Tom Ten Have's contributions to causal inference and biostatistics: review and future research directions.

Tom Ten Have made many contributions to causal inference and biostatistics before his untimely death. This paper reviews Tom's contributions and discusses potential related future research directions. We focus on Tom's contributions to longitudinal/repeated measures categorical data analysis and particularly his contributions to causal inference. Tom's work on causal inference was primarily in the areas of estimating the effect of receiving treatment in randomized trials with nonadherence and mediation analysis. A related area to mediation analysis he was working on at the time of his death was posttreatment effect modification with applications to designing adaptive treatment strategies.

Development and validation of primary graft dysfunction predictive algorithm for lung transplant candidates.

BACKGROUND: Primary graft dysfunction (PGD) is the leading cause of early morbidity and mortality after lung transplantation. Accurate prediction of PGD risk could inform donor approaches and perioperative care planning. We sought to develop a clinically useful, generalizable PGD prediction model to aid in transplant decision-making. METHODS: We derived a predictive model in a prospective cohort study of subjects from 2012 to 2018, followed by a single-center external validation. We used regularized (lasso) logistic regression to evaluate the predictive ability of clinically available PGD predictors and developed a user interface for clinical application. Using decision curve analysis, we quantified the net benefit of the model across a range of PGD risk thresholds and assessed model calibration and discrimination. RESULTS: The PGD predictive model included distance from donor hospital to recipient transplant center, recipient age, predicted total lung capacity, lung allocation score (LAS), body mass index, pulmonary artery mean pressure, sex, and indication for transplant; donor age, sex, mechanism of death, and donor smoking status; and interaction terms for LAS and donor distance. The interface allows for real-time assessment of PGD risk for any donor/recipient combination. The model offers decision-making net benefit in the PGD risk range of 10% to 75% in the derivation centers and 2% to 10% in the validation cohort, a range incorporating the incidence in that cohort. CONCLUSION: We developed a clinically useful PGD predictive algorithm across a range of PGD risk thresholds to support transplant decision-making, posttransplant care, and enrich samples for PGD treatment trials.

Two-Year Results of Think Health! ¡Vive Saludable!: A Primary Care Weight-Management Trial.

OBJECTIVE: Think Health! ¡Vive Saludable! evaluated a moderate-intensity, lifestyle behavior-change weight-loss program in primary care over 2 years of treatment. Final analyses examined weight-change trajectories by treatment group and attendance. METHODS: Adult primary care patients (n = 261; 84% female; 65% black; 16% Hispanic) were randomly assigned to Basic Plus (moderate intensity; counseling by primary care clinician and a lifestyle coach) or Basic (clinician counseling only). Intention-to-treat analyses used all available weight measurements from data collection, treatment, and routine clinical visits. Linear mixed-effects regression models adjusted for treatment site, gender, and age, and sensitivity analyses evaluated treatment attendance and the impact of loss to follow-up. RESULTS: Model-based estimates for 24-month mean (95% CI) weight change from baseline were -1.34 kg (-2.92 to 0.24) in Basic Plus and -1.16 kg (-2.70 to 0.37) in Basic (net difference -0.18 kg [-2.38 to 2.03]; P = 0.874). Larger initial weight loss in Basic Plus was attenuated by a ~0.5-kg rebound at 12 to 16 months. Each additional coaching visit was associated with a 0.37-kg greater estimated 24-month weight loss (P = 0.01). CONCLUSIONS: These findings in mostly black and Hispanic female primary care patients suggest that strategies to improve treatment attendance may improve weight loss resulting from moderate-intensity counseling.

Automated identification of implausible values in growth data from pediatric electronic health records.

OBJECTIVE: Large electronic health record (EHR) datasets are increasingly used to facilitate research on growth, but measurement and recording errors can lead to biased results. We developed and tested an automated method for identifying implausible values in pediatric EHR growth data. MATERIALS AND METHODS: Using deidentified data from 46 primary care sites, we developed an algorithm to identify weight and height values that should be excluded from analysis, including implausible values and values that were recorded repeatedly without remeasurement. The foundation of the algorithm is a comparison of each measurement, expressed as a standard deviation score, with a weighted moving average of a child's other measurements. We evaluated the performance of the algorithm by (1) comparing its results with the judgment of physician reviewers for a stratified random selection of 400 measurements and (2) evaluating its accuracy in a dataset with simulated errors. RESULTS: Of 2 000 595 growth measurements from 280 610 patients 1 to 21 years old, 3.8% of weight and 4.5% of height values were identified as implausible or excluded for other reasons. The proportion excluded varied widely by primary care site. The automated method had a sensitivity of 97% (95% confidence interval [CI], 94-99%) and a specificity of 90% (95% CI, 85-94%) for identifying implausible values compared to physician judgment, and identified 95% (weight) and 98% (height) of simulated errors. DISCUSSION AND CONCLUSION: This automated, flexible, and validated method for preparing large datasets will facilitate the use of pediatric EHR growth datasets for research.

Central Venous Catheter Retention and Mortality in Children With Candidemia: A Retrospective Cohort Analysis.

BACKGROUND: Candidemia causes significant morbidity and mortality among children. Removal of a central venous catheter (CVC) is often recommended for adults with candidemia to reduce persistent and metastatic infection. Pediatric-specific data on the impact of CVC retention are limited. METHODS: A retrospective cohort study of inpatients <19 years with candidemia at the Children's Hospital of Philadelphia between 2000 and 2012 was performed. The final cohort included patients that had a CVC in place at time of blood culture and retained their CVC at least 1 day beyond the blood culture being positive. A structured data collection instrument was used to retrieve patient data. A discrete time failure model, adjusting for age and the complexity of clinical care before onset of candidemia, was used to assess the association of CVC retention and 30-day all-cause mortality. RESULTS: Two hundred eighty-five patients with candidemia and a CVC in place at the time of blood culture were identified. Among these 285 patients, 30 (10%) died within 30 days. Central venous catheter retention was associated with a significant increased risk of death on a given day (odds ratio, 2.50; 95% confidence interval, 1.06-5.91). CONCLUSIONS: Retention of a CVC was associated with an increased risk of death after adjusting for age and complexity of care at candidemia onset. Although there is likely persistence of unmeasured confounding, given the strong association between catheter retention and death, our data suggest that early CVC removal should be strongly considered.

Variation in Antibiotic Prescribing Across a Pediatric Primary Care Network.

BACKGROUND: Outpatient respiratory tract infections are the most common reason for antibiotic prescribing to children. Although prior studies suggest that antibiotic overuse occurs, patient-specific data or data exploring the variability and determinants of variability across practices and practitioners is lacking. METHODS: This study was conducted from a retrospective cohort of encounters to 25 diverse pediatric practices with 222 clinicians, from January 1 to December 31, 2009. Diagnoses, medications, comorbid conditions, antibiotic allergy, and demographic data were obtained from a shared electronic health record and validated by manual review. Practice-specific antibiotic prescription and acute respiratory tract infection diagnosis rates were calculated to assess across-practice differences after adjusting for patient demographics and clustering of encounters within clinicians. RESULTS: A total of 102 102 (28%) of 399 793 acute visits by 208 015 patients resulted in antibiotic prescriptions. After adjusting for patient age, sex, race, and insurance type, and excluding encounters by patients with chronic conditions, antibiotic prescribing by practice ranged from 18% to 36% of acute visits, and the proportion of antibiotic prescriptions that were broad-spectrum ranged from 15% to 58% across practices, despite additional exclusion of patients with antibiotic allergies or prior antibiotic use. Diagnosis of (Dx) and broad-spectrum antibiotic prescribing (Broad) for acute otitis media (Dx: 8%-20%; Broad: 18%-60%), sinusitis (Dx: 0.5%-9%; Broad: 12%-78%), Streptococcal pharyngitis (Dx: 1.8%-6.4%; Broad: 2%-30%), and pneumonia (Dx: 0.4%-2%; Broad: 1%-70%) also varied by practice (P < 0.001 for all comparisons). CONCLUSIONS: Antibiotic prescribing for common pediatric infections varied substantially across practices. This variability could not be explained by patient-specific factors. These data suggest the need for and provide high-impact targets for outpatient antimicrobial stewardship interventions.

Racial disparities in emergency department length of stay for admitted patients in the United States.

OBJECTIVES: Recent studies have demonstrated the adverse effects of prolonged emergency department (ED) boarding times on outcomes. The authors sought to examine racial disparities across U.S. hospitals in ED length of stay (LOS) for admitted patients, which may serve as a proxy for boarding time in data sets where the actual time of admission is unavailable. Specifically, the study estimated both the within- and among-hospital effects of black versus non-black race on LOS for admitted patients. METHODS: The authors studied 14,516 intensive care unit (ICU) and non-ICU admissions in 408 EDs in the National Hospital Ambulatory Medical Care Survey (NHAMCS; 2003-2005). The main outcomes were ED LOS (triage to transfer to inpatient bed) and proportion of patients with prolonged LOS (>6 hours). The effects of black versus non-black race on LOS were decomposed to distinguish racial disparities between patients at the same hospital (within-hospital component) and between hospitals that serve higher proportions of black patients (among-hospital component). RESULTS: In the unadjusted analyses, ED LOS was significantly longer for black patients admitted to ICU beds (367 minutes vs. 290 minutes) and non-ICU beds (397 minutes vs. 345 minutes). For admissions to ICU beds, the within-hospital estimates suggested that blacks were at higher risk for ED LOS of >6 hours (odds ratio [OR] = 1.42, 95% confidence interval [CI] = 1.01 to 2.01), while the among-hospital differences were not significant (OR = 1.08 for each 10% increase in the proportion of black patients, 95% CI = 0.96 to 1.23). By contrast, for non-ICU admissions, the within-hospital racial disparities were not significant (OR = 1.12, 95% CI = 0.94 to 1.23), but the among-hospital differences were significant (OR = 1.13, 95% CI = 1.04 to 1.22) per 10% point increase in the percentage of blacks admitted to a hospital. CONCLUSIONS: Black patients who are admitted to the hospital through the ED have longer ED LOS compared to non-blacks, indicating that racial disparities may exist across U.S. hospitals. The disparity for non-ICU patients might be accounted for by among-hospital differences, where hospitals with a higher proportion of blacks have longer waits. The disparity for ICU patients is better explained by within-hospital differences, where blacks have longer wait times than non-blacks in the same hospital. However, there may be additional unmeasured clinical or socioeconomic factors that explain these results.

Much ado about nothing: a comparison of the performance of meta-analytical methods with rare events.

For rare outcomes, meta-analysis of randomized trials may be the only way to obtain reliable evidence of the effects of healthcare interventions. However, many methods of meta-analysis are based on large sample approximations, and may be unsuitable when events are rare. Through simulation, we evaluated the performance of 12 methods for pooling rare events, considering estimability, bias, coverage and statistical power. Simulations were based on data sets from three case studies with between five and 19 trials, using baseline event rates between 0.1 and 10 per cent and risk ratios of 1, 0.75, 0.5 and 0.2. We found that most of the commonly used meta-analytical methods were biased when data were sparse. The bias was greatest in inverse variance and DerSimonian and Laird odds ratio and risk difference methods, and the Mantel-Haenszel (MH) odds ratio method using a 0.5 zero-cell correction. Risk difference meta-analytical methods tended to show conservative confidence interval coverage and low statistical power at low event rates. At event rates below 1 per cent the Peto one-step odds ratio method was the least biased and most powerful method, and provided the best confidence interval coverage, provided there was no substantial imbalance between treatment and control group sizes within trials, and treatment effects were not exceptionally large. In other circumstances the MH OR without zero-cell corrections, logistic regression and the exact method performed similarly to each other, and were less biased than the Peto method.

Properties of analysis methods that account for clustering in volume-outcome studies when the primary predictor is cluster size.

In recent years health services researchers have conducted 'volume-outcome' studies to evaluate whether providers (hospitals or surgeons) who treat many patients for a specialized condition have better outcomes than those that treat few patients. These studies and the inherent clustering of events by provider present an unusual statistical problem. The volume-outcome setting is unique in that 'volume' reflects both the primary factor under study and also the cluster size. Consequently, the assumptions inherent in the use of available methods that correct for clustering might be violated in this setting. To address this issue, we investigate via simulation the properties of three estimation procedures for the analysis of cluster correlated data, specifically in the context of volume-outcome studies. We examine and compare the validity and efficiency of widely-available statistical techniques that have been used in the context of volume-outcome studies: generalized estimating equations (GEE) using both the independence and exchangeable correlation structures; random effects models; and the weighted GEE approach proposed by Williamson et al. (Biometrics 2003; 59:36-42) to account for informative clustering. Using data generated either from an underlying true random effects model or a cluster correlated model we show that both the random effects and the GEE with an exchangeable correlation structure have generally good properties, with relatively low bias for estimating the volume parameter and its variance. By contrast, the cluster weighted GEE method is inefficient.

Protection of mammography screening against death from breast cancer in women aged 40-64 years.

OBJECTIVE: This study assessed the efficacy of community-based screening mammography in protecting against breast cancer death, asking whether age differences in efficacy persisted in the 1990s. METHODS: In a case-control study with follow-up, odds ratios (OR) were used to estimate the relative mortality rates from invasive breast cancer among women with at least one screening mammogram in the two years prior to a baseline reference date compared to non-screened women, adjusting for potential confounding. The multicenter population-based study included 553 black and white women diagnosed during 1994-1998 who died in the following five years, and 4016 controls without breast cancer. RESULTS: Efficacy for reducing the rate of breast cancer death within five years after diagnosis was greater at ages 50-64 years (OR = 0.47, 95% confidence interval (CI) 0.35-0.63) than at ages 40-49 (OR = 0.89, 95% CI 0.65-1.23), and greater among postmenopausal (OR = 0.45, 95% CI 0.33-0.62) than premenopausal women (OR = 0.74, 95% CI 0.53-1.04). Estimates of efficacy were conservative, as shown by sensitivity analyses addressing whether cancer was discovered by a screening mammogram, age at which screening was received, the length of the screening observation window, and years of follow-up after diagnosis. CONCLUSIONS: Despite the persistence of age differences in efficacy of mammography screening, with greater observed benefit for women aged 50-64 years, these findings support current screening recommendations for women 40-64 years old.

Use of photographs for the identification of pressure ulcers in elderly hospitalized patients: validity and reliability.

To evaluate the ability of research nurses to identify pressure ulcers, the authors assembled digital photographs of the skin of 160 consenting elderly patients (80% African American, 63% women). The series included 39 photos of pressure ulcers, 109 of normal skin, and 12 of other skin conditions, determined by consensus by two experts (D.J.M. and S.H.K.). Photos were packaged electronically into eight blocks of 20, with pressure ulcer prevalence ranging from 20% to 30% per block. The eight blocks were duplicated to create two sets of 160 photos each. Each of six raters (experienced clinical research nurses), working independently, evaluated the 320 photos as if each photo depicted a different patient. For analysis, the ratings were collapsed into binary determinations (any pressure ulcer vs. none). The overall sensitivity and specificity of the ratings were 0.97 (95% confidence interval: 0.94, 0.98) and 0.81 (95% confidence interval: 0.77, 0.86), respectively. Rater-specific prevalence (range: 31.8-47.5%) exceeded the true prevalence (24.4%). Inter- and intrarater reliability coefficients were 0.69 and 0.84, respectively. Trained research nurses can accurately classify pressure ulcers from photographs, even when patients are largely non-White and the photographs depict pressure ulcers spanning all pressure ulcer stages.