A Quantitative Bias Analysis Approach to Informative Presence Bias in Electronic Health Records.

Accurate outcome and exposure ascertainment in electronic health record (EHR) data, referred to as EHR phenotyping, relies on the completeness and accuracy of EHR data for each individual. However, some individuals, such as those with a greater comorbidity burden, visit the health care system more frequently and thus have more complete data, compared with others. Ignoring such dependence of exposure and outcome misclassification on visit frequency can bias estimates of associations in EHR analysis. We developed a framework for describing the structure of outcome and exposure misclassification due to informative visit processes in EHR data and assessed the utility of a quantitative bias analysis approach to adjusting for bias induced by informative visit patterns. Using simulations, we found that this method produced unbiased estimates across all informative visit structures, if the phenotype sensitivity and specificity were correctly specified. We applied this method in an example where the association between diabetes and progression-free survival in metastatic breast cancer patients may be subject to informative presence bias. The quantitative bias analysis approach allowed us to evaluate robustness of results to informative presence bias and indicated that findings were unlikely to change across a range of plausible values for phenotype sensitivity and specificity. Researchers using EHR data should carefully consider the informative visit structure reflected in their data and use appropriate approaches such as the quantitative bias analysis approach described here to evaluate robustness of study findings.

Racial Differences in Germline Genetic Testing Completion Among Males With Pancreatic, Breast, or Metastatic Prostate Cancers.

BACKGROUND: Germline genetic testing is a vital component of guideline-recommended cancer care for males with pancreatic, breast, or metastatic prostate cancers. We sought to determine whether there were racial disparities in germline genetic testing completion in this population. PATIENTS AND METHODS: This retrospective cohort study included non-Hispanic White and Black males with incident pancreatic, breast, or metastatic prostate cancers between January 1, 2019, and September 30, 2021. Two nationwide cohorts were examined: (1) commercially insured individuals in an administrative claims database, and (2) Veterans receiving care in the Veterans Health Administration. One-year germline genetic testing rates were estimated by using Kaplan-Meier methods. Cox proportional hazards regression was used to test the association between race and genetic testing completion. Causal mediation analyses were performed to investigate whether socioeconomic variables contributed to associations between race and germline testing. RESULTS: Our cohort consisted of 7,894 males (5,142 commercially insured; 2,752 Veterans). One-year testing rates were 18.0% (95% CI, 16.8%-19.2%) in commercially insured individuals and 14.2% (95% CI, 11.5%-15.0%) in Veterans. Black race was associated with a lower hazard of testing among commercially insured individuals (adjusted hazard ratio [aHR], 0.73; 95% CI, 0.58-0.91; P=.005) but not among Veterans (aHR, 0.99; 95% CI, 0.75-1.32; P=.960). In commercially insured individuals, income (aHR, 0.90; 95% CI, 0.86-0.96) and net worth (aHR, 0.92; 95% CI, 0.86-0.98) mediated racial disparities, whereas education (aHR, 0.98; 95% CI, 0.94-1.01) did not. CONCLUSIONS: Overall rates of guideline-recommended genetic testing are low in males with pancreatic, breast, or metastatic prostate cancers. Racial disparities in genetic testing among males exist in a commercially insured population, mediated by net worth and household income; these disparities are not seen in the equal-access Veterans Health Administration. Alleviating financial and access barriers may mitigate racial disparities in genetic testing.

Comparing survival in trial- versus routine-care advanced urothelial cancer patients on immune checkpoint blockade.

PURPOSE: Although recent trials involving first-line immune checkpoint inhibitors have expanded treatment options for patients with advanced urothelial carcinoma (aUC) who are ineligible for standard cisplatin-based chemotherapy, there exists limited evidence for whether trial efficacy translates into real-world effectiveness for patients seen in routine care. This retrospective cohort study compares differences in overall survival (OS) between KEYNOTE-052 trial participants and routine-care patients receiving first-line pembrolizumab monotherapy. METHODS: A routine-care patient cohort was constructed from the Flatiron Health database using trial eligibility criteria and was weighted to balance EHR and trial patient characteristics using matching-adjusted indirect comparisons. RESULTS: The routine-care cohort was older, more likely to be female, and more often cisplatin-ineligible due to renal dysfunction. ECOG performance status was comparable between the cohorts. Median OS was 9 months (95% CI 7-16) in the weighted routine-care cohort and 11.3 months (9.7-13.1) in the trial cohort. No significant differences between the Kaplan-Meier OS curves were detected (p = 0.76). Survival probabilities were similar between the weighted routine-care and trial cohorts at 12-, 24-, and 36- months (0.45 vs. 0.47, 0.31 vs. 0.31, 0.26 vs. 0.23, respectively). Notably, routine care patients had modestly lower survival at 3 months compared to trial participants (0.69 vs. 0.83, respectively). CONCLUSION: Our results provide reassurance that cisplatin-ineligible aUC patients receiving first-line immunotherapy in routine care experience similar benefits to those observed in trial patients.

Variation in second breast cancer risk after primary invasive cancer by time since primary cancer diagnosis and estrogen receptor status.

BACKGROUND: In women with previously treated breast cancer, occurrence and timing of second breast cancers have implications for surveillance. The authors examined the timing of second breast cancers by primary cancer estrogen receptor (ER) status in the Breast Cancer Surveillance Consortium. METHODS: Women who were diagnosed with American Joint Commission on Cancer stage I-III breast cancer were identified within six Breast Cancer Surveillance Consortium registries from 2000 to 2017. Characteristics collected at primary breast cancer diagnosis included demographics, ER status, and treatment. Second breast cancer events included subsequent ipsilateral or contralateral breast cancers diagnosed >6 months after primary diagnosis. The authors examined cumulative incidence and second breast cancer rates by primary cancer ER status during 1-5 versus 6-10 years after diagnosis. RESULTS: At 10 years, the cumulative second breast cancer incidence was 11.8% (95% confidence interval [CI], 10.7%-13.1%) for women with ER-negative disease and 7.5% (95% CI, 7.0%-8.0%) for those with ER-positive disease. Women with ER-negative cancer had higher second breast cancer rates than those with ER-positive cancer during the first 5 years of follow-up (16.0 per 1000 person-years [PY]; 95% CI, 14.2-17.9 per 1000 PY; vs. 7.8 per 1000 PY; 95% CI, 7.3-8.4 per 1000 PY, respectively). After 5 years, second breast cancer rates were similar for women with ER-negative versus ER-positive breast cancer (12.1 per 1000 PY; 95% CI, 9.9-14.7; vs. 9.3 per 1000 PY; 95% CI, 8.4-10.3 per 1000 PY, respectively). CONCLUSIONS: ER-negative primary breast cancers are associated with a higher risk of second breast cancers than ER-positive cancers during the first 5 years after diagnosis. Further study is needed to examine the potential benefit of more intensive surveillance targeting these women in the early postdiagnosis period.

Performance of Multiple Imputation Using Modern Machine Learning Methods in Electronic Health Records Data.

BACKGROUND: Missing data are common in studies using electronic health records (EHRs)-derived data. Missingness in EHR data is related to healthcare utilization patterns, resulting in complex and potentially missing not at random missingness mechanisms. Prior research has suggested that machine learning-based multiple imputation methods may outperform traditional methods and may perform well even in settings of missing not at random missingness. METHODS: We used plasmode simulations based on a nationwide EHR-derived de-identified database for patients with metastatic urothelial carcinoma to compare the performance of multiple imputation using chained equations, random forests, and denoising autoencoders in terms of bias and precision of hazard ratio estimates under varying proportions of observations with missing values and missingness mechanisms (missing completely at random, missing at random, and missing not at random). RESULTS: Multiple imputation by chained equations and random forest methods had low bias and similar standard errors for parameter estimates under missingness completely at random. Under missingness at random, denoising autoencoders had higher bias than multiple imputation by chained equations and random forests. Contrary to results of prior studies of denoising autoencoders, all methods exhibited substantial bias under missingness not at random, with bias increasing in direct proportion to the amount of missing data. CONCLUSIONS: We found no advantage of denoising autoencoders for multiple imputation in the setting of an epidemiologic study conducted using EHR data. Results suggested that denoising autoencoders may overfit the data leading to poor confounder control. Use of more flexible imputation approaches does not mitigate bias induced by missingness not at random and can produce estimates with spurious precision.

Inverse Probability of Treatment Weighting and Confounder Missingness in Electronic Health Record-based Analyses: A Comparison of Approaches Using Plasmode Simulation.

BACKGROUND: Electronic health record (EHR) data represent a critical resource for comparative effectiveness research, allowing investigators to study intervention effects in real-world settings with large patient samples. However, high levels of missingness in confounder variables is common, challenging the perceived validity of EHR-based investigations. METHODS: We investigated performance of multiple imputation and propensity score (PS) calibration when conducting inverse probability of treatment weights (IPTW)-based comparative effectiveness research using EHR data with missingness in confounder variables and outcome misclassification. Our motivating example compared effectiveness of immunotherapy versus chemotherapy treatment of advanced bladder cancer with missingness in a key prognostic variable. We captured complexity in EHR data structures using a plasmode simulation approach to spike investigator-defined effects into resamples of a cohort of 4361 patients from a nationwide deidentified EHR-derived database. We characterized statistical properties of IPTW hazard ratio estimates when using multiple imputation or PS calibration missingness approaches. RESULTS: Multiple imputation and PS calibration performed similarly, maintaining ≤0.05 absolute bias in the marginal hazard ratio even when ≥50% of subjects had missing at random or missing not at random confounder data. Multiple imputation required greater computational resources, taking nearly 40 times as long as PS calibration to complete. Outcome misclassification minimally increased bias of both methods. CONCLUSION: Our results support multiple imputation and PS calibration approaches to missingness in missing completely at random or missing at random confounder variables in EHR-based IPTW comparative effectiveness analyses, even with missingness ≥50%. PS calibration represents a computationally efficient alternative to multiple imputation.

Temporal Trends and Factors Associated with Receipt of Post-mastectomy Radiation After Neoadjuvant Chemotherapy in Women with cT3 Breast Cancer.

INTRODUCTION: Given the potential impact of increasingly effective neoadjuvant chemotherapy (NACT) on post-mastectomy radiotherapy (PMRT) recommendations, we examined temporal trends in post-NACT PMRT for cT3 breast cancer. METHODS: We identified women ≥ 18 years in the National Cancer Database (NCDB) diagnosed 2004-2019 with cT3N0-1M0 breast cancer treated with chemotherapy and mastectomy. Multivariable logistic regression and Cox proportional hazards models were used to estimate associations between pathologic NACT response [complete response (CR), partial response (PR), or no response (NR); or disease progression (DP)] and PMRT and between PMRT and overall survival (OS), respectively. RESULTS: We identified 39,901 women (Asian/Pacific Islander 1731, Black 5875, Hispanic 3265, White 27,303). Among cN0 patients with CR, PMRT rates declined from 67% in 2004 to 35% in 2019 but remained unchanged for patients with DP. Relative to NR, CR [odds ratio (OR) 0.36, 95% confidence interval (CI) 0.29-0.46] and PR (OR 0.44, 95% CI 0.36-0.55) in cN0 patients were associated with lower odds of PMRT while DP (OR 1.33, 95% CI 1.05-1.69) was associated with higher odds. Among cN1 patients, PMRT rates decreased from 90% to 73% for CR between 2005 and 2019 and increased from 76% to 82% for DP between 2004 and 2019. Relative to NR, CR (OR 0.78, 95% CI 0.63-0.95) was associated with lower odds of PMRT while DP (OR 1.93, 95% CI 1.58-2.37) was associated with higher odds. PMRT was associated with improved OS among cN1 patients (hazard ratio (HR) 0.77, 95% CI 0.67-0.88). CONCLUSION: CR was associated with decreased PMRT receipt over time, while temporal trends following PR and DP differed by cN status, suggesting that nodal involvement guided PMRT receipt more than in-breast disease.

Estimation and evaluation of individualized treatment rules following multiple imputation.

An individualized treatment rule (ITR) is a function that inputs patient-level information and outputs a recommended treatment. An important focus of precision medicine is to develop optimal ITRs that maximize a population-level distributional summary. However, guidance for estimating and evaluating optimal ITRs in the presence of missing data is limited. Our work is motivated by the Social Incentives to Encourage Physical Activity and Understand Predictors (STEP UP) study. Participants were randomized to a control or one of three interventions designed to increase physical activity and were given wearable devices to record daily steps as a measure of physical activity. Many participants were missing at least one daily step count during the study period. In the primary analysis of the STEP UP trial, multiple imputation (MI) was used to address missingness in daily step counts. Despite ubiquitous use of MI in practice, it has been given relatively little attention in the context of personalized medicine. We fill this gap by describing two frameworks for estimation and evaluation of an optimal ITR following MI and assessing their performance using simulated data. One framework relies on splitting the data into independent training and testing sets for estimation and evaluation, respectively. The other framework estimates an optimal ITR using the full data and constructs an m $$ m $$ -out-of- n $$ n $$ bootstrap confidence interval to evaluate its performance. Finally, we provide an illustrative analysis to estimate and evaluate an optimal ITR from the STEP UP data with a focus on practical considerations such as choosing the number of imputations.

Risk Prediction Models for Post-Operative Mortality in Patients With Cirrhosis.

BACKGROUND AND AIMS: Patients with cirrhosis are at increased risk of postoperative mortality. Currently available tools to predict postoperative risk are suboptimally calibrated and do not account for surgery type. Our objective was to use population-level data to derive and internally validate cirrhosis surgical risk models. APPROACH AND RESULTS: We conducted a retrospective cohort study using data from the Veterans Outcomes and Costs Associated with Liver Disease (VOCAL) cohort, which contains granular data on patients with cirrhosis from 128 U.S. medical centers, merged with the Veterans Affairs Surgical Quality Improvement Program (VASQIP) to identify surgical procedures. We categorized surgeries as abdominal wall, vascular, abdominal, cardiac, chest, or orthopedic and used multivariable logistic regression to model 30-, 90-, and 180-day postoperative mortality (VOCAL-Penn models). We compared model discrimination and calibration of VOCAL-Penn to the Mayo Risk Score (MRS), Model for End-Stage Liver Disease (MELD), Model for End-Stage Liver Disease-Sodium MELD-Na, and Child-Turcotte-Pugh (CTP) scores. We identified 4,712 surgical procedures in 3,785 patients with cirrhosis. The VOCAL-Penn models were derived and internally validated with excellent discrimination (30-day postoperative mortality C-statistic = 0.859; 95% confidence interval [CI], 0.809-0.909). Predictors included age, preoperative albumin, platelet count, bilirubin, surgery category, emergency indication, fatty liver disease, American Society of Anesthesiologists classification, and obesity. Model performance was superior to MELD, MELD-Na, CTP, and MRS at all time points (e.g., 30-day postoperative mortality C-statistic for MRS = 0.766; 95% CI, 0.676-0.855) in terms of discrimination and calibration. CONCLUSIONS: The VOCAL-Penn models substantially improve postoperative mortality predictions in patients with cirrhosis. These models may be applied in practice to improve preoperative risk stratification and optimize patient selection for surgical procedures (www.vocalpennscore.com).

Real-world prospective observational single-centre study: Hybrid closed loop improves HbA1c, time-in-range and quality of life for children, young people and their carers.

Hybrid closed-loop (HCL) systems are characterised by integrating continuous glucose monitoring (CGM) with insulin pumps which automate insulin delivery via specific algorithms and user-initiated insulin delivery. The aim of the study was to evaluate the effectiveness of HCLs on Hba1c, time-in-range (TIR), time in hypoglycaemia, fear of hypoglycaemia, sleep and quality of life measure in children and young people (CYP) with T1D and their carers. Data on HbA1c, TIR and hypoglycaemia frequency were reviewed at baseline prior to starting HCL and 3 months after commencement. As part of clinical care, all patients and carers were provided with key education on the use of the HCL system by trained diabetes healthcare professionals. CYP aged 12 years and above independently completed the validated Hypoglycaemia Fear Survey (HFS). Parents of patients <12 were asked to complete a modified version of the HFS-Parent (HFS-P) survey. There were 39 CYP (22 men) with T1D included with a mean age of 11.8 ± 4.4 at commencement of HCL. Median duration of diabetes was 3.8 years (interquartile range 1.3-6.0). There were 55% of patients who were prepubertal at the time of HCL commencement. 91% were on the Control-IQ system and 9% on the CamAPS FX system. HCL use demonstrated significant improvements at 3 months in the following: HbA1c in mmol/mol (63.0 vs. 56.6, p = 0.03), TIR (50.5 vs. 67.0, p = 0.001) and time in hypoglycaemia (4.3% vs. 2.8%, p = 0.004). HFS scores showed improved behaviour (34.0 vs. 27.5.9, p = 0.02) and worry (40.2 vs. 31.6, p = 0.03), and HFS-P scores also showed improved behaviour (p < 0.001) and worry (p = 0.01). Our study shows that HCL at 3 months improves glucose control, diabetes management and quality of life measures such as fear and worry of hypoglycaemia for CYP and carers.

Nonhomogeneous Markov chain for estimating the cumulative risk of multiple false positive screening tests.

Screening tests are widely recommended for the early detection of disease among asymptomatic individuals. While detecting disease at an earlier stage has the potential to improve outcomes, screening also has negative consequences, including false positive results which may lead to anxiety, unnecessary diagnostic procedures, and increased healthcare costs. In addition, multiple false positive results could discourage participating in subsequent screening rounds. Screening guidelines typically recommend repeated screening over a period of many years, but little prior research has investigated how often individuals receive multiple false positive test results. Estimating the cumulative risk of multiple false positive results over the course of multiple rounds of screening is challenging due to the presence of censoring and competing risks, which may depend on the false positive risk, screening round, and number of prior false positive results. To address the general challenge of estimating the cumulative risk of multiple false positive test results, we propose a nonhomogeneous multistate model to describe the screening process including competing events. We developed alternative approaches for estimating the cumulative risk of multiple false positive results using this multistate model based on existing estimators for the cumulative risk of a single false positive. We compared the performance of the newly proposed models through simulation studies and illustrate model performance using data on screening mammography from the Breast Cancer Surveillance Consortium. Across most simulation scenarios, the multistate extension of a censoring bias model demonstrated lower bias compared to other approaches. In the context of screening mammography, we found that the cumulative risk of multiple false positive results is high. For instance, based on the censoring bias model, for a high-risk individual, the cumulative probability of at least two false positive mammography results after 10 rounds of annual screening is 40.4.

Does Living near Trees and Other Vegetation Affect the Contemporaneous Odds of Asthma Exacerbation among Pediatric Asthma Patients?

Vegetation may influence asthma exacerbation through effects on aeroallergens, localized climates, air pollution, or children's behaviors and stress levels. We investigated the association between residential vegetation and asthma exacerbation by conducting a matched case-control study based on electronic health records of asthma patients, from the Children's Hospital of Philadelphia (CHOP). Our study included 17,639 exacerbation case events and 34,681 controls selected from non-exacerbation clinical visits for asthma, matched to cases by age, sex, race/ethnicity, public payment source, and residential proximity to the CHOP main campus ED and hospital. Overall greenness, tree canopy, grass/shrub cover, and impervious surface were assessed near children's homes (250 m) using satellite imagery and high-resolution landcover data. We used generalized estimating equations to estimate odds ratios (OR) and 95% confidence intervals (CI) for associations between each vegetation/landcover measure and asthma exacerbation, with adjustment for seasonal and sociodemographic factors-for all cases, and for cases defined by diagnosis setting and exacerbation frequency. Lower odds of asthma exacerbation were observed in association with greater levels of tree canopy near the home, but only for children who experienced multiple exacerbations in a year (OR = 0.94 per 10.2% greater tree canopy coverage, 95% CI = 0.90-0.99). Our findings suggest possible protection for asthma patients from tree canopy, but differing results by case frequency suggest that potential benefits may be specific to certain subpopulations of asthmatic children.

Associations between high ambient temperatures and asthma exacerbation among children in Philadelphia, PA: a time series analysis.

OBJECTIVES: High ambient temperatures may contribute to acute asthma exacerbation, a leading cause of morbidity in children. We quantified associations between hot-season ambient temperatures and asthma exacerbation in children ages 0-18 years in Philadelphia, PA. METHODS: We created a time series of daily counts of clinical encounters for asthma exacerbation at the Children's Hospital of Philadelphia linked with daily meteorological data, June-August of 2011-2016. We estimated associations between mean daily temperature (up to a 5-day lag) and asthma exacerbation using generalised quasi-Poisson distributed models, adjusted for seasonal and long-term trends, day of the week, mean relative humidity,and US holiday. In secondary analyses, we ran models with adjustment for aeroallergens, air pollutants and respiratory virus counts. We quantified overall associations, and estimates stratified by encounter location (outpatient, emergency department, inpatient), sociodemographics and comorbidities. RESULTS: The analysis included 7637 asthma exacerbation events. High mean daily temperatures that occurred 5 days before the index date were associated with higher rates of exacerbation (rate ratio (RR) comparing 33°C-13.1°C days: 1.37, 95% CI 1.04 to 1.82). Associations were most substantial for children ages 2 to <5 years and for Hispanic and non-Hispanic black children. Adjustment for air pollutants, aeroallergens and respiratory virus counts did not substantially change RR estimates. CONCLUSIONS: This research contributes to evidence that ambient heat is associated with higher rates of asthma exacerbation in children. Further work is needed to explore the mechanisms underlying these associations.

Patient and tumour characteristics of screening-age adults diagnosed with screen-detected versus symptomatic colon cancer.

AIM: International studies have shown that most colon cancers are diagnosed among people with symptoms, but research is limited in the United States. Here, we conducted a retrospective study of adults aged 50-85 years diagnosed with stage I-IIIA colon cancer between 1995 and 2014 in two US healthcare systems. METHODS: Mode of detection (screening or symptomatic) was ascertained from medical records. We estimated unadjusted odds ratios (OR) and 95% confidence intervals (CI) comparing detection mode by patient factors at diagnosis (year, age, sex, race, smoking status, body mass index [BMI], Charlson score), prediagnostic primary care utilization, and tumour characteristics (stage, location). RESULTS: Of 1,675 people with colon cancer, 38.4% were screen-detected, while 61.6% were diagnosed following symptomatic presentation. Screen-detected cancer was more common among those diagnosed between 2010 and 2014 versus 1995-1999 (OR 1.65, 95% CI: 1.19-2.28), and those with a BMI of 25-<30 kg/m2 (OR 1.54, 95% CI: 1.21-1.98) or ≥30 kg/m2 (OR 1.52, 95% CI: 1.18-1.96) versus <25 kg/m2 . Screen-detected cancer was less common among people aged 76-85 (OR 0.50, 95% CI: 0.39-0.65) versus 50-64, those with comorbidity scores >0 (OR 0.71, 95% CI: 0.56-0.91 for score = 1, OR 0.34, 95% CI: 0.26-0.45 for score = 2+), and those with 2+ prediagnostic primary care visits (OR 0.53, 95% CI: 0.37-0.76) versus 0 visits. The odds of screen detection were lower among patients diagnosed with stage IIA (OR 0.33, 95% CI = 0.27-0.41) or IIB (OR 0.12, 95% CI: 0.06-0.24) cancers versus stage I. CONCLUSIONS: Most colon cancers among screen-eligible adults were diagnosed following symptomatic presentation. Even with increasing screening rates over time, research is needed to better understand why specific groups are more likely to be diagnosed when symptomatic and identify opportunities for interventions.

Very high PD-L1 expression as a prognostic indicator of overall survival among patients with advanced non-small cell lung cancer receiving anti-PD-(L)1 monotherapies in routine practice.

PURPOSE: Programmed death or ligand-1 (PD-(L)1) pathway inhibitors confer improved survival as the first-line treatment for advanced non-small cell lung cancer (aNSCLC) in patients with PD-L1 expression (PD-L1 + e ≥ 50%) compared to platinum-doublet chemotherapy and have become a standard therapy. Some recent evidence suggests that among aNSCLC patients with PD-L1 + e of ≥50% receiving pembrolizumab monotherapy, very high levels of PD-L1 + e (≥90%) may be associated with better outcomes. We sought to assess whether very high PD-L1 + e (≥90%) compared to high PD-L1 + e (50%-89%) is associated with an overall survival benefit in aNSCLC patients receiving anti-PD-(L)1 monotherapies. METHODS: We conducted a single-site retrospective cohort study of aNSCLC patients who initiated PD-(L)1 inhibitor monotherapy as the first-line treatment from October 24, 2016, to August 25, 2021, and had a PD-L1 + e ≥ 50%. The primary outcome was overall survival, measured from the start of the first-line PD-(L)1 inhibitor monotherapy (index date) to date of death or last confirmed activity prior to the cohort exit date. Propensity score-based inverse probability weighting (IPW) was used to control for confounding in Kaplan-Meier curves and Cox proportional hazard regression analysis. RESULTS: One hundred sixty-six patients with aNSCLC receiving PD-(L)1 inhibitor monotherapy met inclusion criteria. 54% were female, 90% received pembrolizumab, median age was 68 years, 70% had non-squamous cell carcinoma, 94% had a history of smoking, 29% had a KRAS mutation, and 37% had very high PD-L1 + e. Unweighted covariates at cohort entry were similar between groups (absolute standardized mean differences [SMDs] <0.1) except for race (SMD = 0.2); age at therapy initiation (SMD = 0.13); smoking status (SMD = 0.13), and BRAF mutation status (SMD = 0.11). After weighting, baseline covariates were well balanced (all absolute SMDs <0.1). In the weighted analysis, having a very high PD-L1 + e was associated with lower mortality (weighted hazard ratio 0.57, 95% CI 0.36-0.90) and longer median survival: 3.85 versus 1.49 years. CONCLUSIONS: Very high PD-L1 + e (≥90%) was associated with an overall survival benefit over high PD-L1 + e (50%-89%) in patients receiving the first-line PD-(L)1 inhibitor monotherapy in a model controlling for potential confounders. These findings should be confirmed in a larger real-world data set.

Comparing LASSO and random forest models for predicting neurological dysfunction among fluoroquinolone users.

BACKGROUND: Fluoroquinolones are associated with central (CNS) and peripheral (PNS) nervous system symptoms, and predicting the risk of these outcomes may have important clinical implications. Both LASSO and random forest are appealing modeling methods, yet it is not clear which method performs better for clinical risk prediction. PURPOSE: To compare models developed using LASSO versus random forest for predicting neurological dysfunction among fluoroquinolone users. METHODS: We developed and validated risk prediction models using claims data from a commercially insured population. The study cohort included adults dispensed an oral fluoroquinolone, and outcomes were CNS and PNS dysfunction. Model predictors included demographic variables, comorbidities and medications known to be associated with neurological symptoms, and several healthcare utilization predictors. We assessed the accuracy and calibration of these models using measures including AUC, calibration curves, and Brier scores. RESULTS: The underlying cohort contained 16 533 (1.18%) individuals with CNS dysfunction and 46 995 (3.34%) individuals with PNS dysfunction during 120 days of follow-up. For CNS dysfunction, LASSO had an AUC of 0.81 (95% CI: 0.80, 0.82), while random forest had an AUC of 0.80 (95% CI: 0.80, 0.81). For PNS dysfunction, LASSO had an AUC of 0.75 (95% CI: 0.74, 0.76) versus an AUC of 0.73 (95% CI: 0.73, 0.74) for random forest. Both LASSO models had better calibration, with Brier scores 0.17 (LASSO) versus 0.20 (random forest) for CNS dysfunction and 0.20 (LASSO) versus 0.25 (random forest) for PNS dysfunction. CONCLUSIONS: LASSO outperformed random forest in predicting CNS and PNS dysfunction among fluoroquinolone users, and should be considered for modeling when the cohort is modest in size, when the number of model predictors is modest, and when predictors are primarily binary.

Integrating real world data and clinical trial results using survival data reconstruction and marginal moment-balancing weights.

Outcomes in electronic health records (EHR)-derived cohorts can be compared to similarly treated clinical trial cohorts to estimate the efficacy-effectiveness gap, the discrepancy in performance of an intervention in a trial compared to the real world. However, because clinical trial data may only be available in the form of published summary statistics and Kaplan-Meier curves, survival data reconstruction methods are needed to recreate individual-level survival data. Additionally, marginal moment-balancing weights can adjust for differences in the distributions of patient characteristics between the trial and EHR cohorts. We evaluated bias in hazard ratio (HR) estimates by comparing trial and EHR cohorts using survival data reconstruction and marginal moment-balancing weights through simulations and analysis of real-world data. This approach produced nearly unbiased HR estimates. In an analysis of overall survival for patients with metastatic urothelial carcinoma treated with gemcitabine-carboplatin captured in the nationwide Flatiron Health EHR-derived de-identified database and patients enrolled in a trial of the same therapy, survival was similar in the EHR and trial cohorts after using weights to balance age, sex, and performance status (HR = 0.93, 95% confidence interval (0.74, 1.18)). Overall, we conclude that this approach is feasible for comparison of trial and EHR cohorts and facilitates evaluation of outcome differences between trial and real-world populations.

Association of COVID-19 vs Influenza With Risk of Arterial and Venous Thrombotic Events Among Hospitalized Patients.

Importance: The incidence of arterial thromboembolism and venous thromboembolism in persons with COVID-19 remains unclear. Objective: To measure the 90-day risk of arterial thromboembolism and venous thromboembolism in patients hospitalized with COVID-19 before or during COVID-19 vaccine availability vs patients hospitalized with influenza. Design, Setting, and Participants: Retrospective cohort study of 41 443 patients hospitalized with COVID-19 before vaccine availability (April-November 2020), 44 194 patients hospitalized with COVID-19 during vaccine availability (December 2020-May 2021), and 8269 patients hospitalized with influenza (October 2018-April 2019) in the US Food and Drug Administration Sentinel System (data from 2 national health insurers and 4 regional integrated health systems). Exposures: COVID-19 or influenza (identified by hospital diagnosis or nucleic acid test). Main Outcomes and Measures: Hospital diagnosis of arterial thromboembolism (acute myocardial infarction or ischemic stroke) and venous thromboembolism (deep vein thrombosis or pulmonary embolism) within 90 days. Outcomes were ascertained through July 2019 for patients with influenza and through August 2021 for patients with COVID-19. Propensity scores with fine stratification were developed to account for differences between the influenza and COVID-19 cohorts. Weighted Cox regression was used to estimate the adjusted hazard ratios (HRs) for outcomes during each COVID-19 vaccine availability period vs the influenza period. Results: A total of 85 637 patients with COVID-19 (mean age, 72 [SD, 13.0] years; 50.5% were male) and 8269 with influenza (mean age, 72 [SD, 13.3] years; 45.0% were male) were included. The 90-day absolute risk of arterial thromboembolism was 14.4% (95% CI, 13.6%-15.2%) in patients with influenza vs 15.8% (95% CI, 15.5%-16.2%) in patients with COVID-19 before vaccine availability (risk difference, 1.4% [95% CI, 1.0%-2.3%]) and 16.3% (95% CI, 16.0%-16.6%) in patients with COVID-19 during vaccine availability (risk difference, 1.9% [95% CI, 1.1%-2.7%]). Compared with patients with influenza, the risk of arterial thromboembolism was not significantly higher among patients with COVID-19 before vaccine availability (adjusted HR, 1.04 [95% CI, 0.97-1.11]) or during vaccine availability (adjusted HR, 1.07 [95% CI, 1.00-1.14]). The 90-day absolute risk of venous thromboembolism was 5.3% (95% CI, 4.9%-5.8%) in patients with influenza vs 9.5% (95% CI, 9.2%-9.7%) in patients with COVID-19 before vaccine availability (risk difference, 4.1% [95% CI, 3.6%-4.7%]) and 10.9% (95% CI, 10.6%-11.1%) in patients with COVID-19 during vaccine availability (risk difference, 5.5% [95% CI, 5.0%-6.1%]). Compared with patients with influenza, the risk of venous thromboembolism was significantly higher among patients with COVID-19 before vaccine availability (adjusted HR, 1.60 [95% CI, 1.43-1.79]) and during vaccine availability (adjusted HR, 1.89 [95% CI, 1.68-2.12]). Conclusions and Relevance: Based on data from a US public health surveillance system, hospitalization with COVID-19 before and during vaccine availability, vs hospitalization with influenza in 2018-2019, was significantly associated with a higher risk of venous thromboembolism within 90 days, but there was no significant difference in the risk of arterial thromboembolism within 90 days.

Effects of Neighborhood-level Data on Performance and Algorithmic Equity of a Model That Predicts 30-day Heart Failure Readmissions at an Urban Academic Medical Center.

BACKGROUND: Socioeconomic data may improve predictions of clinical events. However, owing to structural racism, algorithms may not perform equitably across racial subgroups. Therefore, we sought to compare the predictive performance overall, and by racial subgroup, of commonly used predictor variables for heart failure readmission with and without the area deprivation index (ADI), a neighborhood-level socioeconomic measure. METHODS AND RESULTS: We conducted a retrospective cohort study of 1316 Philadelphia residents discharged with a primary diagnosis of congestive heart failure from the University of Pennsylvania Health System between April 1, 2015, and March 31, 2017. We trained a regression model to predict the probability of a 30-day readmission using clinical and demographic variables. A second model also included the ADI as a predictor variable. We measured predictive performance with the Brier Score (BS) in a held-out test set. The baseline model had moderate performance overall (BS 0.13, 95% CI 0.13-0.14), and among White (BS 0.12, 95% CI 0.12-0.13) and non-White (BS 0.13, 95% CI 0.13-0.14) patients. Neither performance nor algorithmic equity were significantly changed with the addition of the ADI. CONCLUSIONS: The inclusion of neighborhood-level data may not reliably improve performance or algorithmic equity.

Case contamination in electronic health records based case-control studies.

Clinically relevant information from electronic health records (EHRs) permits derivation of a rich collection of phenotypes. Unlike traditionally designed studies where scientific hypotheses are specified a priori before data collection, the true phenotype status of any given individual in EHR-based studies is not directly available. Structured and unstructured data elements need to be queried through preconstructed rules to identify case and control groups. A sufficient number of controls can usually be identified with high accuracy by making the selection criteria stringent. But more relaxed criteria are often necessary for more thorough identification of cases to ensure achievable statistical power. The resulting pool of candidate cases consists of genuine cases contaminated with noncase patients who do not satisfy the control definition. The presence of patients who are neither true cases nor controls among the identified cases is a unique challenge in EHR-based case-control studies. Ignoring case contamination would lead to biased estimation of odds ratio association parameters. We propose an estimating equation approach to bias correction, study its large sample property, and evaluate its performance through extensive simulation studies and an application to a pilot study of aortic stenosis in the Penn medicine EHR. Our method holds the promise of facilitating more efficient EHR studies by accommodating enlarged albeit contaminated case pools.