Data and Analysis for Clinical and Health Research Clinic Notes (2019)


2019 December 19

Raymond Zhou, Natalia Morale, Sean Donahue, Vanderbilt Eye Institute

  • We using two separate approaches to study patients that are forwarded to Vanderbilt Eye Institute for ophthalmologic evaluation after a failed vision screening (a test that estimates refractive error and eye alignment). My colleague is completing a retrospective chart review of patients forwarded to VEI after failed vision screening at two specific pediatrics practices. I am working with VICTR on a data pull to look at all patients forwarded to VEI after failed vision screening. Our primary analysis for both projects will be a multivariate regression analysis, to assess for the effects of various demographic/clinical variables (age, race, zip code, etc.) on the prevalence or absence of Amblyopia Risk Factors (a defined set of eye diseases). Other analyses will try to assess for the individual affect of certain variables, e.g. by comparing the positive predictive value of failed vision screening in hispanic patients vs. non-hispanic patients. How we can best conduct these analyses is another question we want to ask of you. Lastly, I am interested in finding out how many patients I would need to manually review to validate a dataset generated from a data pull. Mentor confirmed.
  • Meeting notes: About 40-50% of referred patients actually have Amblyopia, so the goal is to reduce the number of referrals and no-shows. Risk factors for Amblyopia include myopia, astigmatism, and cataracts. Hispanics tend to have higher rates of astigmatism. Have collected clinical data from patients who failed vision screening (birth weight, gestational age, vision screening measurements, whether referred, etc.). How many patients should be included in the study (currently have 400/500 patients eligible after chart review).
  • Recommend using a multivariable logistic regression model to evaluate association between risk factors and diagnosis of Amblyopia. The risk factors in the model should be pre-specified. The effective sample size is the number of patients with Amblyopia (case), and recommend 10-15 cases per predictor in the model (need 96 cases just to estimate the intercept in the model). Recommend contacting Dr. Cindy Chen regarding Biostatistics collaboration support. If this project does not qualify, then can apply for VICTR Award for biostatistics support (90 hours).

2019 December 12

Amanda Abraham, MPH Program

  • Mentor confirmed
  • My thesis question is looking at the relation between food insecurity and engagement in HIV care in the Care4Life program in Nigeria. The second component of my thesis is looking to see if this association varies among female head of households vs. male head of households in the program. We hypothesize that food insecurity will be associated with poor engagement in care and that female head of households will have worse outcomes compared to male head of households.
    I am applying for the VICTR voucher to get help with my data cleaning and analysis in STATA.
  • Recommendations: Using number of visits or number of gaps in care may provide more information than binary outcome. Number of visits plus the longest gap between any two visits may be helpful as well. To look at relationship between two variables only, try rank correlation Somer’s D or Kendall’s Tau (ex: correlation between food insecurity and female head of household. The main analysis can be divided into one large multivariable logistic regression model or two models depending on the question. We suggest getting in the VICTR queue. Application website (https://starbrite.app.vumc.org/) & research proposal template (https://starbrite.app.vumc.org/funding/templatesforms/)

Spencer Workman, Radiology

  • Mentor confirmed
  • Pilot evaluating PET CT after Y90 for treatment of liver cancer
  • Recommendations: A VICTR voucher is not necessary at this stage, but will be once data collection is complete to determine feasibility. Need to decide what matters most and choose the appropriate measure to be calculated, such as ratio or delta between measurements. Then, we can calculate power and sample size. A larger sample will be needed if there is much variability among patients. It may be helpful to start thinking about graphical approaches like scatterplots visualize the measure.

2019 December 5

Landrew Sevel, PM&R

  • I am in the process of amending an IRB to look at a different set of outcomes in a mindfulness-based intervention. I intend to apply for a VICTR voucher to support participant payment. Our planned analyses is of mediation in a repeated measures setting (See Montoya, A. K., & Hayes, A. F. (2017). Two-condition within-participant statistical mediation analysis: A path-analytic framework. Psychological Methods, 22(1), 6.). I am specifically looking for assistance in identifying how to conduct a power analysis for this analytic technique. As a note, I will be coming from an 11am patient appointment at 3401 West End, and will likely not arrive until 12:15pm.
  • Meeting notes: Subjects are patients with chronic pain. Instruments will be completed at baseline and follow-up to measure mindfulness and self-compassion. Limited funding is available, so do not plan to randomize subjects to intervention.
  • Without randomization, concerned with regression to the mean. Recommend enrolling twice as many subjects as planned and randomizing half to receive the intervention. Can compare change in scores between intervention and control groups. What difference is possible to detect (ex. -2), and is this difference clinically meaningful? Sample size may be calculated using software programs PS (http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize or https://statcomp2.app.vumc.org/ps/) or nQuery Advisor (VU Software). May attend the Learning Healthcare System Workshop under CTSA for support with pragmatic trials.

2019 November 21

Devika Nair, Nephrology

  • This is related to a project aimed to measure the demographic characteristics associated with job satisfaction and burnout. I have a specific question about the inclusion of two variables that I want to include in a regression analysis
  • Meeting notes: Collected demographics and perceptions of burnout with a national REDCap survey of adult nephrologists. Burnout is defined using two scales: emotional exhaustion and depersonalization (outcomes, range 0-6). Would like to fit a regression model with age, gender, race, foreign medical graduate, years of practice, number of patients seen, part of country, percentage of work spent seeing patients, setting of practice (select all that apply), area of practice (select all that apply), patient population served, and marital status.
  • Recommend starting with univariate analyses between the outcomes and each covariate. Variables that permitted 'select all that apply' are included in database as separate indicator variables. Can select which indicator variables to include in the regression model. May want to relabel the indicator variables for easier interpretation.

DROP IN: Brooke Baranosky, VU Undergraduate

  • Meeting notes: Have collected data from Synthetic Derivative to assess relationship between retinopathy of prematurity (ROP) and race in newborns and subgroup of premature babies. Have data on stage of ROP, treatment status,
  • If can get access to the raw data, then recommend formatting the spreadsheet with one row of data for each baby and one column for each variable. Recommend using a chi-square test on ROP (yes/no) and race for all newborns. Can also use a chi-square test for the subgroup of premature babies. A logistic regression model for ROP (yes/no) can be used to calculate odds ratios by race.

2019 November 14

Kianna Jackson, Plastic Surgery

  • We reviewed the charts of ~400 adult patients presenting to the ED with concern for orbital entrapment after traumatic eye injury. We are hoping to learn more about correlating radiologic findings to the clinical signs of entrapment. We are also repeating this project in the pediatric population.
  • Mentor Confirmed
  • Recommendations: Due to limitations in sample size (8 outcomes), we are concerned that the confidence intervals around PPV will be too large. Suggested to start out by describing patient characteristics first to see if there is a signal somewhere. We can also estimate the probability that the radiologist would diagnose entrapment given patient had entrapment, as well as the probability given patient did not have entrapment. For these probabilities, obtain Wilson confidence intervals. We can also obtain proportions in subdivisions for more common outcomes (ex: male versus female); these common outcomes should have at least 96 for any subgroup in order to obtain a CI +/- 0.1. Calculating yearly proportion of positive radiologist calls and plotting time trends is another interesting idea.

NO SHOW: Neil Newman, Radiation Oncology

  • This is going to be help regarding how to best design a phase I trial based of of retrospective work. The idea stems from prior retrospective work and we seek to see how many patients would be required for our endpoint. The project sought to see whether radiation to the vertebral bone marrow results in grade 4 hematologic toxicity, a known negative prognostic marker. Based on our results we did establish criterion to abide by to avoid this toxicity via sparing of dosimetry. We now see to evaluate prospectivley if limiting the dose to our pre-specified values will result in sparing of grade 4 lymphopenia and want to gain a sense of how many to enroll for the initial set.
  • Mentor Confirmed, will attend via phone

2019 November 7

No Show: Colin Walsh, DBMI

  • Comparison of risk predictive algorithm and clinical screening instrument

2019 October 31

Carla Copeland, Pulmonary

  • We are performing a retrospective cohort analysis of patients affected by familial related interstitial lung disease looking at the association of environmental and occupational exposures with disease pattern on both high resolution CT chest and pathologic biopsy. We are determining exposure history by a patient completed questionnaire that encompasses 48 different possible exposures with both planned analysis of both categorical variables (presence or absence of exposure) and continuous variables (time of exposure). Given the complexity of this data, we would like to determine the best statistical method to analyze the data in a cohort of about 250 patients. We have exposure data in electronic redcap database but CTs and biopsy results will need to be determined via clinical chart review. Mentor confirmed.
  • Meeting notes: Can look at correlation between all features and outcomes. Can run chi-square test on each exposure and outcome. It may be possible to do a cluster analysis on the model features (ex. 2-3 big groups) and/or outcomes using principal components analysis or machine learning, but do not recommend clustering the exposures. Recommend fitting a logistic regression model for each outcome. Can split data into training and testing sets to complete internal model validation. May conduct survival analysis using time between birth and disease diagnosis.
  • Recommend applying for VICTR Award for biostatistics support (90 hours).

2019 October 24

Carolyn Ahlers, Department of Ophthalmology and Visual Sciences

  • This project aims to determine what percentage of patients with FEVR (a rare ophthalmology disease in children) have persistent vascular leakage and neovascularization after complete laser photocoagulation (laser is the first line treatment) and determine if anti-vascular endothelial growth factor (anti-VEGF) therapy is associated with lower risk of vascular leakage and neovascularization (separate outcomes). We have a database of a total of 404 patients with FEVR. We estimate that 75% of these patients have been treated with laser for vascular leakage and neovascularization. We will analyze the records from these patients to determine how many have had persistent vascular leakage and neovascularization even after complete laser photocoagulation. We will then determine how many of these have been treated with anti-VEGF therapy (approximately 50 patients). Because anti-VEGF is reserved for patients who “fail” laser therapy and are thus “more sick,” we will use propensity score matching to match patients who did and did not (anti-VEGF yes/no) receive anti-VEGF therapy. Using propensity care matching, our exposure would be anti-VEGF (yes/no) and the outcomes we would like to assess include 1) neovascularization (yes/no), 2) vascular leakage (yes/no), 3) change in snellen score (before anti-VEGF and after anti-VEGF), and 4) days from the last anti-VEGF injection to the first retinal detachment or vitreous hemorrhage.
  • Mentor Confirmed

Recommendations:
  • Use of propensity scores do not seem feasible in this case, since we do not have a valid control group and one group is already more severe than the other (first line versus second line therapies).
  • Conduct binary logistic regression with failure (retinal detachment) as the outcome, controlling for 2-3 key predictors since number of failures is small (n=35). Each eye would be the unit of sample.
  • Conduct Cox PH model examining time to failure from diagnosis controlling for a few key predictors.
  • Since some people are contributing both eyes, there is correlation in the data that needs to be accounted for.

No Show: Lisa Bastarache, Biomedical informatics

  • I am working on a high-throughput method to generate evidence regarding the pathogenicity of rare genetic variants. I’ve come up with a way of interpreting results from linear regressions that I think is good, but might be completely nuts. The most pressing question I have concerns the interpretation of upper and lower confidence intervals.

2019 October 17

Lindsey Knake, Pediatrics

  • Plan to study antibiotic use in the NICU (exposure) and hearing loss in sample of 1000 neonates. Some patients receive antibiotics for 3 or 7 days. Patients who are diagnosed with hearing loss will generally have repeat hearing tests over time. If plan to fit a regression model, suggest calculating a propensity score to match exposed and non-exposed patients. May also look at time to diagnosis of hearing loss.

2019 October 10

Gabriella Glassman, plastic surgery

Project #1:
  • Multicenter Breast Reduction database; Retrospective chart abstraction to collect data on breast reduction surgery, specifically comparing drains vs. no drains; want to recommend use of no drains. We want to look at all patients from VU and eventually become a multi-center registry with other universities. Plans to collect information on demographics, BMI, complications, co-morbidities, and other relevant variables. Mainly interested in the relationship between type of surgery and complications (ordinal scale 0-3).

Our feedback:
  • Reasons for not using drains would be an important component in the database that needs to be clearly understood, because it may be physician’s preference.
  • Ideally we want people in the two groups to have similar covariates (age, BMI, etc..), so that they can be comparable like in a randomized trial.
  • Need to have adequate number of people with no drains; rule of thumb is (10-15 per covariate) to control for in an analysis
  • Perhaps can start out as a descriptive study, or as a pilot study first to see if study is feasible
  • Use synthetic derivative or starpanel to retrieve historical data
  • Brainstorm how to clearly define presence of complications
  • Calculate own BMI using height and weight; collect birthdates and surgery dates to calculate age rather than using number on chart

Project #2:
  • Dermatology study using a device to detect lymphedema, wants to know about this device and then also compare results to another device; both devices test different spots and take multiple measurements for each person at different locations (ex: left arm versus right arm). Will collect information on BMI, age, measurements on affected arm vs. no affected arm, stiffness, temperature etc…

Our feedback:
  • start out by plotting the data first: some ideas are box-and-whiskers plot, scatterplot of all measurements comparing left and right arm values, plots showing absolute differences of measurements and each device separately
  • If plots show a linear relationship, we can see there is some correlation.
  • Need to define what are the values to be considered clinically meaningful
  • Try describing the data first and perhaps start out with within subject differences
  • Next step is to look at sensitivity and specificity

2019 October 3

Patrick Kelly, Neurosurgery

  • MSCI student, drop-in
  • Interested in VICTR voucher
  • don’t have any data right now, but expects about 120 people to be in data linked from EPIC with about 50% to use survey linked from MyHealth at Vanderbilt
  • need help calculating a sample size, don’t know baseline satisfaction
  • ultimately wants to do a survival analysis looking at quality adjusted life years
  • recommendations: conduct a time to death analysis, conduct a time to “absence/reduction of quality of life” (no longer able to work or function..etc as a proxy), surveying family members on behalf of patients in the long-run, work backwards to calculate sample size using nQuery program or PS program (https://statcomp2.app.vumc.org/ps/)

Benjamin Campbell, Otolaryngology

  • drop-in
  • already wrote paper, but rejected because of statistics
  • looking incidence of vocal fold immobility in intubated ICU patients
  • prospectively enrolled 100 patients and found 7 cases, collected demographics, diseases, intubation details
  • want to compare characteristics of cases versus non-cases, and want to know how to calculate effect size
  • use of vasopressors and having hypotension are key predictors
  • recommendations: use Mann-Whitney U test, and Fisher’s exact test, perhaps think about propensity score modeling to adjust for covariates and use to predict main outcome of interest, discourage using Cohen's d for effect sizes and encourage thinking about a clinically meaningful effect size in this setting, recommend requesting VICTR voucher

WITHDREW Carolyn Ahlers, Department of Ophthalmology and Visual Sciences

  • This project aims to determine what percentage of patients with FEVR (a rare ophthalmology disease in children) have persistent vascular leakage and neovascularization after complete laser photocoagulation (laser is the first line treatment) and determine if anti-vascular endothelial growth factor (anti-VEGF) therapy is associated with lower risk of vascular leakage and neovascularization (separate outcomes). We have a database of a total of 404 patients with FEVR. We estimate that 75% of these patients have been treated with laser for vascular leakage and neovascularization. We will analyze the records from these patients to determine how many have had persistent vascular leakage and neovascularization even after complete laser photocoagulation. We will then determine how many of these have been treated with anti-VEGF therapy (approximately 50 patients). Because anti-VEGF is reserved for patients who “fail” laser therapy and are thus “more sick,” we will use propensity score matching to match patients who did and did not (anti-VEGF yes/no) receive anti-VEGF therapy. Using propensity care matching, our exposure would be anti-VEGF (yes/no) and the outcomes we would like to assess include 1) neovascularization (yes/no), 2) vascular leakage (yes/no), 3) change in snellen score (before anti-VEGF and after anti-VEGF), and 4) days from the last anti-VEGF injection to the first retinal detachment or vitreous hemorrhage.
  • Mentor Confirmed

2019 September 26

Lana Boursoulian, Pulmonary / sleep

  • I have collected data from SD DISCOVER aiming find out the recurrence rate for seizures for patient with autism compared to the general population. Have data for 147 patients; of these, 103 experienced a recurrent convulsion. Goal is to assess relationship between EEG category (3 levels) and recurrent convulsion. Patients are further categorized by seizure type (behavioral rest, generalized, or both).
  • Meeting notes: Can specify recurrent convulsion status as primary endpoint and specify EEG category and seizure type as independent variables. Can also conduct a marginal analysis with recurrent convulsion status and seizure type.
  • Recommend applying for VICTR Award for biostatistics support (90 hours).

Whitney Smith, Medical Student, Radiation Oncology

  • Looking at SRS radiation therapy in pediatric cancer patients with CNS tumors. Goal to look at patterns of care in NCDB population. Plan to report descriptive statistics for demographics and treatment characteristics. Would also like to utilize chi-square test for trend. Is this a good project for a VICTR Award?
  • Meeting notes: Recommend resident and their mentor attend another clinic to better understand the aims of this project.

Zac Yoneda, Cardiology

  • Vanderbilt Atrial Fibrillation Ablation Registry has ~3,000 subjects. This procedure is expensive and not always effective. Patient's imaging and rhythm monitoring are collected at 3, 6, and 12 months post-procedure. Primary outcome is time to recurrence of AFib. How should missing data be imputed (~400 data points)? How to handle data points that are missing because the test was never ordered?
  • Meeting notes: Recommend selecting two sets of variables: one set for the imputation method and a second set for the primary analysis. Can run a sensitivity analysis excluding missing data.

WITHDREW Matthew Mart, Pulmonary Disease and Critical Care

  • We’re performing a pilot study evaluating the role of cardiorespiratory fitness (measured through exercise testing) and long-term cognitive impairment following critical illness.
  • Questions: calculating sample size in absence of prior data for a larger trial; best methods for assessing correlation between testing methods
  • Also participating in this as part of MSCI Biostatistic 1 course with Daniel Byrne as instructor

2019 September 19

Yuxi Zheng, Ophthalmology

  • Surgically naive students interested in going into a surgical field will perform a series of pre and post tests to assess for speed and accuracy under the microscopes in the wetlab. They will be randomized to either the 2d or 3d group and assessed on various tasks pre and post intervention.
  • We want to determine what sample size would be needed to have enough power to identify a difference if it exists.
  • We want to make sure that we are evaluating ideal metrics during the study (ie timing, degrees, # times damaged other structures, etc)
  • We will likely want to compare pre and post intervention, and then compare the improvement between the two groups to see if there is a difference.
  • Mentor Confirmed

Our feedback:
  • provide units for time in survey
  • decide on questions to focus on to compare between two groups (speed, accuracy, timing)
  • recommend use of PS software to calculate sample size (http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize)
  • consider creating an overall summary score for pre- and post- tests to compare between subjects and between groups, or try to find a validated scale
  • null hypothesis: 3d is at least as good as 2d video (non-inferiority test), which requires a one-sided test and helps to improve power
  • add exclusion criteria for 3d group for those who have visual problems, good idea to do a pre-screening or ask some questions
  • try to keep study double-blinded
  • try to randomize within a block using pre-screening questions/tasks (ex: how much time spent under a microscope or baseline performance)

Rajiv Agarwal, Medicine/Hematology-Oncology

  • I’m a new faculty member at Vanderbilt, and am part of the MSCI/K12 program. I’m hoping to learn more about how to design effective studies related to my research - on measuring outcomes longitudinally over time from palliative care interventions in patients with cancer.
  • Mentor Confirmed
  • Rajiv and mentor attended clinic to have an idea of Biostatistics clinics. He will come back once he is closer to starting his study.

2019 September 12

Meredith Campbell, Neonatology

  • Sample Size and Power
  • Mentor Confirmed

2019 September 5

Laura Shashy, Neonatology

  • Assessing pre and post anxiety/depression scores in parents after participating in a journaling program. Best way to analyze results between cohorts of parents. Mentor confirmed, VICTR voucher.
  • Meeting notes: Recommend reformatting spreadsheet with variable names in the first row (no spaces, can use underscores) and one row per parent. Add an indicator variable for intervention (0 = control, 1 = intervention). Can recode other indicator variables (0 = no, 1 = yes). Rename variables anxiety_pre, anxiety_post, depression_pre, and depression_post. Test for difference in anxiety_pre between the two groups using the Wilcoxon rank-sum test. Assuming there is no difference, calculate difference between anxiety_pre and anxiety_post for each subject, then use the Wilcoxon rank-sum test to determine whether there is a difference in pre/post score change between the groups. Can fit a linear regression model for anxiety_post including covariates for anxiety_pre, intervention indicator, and parent and child demographics. For the intervention group, may also assess relationship between how often the journal was used and the difference in pre/post score change. Compare mean anxiety_pre for subjects who do vs. do not complete the post survey. Similar analyses can be done for depression.

Melis Sahinoz, Nephrology

  • Meeting notes: Enrolled 20 cancer patients and healthy controls. Goal to match on age, sex, race, history of diabetes, BMI, hypertension, hypercholesterolemia. BMI and SBP measurements were collected at screening and week 0. For 1-to-1 matching, first sort on important factors, then select matches in order and enter match code (ex. 1, 1, 2, 2,..., 20, 20). Can also calculate difference in age, sex, race between a case and every available control, then look at smallest difference to select a match.

2019 August 29

Samantha Brokenshire, Pharmacy

  • Retrospective chart review of the impact of a EHR notification on alteplase use and on infection (likely Chi-Square). Additionally, a retrospective chart review of the effects of spironolactone on serum electrolytes and diuretic use (likely Paired T).
  • Note: scheduled for 8/22 as well. Mentor confirmed
  • Meeting notes: Pediatric patients with central lines are at higher risk of developing blood stream infections. Plan to look at two groups of patients, before and after change to EHR notification. Eligible patients were seen in an inpatient or outpatient setting. Primary outcome is infection diagnosed with positive blood culture. Blood cultures are ordered for patients with a fever. Additionally, several factors lead to central line infection, so plan to build multivariable logistic regression model.
  • Recommend 3 categories for outcome: blood culture positive, blood culture negative, blood culture not ordered. Can use ordinal analysis to look for association between pre- and post-periods. Do not recommend using univariate analysis to build list of model covariates. Recommend a minimum of 20 events per candidate predictor in the multivariable logistic regression model (ex. 10 variables requires 200 events). Select candidate predictors using clinical knowledge. Due to differences in study methods with respect to controls, recommend ignoring controls in the meta-analysis.

Gabriella Glassman, Plastic Surgery

  • I have completed a systematic review. I would like to run a meta- analysis on my data.
  • Mentor confirmed. Possible VICTR voucher.
  • Meeting notes: Reviewed 24 studies and collected summary data on physician and patient satisfaction (scaled all studies to use a 4-point Likert scale for each question). Studies used heterogeneous populations. Summarized mean physician satisfaction score and mean patient satisfaction score.
  • Be aware of floor and ceiling effects when physician/patient is asked to compare satisfaction or improvement between time points. Should stratify studies into two groups based on whether the questions were worded as satisfaction at current point in time or satisfaction compared to previous point in time. Raw data from 1+ studies will provide much more useful information, and combining raw data from multiple studies is necessary to implement meta-analysis methods. Recommend creating scatterplot of individual satisfaction scores and bubble plot of summary scores.

2019 August 22

Ellen Casale, Special Education

  • I am currently working on the randomization excel sheet to support the randomization component to a redcap registration form.
  • Meeting notes: Prefer to discuss the general plan of study to discuss whether the method of randomization is appropriate before loading into redcap. Intervention is a training day. Randomizing to intervention in fall or spring. Hypothesis is that the group that gets the intervention in the fall will show knowledge improvements at various timepoints while the spring group will have constant knowledge (no improvements) while waiting until spring intervention starts. Suggest putting together a statistical analysis plan that lays out what will be analyzed at what timepoints. Recommend reproducible code for producing randomization table. Can work with Chris Lindsell to get a victr voucher to get a statistician one on one to assist.

Samantha Brokenshire, Pharmacy

  • Retrospective chart review of the impact of a EHR notification on alteplase use and on infection (likely Chi-Square). Additionally, a retrospective chart review of the effects of spironolactone on serum electrolytes and diuretic use (likely Paired T).
  • Note: scheduled for 8/29 as well.
  • Mentor confirmed
  • Meeting notes: Looking at spironolactone use. Describe characteristics of patients who are using it. Demographics, other diuretics, etc. Also looking at effect on clinical outcomes such as potassium, other meds. No protocol in place for prescribing, at clinical discretion. Concerns about patient population selection. Beneficial to view graphical information about pre and post using spaghetti plots, AUC, etc. The values will help determine what kind of test to use. Regression can look at the increase based on drug use.

2019 August 15

Willibroad Maimo, Medicine, Health & Society

  • Using longitudinal survey data to test the relationship between educational attainment and metabolic syndrome. Would like to discuss mediation and moderation.
  • Meeting notes: Collected cholesterol levels and biomarkers (TC, LDL, HDL, TG, LpA) before and after initiation (1-year follow-up) of statin treatment for hypercholesterolemia. Due to compliance issues, selected best/minimum value for each cholesterol level to calculate percent reduction. Goal to assess how statin reduces cholesterol for patients with specific biomarkers (compare average percent reduction between groups LpA LpA >30).
  • Recommended statistical software Stata, SPSS, or R. Remove special characters from variable names and rename duplicate variable names. Can plot percent change in cholesterol vs. LpA (continuous variable). May also look at descriptive statistics and plots of cholesterol levels by age, gender, statin intensity, etc. Plot baseline cholesterol vs. percent change in cholesterol. Calculate Spearman-rank correlation coefficient between percent change in cholesterol and LpA level (and age, etc.) to quantify strength and direction of relationship. Can use linear regression to predict percent change in cholesterol adjusted for statin intensity, LpA (continuous), age, gender, and other patient characteristics. Do not recommend dichotomizing continuous variables because valuable information is lost. Selecting the best/minimum value and calculating percent change may not be ideal because it could be associated with the most measurement error. To handle incomplete data, can use interpolation or model available data points. Important to include timing of cholesterol measurement since initiation of statin in the model. Want to be able to show that cholesterol level (biomarker) is not correlated with compliance. Fundamental question is whether two patients who start with the same baseline cholesterol, but different LpA levels, end up with different follow-up cholesterol levels.

2019 August 8

Lauren Gaydosh, Internal medicine

  • This is restropective cohort study comparing the effect of Statin therapy between patient’s with elevated Lpa compared to those with low Lpa. Dan Byrne listed as mentorl.

James Andry, Sleep Medicine

  • Please provide a short description of your project and the questions you’d like to address: I have collected CPAP adherence data from 2 groups of patients with sleep apnea and have performed preliminary analysis on this dataset. I was hoping to run my methodology by a biostatistician to confirm that I’m not making major errors.
  • Meeting notes: Suggest measuring predictors of adherence. Difficulty due to lack of randomization. Learning healthcare resource can be used to design new prospective study.

2019 June 6

Matthew Yandell, Mechanical Engineering

  • This is a follow up from May 16. May 16 notes are below.
  • We are exploring the comfort limits of human subjects as forces are applied to their back, thigh and calf. We are hoping to determine the most appropriate tests to conduct to determine the comfort limits and how these change across days.
  • Enrolled 10 subjects, data collected for 4 consecutive days, randomized order of force applied to back, thigh, and calf (5 fast and 5 slow). Outcome was peak linear pulling force the subject experienced before pressing the button. Subject pressed button when became uncomfortable or stopped at safety limit of study. Wash out period was 15-30 seconds between measurements. The study objectives are to 1) determine whether mean comfort limit changes across days and 2) determine whether mean comfort limit changes due to rate (fast/slow). Have plots of mean comfort level across 4 days (separate for each location), and plan to test whether slopes are different from 0.
  • Meeting notes 5/16/19: Recommend structuring database with rows of Subject ID, Day, Location, Rate Indicator, Trial Number, and Peak Linear Pulling Force. Then can use a mixed effects linear regression model for Peak Linear Pulling Force; treat Subject ID as a random effect and Day, Location, Rate Indicator, and Trial Number as fixed effects. Results will inform whether there are differences with respect to Day, Location, or Rate. Can conduct subgroup analyses for measurements for each specific location by fitting 3 separate models for back, thigh, and calf measurements. If the within subject variance is very small, then the total sample size will be close to N=10. If the within subject variance is very large, then the total sample size will be close to N = 1200. Watch for potential issues with model not converging. Using R software, 'lme4' package can be used to fit mixed effects models. If the outcome is not normally distributed, then recommend using a proportional odds model with robust sandwich estimator.
  • Meeting notes 6/6/19: To assess the effect of Day, recommend using Stata software to conduct a response feature analysis on a vector of observations. Estimate a slope for each patient and test whether slope is different from 1 (see book by William Dupont). Can also run Friedman's test across all four days. Next can compare pairs of days using the Wilcoxon signed rank test for paired data; do not need to include Bonferroni correction for multiple comparisons because it is so conservative. To assess differences by Rate, recommend using the Wilcoxon signed rank test. A parametric analysis would be required to calculate a confidence interval for the difference.

2019 May 30

James England, Infectious Diseases

  • Descriptive epidemiologic study of patients admitted for possible pulmonary tuberculosis. Working to identify factors associated with increased risk of being diagnosed with TB. Main question: what is the best way to handle patients with multiple admissions?
  • Collected data from 662 hospital admissions and 543 unique patients, and 15 of these patients were diagnosed with TB. There were at most 5 admissions per patient. Many of the patients have already been diagnosed with HIV.
  • Meeting notes: Recommend running two analyses: the first using one admission per patient, and the second including multiple admissions. Can run univariate hypothesis tests (chi-square test, Mann-Whitney U test) to determine whether a patient characteristics is associated with TB diagnosis. Recommend placing covariates into buckets such as "known prior to admission", "known for select patients only", etc. Can include more than one covariate in the logistic regression model (overfitting), but the rule-of-thumb is one covariate per 15 cases. Should validate the model on an external dataset (ex. use VUMC data from a different time period or data from another university to replicate the model). Then use model to predict status of future patients and whether to require isolation. Can report reduced costs and outcomes of patients over one year.

Christopher Brett, Radiation Oncology

  • We would like to develop a genetic risk score to assist in patient selection for lung cancer screening. Using BioVU we plan to separate 1) patients meeting screening criteria found to have cancer, 2) patients meeting screening criteria but without cancer, and 3) patients not meeting screening criteria but nonetheless developed lung cancer. We then hope to identify SNPs/mutations correlating with the binning and create a genetic risk score to couple with clinical screening criteria.
  • Can do a chart review in the Synthetic Derivative and match cases (n=3400) and controls. 94% of positive screens are false positives. How can a genetic risk score be created? How can the propensity score matching be completed?
  • Meeting notes: Recommend using a logistic regression model to calculate a probability for developing lung cancer. Include patient demographics and clinical characteristics in the model in addition to the genetic information. Need to create spreadsheet with cases and controls and patient demographics, clinical characteristics, and genetic information.
  • Recommend applying for VICTR Award for biostatistics support (90 hours). Cornelius meetings are held Thursdays at 3:00pm to assist in model fitting.

2019 May 23

Kathryn Gayle, Cardiovascular Medicine

  • Evaluate utilization rates of nuclear stress tests, determine correlation of findings on stress imaging with adverse cardiac outcomes.
  • VICTR Voucher, Mentor confirmed
  • Meeting notes: Interested in relationship of stress test findings vs composite score of outcomes. Estimated 300 subjects with event rate of 15-30. The strength of the correlation will depend on the stress test findings. Interested in VA system using current stress algorithms. Recommend focusing on one group of patients and comparing those who received stress test vs not and looking at outcomes. Will apply for VICTR voucher for pilot study and then work on getting a collaboration for a larger study.

2019 May 16

Matthew Yandell, Mechanical Engineering

  • We are exploring the comfort limits of human subjects as forces are applied to their back, thigh and calf. We are hoping to determine the most appropriate tests to conduct to determine the comfort limits and how these change across days.
  • Enrolled 10 subjects, data collected for 4 consecutive days, randomized order of force applied to back, thigh, and calf (5 fast and 5 slow). Outcome was peak linear pulling force the subject experienced before pressing the button. Subject pressed button when became uncomfortable or stopped at safety limit of study. Wash out period was 15-30 seconds between measurements. The study objectives are to 1) determine whether mean comfort limit changes across days and 2) determine whether mean comfort limit changes due to rate (fast/slow). Have plots of mean comfort level across 4 days (separate for each location), and plan to test whether slopes are different from 0.
  • Meeting notes: Recommend structuring database with rows of Subject ID, Day, Location, Rate Indicator, Trial Number, and Peak Linear Pulling Force. Then can use a mixed effects linear regression model for Peak Linear Pulling Force; treat Subject ID as a random effect and Day, Location, Rate Indicator, and Trial Number as fixed effects. Results will inform whether there are differences with respect to Day, Location, or Rate. Can conduct subgroup analyses for measurements for each specific location by fitting 3 separate models for back, thigh, and calf measurements. If the within subject variance is very small, then the total sample size will be close to N=10. If the within subject variance is very large, then the total sample size will be close to N = 1200. Watch for potential issues with model not converging. Using R software, 'lme4' package can be used to fit mixed effects models. If the outcome is not normally distributed, then recommend using a proportional odds model with robust sandwich estimator.

2019 April 25

Shawniqua Williams Roberson, Neurology

  • The overall objective of this study is to establish pilot data to support a more comprehensive study evaluating quantitative electroencephalography (qEEG) as a biomarker of ICU delirium. The specific aims are as follows:
    Aim 1: Determine the quantitative EEG characteristics of delirium in mechanically ventilated ICU patients. Aim 2: Determine the predictive value of quantitative EEG features associated with delirium with regard to cognitive and functional deficits at discharge.
  • Meeting notes: Primary question, which features best indicate delirium in mechanically ventilated icu patients and do these features predict enduring brain dysfunction? Study design: perform eegs and do delirium screening. Target recruitment 25 patients. Discussed scope of project. Recommend applying for voucher to get statistical advice for initial study submission and then a larger grant for analysis of study. PS software for sample size: http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize

Kate Humphreys, Peabody

  • The overall objective of this study is to look at pregnant mom's stress and baby's brain.
  • Meeting notes: Stress questionnaire (self-reported) for mom regarding stress and cortisol levels. Need sample size justification for preliminary analysis. Recommend applying for voucher to get statistical advice for initial study submission and then a larger grant for analysis of study. With 24 subjects, the minimal detectable correlation is _. change to "in order a detectable correlation of _, we will need a sample size of 24" Recommend nquery software. https://www.vumc.org/it/softwarestoresearch or applying for a victr voucher to get a sample size justification.
.

2019 April 4

Audrey Bowden, Biomedical Engineering

  • I’m developing a new device for retinal imaging of diabetic eye disease. Want to compare results with standard of care. Need to know how many patients to power study and what analysis to do.
  • Proposed 10 patients for pilot study which will determine if images are of high enough quality (resolution, signal level) or if the instrument should be refined. Next plan to conduct study to validate new device in patients who have diabetic retinopathy or edema. The standard of care instrument outputs retinal images from which the thickness of layers can be measured.
  • If the new device is not calibrated to give exact same measurements as the gold standard, then will have to conduct correlation analysis to obtain an absolute calibrator. To estimate a correlation coefficient requires sample size of 300 patients (or 75 patients to be within +/- 2 SE). Can find course notes at hbiostat.org/doc/bbr.pdf and search keyword "plotCorr". If the absolute calibrator is linear, then can have equal number of diseased and healthy patients. Otherwise, will need to have more uniform distribution of patients. Should obtain images from standard of care and new device for each patient. First need to determine level of noise in the standard of care by collecting repeat images in the same setting; this may already be published. Recommend collecting 10 repeated measures on 10 patients in the pilot study. Can research whether a large medical society has made publicly available a retinal imaging database; this could serve as pilot control data. Remember to collect measurements on new patients to validate changes to the new device.

Lindsay Mayberry, Medicine/Gen Int Med&PH

  • Discuss issues with using change in HbA1c as an outcome variable, per suggestion of Frank Harrell.
  • A1c is measured every 3 months. Goal is to estimate control at 3m, 6m, and 9m. Have several sources for A1c measurements, mail in kit, EHR venipuncture blood draw, and point of care. Missing A1c values have been imputed. Patients with higher starting A1c have greater reduction in A1c by 3m (1% vs. 0.5%). Outcome is A1c at 12m because want to determine the long-term effect.
  • A1c measurements with more measurement error will have an attenuated effect. Smaller A1c values are known to have lower measurement error than large A1c values. The ERIC study has good data on long-term changes in A1c. There is a linear relationship in general, but can turn up when the risk is higher. Related to cardiovascular events, it does not behave in a percent-change basis. When you subtract two numbers, the numbers first have to be perfectly transformed (ex. log, A1c^-1.5). In general, subtracting does not work well enough; adjusting for baseline A1c in the model works better. Can then conduct joint tests on interaction terms. Difficult to model the mean when the variance keeps changing. Log-log survival curves will be parallel (ex. vs. Age). Freedman's article states that obtaining the wrong standard error for the model does not necessitate use of a robust sandwich estimator. In a Cox model, all covariates would be related to the outcome in a non-linear way. Can present marginal (or conditional) estimate of difference in means between treatments.

2019 March 28

Bruce Damon, Radiology/VUIIS

  • Meet to discuss study design for upcoming experiment, including VICTR quote

    VICTR Voucher
  • Meeting notes: Discussed research question regarding FSF and response to cooling, meals, body composition. Viewed response by decades and reviewed variable functions depending on baseline. Viewed 3d mapping and discussed tensorsplines. Discussed inflection point and its relationship to obesity. Questioned sample size for future study.

2019 March 21

Sara Duffus, Pediatric Endocrinology

  • This project is a pilot study focused on adolescents with type 1 diabetes. We are planning a randomized clinical trial investigating the effect of a low carbohydrate diet intervention vs standard carbohydrate diet on outcomes including HbA1c, glycemic variability, lipid profiles and quality of life. Our primary question is regarding methods for sample size determination for pilot studies. We also anticipate that the low carbohydrate intervention will lower Hba1c more than the standard carbohydrate diet (aim 1). However, it is possible that the low carbohydrate diet will raise LDL cholesterol. We think this concern is unlikely to play out, so our hypothesis is that the difference in the change in LDL cholesterol between the groups will be minimal (aim 2). We would like to discuss statistical methods for analyzing aim 2. VICTR Voucher, Mentor confirmed.
  • Expect to enroll 20 subjects in each arm (40 total). Intervention and data collection will last 12 weeks. Do not plan to implement a crossover design due to time constraints. Insulin pump continuously injects a base level of insulin. The patient has to enter blood sugar level and number of carbs eaten throughout the day, and the insulin pump will adjust the insulin level.
  • Recommend using PS or nQuery software (from VU software) for sample size calculations. Can vary the standard deviation of delta to calculate range for sample size. Suggest including a pediatric health risk assessment questionnaire. At the end of the intervention, may also want to ask parent/child about adherence. Recommend including 10-15 subjects per 1 degree of freedom in the model.

Aparna Singh, Biomedical Engineering

  • I am currently working on high intensity focused ultrasound ablation research wherein I collected phantom data and I would like to know if my stats make sense. Mentor confirmed.
  • Studying a non-invasive treatment of hepatocellular carcinoma by using two types of phantoms to ablate regions (with nano droplets to enhance ablation vs. without nano droplets). Goal to compare efficiency of ablation between phantom types. Collected 4 measurements at each of 4 sites (16 total). Calculated volumetric ablation and CEM 240 and conducted a rank sum analysis.
  • Recommend using PS or nQuery software (from VU software) for sample size calculations. The larger the effect size, the smaller the required sample size. If have at least 40 total measurements, could conduct ANOVA to compare 4 sites. This is not as useful with only 16 measurements due to the variability. Recommend reporting plot of max ablation efficiency.

2019 March 7

Lindsey Knake, Neonatology

  • Improving volume targeted ventilation use in the NICU using clinical decision support. Would like to get an estimate of sample size needed for a grant that I am writing. VICTR Voucher, Mentor confirmed.
  • Goal to develop algorithms in EPIC to improve use of volume targeted ventilation. Currently have 65% use in NICU, and goal is to increase to 85%. Outcome is improved volume targeted ventilation. Plan to use an interrupted time series design; this design will require a very large sample size. Estimate there are 20 NICU babies who are intubated on any given day. Also plan to use retrospective data to build predictive model.
  • It may be possible to randomize patients in EHR; patients would be randomized to intervention or standard of care. With an interrupted time series design, there are a lot of factors that can change between the pre and post time frames. Can download PS software program (or nQuery from VU software) for sample size calculations. Recommend applying for VICTR Award for biostatistics support (90 hours). Also recommend presenting study to Learning Healthcare System with Gordon Bernard, which is part of VICTR; email Marylynn Dear to be added to the schedule.

Rena Robinson, Chemistry

  • We are interested in studying variants of a genetic risk factor for Alzheimer's Disease (AD) using proteomics and lipidomics. VICTR Voucher.
  • Plan to submit grant proposal tomorrow for the Alliance on Health Disparities between Vanderbilt and Meharry. Interested in specific variants in two genes. Will look at plasma and lipid samples to assess variants and whether there are differences in race. For pilot study, plan to utilize BioVU data and collect 20 samples for each of the five gene groups and 20 controls. Will randomize order of sample runs. Plan to generate volcano plots to determine significance. Results of the pilot study will be used in R21 grant proposal.
  • Recommend applying for VICTR Award for biostatistics support (90 hours). Can finalize statistical analysis plan and sample size requirements for a linear or generalized linear regression model. Outcome is the plasma (or lipid) measurement, the exposure is the genetic variant, and the additional covariates in the model will include race, gender, age, clinical characteristics, etc.

2019 February 28

Crystal Coolbaugh, Radiology

  • Please provide a short description of your project and the questions you’d like to address: Our group is imaging brown adipose tissue in humans. We would like to design a study to understand the responses of brown adipose tissue, white adipose tissue, muscle, and liver to cold exposure and feeding using MRI. In addition, we would like to study how these responses vary with age, body composition, sex, and physical activity.

  • VICTR Voucher, Mentor confirmed

  • Meeting notes: FSF is ratio of fat tissue seen in MRI. The activation of the BAT is usually measured by an average of voxole over certain threshhold . Discussed how best to show change in fsf aong different tissues. Suggested going back to raw data and considering 3d tensor splines. Suggest focusing on inflection points within individuals.

2019 February 7

Wendy Bottinor, Cardiology

  • Childhood cancer survivors are known to be at increased risk for cardiovascular disease. As a result, period, serial, cardiac screening is recommended. Currently, cardiac screening focuses on LVEF, a late marker of cardiac dysfunction. My goal is to determine is myocardial strain can be used as an early marker of cardiovascular dysfunction and thereby provide an earlier opportunity of clinical intervention. I would like to determine 1) The prevalence of abnormal strain in early off therapy childhood cancer survivors, 2) the natural history of strain on serial imaging, and 3) the short and long term cardiovascular outcomes associated with abnormal strain.

  • VICTR Voucher, Mentor confirmed

  • Has worked with Chang, requested to work with him again, if possible.
  • Meeting notes: 80% survival at 5 years. 2 leading causes of death are cancer and heart disease. Overall goal to look at identifying risk and if screening is done appropriately. Approximately 300 patients, want to go back and measure strain which could be a precursor to other dysfunction. Recommend using more than just baseline, other measurements to look at progression. Discussed excluding those who have repeated malignancies and determining reasons for why echo prescribed. Currently has a voucher obtaining survery results. Will apply for another voucher for this project.

Pingsheng Wu, Medicine

  • Walk-in: looking at inflammation leading to physiological changes in organs which causes function changes. Specifically lung function. Interested in which changes such as airways, attachment, thickness

  • Meeting notes: Discussed regression techniques. Recommended sorting out identifiying which pre-dates which and the pathways involved in the process to assist in mediation analysis.

2019 January 31

Michael Pridmore, Chemical & Physical Biology & Richard Dortch

  • Individuals with Hereditary Neuropathy with Liabilities to Pressure Palsies (HNPP) experience a “dying back” neuropathy that begins with loss of sensory and motor functions in the distal ends of the leg and spreads proximally towards the spine. Research on mechanisms underlying degeneration of the sciatic and tibial nerves is limited primarily to nerve conduction studies in distal nerves, which are often unavailable in patients with HNPP. Here, we are using a longitudinal MRI study to assess sciatic and tibial nerve health in HNPP individuals. A Magnetization Transfer (MT) sequence will be used to provide information of myelination and axonal integrity of the nerves in the thigh and ankle. Dixon (Fat/Water) sequence will be conducted on the thigh and ankle to probe fat infiltration following muscle atrophy, a symptom of HNPP. Clinical scores (such as Charcot-Marie Tooth Neuropathy Score) were also collected. The questions we’d like to address with this data are: (i.) What length-dependent differences (thigh vs. ankle) exist? (ii.) How do healthy controls and HNPP subject compare? (iii.) Is there a relationships between MRI and clinical scores? and (iv.) Is there an age related effect? Mentor confirmed
  • Goal is to assess correlation between clinical and MRI data for 10 HNPP patients. Also have MRI data for 10 healthy controls.
  • Can use principal component analysis data reduction to combine data elements. The first principal component essentially scales by standard deviation units and adds them up. This will be useful given the small sample size. The margin of error will be around +/- 0.5. Would need about 400 subjects to have margin of error around +/- 0.1.

Amalie Chen, Neurology/Neuroimmunology & Francesca Bagnato

  • My project is looking at measuring white matter axonal injury in patients with multiple sclerosis. Things I need help on: how to analyze differences in 6 MRI metrics between 4 pairs of different tissue types (i.e., what kind of stats to run) and how to compare the magnitude of the effect size among these variables. I have all the data already, I’m just not sure what are the proper stats to run. Mentor confirmed
  • Goal is to determine differences in imaging modalities. Have 18 cases and 9 controls, but the number of images/lesions per tissue type varies between patients (ex. 35, 2, or 0). How should the metrics be averaged within patient?
  • Faculty with imaging expertise, Hakmook Kang and Simon Vandekar, attend Biostatistics clinic on Tuesdays. Recommend scheduling another clinic appointment to address follow-up questions.

Alicia Beeghly-Fadiel, Epidemiology

  • Pilot study in BioVU looking at exposure to PPI (y/n) in pre- and post-bariatric surgery time points. Have 200 patients in 4 groups (+PPI/Pre, -PPI/Pre, +PPI/Post, -PPI/Post); do not plan to collect longitudinal data (only cross-sectional). Outcome is methylation of DNA at 15 sites of interest. Goal is to test whether methylation differs by medication exposure or pre/post surgery. Also plan to analyze obesity (BMI measured just prior to surgery and 1 year after surgery) and methylation. Expect BMI affects methylation and surgery affects BMI.
  • Recommend using a linear regression model (i.e. two-way ANOVA) with PPI effect, pre/post effect, and interaction effect between PPI and pre/post. For a multivariable linear regression model, can think of 15 methylation sites as independent variables to predict BMI (already multiplicity adjusted with 15 degrees of freedom); make sure to transform methylation sites if needed. Adjusting weights for height will make fewer assumptions than using BMI.

2019 January 17

Daniel Tilden, Pediatric Endocrinology

  • Using the RD to assess the effect of pediatric to adult transition on control of T1DM. Mentor confirmed, mentor will attend via phone. VICTR voucher request.
  • Transition system guidelines are vague and inconsistently applied. Current knowledge gap for A1c and clinic attendance. Goal of study is to quantify impact of current transition system on patient disease control. Hypothesize a significant rise (>1%) in A1c in the 2 years after transitioning to adult care. Will use ICD-9 and ICD-10 diagnosis codes to identify T1DM patients. Will collect data on demographics (zip code, SES, distance to VUMC, insurance/TennCare), all A1c measurements, and dates of clinic visits and inpatient admissions.
  • This is an estimating study, so recommend using margin of error rather than calculating power. Can plot A1c trajectories for 500 patients, then estimate starting A1c (just prior to transition). Use quantile regression to estimate quantiles of A1c as function of beginning transition. Use ordinal regression to estimate probability of A1c < 7% (8% or 9%). Compliance may be a factor in A1c trajectory. Need to address how to handle duplicate lab values from RD.
  • Recommend applying for VICTR Award for biostatistics support (90 hours). Contact Chang Yu and Dan Byrne for assistance with writing statistics portion of application.

Maureen Saint Georges, Pediatric Emergency Medicine Fellow

  • My project is a retrospective chart review of all children 0-18.99 years presenting to the VCH Emergency Department in cardiac arrest between Oct 2009 and Oct 2017. The goal of this study is to identify factors in the patients’ history and management that might predict survival to PICU admission. Data has already been collected. Possible VICTR voucher request. Mentor cannot attend, but will attend if voucher needed. * Sample size is 170. How many covariates should be included in the logistic regression model?
  • Analysis should use periodicity to allow 11:00pm to be closer to 12:00 midnight. Do not recommend using univariate analysis to identify candidate variables for logistic regression model. Need 96 patients just to estimate intercept and another 100-10,000 patients to estimate the effect of one variable. Using the 15:1 rule-of-thumb, 3 pre-specified variables can be included in the model. Recommend ranking 40 candidate variables based on correlation with death (Somer's D) and using bootstrapping to calculate confidence intervals. Will determine the potential each variable has (ex. Top 5, Top 15). Need to decide how to handle missing data values. Recommend contacting James (Chris) Slaughter regarding statistical analysis as part of Biostatistics collaboration plan or contacting Chang Yu and Dan Byrne to apply for VICTR Award for biostatistics support (90 hours).

2019 January 10

Renee Rosati, Physical Medicine and Rehabilitation

  • Provide live musical concerts regularly for patients in the rehab hospital and take pre and post assessments of anxiety and depression using the Hospital Anxiety and Depression Scale. I would like to address if mood is improved immediately after the musical performance compared to right before.
  • Meeting notes: Applying for osteopathic grant focused on mind body and spirit improving treatment. Have medical information about patients. "How's your mood?" likert scale? Pain 1-5. Satisfaction 1-5. Timing of surveys. Discussed difficulty of singling out the intervention benefit as the measurable effect. Importance of control group and validated testing method. Because there's not a way logistically to have a control group with a different intervention, there will be limitations to the write-up. Focus on acute effect, the change in mood and satisfaction for this group. Suggest a non-concurrent control activity: send survey pre and post activity but have activity be concert sometimes and non-community event (watch TV).
  • Mentor confirmed, mentor will attend via phone. VICTR voucher request.

Parisa Samimi, OBGYN

  • Opioid use following passage of TN law
  • Meeting notes: Uro-gyno surgical patients. July 1 change in process for subscribing medications. Providers must consent to being prescribed meds. (All consent). Did the number of morphine equivalents change between the first 6 months and second 6 months (2019). Same 4 surgeons, 18+ patients, exclude combined surgery. Also interested in overall trend of prescriptions. Estimated ~300 surgeries. Power calc: alpha 0.05, Power 0.80. comparison: 260 compared to 210 units. Patients are prescribed typical amounts. Suggestion interupted time series looking at trends and abrupt change points. ARIMA model time series techniques can examine change point.
  • Mentor confirmed, VICTR voucher request.
Topic revision: r1 - 18 Jan 2021, 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