Biostatistics applications in surgery, anesthesiology, and emergency and critical care medicine Clinic Notes (2018)

2018 December 19

Ryan Hsi, Urology

• Sample size calculation
• Planing a prospective RCT on patients undergoing ureteroscopy, randomizing to stent or no stent. The main outcome is 30 day complications. The literature indicates the outcome has a ~15% rate. There have been two *somewhat similar studies from 2001 and 2007 to help estimate SD - and need help determining best way to calculate sample size.

2018 December 12

Justin Banerdt, Department of Medicine, Division of Pulmonary and Critical Care Medicine, CIBS Center

• Our project is a prospective cohort study of delirium incidence and outcomes at a resource-limited referral hospital in Lusaka, Zambia with a high burden of critical illness and HIV. 820 new medical and surgical patients were evaluated for delirium using the brief confusion assessment method (B-CAM). 28-day and 6-month outcomes include survival and functional status. The primary questions we would like to address include: What is the prevalence of delirium and risk factors for delirium in this cohort? Is delirium an independent predictor of mortality and long-term functional impairment in this cohort?
• Outcome: VICTR Voucher

Sydney Payne, Plastic Surgery

• Clinical trial investigating traditional surgical follow up vs. no follow-up. Would like to discuss general approach to analysis based on data that we are collecting.
• Outcome: Protocol with no expected funding support

2018 December 5

Kelli Rumbaugh, Pharmacy/ Surgical ICU

• We are examining the incidence of acute kidney injury among SICU patients who received either vancomycin (+) zosyn, vancomycin (+) cefepime, vancomycin (+) levofloxacin, and vancomycin (+) meropenem. I would like help on the best statistical tests to conduct to detect differences among the four groups.
  
  • Outcome: VICTR Voucher

Ryan Brown, Allergy, Pulmonary and Critical Care Medicine

• We are comparing the use of central catheters and vasopressors 6 mo before and 6 mo after the institution of a protocol within the MICU to allow for peripheral administration of vasopressors. Primary outcome is time from ICU admission to vasopressor initiation. Secondary outcomes are safety outcomes. Cohort - patients receiving vasopressors within 24 hours of ICU admission.
  
  • Outcome: VICTR Voucher
  • Mentor confirmed

2018 November 28

Yuxi Zheng, Ophthalmology

• Outcomes for long-term followup of surgical correction of head positioning (ordinal variable) associated with infantile nystagmus syndrome

  Question: Best statistical test for evaluation of improvement of head positioning (ordinal variable) post-operatively and over time. Is ordinal logistic regression appropriate? I appreciate your help and guidance!

  My mentor was present during first joint meeting with James Law, Cathy Jenkins, and Li Wang on 11/7.

2018 November 14

Sabina Dang, Otolaryngology

• Social determinants of health in a population of airway stenosis patients. I would like to address multivariate analysis.
• Outcome: Other
• Mentor confirmed

James Law, Vanderbilt School of Medicine; Vanderbilt Eye Institute

• Atttended clinic 11/7, want to attend on 11/14 as follow up. Mentor will not attend (did attend 11/7)
• Outcome: Protocol with no expected funding support

2018 November 7

James Law, Vanderbilt School of Medicine; Vanderbilt Eye Institute

• Have collected data for medical student research project - hoping to present this as a retrospective interventional comparative case series, looking for general advice for analyzing the data.
• Outcome: Protocol with no expected funding support
• Mentor confirmed

2018 October 31

Rachel Forbes, Surgery-renal transplant

• Learning curve for vascular anastomosis based on times for various residents
• Outcome: Abstract

2018 October 10

Christine Helou, OBGYN

• Study looking at effect of BMI on endometrial hyperplasia/malignancy in premenopausal women. Can we identify a specific BMI cut off that would warrant biopsy. Study: retrospective cohort comparing obese premenopausal women who had endometrial sampling to non-obese premenopausal women. Would like to ID cases using SD. Question for clinic: is this feasible, help calculating sample size of cases needed.
• VICTR voucher

Judd Heideman, Internal Medicine

• Assistance with development of logistic regression to assess whether periprocedural hypoxemia causes in-hospital mortality, from database of 1000 hospital intubation procedures
• VICTR voucher

2018 October 3

Svetlana Avulova, Urology

• Obesity paradox and its impact on cancer related mortality has been previously observed
  • Obesity is associated with aggressive prostate cancer (increases risk of high-grade CaP) and worse long term outcomes
  • In localized prostate cancer, obesity is associated with greater risk of aggressive pathology and worse overall survival
  • In metastatic castrate resistant prostate cancer, obesity is associated with improved overall survival and prostate cancer specific survival (Halabi et al. 2007)
  • In nonmetastatic CRPC, it is associated with reduced all cause mortality (ie improved overall survival) but not prostate cancer specific mortality (Vidal et al. 2018)
  • In metastatic HSPC, BMI is associated with improved OS (Montgomery et al. 2007)
  We aimed to identify if sarcopenia (muscle mass < 5.5 cm2/m2) rather than obesity (BMI ≥ 30kg/m2) is associated with overall survival in men with mPCa/CRPC. Our goal is to elucidate whether muscle mass independently predicts overall survival by utilizing a validated software package provided by the Diet, Body Composition and Human Metabolism Core led by Dr. Silver to measure the muscle mass index. In addition, we will incorporate the Charlson Comorbidity Index in our multivariate analysis as patient comorbidity may predict overall survival regardless of muscle mass or BMI.
• VICTR voucher

2018 September 26

Rosemarie Dudenhofer, Medicine / Allergy, Pulmonary, and Critial Care

• We want to assess if there are benefits of positive airway pressure (pap) in the lung transplant population:

  • Primary Endpoint: whether or not pap decreases the incidence of transplant rejection and whether pap improves graft survival
  • Secondary Endpoint: Prevalence of OSA in the lung transplant population
    Questions: how many patients should we plan to enroll in the 3 arms:
    a. no OSA with no pap tx (control)
    b. no OSA with pap tx
    c. OSA with pap tx
    Question: ideal length of follow up to determine endpoints.
• VICTR voucher

Ehtesham Khalid, Neuromuscular Medicine

• Retrospective data analysis for IVIG and PLEX treatmentPrimary Endpoint: whether or not pap decreases the incidence of transplant rejection and whether pap improves graft survival

• VICTR voucher

2018 September 19

Lauren Schmidt, Pharmaceutical Services

• The objective of this retrospective review will examine if intravenous lipid emulsion(s) (ILE) from either propofol or a parenteral nutrition source is associated with adverse effects on outcomes in critically ill patients. We are interested in identifying the incidence of infection, hospital and ICU lengths of stay, ventilator free days and rates of mortality between patients who received ILE versus patients who did not receive ILE.
• VICTR voucher

Chi Le, OB/Gyn

• The project aims to identify drivers of 30 days unplanned re-admissions after laparoscopic hysterectomies. I have done some preliminary analysis in R, and we would like to receive some input from Biostatistics. I will send the R code and de-identified data to the clinic email a few days in advance of the meeting.
• VICTR voucher

2018 September 12

Timothy Hopper, Critical Care Medicine

• Proof-of-concept for new medical device, seeking to compare device measurements to clinical gold standard for intravascular volume status: PCWP during RH catheterization.
• Outcome: abstract

Rob Freundlich, Anesthesiology

• K23 grant application
• Outcome: grant

2018 August 22

Lindsey Safley, Emergency Medicine/ Pharmacy

• What are the safety and efficacy outcomes of patients discharged from the ED on apixaban for VTE in a real-world setting?
• Retrospective study of patients discharged from the ED from June 2016 –June 2018
• VICTR voucher

Ryan Stark, Pediatric Critical Care

• We have submitted a small descriptive paper on 15 subjects that suggests a novel finding. The reviewer requested a “power calculation” even though this is not a prospective study. I would like some input on where they are just asking for us to provide an effect size of our finding and if there would be a way to do a reverse power calculator to estimate effect.

2018 August 15

Shayan Rakhit, Trauma

• We are evaluating educational outcomes of a School of Medicine course [ISC: Injury, Repair, Rehabilitation] that primarily deals with trauma. We have Likert scale before/after data for the course as well as numeric and qualitative assessments of the students. We would like to meet in order to develop an analysis plan.
• VICTR voucher

2018 August 8

Nicholas Kavoussi , Urology

• Retrospective study of SD data to identify patients who have had kidney stone surgery and determine whether routine imaging is helpful in preventing additional stone events as well as optimal modality.
  PI for project Ryan Hsi met with Dr. Shyr to discuss the above.
• Have discussed the project. The aim is to look at association between post-op imaging the 5-year recurrence after kidney stone removal surgery.
• Recurrence is defined as ER visits with pre-defined CPT codes, or ER visits with documentation of "stone", or surgery including pre-defined CPT codes.
• Want to compare imaging pattern (frequency and timeing) between patients with and without recurrence.
• $5000 biostatistics voucher is suggested.

2018 August 8

Maya Yiadom, Emergency Medicine

• I’m looking for the most appropriate way to analyze concordance (or disagreement) in a sample of 69 patients. We collected EKG results (a test) for each of these patients and documented the physician’s clinical interpretation (ResGMO _Read_Final) using a coded rubric (ResGMO _Read_Coded). We then had experts review the same EKGs (VUMC_Attd_Final) and provide an interpretation as part of a quality assurance pilot using the same rubric (Att_Final_Coded). The coding rubric includes 47 unique diagnoses coded (1-47). From manual assessment in excel, we found the interpretations disagreed 49.3% of the time, but are looking for a test statistic to replicate this that will give us an estimate of variation (SD) or confidence (CI).

2018 August 1

David Isaacs, Neurology

• Deep brain stimulation (DBS) is an FDA-approved treatment to address uncontrolled motor symptoms in Parkinson’s disease (PD). Vanderbilt University Medical Center is one of the highest-volume DBS centers in the country. From 2007 through October 2017, 265 Parkinson’s disease patients underwent DBS implantation at Vanderbilt, with electrodes placed in one of two anatomical targets: 168 in STN and 97 in GPi. Pre-operatively, each patient is extensively evaluated with a battery of validated motor, cognitive, and mood instruments. The majority of these patients continue following with their Vanderbilt neurologist. In an attempt to capture longitudinal outcomes in this population of interest, we will recruit all PD patients two years or more status post DBS who are receiving regular care at Vanderbilt University Medical Center. Study participants will undergo a condensed evaluation of motor function (Unified Parkinson’s Disease Rating Scale Part III), cognitive performance (Montreal Cognitive Assessment), mood (Beck Depression Inventory), and quality of life (Parkinson’s Disease Questionnaire-39). These results will be compared to baseline measures performed pre-operatively, allowing for assessment of interval change. STN and GPi DBS patients will be analyzed separately. Goals of attending the Biostatistics Clinic are as follows:
  – appropriateness of selected statistical methods, as enrolled patients will have variable follow-up time
  – guidance for developing a regression model to address confounding variables (disease duration, age of surgery, etc)
  – methods for addressing baseline differences between comparison groups (STN and GPi groups)
  – feasibility / reasonableness of VICTR voucher for ongoing statistical assistance
• Design complete but no enrollment/data collection
• Has a total ~250 patients who underwent DBS surgery. Estimated ~150 patients to be enrolled in the study.
• The goal of the study is to compare long term longitudinal outcomes between two groups.
• $5000 VICTR biostat support is suggested.

Ashley Nassiri, Otolaryngology

• We suspect that there are specific proliferative factors that can be identified in vestibular schwannoma histology that can be indicative to future growth rates thus impacting the clinical decision to radiate postoperatively. This project involves a retrospective review of pathology slides (with new staining for proliferative factor) correlated with growth noted on imaging postoperatively.
• Questions for session: 1) Power analysis for estimate of needed participants & planning for statistical analysis. 2) Stats budgeting for funding request.
• Tumor cells were not removed completely to avoid nerve damage. Some patients will undergo radition after sugery. However, radition can cause complications.
• The goal of this study is to look fro clinical factors that can predict tumor growth, which can help decide whether the patient should be given radition or not.
• MRI was performed prior surgery, immediately after surgery, 1 year after and 2 year after surgery. The left behind tumor growth can be assessed from MRI scans.
• The interested marker is Ki-67 index. Can fit separate linear models for 1 year and 2 year post surgery.
• $5000 VICTR biostat support is suggested.

Robert Yawn, Otolaryngology

• Same patient cohort as above, but look at facial function at 1 year post surgery. Facial function is on a 1-6 scale, with 1 means normal and 6 means complete paralysis.
• The interested clinical factor is time the patient spent while glucose falls in a certain range. Other confounders are available including tumor size, age, commorbidity, etc.
• Suggested ordinal logisitc regression model.
• $5000 VICTR biostat support is suggested.

2018 July 25

Diane Haddad, Surgery

• Health inequity in the US is seen and documented across all healthcare fields with advances in medicine not experienced by all racial and ethnic groups. Outcomes in critical illness are no exception with documented disparities existing in severity, mortality and hospitalization rates attributed to race, access and poverty. We designed a study, using the BRAIN-ICU cohort to evaluate effects of socioeconomic and insurance status on in-hospital delirium and long-term cognitive impairment. Each participant in the BRAIN-ICU cohort had a socioeconomic score geocoded using census data and zip code. We similarly have data on insurance status and post-hospital disposition. We plan to look at effects of socioeconomic status, insurance and post-critical illness disposition on duration of in-hospital delirium and long-term cognitive impairment. Other covariates we plan to include in our model include age, race, sex, education level, co-morbidities (Charlson, Framingham), disease severity (APACHE, SOFA, sepsis, hypoxemia, coma), benzos/opioid equivalents, EtOH abuse. Our hypothesis is that socioeconomic and insurance status will have no significant effect on in-hospital delirium but a significant effect on LTCI and post-hospital disposition.

2018 July 11

Kala Dixon, Surgery

• The study purpose will consists of participants undergoing RYGB Surgery at the surgical weight loss clinic at Vanderbilt. To cover surgical costs, Vanderbilt accepts Aetna, Cigna, and Blue Cross Blue Shield. Aetna requires pre-op patients to go through a program that is interprofessionally driven in which they meet with the surgeon/nurse practitioner/ and dietitian. They meet with this team once a month every 90 days before surgery begins. BlueCross Blue Shield only require 6 months of uniprofessional intervention, prior to surgery, consisting of only a primary care office visit. Therefore, once IRB approval has been obtained, I will receive that list of patients, contact them, and ask, via telephone, if they would be interested in partaking in a telephone survey (short form 36 and a diet habit survey). This initial contact will be the baseline survey results. Following the telephone survey they will undergo their intervention method, and will be re-contacted right before surgery begins to re-take the survey. The goal is to compare the quality of life survey results between the 2 groups (control group:uniprofessional patients who are only required a pre-op primary office care visits and compare them to the experimental group of those who undergo the interprofessional intervention with a surgeon/nurse practitioner/ dietician). With the aim of the project being to compare the two groups to see if there was a change in results in one’s quality of life between those who undergo Vanderbilt’s interprofessional intervention pre-operatively to those who only undergo uniprofessional intervention pre-operatively.
• Questions:
• 1) How to conduct the power of analysis to determine the number of participants needed to conduct the project.
• 2) The current statistical methods that will be proposed to analyze the results are: t-test, multiple linear and logistic regression. Are these the best methods to analyze comparative results and outcome improvements?
• 3) What statistical data do we need to collect to determine study significance?
• Design complete but no enrollment/data collection

2018 June 27

James Patrinely, Plastic Surgery

• Our prospective, randomized controlled trial compared pain outcomes using the visual analog scale (1-10) for two types of injections for trigger finger (steroid + saline vs steroid + lidocaine). We have completed patient enrollment and need recommendations for final data analysis.
• Study design is non-inferiority.
• Data have been unblinded and reviewed, final analysis not complete. Data are not normally distributed. Suggest use of median and IQR in table 1.
• Discussed if there is a need to revise power calculation, given that data collected have a different SD than was estimated. There is no need to revise, but could discuss implications of a larger sample in the manuscript.
• Suggest Wilcoxen test for primary outcome and Fishers (or Chi-Square) for secondary outcome of treatment efficacy.
• Suggest visual display of data in a strip chart.

2018 June 13

Justin Shinn, Otolaryngology

• "Endoscopic evaluation of critically ill patients who are extubated in the ICU to determine if they have acute laryngeal trauma. Binary outcome, also with ordinal values within acute laryngeal injury. Phone follow up data at 3 months using a voice and breathing survey to begin assessment of longterm outcomes. Data collection is complete"
• Justin's primary question of interest is describing the incidence of laryngeal trauma in patients who were intubated. We discussed providing an overall estimate as well as among those who were intubated multiple times.
• We also discussed logistic regression (gives odds ratios) versus a modified Poisson model (gives relative risks) for investigating the risk factors of trauma. We discussed creating a list of clinically important predictors and ranking them in order of importance. The complexity of the model that can be fit is driven by the number of non-trauma patients in this case (~40 without trauma vs ~60 with trauma). The rule of thumb is 1 parameter can be estimated for every 10-20 events. For continuous variables without splines and dichotomous variables, 1 parameter is estimated. If splines are fit with continuous variables, then # knots - 1 are the number of parameters estimated for that variable. For categorical variables with more than 2 levels, # levels - 1 parameters are estimated.
• For univariate tests of association, the Wilcoxon Rank Sum test is preferred over a t-test for continuous variables. For categorical variables, the Pearson chi-square test is appropriate.
• For particular types of injury, rather than fitting some kind of model, it was suggested to graphically display or illustrate in tables the distributions.

2018 June 6

Jeff Heimiller, Emergency Medicine

• "We built a cricothyrotomy model that we used to test our EM residents on. We wanted to see if it could differentiate between the novice and more experienced residents."

2018 June 6

Austin Adair, Pediatric Critical Care

• "We need help in initiating a study involving pre- and post-utilization of electrolyte monitoring using a specific machine for retrospective study. We need help determining effectiveness based off of multiple patient variables."

2018 May 23

Lara Harvey, OB/GYN

• "VICTR voucher."

2018 May 16

Nicola White, OB/GYN

• "Comparison of outcomes for women who have primary c sections vs higher order lacerations after delivery. We would like assistance with creating a case matched comparative group."
• Women with 1st pregnancy are eligible. There are ~10,000 women with C-section and ~500 women with higher order lacerations. Primary outcome is depression scale and want to compare depression between two groups.
• Chart review is effort consuming and are planning propensity matching on the 500 cases. Will check to see what available information will have for the matching.

2018 April 25

Joy Carroll, Ophthalmology

• "I have a dataset with pre-operative factors and post-operative outcomes for cataract surgery. I would like to compare several measures and outcomes between a group taking a type of medication and everyone not taking the medication to determine whether the outcomes vary. I used a student T test for a smaller version of this study that looked at one factor but not the overall outcomes. I would like to discuss which statistical tests to use and what programs may be acceptable (i.e. may I continue to use excel, or would I need to learn R?)"

2018 April 18

Luis Huerta, Pulmonary/Critical Care

• "I am attempting to validate an electronic severity of illness calculator (SOFA score) versus the gold standard of manual chart review. There are 6 components to the SOFA score, and each is on an ordinal scale of 0-4 (for a total of 24 possible points). My primary question for the clinics relates a secondary analysis: I am attempting to determine the best statistical method to use to compare the agreement between manual and electronic collection of the six individual components, as 1-2 components are likely to be more difficult to collect electronically. Data colelction is complete."

James Zhang, Surgery

• "Our project investigates the incidence of new onset orthostatic intolerance after bariatric surgery. We are also interested in whether onset of these symptoms is related to weight loss."

2018 April 4

Muhammad Aanish Raees, Pediatric Cardiac Surgery

• "We are applying for a VICTR grant for the project titled: Evaluation of different patching materials for vascular reconstruction in surgical repair of congenital heart disease and clinical outcomes. I would like to discuss the statistical plan for this study, and how many statistical hours this study will require."

2018 April 4

Andrew Medvecz, General surgery

• Previous clinic notes
• "The project is a retrospective cohort study evaluating long-term outcomes (readmission in particular) of small bowel obstruction, comparing operative versus non-operative management. We plan to apply to VICTR for biostatistical voucher support as well as data support. We would like biostatistical support on the project and are hoping to meet with biostatisticians that would be associated with the analysis if VICTR funding is granted. In particular, we would like assistance with developing an analysis plan."
• One suggestion: multi-state models/competing risks, because patients can move between surgical and non-surgical management if they have recurring SBO (which many patients do, in highly varying numbers); could also do survival analysis with recurring events
• Hospitalizations and surgeries can be recurring events (at different hospitals, though having THA data will be very helpful here)
• Max of ten years followup on ~15,000 SBOs (2007-2009); estimate 3:1 nonoperative vs operative (no data yet on individual patients, but is coming)
• This is a challenging statistical problem; Li Wang and Chris Lindsell (VICTR statisticians) will work on analysis plan for VICTR application
• Second aim: for an individual hospitalization, how likely are patients to be readmitted? other outcomes? This gets extra complicated because of not only bounce-backs, but "bounce-aways" who go to a different hospital

2018 March 21

Katherine Riera, General surgery

• "Peds trauma QI project. Will be large database before/after QI interventions. Will be patient demographics, outcome measures, and trauma activation information. Also will be Redcap staff survey data. Planning to apply for VICTR funding and needed to come to a clinic prior."
• Retrospective data: possible to collect data on level I/II trauma cases, 2015-2018 (~1500 total, with ~360 level I); main outcomes are times to events (time in trauma bay, time to intubation, etc), which will need to be extracted by hand. Recommend getting data from roughly the same seasonal time period as prospective data (eg, April-October both years), since trauma rates vary a lot seasonally.
• Mortality rate is unknown but low
• Interventions (all done together, "on" clinical staff, not patients): heads up, prebrief, wall poster with relevant components; will begin after IRB approval, then post-intervention data collection will begin a month later (likely ~280 patients in those six months)
• Times are available both via trauma flowsheet and video recordings
• 15-question pre-intervention survey to ~150 staff will be done once IRB approves study (before intervention); post survey done later; goal is to describe staff satisfaction with "how things are going" for various types of patients (some questions are VAS; demographics are categorical). Shoot for 90% response rate; might want to do a simple random sample of staff population to be able to better target followup efforts.
• Send current protocol to li.wang@vanderbilt.edu; Li (VICTR statistician) can help draft a statistical analysis plan for application
• Plan to apply VICTR voucher for biostatistical support. $5000 is suggested.

Josh Latner, Emergency Medicine/Medical Student

• "Survey of medical students on their opinions/attitudes towards point-of-care ultrasound. I would like to discuss the best ways to create a summary score for a series of Likert-type questions and ways to validate it." (mentor will be attending; data collection completed in REDCap)
• Creating a summary statistic won't be very helpful here - what does a "score" of 10 mean? Validating an instrument will be a long process; not within the scope of the descriptive study.
• Response rate about 56%; good to provide information about/compare nonresponders to survey responders (whatever is available) to help determine whether there is likely to be bias in the observed results vs actual views of population.
• Stacked barcharts can be hard to compare across groups (if comparison is what the goal of the visualization is); consider side-by-side barcharts instead ( position_dodge in ggplot2)

2018 March 7

Jackson Cabo, medical student/Urology

• "I am working on a quality improvement study looking at post-operative disposal of opiates as well as prescribing practices. I have already performed preliminary analysis, consisting primarily of Fisher’s exact tests and Kruskall Wallis tests. I only require verification that my code in R looks correct as we plan to submit an abstract soon."
• Rather than binary classification, preserve as much information as possible. eg, rather than "keeper" vs "non-keeper," use "used all of prescription," "disposed of leftovers properly," "kept leftovers."
• Descriptive statistics for, say, morphine equivalents left over after postop, will give you a lot of bang for your buck.
• Consider interrupted time series for continuing PDSA interventions.
• No p-hacking

Christodoulos Kaoutzanis, Plastic Surgery

• "The purpose of this study is to compare the different autologous fat grafting techniques Telfa Rolling and REVOLVE System in patients undergoing postmastectomy breast reconstruction. Data has been collected and analyzed but we would like to do a multivariate analysis to identify risk factors for 2 of our outcomes and we need assistance with that analysis."
• Primary outcomes = differences in number of times surveillance imaging performed (secondary/exploratory), amount of fat necrosis
• Techniques are not randomized; Telfa was used most commonly earlier, REVOLVE more common over the last few years, but Telfa still used sometimes (mostly chosen by surgeon preference)
• All cases are from a single surgeon, which could confound results (surgeons hopefully perform better over time; does REVOLVE look better just because of surgeon experience?). This is a straightforward procedure and the surgeon had prior experience, so hopefully not a huge factor, but definitely a limitation.
• Data is currently recorded as number of breasts (~188 pre, 131 post) over four years; important to recognize that there are some within-patient factors that will lead to correlation between the same patient's breasts
• Plan to apply for VICTR voucher; recommend summarizing project (mini-protocol; manuscript shell even better) and contacting Chris Lindsell/Li Wang with VICTR biostats to work on analysis plan prior to voucher application
• Chris suggested interrupted time series

2018 February 28

Aaron Bolduc, surgery/GI-Lap -- Cancelled

• "Followup for my statistics results based on the 2/21/18 clinic plan."

2018 February 21

Aaron Bolduc, surgery/GI-Lap

• "Bowel length study on outcomes for bariatric surgical patients. Does bowel length correlate with height, weight, age, sex, or race? Are the groups of data recorded by two different observers homogeneous?"
• Four surgeons measured bowel length during separate procedures; total N is currently about 300. Two surgeons account for vast majority of cases.
• Will not be able to discuss in terms of interrater reliability/agreement, since only one surgeon measured each bowel. Goal is to see if distributions are roughly equivalent between the surgeons.
• All gastric bypass patients are currently treated the same way - bypassing 150cm (for patients with BMI<45) or 200cm (others) of small bowel. Eventual goal (much later) is to determine whether outcomes would improve if amount of small bowel bypass is specified by patient characteristics. Intermediate goal (later) is to determine whether bowel length is associated with surgery outcomes under current practice. Current goal: is bowel length associated with baseline/demographic characteristics?
• Suggest multivariable regression model with bowel length (which is normally distributed, wonder of wonders) vs patient characteristics of interest listed above.
• Also interest in describing characteristics of two surgeons' patient populations; hope is to be able to state that patient characteristics don't differ strongly by surgeon.
• Plan is to apply for VICTR voucher; this will fit within the 90-hour time frame, and plan is for a manuscript. Strongly suggest creating a data dictionary to go along with deidentified data, including variable definitions, coding (eg 0 = male, 1 = female), other variable info. (No time for VICTR prior to next Thursday's abstract deadline, but feasible for June conference/manuscript.)
• If there is strong interest in interrater reliability, could do a small subsample (10-20%?) of patients who get measured by both surgeons; this adds time and logistical complexity. Important to make sure each surgeon is blinded to the other's measurements.

2018 February 14

Chelsea Isom, resident, General Surgery

• "Retrospective chart review trying to answer the question: Do ipsilateral central venous ports in breast cancer increase the risk of complications? Question: I wanted to make sure that I did my power calculations in the PS software correctly."

2018 February 7

Christin Giordano, resident, Department of Medicine

• "We will be trialing a new rounding method on 2 of the 5 general internal medicine resident teams and measuring burnout as well as percent early discharges and conference attendance (comparing the intervention teams vs. those with usual rounding plans). We can bring our survey questions but one is the Masloch burnout scale which has its own scoring. We would like help developing a statistical plan. Ultimately, we would like to qualify for a voucher for VICTR funding for biostatistics and Dr. Lindsell said this was the first step."
• Total of ~60 people, 24 in intervention teams and 36 in control teams.
• Recommend trying to do burnout scale at beginning and end of rotation (two-week period); depending on barriers, may or may not work out well, but if cost is acceptable, doesn't hurt to try and could definitely help when analyzing and discussing results/limitations.
• Primary analysis should not categorize subscores; this will result in less power and ability to see a difference between groups.
• Potential confounders: prior rotation (but in such a small group, this could potentially identify people - get as much detail as possible while preserving anonymity)
• Plan to collect data in REDCap; suggest REDCap clinic to help with survey-specific design questions
• "Early discharge" will be measured only for patients who are discharged alive from the hospital (patients who are transferred, etc will not count toward this)
• For burnout scale, it'll depend on the distribution of the data, but nonparametric Wilcoxon test is likely a good idea (or proportional odds logistic regression for multivariable analysis, adjusting for baseline, if possible)
• This should fit into a standard 90-hour VICTR project

2018 January 24

Jackson Cabo, medical student

• "From my prior statics clinic visit: "“I am doing a project with Dr. Bailey (Surgical Oncology) regarding healthcare disparities in colorectal cancer and how they may vary depending on setting of care (i.e. type of treating hospital). A Cox Proportional Hazard Regression will be used to estimate the effects of race, income status, and insurance status in the context of hospital facility (type) on overall survival.”
• This will be my second visit to statistics clinic for assistance with my project. I would like some assistance with preliminary data analysis in order to aid in my preparation for presenting my work at the U54 Cancer disparities symposium at the beginning of next month. I would like assistance with selection of the appropriate statistical test (ideally in R) for a couple of comparisons I am making between median survival depending on race, insurance status, and facility type. I also had some questions as to which tests I should perform to accompany Kaplan-Meier curves. I will attach the document with relevant comparisons to an email and send it to the statistics clinic staff."

2018 January 24

Benjamin Weisenthal, orthopedic resident

• "I have NSQUIP data from 2005-2013. For each year, I would like to pull a certain group of patients with certain ICD-9 codes and then group these patients by CPT codes. I have the NSQUIP data in SPSS but just need help extracting the information."

Joshua Arenth, Pediatric Critical Care

• "Addressing optimal data formatting for analysis re: previously discussed project."

2018 January 17

Andrew Medvecz, general surgery/trauma resident

• "The project is a retrospective cohort study comparing operative vs nonoperative management of small bowel obstruction. The primary outcomes are readmission with small bowel obstruction and cost associated with the all the subsequent admissions. We are applying for VICTR funding and would like to discuss our statistical plan with the biostatistician. Oscar Guillamondegui is the mentor for this project and will be in attendance."
• Estimated recurrence rate for nonsurgical patients is about 30-35% within two years; data more long-term is hard to ascertain due to usual factors + patient movement between physicians/locations
• This study will use data from the Tennessee Hospital Association, so is better able to track more patients even between hospitals (exceptions: Indian hospitals, VAs); this data often used for business purposes, but can be helpful for longitudinal studies
• Uses ICD9/10 to determine reason for admission (we are interested in three, any of which identify this cohort)
• Plan is to take cohort with first admission for small bowel obstruction in 2007, 2008, 2009; send those patient IDs to THA; THA will provide deidentified dataset of future admissions for any reason, CPT codes and charges for admission-related procedures, as well as death dates and some other info (eg, hospital type)
• Nonsurgical management should be pretty consistent between patients
• Analysis approaches for primary outcome could include a cumulative incidence model (Bryan Shepherd has experience with this), Cox models with recurring events [readmissions]; have data on confounders (age, comorbidities)
• Given the nature of the database, we're hopeful that data will be in good shape from the beginning (CFOs/CEOs use this), sent in a CSV
• Also interested in cumulative costs over study time, acknowledging issues with charges vs costs, different charges for different patients...
• For primary outcome plus straightforward cost descriptives, this should fit into a standard 90-hour VICTR voucher. (If cost data gets much more complicated or interesting, might recommend another voucher or talking with health policy folks.)

Mitchell Hayes, medical student

• "Our question is whether or not cancer detection rates change over time adjusting for the individual radiologists who read the MRIs and individual urologists who performed the biopsies. We would like to create a logistic regression model in R that analyzes cancer detection rate as a function of at least two interaction variables: urologisttime(1) + radiologisttime(2). Time(1) and time(2) are two different continuous variables (time in this case is more like experience with MRI-US fusion biopsy). Urologist and radiologist are two categorical variables. I need assistance determining whether we have sufficient power and analyzing the output in R."
• Dataset of ~350 patients, January 2015-July 2016, some done with new method and some with previous standard of care; 200 are active surveillance, meaning they've already had one positive biopsy, so final sample size will be ~150 patients with no known signs of cancer
• There is a recognized learning curve for radiologists with this technology such that cancer detection rate goes up over time; suspect that urologists also have a learning curve
• Don't currently know for sure whether a patient has cancer or not; best proxy is the fusion biopsy result, but best we can do is "were there more/less positive biopsies," not "did things get more/less accurate", because we don't know which ones are false negatives/false positives
• Degrees of freedom for logistic regression model: Total df you can reliably fit = 1 for every 10-20 [limiting sample size], where limiting sample size is the minimum of [events, non-events]. Eg, with 100 patients and a 30% positive biopsy rate, limiting sample size is 100 * 0.3 = 30, so number of df you can reliably fit is ~2-3.
• Recommendation for this study: keep things straightforward; there will always be more questions you want to answer, but trying to do too much will limit the reliability of your results (don't trust a model that's very overfit) as well as the generalizability (this model perfectly predicts results at VUMC for these 18 months, but once a new radiologist or two comes on board, results no longer apply).