improvement in sensation in patients that have been treated with polyethylene-glycol(PEG) assisted nerve coaptation in phalloplasties. We will have 3 cohorts with an equal number of patients: control, PEG-only, and PEG with an additional 10 minutes of electrical nerve stimulation. We will have 5 post-op follow up visits with the patients to test for improved sensation. Mentor confirmed.

Multiple tests of sensory function will be administered during the follow up period. QOL tests will also be administered. All measured should be collected in the finest form (not categorized). Outcome scale is ordinal. Repeated measures are present.

Sample size question. Limited prior data. Hard to answer without knowing how much change we will see at each visit. For a pilot study we plan for precision rather than power. Idea is to identify precision and effect size. And what is practical. ~24 patients a year have this surgery. With this few of patients, precision is key. Good training, standard protocols for testing, precise recording of measures.

2022 July 14

Amy Guidera (Deonni Stolldorf), VAQS Fellowship

We conducted retrospective chart reviews on all Veterans who tested positive for at least one STI from 12/06/2020-12/06/2021. Outcomes sexual health history taking and STI testing. 1. Understanding health equity within STI testing and sexual health history taking 2. Exposures age, gender, race, STI location 3. Understanding the relationship between the type of provider (MD/NP/DO) and sexual health history taken 4. Examining if Veterans who had symptomatic (reactionary) testing versus preventative testing had any correlation with the amount of sexual health history taken. Mentor confirmed.

Clinic Notes:
  • Retrospective chart reviews (n = 285) on all veterans who tested positive for at least one sexually transmitted infection. Have been doing Chi-squared tests and would like to know if they should use any other test.
  • Recommendations:
    • Need to know the degrees of freedom for Chi-squared test. Since White and Black participants are the vast majority, the p-value from Chi-squared test is mainly picking up the difference between these two racial groups.
    • It's better to use the order of the outcome variable instead of using it as a categorical variable.
    • Looking at the rank correlation between raw age and category of history is a more powerful assessment of their relationship. This will produce a correlation coefficient and a p-value for the correlation coefficient.
    • For descriptive statistics, could look at age distribution by racial group, sex difference be racial group, and history-taking by racial group. Could also do an ordinal regression to look at adjusted effects. May want to pull all the non-White participants into one category for the regression.
    • When writing up conclusions, note that patients are only included if they have at least one positive STI test.

Yaa Kumah-Crystal, Adonaye Woldegebriel, Mihir Kulkarni, DBMI

We sent a survey to 2000 users asking them to select if they wanted to make a particular change in Epic. They could select one of 3 categories: Yes, no, unsure We would like to know if the time they spend in Epic is correlated with whether they selected yes, no, or unsure. Based on the flowsheet below, I believe my predictor (time) is quantitative and that my outcome variable (yes/no/unsure) is categorical... so would that mean the best test us use would use logistic regression?

Clinic Notes:
  • Would like to know the best analysis tool for an InBasket survey response. Hypothesis is that the busiest people are not responding to the survey. Would like to investigate the amount of messages and the response from survey. Total response is ~2600.
  • Recommendations:
    • Could do two separate regressions to answer two questions: 1. What factors contribute to whether or not people respond to the survey, and 2. Among the people who responded to the survey, what factors affect which response people choose (yes/unsure/no). For the first regression, could use a binary logistic regression with whether or not a person responded to the survey as the outcome, and age, sex, volume (maybe others) as independent variables. For this regression, could use the lrm function from R package rms (helpful link: hbiostat.org/rms). For the second regression, could use an ordinal logistic regression or multinomial logistic regression. An ordinal logistic regression may be easier to use in this case and need to order the outcome as yes/unsure/no. Logistic regression will give a Chi-squared statistic, answering the question "is the model significant overall". And if we fit a model with age, sex, and volume, we could answer the question "holding age and sex, how would volume affect the response".
    • Could consider adding work schedule (day/night shift) and time of day of responding as independent variables.
    • Some independent variables could be U-shaped and this requires a non-linear term. Could consider a VICTR voucher.
    • Possible VICTR voucher. Application website (https://starbrite.app.vumc.org/) and research proposal template (https://starbrite.app.vumc.org/funding/templatesforms/).

2022 June 9

Steven Allon, Medicine/General Internal Medicine

This is a replication of a previous study that had biostats clinic helped us with (data analysis). It is a multicenter study examining the effects of a gamified format for an internal medicine residency journal club curriculum on resident engagement. Residents are surveyed pre/post using a survey measure that captures multiple dimensions of engagement with Likert-type responses. Questions to address: Review the survey instrument to determine its psychometric properties, or what additional information is needed to do so. The survey was developed by our group, and we have a dataset of pre/post responses from the initial study. We will collect data during this replication.

Clinic Notes:
  • Pre/post design. Need to have everyone (100%) get pre & post surveys, no room for error. For next study, looking at multiple formats of journal club. Wanting to see engagement levels of each format.
  • Recommendations:
    • Suggest having a negative control (Questions that are completely unrelated to what you're studying - something that would not change).
    • Not a lot of people go for extreme values on likert scale, might be better to have slider (0-100) instead of "much". To combine data points from asking the survey 2 different ways (likert/slider) collect data points both ways to calibrate slider. Matrix format.
    • Psychometric properties are difficult - need test/retest to determine. Could take small random sample to ask for retest with almost 100% agree. This provides more confidence in survey, something to refer to when submitting to journal.
    • Consider randomization for types of format. Try to get response as soon as possible (before leaving room if possible).

2022 May 19

Patrick Reardon, Orthopedic Surgery, Sports

I am doing a project on the innervation of the hip labrum and knee meniscus. We are getting 10 cadavers and trying to figure out how many samples of tissue we need to take to be appropriately powered with our comparison between the two. I was hoping you would be able to assist us with this. The unfortunate part is that this is a fairly novel study and there isn't much data out there to give any baseline numbers to help with the power analysis. Measure is quant measure of density. Could take replicate samples, in effort to increase precision. Would suggest do all samples twice, if possible. The more the better. May also consider not doing stat testing. Do estimation study instead, with CIs. Not a setting for power, setting for precision. No power analysis needed.

Suggest randomizing order of cadaver analysis and blinding the people analyzing the results (reduce bias).

2022 May 5

Douglas Bryant (W. Evan Rivers), Physical Medicine and Rehabilitation

This systematic review will examine the effect of endoscopic facet joint denervation on pain and pain-related disability among adults with low back pain. It will also examine screening criteria for offering the procedure, reported complications, time to perform procedure, duration of effect, success rate (based on pain scores, disability scores, or global scales such as MacNab criteria), and patient satisfaction. Mentor confirmed.

VICTR voucher request. Kim and Frank lack experience with this type of analysis--may need to seek someone else in the VICTR group.

Systematic review and meta analysis. Team has met with librarian to help will article identification. Excel proposed to collect data, strongly suggest Redcap. Team can return to clinic at any time to discuss redcap design and what an be reliably coded.

Limitation of such a review can be considerable. In this case, we are probably looking at between 4 and 10 studies. Long term, development of a registry would be useful for this topic.

Discussion of modeling, random effects model, bayesian random effects may be the way to go.

2022 April 28

Emily Mason, Pathology/Division of Hematopathology

Small project looking at immunohistochemical stains in marginal zone lymphoma. Our VICTR protocol has already been reviewed by Frank Harrell, who recommended we attend a Thursday clinic.

Pilot study looking at stomach lymphoma, which can be difficult to differentiate. Stains are "scored", but this is not terribly exact as I understand it. But this is preferred over a binary score (pos/neg). These samples are from patients with known lymphoma, confirmed no lymphoma (gold standard). Suggestion: consider the degree to which lymphoma patients had a higher score (Wilcoxen test). Suggest 40 samples, may be too small, but reasonable. Voucher requested for staining only.

Courtney Tomblinson, Radiology

Statistical analysis of learner attitudes toward a curriculum they've participated in. Challenge is low response in baseline control group, non response rate is very high (95% would be preferred). We don't have the ability to examine those who did not respond. This is a problem in many surveys. In this case we don't have a random sample, this is conditional on response. This is a "pilot" study, although not what we would typically call a pilot study. Education methods are challenging. Sidebar: adaptive testing may be useful as we study education methods.

2022 April 21

Kristina Niehoff, Vanderbilt Home Care Services

There is an increased risk of venous thromboembolism (VTE) associated with SARS-CoV-2 infection. As a result, patients enrolled in Vanderbilt’s COVID to Home program (high intensity) receive anticoagulation. To qualify for the high intensity the patient may have: 1) a hypoxia requirement new or increased home oxygen; 2) required ICU level care while admitted; or 3) have significant comorbidities (particularly age 65 or older, severe obesity, underlying heart or lung disease, is within one year of organ transplantation or is severely immunocompromised). The preferred anticoagulant was chosen as apixaban since it can be taken orally and given its lower bleeding profile compared to other oral agents. However, literature does not exist to know if low dose apixaban is safe and effective for VTE prophylaxis. We seek to understand the safety outcomes of low dose apixaban 2.5mg twice daily for 7 days in a post-hospitalization population of individuals diagnosed with SARS-CoV-2 and understand the risk for VTE event post discharge.

Clinic Notes:
  • Vanderbilt home health (covid to home program). Patients who are deemed high risk were enrolled. NP would go out to home day after discharge, implemented home health services to determine future admission. Covid makes you clot easier, so patients were given blood thinners. 400 patients. No control group. Still collecting data until end of June. Patients monitored for 4 weeks post discharged. Hypotheses: was it tolerated? was is safe? did anyone have heart attack/stroke/clot while taking blood thinners? Want to look at frequencies/descriptives.
  • Recommendations

Andrew Yi, Quality Safety & Risk Prevention (QSRP)

I have a project on children overweight/obesity study and draft of results had been accomplished. I would like to ask a few questions on data analysis and review our results for comments.

Clinic Notes:
  • Finished data analysis, wrote results section and requesting feedback on design/interpretation. Cross sectional study of children - epidemiology of obesity and their influential factors. Aims - prevalence of obesity and identifying/quantifying associations with obesity. N=16,640.
  • Recommendations
    • Not valid to dichotomize BMI. Also, don't want to analysis BMI as outcome, best way is to analyze weight as outcome with height (and age, sex) as adjustment. Next best would be to analyze log-transformed BMI as outcome.
    • Anytime you have a criterion that is based on a continuous trusted outcome (ex. weight, height), then you need to use height and weight not that calculated variable (bmi). Analyzing continuous variables gives you more statistical power.
    • Analyze data as close to raw form as you can. Do not analyze raw data as dichotomized, you will lose too much information from the raw data. Analyzing raw data (continuous) allows you to to find mean, median, etc. as well as making statements about cut points later. Want to convert to dichotomous after analysis is finished. Can use the regression model to estimate the chance of being obese for a certain threshold.
    • For continuous variable, you can use ordinary/linear regression.

2022 April 7

Hannah Fish-Trotter (Jeffrey Dendy), Cardiology

We will be investigating both the safety and image quality of MRI in post-cardiac transplant patients who still have abandoned cardiac device leads in situ. We plan to conduct a controlled prospective MRI study of post-transplant patients with abandoned leads versus controls (post-transplant patients with no devices or leads). We have discussed with the cardiac transplant team, who estimate about 25-30% of their transplant patients having abandoned leads. We would like to discuss numbers needed for power and other study design questions prior to applying for a VICTR voucher. Mentor confirmed.

Hypothesis: MRI will not cause AEs or image interpretation. The team does 6-8 transplant studies per week, ~200+ controls.20-30% of transplants have abandoned leads.

Suggest not matching, but controlling for age, sex, time from transplant in a regression model. Propose goal as an estimation study. Estimate difference in adverse events, adjusted for confounders. Or could flip the equation, predict abandoned leads with number of adverse events. For ordinal outcome scale, suggest at least three categories that are well populated.

Re: Sample size. Consider margin of error. Minimum would be 96, with relatively similar distribution of abandoned leads.

This will fit in the framework of a voucher.

2022 March 31

Amanda Peltier, Neurology

We have collected data on neuropathy phenotype, nerve conduction study results, and cardiac evaluations. we would like to verify the correct analysis and ask regarding skewed data/ graphs for publication. There are 37 patients, with ~ 28 with nerve connection studies.

2022 March 17

Milner Staub, Infectious Diseases

Practical study comparing respiratory tract infection clinical decision algorithm implementations: AgileMD vs ExpressLane. Would like input on design/analysis and ability to evaluate differences in performance as regards primary outcome of reducing antibiotic prescriptions for upper respiratory tract infections and secondary outcomes of adoption and usability.

Agile MD available to all, but Express Lane an be restricted. Providers can use either or none of these similar decision tools. Agile MD may be memorized and used without accessing the tool. Need to build in decay possibility. Time trends will be important. These are not live until April. Providers cross all clinics.

One outcome is the decision to use tool, then was the antibiotic prescribed or not prescribed.

Can we capture patient level data, especially complicated cases? Can capture allergies, diagnoses. We must capture instrument use data wherever possible. How many patients, minutes per use. Provider and patient factors are important.

2022 March 03

Alex Foy (Stacy Killen), Pediatric Cardiology

Prior Notes:

Retrospective chart review to identify mitral valve or dimensional differences apparent by fetal echocardiography in those fetuses with postnatally confirmed coarctation of the aorta who undergo biventricular repair compared to those who ultimately require univentricular palliative surgery. VICTR Biostatistics voucher. Mentor confirmed.

10 Years of data, about 100 with coarc and fetal echo, see if measures during echo can predict bi vs uni repair. Patient may have between 1-5 fetal echos. Also have multiple echos before interventions. Actual repair decision is made during surgery. Measure of mitral valve (Z score) to predict outcome. Gold standard may be the measure before surgery. Repeated measures issues-coud think of these as updated baselines. This may fit that issue, rather that a true repeated measures. A landmark analysis. This would fit in the scope of a voucher.

Return after first attending the clinic on January 6 2022. Need to discuss the statistics section for the VICTR voucher application.

New Notes: Goals: Do differences in mitral valves predict surgical outcome? Outcome: single ventrical repair or bi-ventricular repair. All patients will proceed to surgery. We have multiple baselines, and are modeling type of surgery. About 100 surgeries, unsure of outcome breakdown. Rule of thumb, 15:1 rule, based on smallest outcome group. Binary logistic regression model can be framework. Need to identify co-variates, and how they will be measured. Perhaps include the shape of the relationship in the model. Could create two (or more) models. VICTR application should capture the spirit of the project, details will be fleshed out after award.

2022 February 17

Candace Grisham (Marjan Rafat), Biomedical Engineering

We are investigating immune cell markers and the predictive power of these markers on recurrence of laryngeal head and neck cancer after radiation therapy. VICTR Biostatistics voucher. Mentor confirmed.

WBC are the marker of interest. Possible + HPV as well. Prior work: Systemic inflammation on cancer return, in breast cancer that used Cox model.

Outcomes: Death, distal met.

Could use longitudinal analysis with a a scaled outcome (death, loss of remission, etc). Define "worst thing that happened in a given week". How to handle CBC (with diff) data? Timing is not the same, and treatments may change over time. Data structure (ideal) string of dates and events on each date (long format). Also, include treatment plans, biometric data, etc. Monitor drop outs. Have drop outs, but also people who enter "late state" after diagnosis elsewhere. Establishing full timeline for everyone is important.VUMC also has expected/observed predicted mortality,

Big picture: How does immune inflammation impact cancer return.

This project should fit into the bounds of a VICTR vouched, would reccomend applying for the voucher

Tomas Bermudez (Maria Hadjifrangiskou), Microbe-Host Interactions

I have measured the appearance of antibiotic resistant subpopulations within clinical urinary isolates banked by microVU. The individual isolates are paired with de-identified patient data that I would like to be able to use and see if there are any trends that could be observed in terms of particular aspects of the patient info correlating with increased counts of resistant subpopulations.Mentor confirmed.

Best method for this type of correlative study? Spearman's Rho is an option, Somer D. Somers' D = Wilcoxon-Mann-Whitney two-sample rank sum test comparing two groups

What is a more appropriate way to display this type of data, and a more appropriate way to display the distributions of subpopulation counts with my current set of screened isolates (ex. low, avg, and high producers of a subpopulation). Use rawest form of data, do not categorize. Histogram, 100 bin bar graph, rug plot, spike histogram

2022 February 10

Mryia Hubert (Kate Mittendorf, Gillian Hooker), Genetic Counseling

We report on the early phases of the Family History and Cancer Risk Study (FOREST) which was initiated to improve identification of high-risk patients and access to hereditary cancer services for the general adult patient population at Vanderbilt University Medical Center (VUMC). We are asking: what variables predict high engagement and whether engagement mediates the predictive effect of the variables on our outcome measures. Mentor confirmed.

Baseline Survey

Start health history tool

Finish health history tool

Engagement is how often one thinks about cancer (dichotomized). Suggest treating as ordinal.

Aim1: Suggest proportional odds model. Small cells sizes are an issue in these data, but ordinal will still be a better path.

Aim2: Binary

Could make initiation and completion an ordinal variable, and do a single model. Don't really have a solid measure of proportion completed.

For an intro to ordinal modeling with R go to the nonparametrics chapter of https://hbiostat.org/doc/bbr.pdf

2022 January 20

Christopher Kalmar (Galen Perdikis), Plastic Surgery

Frank Harrell requested to discuss the biostatistics during pre-review of VICTR resource request VR56032 titled ‘Viscoelastic Hemostatic Assays for Microsurgical Procedures’. VICTR voucher not including biostatistics. Mentor confirmed.

Using viscoelastic parameters for a measure of flap healing/potential low perfusion. Hypercoag is the cause of issue. Multiple measures over time, pre surgery, post surgery, and daily. Using four markers from TEG.

Advise against sensitivity/specificity. Trying to find inflection point is problematic. Risk model may be better. Flap failure is fairly rare (2-3%), and it is a yes/no, typically.

Goals for this study could be; estimate of failure. Would need about 15 failures per variable. Sample size is problematic here. Could reframe as proof of concept/proof of information. Can this study demonstrate value in using the TEG device--will the device measure correlate with flap failure. Look at correlations. Can also look at relationship with existing clinical variables, demonstate that TEG can contribute (redundancy). Suggest understanding precision of TEG device.

Dr. Perdikis could not attend.

2022 January 13

Garrett Booth, Pathology/Laboratory Medicine

We have created a database of US medical boards and their composition, including board member title (eg CEO) and gender. We have annual data related to individuals within 24 US medical boards. Our primary hypothesis is that women physicians are underrepresented in US medical boards. Question for biostats clinic? The gender composition for each medical board appears to be non-parametric (lacks a normal distribution) and I would like to compare the percent of women physicians in US medical boards between two separate years. Can I compare percentages using the Mann-Whitney test?

Are women represented at parity? Absolute parity. This is all boards combined.

Are women represented in the same proportion as included in the specialty? Relative parity by specialty. Board size has changed, some increased, some decreased.

Has there been an improvement since 2016? Defined as increase in women (usually).

We have all data (population), so no inference issue here

Suggest multi panel dot chart, aka Cleveland dot charts. Suggest Google. Also: https://uc-r.github.io/cleveland-dot-plots

Sort in descending order, proportion of females.

2022 January 06

Alex Foy (Stacy Killen), Pediatric Cardiology

Retrospective chart review to identify mitral valve or dimensional differences apparent by fetal echocardiography in those fetuses with postnatally confirmed coarctation of the aorta who undergo biventricular repair compared to those who ultimately require univentricular palliative surgery. VICTR Biostatistics voucher. Mentor confirmed.

10 Years of data, about 100 with coarc and fetal echo, see if measures during echo can predict bi vs uni repair. Patient may have between 1-5 fetal echos. Also have multiple echos before interventions. Actual repair decision is made during surgery. Measure of mitral valve (Z score) to predict outcome. Gold standard may be the measure before surgery. Repeated measures issues-coud think of these as updated baselines. This may fit that issue, rather that a true repeated measures. A landmark analysis. This would fit in the scope of a voucher.

Zachary Bressman (Heather Pua), Pathology, Microbiology, and Immunology

The project is a collaboration between the Heather Pua lab and Dr. Katherine Cahill and aims to examine changes in extracellular vesicle (EV) abundance in biofluids collected from patients with aspirin exacerbated respiratory disease (AERD) during different timepoints in controlled aspirin challenges that cause them inflammation. We hope to identify potential EV biomarkers in the context of AERD progression. Currently, we track the abundance of different EV populations in AERD patient biofluids with bead-based flow cytometry. As part of the project, we want to figure out how semi-quantitative bead flow cytometry is. To help answer this question, I’ve collected a series of cell-derived EV serial dilution bead flow cytometry measurements. I collected flow cytometry data for 1/2 diluted EVs, 1/4, 1/8, 1/16, and 1/32, and because the changes between subsequent dilutions were consistent and linear, I wanted to see how linear the changes in flow cytometry measurements were. It has become increasingly clear that there are upper and lower bounds outside which bead flow cytometry measurements is unreliable and changes between different dilution measurements is nonlinear. I would like the biostatistics clinic’s help with finding a statistical test to pinpoint the upper and lower limits of linearity for the data sets I’ve collected, so I can better filter out unreliable measurements. Mentor confirmed.

We can see on a scatter plot that some values do not correlate well. But is there a way to quantify this poor match. Trying to establish the true lower limit of the assay. Consider geometric median, rather than mean
Topic revision: r2 - 18 Dec 2023, IneSohn
 

This site is powered by FoswikiCopyright © 2013-2022 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