High-dimensional generalized median adaptive lasso with application to omics data.

Recently, there has been a growing interest in variable selection for causal inference within the context of high-dimensional data. However, when the outcome exhibits a skewed distribution, ensuring the accuracy of variable selection and causal effect estimation might be challenging. Here, we introduce the generalized median adaptive lasso (GMAL) for covariate selection to achieve an accurate estimation of causal effect even when the outcome follows skewed distributions. A distinctive feature of our proposed method is that we utilize a linear median regression model for constructing penalty weights, thereby maintaining the accuracy of variable selection and causal effect estimation even when the outcome presents extremely skewed distributions. Simulation results showed that our proposed method performs comparably to existing methods in variable selection when the outcome follows a symmetric distribution. Besides, the proposed method exhibited obvious superiority over the existing methods when the outcome follows a skewed distribution. Meanwhile, our proposed method consistently outperformed the existing methods in causal estimation, as indicated by smaller root-mean-square error. We also utilized the GMAL method on a deoxyribonucleic acid methylation dataset from the Alzheimer's disease (AD) neuroimaging initiative database to investigate the association between cerebrospinal fluid tau protein levels and the severity of AD.

Scalable kernel balancing weights in a nationwide observational study of hospital profit status and heart attack outcomes.

Weighting is a general and often-used method for statistical adjustment. Weighting has two objectives: first, to balance covariate distributions, and second, to ensure that the weights have minimal dispersion and thus produce a more stable estimator. A recent, increasingly common approach directly optimizes the weights toward these two objectives. However, this approach has not yet been feasible in large-scale datasets when investigators wish to flexibly balance general basis functions in an extended feature space. To address this practical problem, we describe a scalable and flexible approach to weighting that integrates a basis expansion in a reproducing kernel Hilbert space with state-of-the-art convex optimization techniques. Specifically, we use the rank-restricted Nyström method to efficiently compute a kernel basis for balancing in nearly linear time and space, and then use the specialized first-order alternating direction method of multipliers to rapidly find the optimal weights. In an extensive simulation study, we provide new insights into the performance of weighting estimators in large datasets, showing that the proposed approach substantially outperforms others in terms of accuracy and speed. Finally, we use this weighting approach to conduct a national study of the relationship between hospital profit status and heart attack outcomes in a comprehensive dataset of 1.27 million patients. We find that for-profit hospitals use interventional cardiology to treat heart attacks at similar rates as other hospitals but have higher mortality and readmission rates.

Heterogeneity in the Association Between the Presence of Coronary Artery Calcium and Cardiovascular Events: A Machine-Learning Approach in the MESA Study.

BACKGROUND: Coronary artery calcium (CAC) has been widely recognized as an important predictor of cardiovascular disease (CVD). Given the finite resources, it is important to identify individuals who would receive the most benefit from detecting positive CAC by screening. However, the evidence is limited as to whether the burden of positive CAC on CVD differs by multidimensional individual characteristics. We sought to investigate the heterogeneity in the association between positive CAC and incident CVD. METHODS: This cohort study included adults from MESA (Multi-Ethnic Study of Atherosclerosis) ages ≥45 years and free of cardiovascular disease. After propensity score matching in a 1:1 ratio, we applied a machine learning causal forest model to (1) evaluate the heterogeneity in the association between positive CAC and incident CVD, and (2) predict the increase in CVD risk at 10-years when CAC>0 (versus CAC=0) at the individual level. We then compared the estimated increase in CVD risk when CAC>0 to the absolute 10-year atherosclerotic CVD (ASCVD) risk calculated by the 2013 American College of Cardiology/American Heart Association pooled cohort equations. RESULTS: Across 3328 adults in our propensity score-matched analysis, our causal forest model showed the heterogeneity in the association between CAC>0 and incident CVD. We found a dose-response relationship of the estimated increase in CVD risk when CAC>0 with higher 10-year ASCVD risk. Almost all individuals (2293 of 2428 [94.4%]) with borderline risk of ASCVD or higher showed ≥2.5% increase in CVD risk when CAC>0. Even among 900 adults with low ASCVD risk, 689 (69.2%) showed ≥2.5% increase in CVD risk when CAC>0; these individuals were more likely to be male, Hispanic, and have unfavorable CVD risk factors than others. CONCLUSIONS: The expected increases in CVD risk when CAC>0 were heterogeneous across individuals. Moreover, nearly 70% of people with low ASCVD risk showed a large increase in CVD risk when CAC>0, highlighting the need for CAC screening among such low-risk individuals. Future studies are needed to assess whether targeting individuals for CAC measurements based on not only the absolute ASCVD risk but also the expected increase in CVD risk when CAC>0 improves cardiovascular outcomes.

Treatment Modalities and Outcomes in Brainstem Cavernous Malformations: A Large Multicenter Observational Cohort Study.

BACKGROUND: Symptomatic brainstem cavernous malformations (BSCMs) pose a high risk of morbidity and mortality due to recurrent hemorrhage, warranting aggressive management. However, few studies have compared the effectiveness of different treatment modalities for BSCMs. We aimed to assess the association of treatment modalities with recurrent hemorrhage and neurological outcomes in patients with BSCM. METHODS: We conducted a retrospective cohort study using an observational registry database covering population of southwest and southeast China. Adult patients with BSCM were included and followed up between March 1, 2011, to March 31, 2023. We compared outcomes between microsurgery and stereotactic radiosurgery (SRS) in propensity score-matched case pairs, incorporating demographic, medical history, and lesion characteristics. The outcomes studied included recurrent hemorrhage and poor prognosis (defined as a Glasgow Outcome Scale score, <4). Absolute rate differences and hazard ratios (HRs) with 95% CIs were calculated using Cox models. RESULTS: Among 736 diagnosed patients with BSCM, 96 (48 matched pairs) were included after exclusions and propensity score matching (mean age, 43.1 [SD, 12.1] years; 50% women). During the median 5-year follow-up, no significant differences in recurrent hemorrhage (4.2% [microsurgery] versus 14.6% [SRS], HR, 3.90 [95% CI, 0.46-32.65]; P=0.21) and poor prognosis (12.5% [microsurgery] versus 8.3% [SRS], HR, 0.29 [95% CI, 0.08-1.08]; P=0.07) were observed between microsurgery and SRS recipients. Furthermore, either microsurgery or SRS correlated with fewer recurrent hemorrhage (HR, 0.09 [95% CI, 0.02-0.39]; P=0.001; HR, 0.21 [95% CI, 0.07-0.69]; P=0.01) compared with conservative treatment. CONCLUSIONS: In this study, both microsurgery and SRS were safe and effective for BSCM, demonstrated comparable outcomes in recurrent hemorrhage and poor prognosis. However, interpretation should be cautious due to the potential for residual confounding. REGISTRATION: URL: https://www.chictr.org.cn/; Unique identifier: ChiCTR2300070907.

Higher Risk of Mental Illness in Patients With Diagnosed and Untreated Unruptured Intracranial Aneurysm: Findings From a Nationwide Cohort Study.

BACKGROUND: We aimed to investigate the association between a diagnosis of untreated unruptured intracranial aneurysms (UIAs) and the development of mental illness. METHODS: This retrospective, propensity-score-matched cohort study was based on the nationwide South Korean database. The UIA diagnosis group included participants newly diagnosed with UIA between 2011 and 2019. For a well-matched control group, patients diagnosed with an acute upper respiratory infection but without UIA during the same period were selected through 1:4 matching based on propensity scores, which were calculated using age, sex, economic status, and comorbidities. The study's outcome measure encompassed the incidence of mental illnesses over a 10-year period, using International Classification of Diseases-Tenth Revision codes for anxiety, stress, depressive, bipolar, and eating disorders, insomnia, and alcohol or drug misuse. RESULTS: After propensity score matching, 85 438 participants with untreated UIAs (50.75% male; average age, 56.41 [±13.82] years; follow-up, 4.21 [±2.56] years) and 331 123 controls (49.44% males; average age, 56.69 [±12.92] years; follow-up, 7.48 [±2.12] years) were compared. Incidence rate of mental illness was higher in the UIA group (113.07 versus 90.41 per 1000 person-years; hazard ratio, 1.104 [95% CI, 1.089-1.119]). The risk of mental illness varied slightly by sex (males: hazard ratio, 1.131 [95% CI, 1.108-1.155]; females: hazard ratio, 1.082 [95% CI, 1.063-1.103]). Hazard ratios showed a U-shaped relationship with age, peaking in younger age groups, decreasing in middle-aged groups, and slightly increasing in older age groups, especially in patients with severe mental illness receiving psychotherapy. CONCLUSIONS: Our findings indicate a higher risk of mental illness in patients with UIA diagnosis in specific demographic groups, suggesting a possible psychological burden associated with UIAs. Clinicians treating cerebral aneurysms should be aware that the psychological burden caused by the diagnosis of UIA itself could contribute to mental illness and strive to provide comprehensive care for these patients.

Propensity score methods for confounding control in observational studies of therapeutics for COVID-19 infection.

The authors provide a brief overview of different propensity score methods that can be used in observational research studies that lack randomization. Under specific assumptions, these methods result in unbiased estimates of causal effects, but the different ways propensity score are used may require different assumptions and result in estimated treatment effects that can have meaningfully different interpretations. The authors review these issues and consider their implications for studies of therapeutics for COVID-19.

Management of vulnerable patients hospitalized for COVID-19 with remdesivir: a retrospective comparative effectiveness study of mortality in US hospitals.

BACKGROUND: COVID-19 remains a major public health concern, with continued resurgences of cases and substantial risk of mortality for hospitalized patients. Remdesivir has become standard-of-care for hospitalized COVID-19 patients. Given the continued evolution of the disease, clinical management relies on evidence from the current endemic period. METHODS: Using the PINC AI Healthcare database, effectiveness of remdesivir was evaluated among adults hospitalized with a primary diagnosis of COVID-19 between December 2021 and February 2024. Three cohorts were analysed: adults, elderly (≥65 years), and those with documented COVID-19 pneumonia. Analyses were stratified by oxygen requirements. Patients receiving remdesivir were matched to those not receiving remdesivir using propensity score matching. Cox proportional hazards models were used to examine in-hospital mortality. RESULTS: 169,965 adults hospitalized for COVID-19 were included, of which 94,129 (55.4%) initiated remdesivir in the first two days of hospitalization. Remdesivir was associated with a significantly lower mortality rate as compared to no remdesivir among patients with no supplemental oxygen charges (NSOc) (aHR [95% CI]: 14-day, 0.75 [0.69-0.82]; 28-day, 0.77 [0.72-0.83]) and among those with supplemental oxygen charges (SOc): 14-day, 0.76 [0.72-0.81]; 28-day, 0.79 [0.74-0.83]) (p<0.0001, for all). Similar findings were observed for elderly patients and those hospitalized with COVID-19 pneumonia. CONCLUSIONS: This evidence builds on learnings from randomized controlled trials from the pandemic era to inform clinical practices. Remdesivir was associated with significant reduction in mortality for hospitalized patients including the elderly and those with COVID-19 pneumonia.

Lower mortality risk associated with remdesivir + dexamethasone versus dexamethasone alone for the treatment of patients hospitalized for COVID-19.

BACKGROUND: Treatment guidelines were developed early in the pandemic when much about COVID-19 was unknown. Given the evolution of SARS-CoV-2, real-world data can provide clinicians with updated information. The objective of this analysis was to assess mortality risk in patients hospitalized for COVID-19 during the Omicron period receiving remdesivir+dexamethasone versus dexamethasone alone. METHODS: A large, multicenter US hospital database was used to identify hospitalized adult patients, with a primary discharge diagnosis of COVID-19 also flagged as "present on admission" treated with remdesivir+dexamethasone or dexamethasone alone from December 2021 to April 2023. Patients were matched 1:1 using propensity score matching and stratified by baseline oxygen requirements. Cox proportional hazards model was used to assess time to 14- and 28-day in-hospital all-cause mortality. RESULTS: A total of 33 037 patients were matched, with most patients ≥65 years old (72%), White (78%), and non-Hispanic (84%). Remdesivir+dexamethasone was associated with lower mortality risk versus dexamethasone alone across all baseline oxygen requirements at 14 days (no supplemental oxygen charges: adjusted hazard ratio [95% CI]: 0.79 [0.72-0.87], low flow oxygen: 0.70 [0.64-0.77], high flow oxygen/non-invasive ventilation: 0.69 [0.62-0.76], invasive mechanical ventilation/extracorporeal membrane oxygen (IMV/ECMO): 0.78 [0.64-0.94]), with similar results at 28 days. CONCLUSIONS: Remdesivir+dexamethasone was associated with a significant reduction in 14- and 28-day mortality compared to dexamethasone alone in patients hospitalized for COVID-19 across all levels of baseline respiratory support, including IMV/ECMO. However, the use of remdesivir+dexamethasone still has low clinical practice uptake. In addition, these data suggest a need to update the existing guidelines.

Remdesivir-associated survival outcomes among immunocompromised patients hospitalized for COVID-19: real-world evidence from the Omicron dominant era.

BACKGROUND: Patients with immunocompromising conditions are at an increased risk for coronavirus disease 2019 (COVID-19)-related hospitalizations and mortality. Randomized clinical trials provide limited enrollment, if any, to inform outcomes of such patients treated with remdesivir. METHODS: Using the US PINC AI Healthcare Database, we identified adult patients with immunocompromising conditions, hospitalized for COVID-19 between December 2021 and February 2024. Primary outcome was all-cause inpatient mortality examined in propensity score (PS) matched patients in remdesivir versus non-remdesivir groups. Subgroup analyses were performed for patients with cancer, hematologic malignancies, and solid organ/hematopoietic stem cell transplant recipients. RESULTS: Of 28,966 patients included in the study, 16,730 (58%) received remdesivir during first two days of hospitalization. After PS matching, 8,822 patients in remdesivir and 8,822 patients in non-remdesivir group were analyzed. Remdesivir was associated with a significantly lower mortality among patients with no supplemental oxygen (aHR [95% CI]: 14-day, 0.73 [0.62-0.86]; 28-day, 0.79 [0.68-0.91]) and among those with supplemental oxygen (14-day, 0.75 [0.67-0.85]; 28-day, 0.78 [0.70-0.86]). Remdesivir was also associated with lower mortality in subgroups of patients with cancer, hematological malignancies (including leukemia, lymphoma, and multiple myeloma), and solid organ/hematopoietic stem cell transplantation. CONCLUSIONS: In this large cohort of patients with immunocompromising conditions hospitalized for COVID-19, remdesivir was associated with significant improvement in survival, including patients with varied underlying immunocompromising conditions. The integration of current real-world evidence into clinical guideline recommendations can inform clinical communities to optimize treatment decisions in the evolving COVID-19 era, extending beyond the conclusion of the public health emergency declaration.

Use of menopausal hormone therapy before and after diagnosis and ovarian cancer survival-A prospective cohort study in Australia.

Menopausal hormone therapy (MHT) use before ovarian cancer diagnosis has been associated with improved survival but whether the association varies by type and duration of use is inconclusive; data on MHT use after treatment, particularly the effect on health-related quality of life (HRQOL), are scarce. We investigated survival in women with ovarian cancer according to MHT use before and after diagnosis, and post-treatment MHT use and its association with HRQOL in a prospective nationwide cohort in Australia. We used Cox proportional hazards regression to estimate hazard ratios (HR) and 95% confidence intervals (CI) and propensity scores to reduce confounding by indication. Among 690 women who were peri-/postmenopausal at diagnosis, pre-diagnosis MHT use was associated with a significant 26% improvement in ovarian cancer-specific survival; with a slightly stronger association for high-grade serous carcinoma (HGSC, HR = 0.69, 95%CI 0.54-0.87). The associations did not differ by recency or duration of use. Among women with HGSC who were pre-/perimenopausal or aged ≤55 years at diagnosis (n = 259), MHT use after treatment was not associated with a difference in survival (HR = 1.04, 95%CI 0.48-2.22). Compared to non-users, women who started MHT after treatment reported poorer overall HRQOL before starting MHT and this difference was still seen 1-3 months after starting MHT. In conclusion, pre-diagnosis MHT use was associated with improved survival, particularly in HGSC. Among women ≤55 years, use of MHT following treatment was not associated with poorer survival for HGSC. Further large-scale studies are needed to understand menopause-specific HRQOL issues in ovarian cancer.

The role of multiple birth and birth complications in the association between assisted reproductive technology conception and autism diagnosis.

In recent decades, the use of assisted reproductive technology (ART) has increased rapidly. To assess the relationship between ART and autism diagnosis, we linked California birth records from 2000 through 2016 with contemporaneous records from the National ART Surveillance System (NASS) and autism caseload records from California's Department of Developmental Services from 2000 through November 2019. All 95 149 birth records that were successfully linked to a NASS record, indicating an ART birth, were matched 1:1 using propensity scores to non-ART births. We calculated the hazard risk ratio for autism diagnosis and the proportions of the relationship between ART conception and autism diagnosis mediated by multiple birth pregnancy and related birth complications. The hazard risk ratio for autism diagnosis following ART compared with non-ART conception is 1.26 (95% CI, 1.17-1.35). Multiple birth, preterm birth, and cesarean delivery jointly mediate 77.9% of the relationship between ART conception and autism diagnosis. Thus, increased use of single embryo transfer in the United States to reduce multiple births and related birth complications may be a strategy to address the risk of autism diagnosis among ART-conceived children.

Multiple imputation for propensity score analysis with covariates missing at random: some clarity on "within" and "across" methods.

In epidemiology and the social sciences, propensity score methods are popular for estimating treatment effects using observational data, and multiple imputation is popular for handling covariate missingness. However, how to appropriately use multiple imputation for propensity score analysis is not completely clear. This paper aims to bring clarity on the consistency (or lack thereof) of methods that have been proposed, focusing on the "within" approach (where the effect is estimated separately in each imputed dataset and then the multiple estimates are combined) and the "across" approach (where typically propensity scores are averaged across imputed datasets before being used for effect estimation). We show that the within method is valid and can be used with any causal effect estimator that is consistent in the full-data setting. Existing across methods are inconsistent, but a different across method that averages the inverse probability weights across imputed datasets is consistent for propensity score weighting. We also comment on methods that rely on imputing a function of the missing covariate rather than the covariate itself, including imputation of the propensity score and of the probability weight. Based on consistency results and practical flexibility, we recommend generally using the standard within method. Throughout, we provide intuition to make the results meaningful to the broad audience of applied researchers.

Invited commentary: mixing multiple imputation and bootstrapping for variance estimation.

Multiple imputation (MI) is commonly implemented to mitigate potential selection bias due to missing data. The accompanying article by Nguyen and Stuart (Am J Epidemiol. 2024;193(10):1470-1476) examines the statistical consistency of several ways of integrating MI with propensity scores. As Nguyen and Stuart noted, variance estimation for these different approaches remains to be developed. One common option is the nonparametric bootstrap, which can provide valid inference when closed-form variance estimators are not available. However, there is no consensus on how to implement MI and nonparametric bootstrapping in analyses. To complement Nguyen and Stuart's article on MI and propensity score analyses, we review some currently available approaches on variance estimation with MI and nonparametric bootstrapping.

Targeted learning with an undersmoothed LASSO propensity score model for large-scale covariate adjustment in health-care database studies.

Least absolute shrinkage and selection operator (LASSO) regression is widely used for large-scale propensity score (PS) estimation in health-care database studies. In these settings, previous work has shown that undersmoothing (overfitting) LASSO PS models can improve confounding control, but it can also cause problems of nonoverlap in covariate distributions. It remains unclear how to select the degree of undersmoothing when fitting large-scale LASSO PS models to improve confounding control while avoiding issues that can result from reduced covariate overlap. Here, we used simulations to evaluate the performance of using collaborative-controlled targeted learning to data-adaptively select the degree of undersmoothing when fitting large-scale PS models within both singly and doubly robust frameworks to reduce bias in causal estimators. Simulations showed that collaborative learning can data-adaptively select the degree of undersmoothing to reduce bias in estimated treatment effects. Results further showed that when fitting undersmoothed LASSO PS models, the use of cross-fitting was important for avoiding nonoverlap in covariate distributions and reducing bias in causal estimates.

Thiazide Diuretics and Risk of Colorectal Cancer: A Population-Based Cohort Study.

Evidence from clinical trials and observational studies on the association between thiazide diuretics and colorectal cancer risk is conflicting. We aimed to determine whether thiazide diuretics are associated with an increased colorectal cancer risk compared with dihydropyridine calcium channel blockers (dCCBs). A population-based, new-user cohort was assembled using the UK Clinical Practice Research Datalink. Between 1990-2018, we compared thiazide diuretic initiators with dCCB initiators and estimated hazard ratios (HR) with 95% confidence intervals (CIs) of colorectal cancer using Cox proportional hazard models. Models were weighted using standardized morbidity ratio weights generated from calendar time-specific propensity scores. The cohort included 377,760 thiazide diuretic initiators and 364,300 dCCB initiators, generating 3,619,883 person-years of follow-up. Compared with dCCBs, thiazide diuretics were not associated with colorectal cancer (weighted HR = 0.97, 95% CI: 0.90, 1.04). Secondary analyses yielded similar results, although an increased risk was observed among patients with inflammatory bowel disease (weighted HR = 2.45, 95% CI: 1.13, 5.35) and potentially polyps (weighted HR = 1.46, 95% CI: 0.93, 2.30). Compared with dCCBs, thiazide diuretics were not associated with an overall increased colorectal cancer risk. While these findings provide some reassurance, research is needed to corroborate the elevated risks observed among patients with inflammatory bowel disease and history of polyps.

Characterizing Imbalance in the Tails of the Propensity Score Distribution.

Understanding characteristics of patients with propensity scores in the tails of the propensity score (PS) distribution has relevance for inverse-probability-of-treatment-weighted and PS-based estimation in observational studies. Here we outline a method for identifying variables most responsible for extreme propensity scores. The approach is illustrated in 3 scenarios: 1) a plasmode simulation of adult patients in the National Ambulatory Medical Care Survey (2011-2015) and 2) timing of dexamethasone initiation and 3) timing of remdesivir initiation in patients hospitalized for coronavirus disease 2019 from February 2020 through January 2021. PS models were fitted using relevant baseline covariates, and tails of the PS distribution were defined using asymmetric first and 99th percentiles. After fitting of the PS model in each original data set, values of each key covariate were permuted and model-agnostic variable importance measures were examined. Visualization and variable importance techniques were helpful in identifying variables most responsible for extreme propensity scores and may help identify individual characteristics that might make patients inappropriate for inclusion in a study (e.g., off-label use). Subsetting or restricting the study sample based on variables identified using this approach may help investigators avoid the need for trimming or overlap weights in studies.

A Cautionary Note on Using Propensity Score Calibration to Control for Unmeasured Confounding Bias When the Surrogacy Assumption Is Absent.

Conventional propensity score methods encounter challenges when unmeasured confounding is present, as it becomes impossible to accurately estimate the gold-standard propensity score when data on certain confounders are unavailable. Propensity score calibration (PSC) addresses this issue by constructing a surrogate for the gold-standard propensity score under the surrogacy assumption. This assumption posits that the error-prone propensity score, based on observed confounders, is independent of the outcome when conditioned on the gold-standard propensity score and the exposure. However, this assumption implies that confounders cannot directly impact the outcome and that their effects on the outcome are solely mediated through the propensity score. This raises concerns regarding the applicability of PSC in practical settings where confounders can directly affect the outcome. While PSC aims to target a conditional treatment effect by conditioning on a subject's unobservable propensity score, the causal interest in the latter case lies in a conditional treatment effect conditioned on a subject's baseline characteristics. Our analysis reveals that PSC is generally biased unless the effects of confounders on the outcome and treatment are proportional to each other. Furthermore, we identify 2 sources of bias: 1) the noncollapsibility of effect measures, such as the odds ratio or hazard ratio and 2) residual confounding, as the calibrated propensity score may not possess the properties of a valid propensity score.

Generalized Propensity Score Methods to Assess CKD-Associated Physiologic Factors and Risk of Kidney Failure in the Chronic Renal Insufficiency Cohort (CRIC) Study.

Epidemiologic studies have identified many biochemical risk factors for chronic kidney disease (CKD) progression that are correlates of kidney function, termed here 'CKD-associated physiologic factors'. Uncertainty remains if these factors are risk factors or risk markers accounting for aspects of kidney function not otherwise captured. We aimed to use flexible machine learning, a dynamic covariate history including kidney function informative markers, and generalized propensity score (GPS) weighting, to better control confounding for such exposures. We studied 3,052 adults with CKD in the Chronic Renal Insufficiency Cohort Study. We established a 2-year run-in period and assembled 90 variables that characterize variability and trends of selected CKD-associated physiologic factors and confounders. Using SuperLearner, we created a GPS for each CKD-associated physiologic factor and performed GPS-weighted Cox regressions. For context, we also evaluated results from traditional multivariable Cox proportional hazards models as in prior studies. Similar to traditional approaches, bicarbonate, calcium, potassium, hemoglobin, and PTH were each associated with risk of kidney failure using GPS weighting. The GPS approach detected non-linear associations in many factors, some of which were not detected with traditional models. We conclude that many associations between CKD-associated physiologic factors and kidney outcomes remain strong after GPS weighting.

Comparing traditional and causal inference methodologies for evaluating impacts of long-term air pollution exposure on hospitalization with Alzheimer's disease and related dementias.

Alzheimer's disease and related dementias (ADRD) present a growing public health burden in the United States. One actionable risk factor for ADRD is air pollution: multiple studies have found associations between air pollution and exacerbation of ADRD. Our study builds on previous studies by applying modern statistical causal inference methodologies-generalized propensity score (GPS) weighting and matching-on a large, longitudinal dataset. We follow 50 million Medicare enrollees to investigate impacts of three air pollutants-fine particular matter (PM${}_{2.5}$), nitrogen dioxide (NO${}_2$), and summer ozone (O${}_3$)-on elderly patients' rate of first hospitalization with ADRD diagnosis. Similar to previous studies using traditional statistical models, our results found increased hospitalization risks due to increased PM${}_{2.5}$ and NO${}_2$ exposure, with less conclusive results for O${}_3$. In particular, our GPS weighting analysis finds IQR increases in PM${}_{2.5}$, NO${}_2$, or O${}_3$ exposure results in hazard ratios of 1.108 (95% CI: 1.097-1.119), 1.058 (1.049-1.067), or 1.045 (1.036-1.054), respectively. GPS matching results are similar for PM${}_{2.5}$ and NO${}_2$ with attenuated effects for O${}_3$. Our results strengthen arguments that long-term PM${}_{2.5}$ and NO${}_2$ exposure increases risk of hospitalization with ADRD diagnosis. Additionally, we highlight strengths and limitations of causal inference methodologies in observational studies with continuous treatments. Keywords: Alzheimer's disease and related dementias, air pollution, Medicare, causal inference, generalized propensity score.

Using Overlap Weights to Address Extreme Propensity Scores in Estimating Restricted Mean Counterfactual Survival Times.

While inverse probability of treatment weighting (IPTW) is a commonly used approach for treatment comparisons in observational data, the resulting estimates may be subject to bias and excessively large variance under lack of overlap. By smoothly down-weighting units with extreme propensity scores, i.e., those that are close (or equal) to zero or one, overlap weighting (OW) can help mitigate the bias and variance issues associated with IPTW. Although theoretical and simulation results have supported the use of OW with continuous and binary outcomes, its performance with survival outcomes remains to be further investigated, especially when the target estimand is defined based on the restricted mean survival time (RMST). We combine propensity score weighting and inverse probability of censoring weighting to estimate the restricted mean counterfactual survival times, and provide computationally-efficient variance estimators when the propensity scores are estimated by logistic regression and the censoring process is estimated by Cox regression. We conduct simulations to compare the performance of weighting methods in terms of bias, variance, and 95% interval coverage, under various degrees of overlap. Under moderate and weak overlap, we demonstrate the advantage of OW over IPTW, trimming and truncation, with respect to bias, variance, and coverage when estimating RMST.