We propose an approach for constructing and evaluating the performance of inverse probability weighted robust mixture priors (IPW-RMP) which are applied to the parameters in treatment group-specific marginal models. Our framework allows practitioners to systematically study the robustness of Bayesian dynamic borrowing using the IPW-RMP to enhance the efficiency of inferences on marginal treatment effects (e.g. marginal risk difference) in a target study being planned. A key assumption motivating our work is that the data generation processes for the target study and external data source (e.g. historical study) will not be the same, likely having different distributions for key prognostic factors and possibly different outcome distributions even for individuals who have identical prognostic factors (e.g. different outcome model parameters). We demonstrate the approach using simulation studies based on both binary and time-to-event outcomes, and via a case study based on actual clinical trial data for a solid tumor cancer program. Our simulation results show that when the distribution of risk factors does in fact differ, the IPW-RMP provides improved performance compared to a standard RMP (e.g. increased power and reduced bias of the posterior mean point estimator) with essentially no loss of performance when the risk factor distributions do not differ. Thus, the IPW-RMP can safely be used in any situation where a standard RMP is appropriate.

Observational cohort studies are increasingly being used for comparative effectiveness research to assess the safety of therapeutics. Recently, various doubly robust methods have been proposed for average treatment effect estimation by combining the treatment model and the outcome model via different vehicles, such as matching, weighting, and regression. The key advantage of doubly robust estimators is that they require either the treatment model or the outcome model to be correctly specified to obtain a consistent estimator of average treatment effects, and therefore lead to a more accurate and often more precise inference. However, little work has been done to understand how doubly robust estimators differ due to their unique strategies of using the treatment and outcome models and how machine learning techniques can be combined to boost their performance, which we call double machine learning estimators. Here, we examine multiple popular doubly robust methods and compare their performance using different treatment and outcome modeling via extensive simulations and a real-world application. We found that incorporating machine learning with doubly robust estimators such as the targeted maximum likelihood estimator gives the best overall performance. Practical guidance on how to apply doubly robust estimators is provided.

This study explores the performance of deep learning models, specifically Convolutional Neural Networks (CNN) and XGBoost, in predicting alpha and beta thalassemia using both public and private datasets. Thalassemia is a genetic disorder that impairs hemoglobin production, leading to anemia and other health complications. Early diagnosis is essential for effective management and prevention of severe health issues. The study applied CNN and XGBoost to two case studies: one for alpha-thalassemia and the other for beta-thalassemia. Public datasets were sourced from medical databases, while private datasets were collected from clinical records, offering a more comprehensive feature set and larger sample sizes. After data preprocessing and splitting, model performance was evaluated. XGBoost achieved 99.34% accuracy on the private dataset for alpha thalassemia, while CNN reached 98.10% accuracy on the private dataset for beta-thalassemia. The superior performance on private datasets was attributed to better data quality and volume. This study highlights the effectiveness of deep learning in medical diagnostics, demonstrating that high-quality data can significantly enhance the predictive capabilities of AI models. By integrating CNN and XGBoost, this approach offers a robust method for detecting thalassemia, potentially improving early diagnosis and reducing disease-related mortality.

Limited evidence exists on the prognostic role of continuing medical therapy in patients with heart failure (HF) and an ejection fraction (EF) that has improved over time. This study assessed rates of, patient profiles, and associations with morbidity/mortality of renin-angiotensin inhibitors (RASi), angiotensin receptor-neprilysin inhibitors (ARNi), beta-blockers (BBL), and mineralocorticoid receptor antagonists (MRA) withdrawal in patients with HF with improved EF.

Patients with a first recorded EF <40% and a later EF ≥40% from the Swedish HF registry between June 11, 2000, and December 31, 2023, were included in this retrospective observational study. Withdrawal was defined as a patient on treatment at the first (reduced) but not at the second (improved) registration. The association between withdrawal and time to first cardiovascular mortality/hospitalization for HF with censoring at 1 year was assessed by Cox regression model using overlap weighting.

Of 8728 patients with HF with improved EF (median age, 70 years [25th to 75th percentile, 61-78], 2611 [29.9%] women), 96%, 94%, and 46% received RASi/ARNi, BBL, and MRA, respectively, when EF was <40%. The withdrawal rates at the time of the improved EF registration were 4.4% for RASi/ARNi, 3.3% for BBL, and 17.2% for MRA. Predictors of withdrawal included lower use of other HF medications, higher EF at the later EF registration, and a longer time between the 2 EF assessments. After weighting, withdrawal was independently associated with a higher risk of cardiovascular mortality/hospitalization for HF by 38% for RASi/ARNi and 36% for MRA, but not for BBL. Withdrawal of BBL was associated with a higher risk of the primary outcome in the subgroup of patients with an improved EF of 40% to 49% versus ≥50% (P-interaction 0.03).

In patients with HF with improved EF, HF therapy withdrawal was rare. Withdrawing RASi/ARNi and MRA was associated with higher mortality/morbidity at 1 year. No association was found for BBL withdrawal, albeit with a significant heterogeneity for EF at improvement, suggesting better outcomes with continuing BBL only until EF improves up to 50%. These results are hypothesis-generating and highlight the need for randomized controlled trials testing BBL withdrawal in patients with HF with improved EF.

Does endometrial preparation using a natural cycle lead to higher live birth rates (LBR) in single euploid frozen embryo transfers (FET) compared with programmed cycles, for women who are normal weight, overweight and obese.

Retrospective study of 845 single euploid FETs from 688 couples. Patients were stratified by body mass index (BMI) into normal weight, overweight and obesity class I/II categories. Outcome was LBR.

After achieving covariate (female age, anti-Müllerian hormone, embryo quality and infertility type) balance in each stratum, the effective sample size was 481 and 262 for the programmed cycles and natural cycles, respectively. The programmed cycle approach (vaginal luteal phase support with 3 × 100 mg micronized vaginal progesterone per day) was associated with significantly lower LBR in the weighted regression analysis of the cohort (RR 0.80, 95% CI 0.73 to 0.88, P < 0.001), compared with the natural cycle approach. The effect was significantly modified by BMI (P = 0.003 but was significant for all BMI categories. Reduction in live birth was less pronounced in patients with normal weight or who were overweight BMI (RR 0.87, 95% CI 0.78 to 0.97, P = 0.014) compared with patients with class I/II obesity (RR 0.61, 95% CI 0.49 to 0.75, P < 0.001).

A natural cycle endometrial preparation approach leads to overall better LBR in single euploid FET. The most significant difference is observed in women with higher BMI. Overweight or obese patients undergoing hormone replacement therapy may require a higher dosage of progesterone for luteal phase support.

There is mounting interest in the possibility that metformin, indicated for glycemic control in type 2 diabetes, has a range of additional beneficial effects. Randomized trials have shown that metformin prevents adverse cardiovascular events, and metformin use has also been associated with reduced cognitive decline and cancer incidence. In this paper, we dig more deeply into whether metformin prevents cancer by emulating target randomized trials comparing metformin to sulfonylureas as first-line diabetes therapy using data from the Clinical Practice Research Datalink, a UK primary-care database (1987-2018). We included 93 353 individuals with diabetes, no prior cancer diagnosis, no chronic kidney disease, and no prior diabetes therapy who initiated use of metformin (n = 79 489) or a sulfonylurea (n = 13 864). In our cohort, the estimated overlap-weighted additive separable direct effect of metformin compared with sulfonylureas on cancer risk at 6 years was -1 percentage point (95% CI, -2.2 to 0.1), which is consistent with metformin's providing no direct protection against cancer incidence or substantial protection. The analysis faced 2 methodological challenges: (1) poor overlap and (2) precancer death as a competing risk. To address these issues while minimizing nuisance model misspecification, we develop and apply double/debiased machine learning estimators of overlap-weighted separable effects in addition to more traditional effect estimates. This article is part of a Special Collection on Pharmacoepidemiology.

Prediction-powered inference (PPI) [1] and its subsequent development called PPI++ [2] provide a novel approach to standard statistical estimation leveraging machine learning systems to enhance unlabeled data with predictions. We use this paradigm in clinical trials. The predictions are provided by disease progression models, providing prognostic scores for all the participants as a function of baseline covariates. The proposed method would empower clinical trials by providing untreated digital twins of the treated patients while remaining statistically valid. The potential implications of this new estimator of the treatment effect in a two-arm randomized clinical trial (RCT) are manifold. First, it leads to an overall reduction of the sample size required to reach the same power as a standard RCT. Secondly, it advocates for an imbalance of controls and treated patients, requiring fewer controls to achieve the same power. Finally, this technique directly transfers any disease prediction model trained on large cohorts to practical and scientifically valid use. In this paper, we demonstrate the theoretical properties of this estimator and illustrate them through simulations. We show that it is asymptotically unbiased for the Average Treatment Effect and derive an explicit formula for its variance. An application to an Alzheimer's disease clinical trial showcases the potential to reduce the sample size.

In observational studies, identifying and adjusting for a sufficient set of confounders is crucial for accurately estimating the causal effect of the exposure on the outcome. Even in studies with large sample sizes, which typically benefit from small variances in estimates, there is a risk of producing estimates that are precisely inaccurate if the study suffers from systematic errors or biases, including confounding bias. To date, several approaches have been developed for selecting confounders. In this article, we first summarize the epidemiological and statistical approaches to identifying a sufficient set of confounders. Particularly, we introduce the modified disjunctive cause criterion as one of the most useful approaches, which involves controlling for any pre-exposure covariate that affects the exposure, outcome, or both. It then excludes instrumental variables but includes proxies for the shared common cause of exposure and outcome. Statistical confounder selection is also useful when dealing with a large number of covariates, even in studies with small sample sizes. After introducing several approaches, we discuss some pitfalls and considerations in confounder selection, such as the adjustment for instrumental variables, intermediate variables, and baseline outcome variables. Lastly, as it is often difficult to comprehensively measure key confounders, we introduce two statistics, E-value and robustness value, for assessing sensitivity to unmeasured confounders. Illustrated examples are provided using the National Health and Nutritional Examination Survey Epidemiologic Follow-up Study. Integrating these principles and approaches will enhance our understanding of confounder selection and facilitate better reporting and interpretation of future epidemiological studies.

With the rapid increase in the number of commercial chemicals, testing methods regarding on median lethal dose (LD50) relying animal experiments face challenges such as high costs and ethical concerns. Classical quantitative structure-activity relationship models relying on single algorithm always lack interpretability and precision, given the complexity of the mechanisms underlying acute toxicity. To address these issues, this study has developed a predictive framework using an ensemble learning model based on Super-learner. Particularly, we first obtained LD50 data for 9843 compounds and constructed 16 meta models using 4 molecular descriptors and machine learning algorithms. The Super-learner model performed well, achieving R² values of 0.61 and 0.64 in five-fold cross-validation and test sets, respectively, with corresponding root mean square errors of 0.55 and 0.64, significantly outperforming the results of individual model. Additionally, we incorporated data filtering and applicability domain methods, which demonstrated that the Super-learner can mitigate the impact of dataset noise to some extent. The model achieved an R² of 0.76 within an applicability domain, ensuring prediction accuracy within the chemical space. Compared to previous studies, the model developed here using Super-learner generally achieved better performance across a larger applicability domain. Finally, we has launched an online tool (http://sltox.hhra.net), allowing users to quickly predict LD50 of compounds, greatly simplifying the chemical safety assessment process. This study not only provides an effective and cost-efficient method for predicting chemical toxicity but also offers technical support and data for risk assessments of chemicals.

Estimating the causal effects of health policy interventions is crucial for policymaking but is challenging when using real-world administrative health care data due to a lack of methodological guidance. To help fill this gap, we conducted a plasmode simulation using such data from a recent policy initiative launched in a deprived urban area in Germany. Our aim was to evaluate and compare the following methods for estimating causal effects: propensity score matching, inverse probability of treatment weighting, and entropy balancing, all combined with difference-in-differences analysis, augmented inverse probability weighting, and targeted maximum likelihood estimation. Additionally, we estimated nuisance parameters using regression models and an ensemble learner called superlearner. We focused on treatment effects related to the number of physician visits, total health care cost, and hospitalization. While each approach has its strengths and weaknesses, our results demonstrate that the superlearner generally worked well for handling nuisance terms in large covariate sets when combined with doubly robust estimation methods to estimate the causal contrast of interest. In contrast, regression-based nuisance parameter estimation worked best in small covariate sets when combined with singly robust methods.

Machine learning (ML) methods are promising and scalable alternatives for propensity score (PS) estimation, but their comparative performance in disease risk score (DRS) estimation remains unexplored.

We used real-world data comparing antihypertensive users to non-users with 69 negative control outcomes, and plasmode simulations to study the performance of ML methods in PS and DRS estimation. We conducted a cohort study using UK primary care records. Further, we conducted a plasmode simulation with synthetic treatment and outcome mimicking empirical data distributions. We compared four PS and DRS estimation methods: 1. Reference: Logistic regression including clinically chosen confounders. 2. Logistic regression with L1 regularisation (LASSO). 3. Multi-layer perceptron (MLP). 4. Extreme Gradient Boosting (XgBoost). Covariate balance, coverage of the null effect of negative control outcomes (real-world data) and bias based on the absolute difference between observed and true effects (for plasmode) were estimated. 632,201 antihypertensive users and nonusers were included.

ML methods outperformed the reference method for PS estimation in some scenarios, both in terms of covariate balance and coverage/bias. Specifically, XgBoost achieved the best performance. DRS-based methods performed worse than PS in all tested scenarios.

We found that ML methods could be reliable alternatives for PS estimation. ML-based DRS methods performed worse than PS ones, likely given the rarity of outcomes.

Marginal structural models have been increasingly used by analysts in recent years to account for confounding bias in studies with time-varying treatments. The parameters of these models are often estimated using inverse probability of treatment weighting. To ensure that the estimated weights adequately control confounding, it is possible to check for residual imbalance between treatment groups in the weighted data. Several balance metrics have been developed and compared in the cross-sectional case but have not yet been evaluated and compared in longitudinal studies with time-varying treatment. We have first extended the definition of several balance metrics to the case of a time-varying treatment, with or without censoring. We then compared the performance of these balance metrics in a simulation study by assessing the strength of the association between their estimated level of imbalance and bias. We found that the Mahalanobis balance performed best. Finally, the method was illustrated for estimating the cumulative effect of statins exposure over one year on the risk of cardiovascular disease or death in people aged 65 and over in population-wide administrative data. This illustration confirms the feasibility of employing our proposed metrics in large databases with multiple time-points.

Propensity scores (PS) are typically evaluated using balance metrics that focus on covariate balance, often without considering their predictive power for the outcome. This approach may not always result in optimal bias reduction in the treatment effect estimate. To address this issue, evaluating covariate balance through prognostic scores, which account for the relationship between covariates and the outcome, has been proposed. Similarly, using a typical model averaging approach for PS estimation that minimizes prediction error for treatment status and covariate imbalance does not necessarily optimize PS-based confounding adjustment. As an alternative approach, using the averaged PS model that minimizes inter-group differences in the prognostic score may further reduce bias in the treatment effect estimate. Moreover, since the prognostic score is also an estimated quantity, model averaging in the prognostic scores can help identify a better prognostic score model. Utilizing the model-averaged prognostic scores as the balance metric for constructing the averaged PS model can contribute to further decreasing bias in treatment effect estimates. This paper demonstrates the effectiveness of the PS model averaging approach based on prognostic score balance and proposes a method that uses the model-averaged prognostic score as a balance metric, evaluating its performance through simulations and empirical analysis.

We conduct a series of simulations alongside an analysis of empirical observational data to compare the performances of weighted treatment effect estimates using the proposed and existing approaches. In our examination, we separately provid four candidate estimates for the PS and prognostic score models using traditional regression and machine learning methods. The model averaging of PS based on these candidate estimators is performed to either maximize the prediction accuracy of the treatment or to minimize intergroup differences in covariate distributions or prognostic scores. We also utilize not only the prognostic scores from each candidate model but also an averaged score that best predicted the outcome, for the balance assessment.

The simulation and empirical data analysis reveal that our proposed model-averaging approaches for PS estimation consistently yield lower bias and less variability in treatment effect estimates across various scenarios compared to existing methods. Specifically, using the optimally averaged prognostic scores as a balance metric significantly improves the robustness of the weighted treatment effect estimates.

The prognostic score-based model averaging approach for estimating PS can outperform existing model averaging methods. In particular, the estimator using the model averaging prognostic score as a balance metric can produce more robust estimates. Since our results are obtained under relatively simple conditions, applying them to real data analysis requires adjustments to obtain accurate estimates according to the complexity and dimensionality of the data.

Using the prognostic score as the balance metric for the PS model averaging enhances the performance of the treatment effect estimator, which can be recommended for a wide variety of situations. When applying the proposed method to real-world data, it is important to use it in conjunction with techniques that mitigate issues arising from the complexity and high dimensionality of the data.

We assessed the association of early statin initiation with inpatient mortality among hospitalised COVID-19 patients.

This observational study emulated a hypothetical target trial using electronic health records data from Northwestern Medicine Health System, Illinois, 2020-2022. We included patients who were ≥40 years, admitted ≥48 hours for COVID-19 from March 2020 to August 2022 and had no evidence of statin use before admission.

Individuals who initiated any statins within 48 hours of admission were compared with individuals who did not initiate statins during this period.

Inpatient mortality at hospital days 7, 14, 21 and 28 were determined using hospital records. Risk differences between exposure groups were calculated using augmented inverse propensity weighting (AIPW) with SuperLearner.

A total of 8893 individuals (24.5% early statin initiators) were included. Early initiators tended to be older, male and have higher comorbidity burdens. Unadjusted day 28 mortality was higher in early initiators (6.0% vs 3.6%). Adjusted analysis showed slightly higher inpatient mortality risk at days 7 (RD: 0.5%, 95% CI: 0.2 to 0.8) and 21 (RD: 0.6%, 95% CI: 0.04 to 1.1), but not days 14 (RD: 0.4%, 95% CI: -0.03 to 0.9) and 28 (RD: 0.4%, 95% CI: -0.2 to 1.1). Sensitivity analyses using alternative modelling approaches showed no difference between groups.

Early statin initiation was not associated with lower mortality contrasting with findings of previous observational studies. Trial emulation helped in identifying and addressing sources of bias incompletely addressed by previous work. Statin use may be indicated for other conditions but not COVID-19.

Critical care medicine has undergone significant evaluation in the 21st century, primarily driven by advancements in technology, changes in healthcare delivery, and a deeper understanding of disease processes. Advancements in technology have revolutionized patient monitoring, diagnosis, and treatment in the critical care setting. From minimally invasive procedures to advances imaging techniques, clinicians now have access to a wide array of tools to assess and manage critically ill patients more effectively. In this editorial we comment on the review article published by Padte S et al wherein they concisely describe the latest developments in critical care medicine.

Propensity score matching is vital in epidemiological studies using observational data, yet its estimates relies on correct model-specification. This study assesses supervised deep learning models and unsupervised autoencoders for propensity score estimation, comparing them with traditional methods for bias and variance accuracy in treatment effect estimations.

Utilizing a plasmode simulation based on the Right Heart Catheterization dataset, under a variety of settings, we evaluated (1) a supervised deep learning architecture and (2) an unsupervised autoencoder, alongside two traditional methods: logistic regression and a spline-based method in estimating propensity scores for matching. Performance metrics included bias, standard errors, and coverage probability. The analysis was also extended to real-world data, with estimates compared to those obtained via a double robust approach.

The analysis revealed that supervised deep learning models outperformed unsupervised autoencoders in variance estimation while maintaining comparable levels of bias. These results were supported by analyses of real-world data, where the supervised model's estimates closely matched those derived from conventional methods. Additionally, deep learning models performed well compared to traditional methods in settings where exposure was rare.

Supervised deep learning models hold promise in refining propensity score estimations in epidemiological research, offering nuanced confounder adjustment, especially in complex datasets. We endorse integrating supervised deep learning into epidemiological research and share reproducible codes for widespread use and methodological transparency.

We investigate estimation of causal effects of multiple competing (multi-valued) treatments in the absence of randomization. Our work is motivated by an intention-to-treat study of the relative cardiometabolic risk of assignment to one of six commonly prescribed antipsychotic drugs in a cohort of nearly 39 000 adults with serious mental illnesses. Doubly-robust estimators, such as targeted minimum loss-based estimation (TMLE), require correct specification of either the treatment model or outcome model to ensure consistent estimation; however, common TMLE implementations estimate treatment probabilities using multiple binomial regressions rather than multinomial regression. We implement a TMLE estimator that uses multinomial treatment assignment and ensemble machine learning to estimate average treatment effects. Our multinomial implementation improves coverage, but does not necessarily reduce bias, relative to the binomial implementation in simulation experiments with varying treatment propensity overlap and event rates. Evaluating the causal effects of the antipsychotics on 3-year diabetes risk or death, we find a safety benefit of moving from a second-generation drug considered among the safest of the second-generation drugs to an infrequently prescribed first-generation drug known for having low cardiometabolic risk.

Practical problems with missing data are common, and many methods have been developed concerning the validity and/or efficiency of statistical procedures. On a central focus, there have been longstanding interests on the mechanism governing data missingness, and correctly deciding the appropriate mechanism is crucially relevant for conducting proper practical investigations. In this paper, we present a new hypothesis testing approach for deciding between the conventional notions of missing at random and missing not at random in generalized linear models in the presence of instrumental variables. The foundational idea is to develop appropriate discrepancy measures between estimators whose properties significantly differ only when missing at random does not hold. We show that our testing approach achieves an objective data-oriented choice between missing at random or not. We demonstrate the feasibility, validity, and efficacy of the new test by theoretical analysis, simulation studies, and a real data analysis.

We investigated the impact of an integrated care initiative in a socially deprived urban area in Germany. Using administrative data, we empirically assessed the causal effect of its two sub-interventions, which differed by the extent to which their instruments targeted the supply and demand side of healthcare provision. We addressed confounding using propensity score matching via the Super Learner machine learning algorithm. For our baseline model, we used a two-way fixed-effects difference-in-differences approach to identify causal effects. We then employed difference-in-differences analyses within an event-study framework to explore the heterogeneity of treatment effects over time, allowing us to disentangle the effects of the sub-interventions and improve causal interpretation and generalizability. The initiative led to a significant increase in hospital and emergency admissions and non-hospital outpatient visits, as well as inpatient, non-hospital outpatient, and total costs. Increased utilization may indicate that the intervention improved access to care or identified unmet need.

One of the most common ways researchers compare cancer survival outcomes across treatments from observational data is using Cox regression. This model depends on its underlying assumption of proportional hazards, but in some real-world cases, such as when comparing different classes of cancer therapies, substantial violations may occur. In this situation, researchers have several alternative methods to choose from, including Cox models with time-varying hazard ratios; parametric accelerated failure time models; Kaplan-Meier curves; and pseudo-observations. It is unclear which of these models are likely to perform best in practice. To fill this gap in the literature, we perform a neutral comparison study of candidate approaches. We examine clinically meaningful outcome measures that can be computed and directly compared across each method, namely, survival probability at time T, median survival, and restricted mean survival. To adjust for differences between treatment groups, we use inverse probability of treatment weighting based on the propensity score. We conduct simulation studies under a range of scenarios, and determine the biases, coverages, and standard errors of the average treatment effects for each method. We then demonstrate the use of these approaches using two published observational studies of survival after cancer treatment. The first examines chemotherapy in sarcoma, which has a late treatment effect (i.e., similar survival initially, but after 2 years the chemotherapy group shows a benefit). The other study is a comparison of surgical techniques for kidney cancer, where survival differences are attenuated over time.

Individuals with schizophrenia exposed to second-generation antipsychotics (SGA) have an increased risk for diabetes, with aripiprazole purportedly a safer drug. Less is known about the drugs' mortality risk or whether serious mental illness (SMI) diagnosis or race/ethnicity modify these effects.

Authors created a retrospective cohort of non-elderly adults with SMI initiating monotherapy with an SGA (olanzapine, quetiapine, risperidone, and ziprasidone, aripiprazole) or haloperidol during 2008-2013. Three-year diabetes incidence or all-cause death risk differences were estimated between each drug and aripiprazole, the comparator, as well as effects within SMI diagnosis and race/ethnicity. Sensitivity analyses evaluated potential confounding by indication.

38 762 adults, 65% White and 55% with schizophrenia, initiated monotherapy, with haloperidol least (6%) and quetiapine most (26·5%) frequent. Three-year mortality was 5% and diabetes incidence 9.3%. Compared with aripiprazole, haloperidol and olanzapine reduced diabetes risk by 1.9 (95% CI 1.2-2.6) percentage points, or a 18.6 percentage point reduction relative to aripiprazole users' unadjusted risk (10.2%), with risperidone having a smaller advantage. Relative to aripiprazole users' unadjusted risk (3.4%), all antipsychotics increased mortality risk by 1.1-2.2 percentage points, representing 32.4-64.7 percentage point increases. Findings within diagnosis and race/ethnicity were generally consistent with overall findings. Only quetiapine's higher mortality risk held in sensitivity analyses.

Haloperidol's, olanzapine's, and risperidone's lower diabetes risks relative to aripiprazole were not robust in sensitivity analyses but quetiapine's higher mortality risk proved robust. Findings expand the evidence on antipsychotics' risks, suggesting a need for caution in the use of quetiapine among individuals with SMI.

Bladder cancer poses a significant public health burden, with high recurrence and progression rates in patients with non-muscle-invasive bladder cancer (NMIBC). Current treatment options include bladder-sparing therapies (BST) and radical cystectomy, both with associated risks and benefits. However, evidence supporting optimal management decisions for patients with recurrent high-grade NMIBC remains limited, leading to uncertainty for patients and clinicians. The CISTO (Comparison of Intravesical Therapy and Surgery as Treatment Options) Study aims to address this critical knowledge gap by comparing outcomes between patients undergoing BST and radical cystectomy.

The CISTO Study is a pragmatic, prospective observational cohort trial across 36 academic and community urology practices in the US. The study will enroll 572 patients with a diagnosis of recurrent high-grade NMIBC who select management with either BST or radical cystectomy. The primary outcome is health-related quality of life (QOL) at 12 months as measured with the EORTC-QLQ-C30. Secondary outcomes include bladder cancer-specific QOL, progression-free survival, cancer-specific survival, and financial toxicity. The study will also assess patient preferences for treatment outcomes. Statistical analyses will employ targeted maximum likelihood estimation (TMLE) to address treatment selection bias and confounding by indication.

The CISTO Study is powered to detect clinically important differences in QOL and cancer-specific survival between the two treatment approaches. By including a diverse patient population, the study also aims to assess outcomes across the following patient characteristics: age, gender, race, burden of comorbid health conditions, cancer severity, caregiver status, social determinants of health, and rurality. Treatment outcomes may also vary by patient preferences, health literacy, and baseline QOL. The CISTO Study will fill a crucial evidence gap in the management of recurrent high-grade NMIBC, providing evidence-based guidance for patients and clinicians in choosing between BST and radical cystectomy. The CISTO study will provide an evidence-based approach to identifying the right treatment for the right patient at the right time in the challenging clinical setting of recurrent high-grade NMIBC.

ClinicalTrials.gov, NCT03933826. Registered on May 1, 2019.

Plasmode simulations are a type of simulations that use real data to determine the synthetic data-generating equations. Such simulations thus allow evaluating statistical methods under realistic conditions. As far as we know, no plasmode algorithm has been proposed for simulating longitudinal data. In this paper, we propose a longitudinal plasmode framework to generate realistic data with both a time-varying exposure and time-varying covariates. This work was motivated by the objective of comparing different methods for estimating the causal effect of a cumulative exposure to psychosocial stressors at work over time.

We developed two longitudinal plasmode algorithms: a parametric and a nonparametric algorithms. Data from the PROspective Québec (PROQ) Study on Work and Health were used as an input to generate data with the proposed plasmode algorithms. We evaluated the performance of multiple estimators of the parameters of marginal structural models (MSMs): inverse probability of treatment weighting, g-computation and targeted maximum likelihood estimation. These estimators were also compared to standard regression approaches with either adjustment for baseline covariates only or with adjustment for both baseline and time-varying covariates.

Standard regression methods were susceptible to yield biased estimates with confidence intervals having coverage probability lower than their nominal level. The bias was much lower and coverage of confidence intervals was much closer to the nominal level when considering MSMs. Among MSM estimators, g-computation overall produced the best results relative to bias, root mean squared error and coverage of confidence intervals. No method produced unbiased estimates with adequate coverage for all parameters in the more realistic nonparametric plasmode simulation.

The proposed longitudinal plasmode algorithms can be important methodological tools for evaluating and comparing analytical methods in realistic simulation scenarios. To facilitate the use of these algorithms, we provide R functions on GitHub. We also recommend using MSMs when estimating the effect of cumulative exposure to psychosocial stressors at work.

There has been growing interest in using nonparametric machine learning approaches for propensity score estimation in order to foster robustness against misspecification of the propensity score model. However, the vast majority of studies focused on single-level data settings, and research on nonparametric propensity score estimation in clustered data settings is scarce. In this article, we extend existing research by describing a general algorithm for incorporating random effects into a machine learning model, which we implemented for generalized boosted modeling (GBM). In a simulation study, we investigated the performance of logistic regression, GBM, and Bayesian additive regression trees for inverse probability of treatment weighting (IPW) when the data are clustered, the treatment exposure mechanism is nonlinear, and unmeasured cluster-level confounding is present. For each approach, we compared fixed and random effects propensity score models to single-level models and evaluated their use in both marginal and clustered IPW. We additionally investigated the performance of the standard Super Learner and the balance Super Learner. The results showed that when there was no unmeasured confounding, logistic regression resulted in moderate bias in both marginal and clustered IPW, whereas the nonparametric approaches were unbiased. In presence of cluster-level confounding, fixed and random effects models greatly reduced bias compared to single-level models in marginal IPW, with fixed effects GBM and fixed effects logistic regression performing best. Finally, clustered IPW was overall preferable to marginal IPW and the balance Super Learner outperformed the standard Super Learner, though neither worked as well as their best candidate model.

In recent decades, there has been growing interest in leveraging external data information for clinical development as it improves the efficiency of the design and inference of clinical trials when utilized properly and more importantly, alleviates potential ethical and recruitment challenges. When it is of interest to augment the concurrent study's control arm using external control data, the potential outcome heterogeneity across data sources, also known as prior-data conflict, should be accounted for. In addition, in the outcome modeling, inclusion of prognostic covariates that may have impact on the outcome can avoid efficiency loss or potential bias. In this paper, we propose a Bayesian hierarchical modeling strategy incorporating covariate-adjusted meta-analytic predictive approach (cMAP) and also introduce a propensity score (PS) based sequential procedure that integrates the cMAP. In the simulation study, the proposed methods are found to have advantages in the estimation, power, and type I error control over the standard methods such as PS matching alone and hierarchical modeling that ignores the covariates. An illustrative example is used to illustrate the procedure.

To examine associations of pulmonary artery catheter (PAC) use with in-hospital death and hospital length of stay (days) overall and within subgroups of hospitalized cardiac surgery patients. Secondary analyses of 1999-2019 National Inpatient Sample data were performed using 969,034 records (68% male, mean age: 65 years) representing adult cardiac surgery patients in the United States. A subgroup of 323,929 records corresponded to patients with congestive heart failure, pulmonary hypertension, mitral/tricuspid valve disease and/or combined surgeries. We evaluated PAC in relation to clinical outcomes using regression and targeted maximum likelihood estimation (TMLE). Hospitalized cardiac surgery patients experienced more in-hospital deaths and longer stays if they had ≥ 1 subgroup characteristics. For risk-adjusted models, in-hospital deaths were similar among recipients and non-recipients of PAC (odds ratio [OR] 1.04, 95% confidence interval [CI] 0.96, 1.12), although PAC was associated with more in-hospital deaths among the subgroup with congestive heart failure (OR 1.14, 95% CI 1.03, 1.26). PAC recipients experienced shorter stays than non-recipients (β =  - 0.40, 95% CI - 0.64, - 0.15), with variations by subgroup. We obtained comparable results using TMLE. In this retrospective cohort study, PAC was associated with shorter stays and similar in-hospital death rates among cardiac surgery patients. Worse clinical outcomes associated with PAC were observed only among patients with congestive heart failure. Prospective cohort studies and randomized controlled trials are needed to confirm and extend these preliminary findings.

Common tasks encountered in epidemiology, including disease incidence estimation and causal inference, rely on predictive modelling. Constructing a predictive model can be thought of as learning a prediction function (a function that takes as input covariate data and outputs a predicted value). Many strategies for learning prediction functions from data (learners) are available, from parametric regressions to machine learning algorithms. It can be challenging to choose a learner, as it is impossible to know in advance which one is the most suitable for a particular dataset and prediction task. The super learner (SL) is an algorithm that alleviates concerns over selecting the one 'right' learner by providing the freedom to consider many, such as those recommended by collaborators, used in related research or specified by subject-matter experts. Also known as stacking, SL is an entirely prespecified and flexible approach for predictive modelling. To ensure the SL is well specified for learning the desired prediction function, the analyst does need to make a few important choices. In this educational article, we provide step-by-step guidelines for making these decisions, walking the reader through each of them and providing intuition along the way. In doing so, we aim to empower the analyst to tailor the SL specification to their prediction task, thereby ensuring their SL performs as well as possible. A flowchart provides a concise, easy-to-follow summary of key suggestions and heuristics, based on our accumulated experience and guided by SL optimality theory.

Since rapid population growth challenges longitudinal population-based HIV cohorts in Africa to maintain coverage of their target populations, this study evaluated whether the exclusion of some residents due to growing population size biases key HIV metrics like prevalence and population-level viremia.

Data were obtained from the Rakai Community Cohort Study (RCCS) in south central Uganda, an open population-based cohort which began excluding some residents of newly constructed household structures within its surveillance boundaries in 2008. The study includes adults aged 15-49 years who were censused from 2019 to 2020.

We fit ensemble machine learning models to RCCS census and survey data to predict HIV seroprevalence and viremia (prevalence of those with viral load >1000 copies/mL) in the excluded population and evaluated whether their inclusion would change overall estimates.

Of the 24 729 census-eligible residents, 2920 (12%) residents were excluded from the RCCS because they were living in new households. The predicted seroprevalence for these excluded residents was 10.8% (95% CI: 9.6% to 11.8%)-somewhat lower than 11.7% (95% CI: 11.2% to 12.3%) in the observed sample. Predicted seroprevalence for younger excluded residents aged 15-24 years was 4.9% (95% CI: 3.6% to 6.1%)-significantly higher than that in the observed sample for the same age group (2.6% (95% CI: 2.2% to 3.1%)), while predicted seroprevalence for older excluded residents aged 25-49 years was 15.0% (95% CI: 13.3% to 16.4%)-significantly lower than their counterparts in the observed sample (17.2% (95% CI: 16.4% to 18.1%)). Over all ages, the predicted prevalence of viremia in excluded residents (3.7% (95% CI: 3.0% to 4.5%)) was significantly higher than that in the observed sample (1.7% (95% CI: 1.5% to 1.9%)), resulting in a higher overall population-level viremia estimate of 2.1% (95% CI: 1.8% to 2.4%).

Exclusion of residents in new households may modestly bias HIV viremia estimates and some age-specific seroprevalence estimates in the RCCS. Overall, HIV seroprevalence estimates were not significantly affected.

The optimal dynamic treatment rule (ODTR) framework offers an approach for understanding which kinds of patients respond best to specific treatments - in other words, treatment effect heterogeneity. Recently, there has been a proliferation of methods for estimating the ODTR. One such method is an extension of the SuperLearner algorithm - an ensemble method to optimally combine candidate algorithms extensively used in prediction problems - to ODTRs. Following the ``causal roadmap," we causally and statistically define the ODTR and provide an introduction to estimating it using the ODTR SuperLearner. Additionally, we highlight practical choices when implementing the algorithm, including choice of candidate algorithms, metalearners to combine the candidates, and risk functions to select the best combination of algorithms. Using simulations, we illustrate how estimating the ODTR using this SuperLearner approach can uncover treatment effect heterogeneity more effectively than traditional approaches based on fitting a parametric regression of the outcome on the treatment, covariates and treatment-covariate interactions. We investigate the implications of choices in implementing an ODTR SuperLearner at various sample sizes. Our results show the advantages of: (1) including a combination of both flexible machine learning algorithms and simple parametric estimators in the library of candidate algorithms; (2) using an ensemble metalearner to combine candidates rather than selecting only the best-performing candidate; (3) using the mean outcome under the rule as a risk function. Finally, we apply the ODTR SuperLearner to the ``Interventions" study, an ongoing randomized controlled trial, to identify which justice-involved adults with mental illness benefit most from cognitive behavioral therapy to reduce criminal re-offending.

Understanding the impact of national public expenditure and its allocation on child mortality may help governments move towards target 3.2 proposed in the 2030 Agenda. The objective of this study was to estimate the impacts of governmental expenditures, total, on health, and on other sectors, on neonatal mortality and mortality of children aged between 28 days and five years.

This study has an ecological design with a population of 147 countries, with data between 2012 and 2019. Two steps were used: first, the Generalized Propensity Score of public spending was calculated; afterward, the Generalized Propensity Score was used to estimate the expenditures' association with mortality rates. The primary outcomes were neonatal mortality rates (NeoRt) and mortality rates in children between 28 days and 5 years (NeoU5Rt).

The 1% variation in Int$ Purchasing Power Parity (Int$ PPP) per capita in total public expenditures, expenditure in health, and in other sectors were associated with a variation of -0.635 (95% CI -1.176, -0.095), -2.17 (95% CI -3.051, -1.289) -0.632 (95% CI -1.169, -0.095) in NeoRt, respectively The same variation in public expenditures in sectors other than health, was associates with a variation of -1.772 (95% CI -6.219, -1.459) on NeoU5Rt. The results regarding the impact of total and health public spending on NeoU5Rt were not consistent.

Public investments impact mortality in children under 5 years of age. Likely, the allocation of expenditures between the health sector and the other social sectors will have different impacts on mortality between the NeoRt and the NeoU5Rt.

Multimorbidity is the presence of two or more chronic health conditions. Tuberculosis (TB) survivors are known to have higher prevalence of multimorbidity, although prevalence estimates from high-income low-TB incidence jurisdictions are not available and potential differences in the patterns of chronic disease among TB survivors with multimorbidity are poorly understood. In this study, we aimed to (1) compare the prevalence of multimorbidity among TB survivors with matched non-TB controls in a high-income setting; (2) assess the robustness of aim 1 analyses to different modelling strategies, unmeasured confounding, and misclassification bias; and (3) among people with multimorbidity, elucidate chronic disease patterns specific to TB survivors.

A population-based cohort study of people immigrating to British Columbia, Canada, 1985-2015, using health administrative data. Participants were divided into two groups: people diagnosed with TB (TB survivors) and people not diagnosed with TB (non-TB controls) in British Columbia. Coarsened exact matching (CEM) balanced demographic, immigration, and socioeconomic covariates between TB survivors and matched non-TB controls. Our primary outcome was multimorbidity, defined as ≥2 chronic diseases from the Elixhauser comorbidity index.

In the CEM-matched sample (n=1962 TB survivors; n=1962 non-TB controls), we estimated that 21.2% of TB survivors (n=416), compared with 12% of non-TB controls (n=236), had multimorbidity. In our primary analysis, we found a double-adjusted prevalence ratio of 1.74 (95% CI: 1.49-2.05) between TB survivors and matched non-TB controls for multimorbidity. Among people with multimorbidity, differences were observed in chronic disease frequencies between TB survivors and matched controls.

TB survivors had a 74% higher prevalence of multimorbidity compared with CEM-matched non-TB controls. TB-specific multimorbidity patterns were observed through differences in chronic disease frequencies between the matched samples. These findings suggest a need for TB-specific multimorbidity interventions in high-income settings such as Canada. We suggest TB survivorship as a framework for developing person-centred interventions for multimorbidity among TB survivors.

Covariate balance is crucial in obtaining unbiased estimates of treatment effects in observational studies. Methods that target covariate balance have been successfully proposed and largely applied to estimate treatment effects on continuous outcomes. However, in many medical and epidemiological applications, the interest lies in estimating treatment effects on time-to-event outcomes. With this type of data, one of the most common estimands of interest is the marginal hazard ratio of the Cox proportional hazards model. In this article, we start by presenting robust orthogonality weights, a set of weights obtained by solving a quadratic constrained optimization problem that maximizes precision while constraining covariate balance defined as the correlation between confounders and treatment. By doing so, robust orthogonality weights optimally deal with both binary and continuous treatments. We then evaluate the performance of the proposed weights in estimating marginal hazard ratios of binary and continuous treatments with time-to-event outcomes in a simulation study. We finally apply robust orthogonality weights in the evaluation of the effect of hormone therapy on time to coronary heart disease and on the effect of red meat consumption on time to colon cancer among 24,069 postmenopausal women enrolled in the Women's Health Initiative observational study.

Statin use prior to hospitalization for Coronavirus Disease 2019 (COVID-19) is hypothesized to improve inpatient outcomes including mortality, but prior findings from large observational studies have been inconsistent, due in part to confounding. Recent advances in statistics, including incorporation of machine learning techniques into augmented inverse probability weighting with targeted maximum likelihood estimation, address baseline covariate imbalance while maximizing statistical efficiency.

To estimate the association of antecedent statin use with progression to severe inpatient outcomes among patients admitted for COVD-19.

We retrospectively analyzed electronic health records (EHR) from individuals ≥ 40-years-old who were admitted between March 2020 and September 2022 for ≥ 24 h and tested positive for SARS-CoV-2 infection in the 30 days before to 7 days after admission.

Antecedent statin use-statin prescription ≥ 30 days prior to COVID-19 admission.

Composite end point of in-hospital death, intubation, and intensive care unit (ICU) admission.

Of 15,524 eligible COVID-19 patients, 4412 (20%) were antecedent statin users. Compared with non-users, statin users were older (72.9 (SD: 12.6) versus 65.6 (SD: 14.5) years) and more likely to be male (54% vs. 51%), White (76% vs. 71%), and have ≥ 1 medical comorbidity (99% vs. 86%). Unadjusted analysis demonstrated that a lower proportion of antecedent users experienced the composite outcome (14.8% vs 19.3%), ICU admission (13.9% vs 18.3%), intubation (5.1% vs 8.3%) and inpatient deaths (4.4% vs 5.2%) compared with non-users. Risk differences adjusted for labs and demographics were estimated using augmented inverse probability weighting with targeted maximum likelihood estimation using Super Learner. Statin users still had lower rates of the composite outcome (adjusted risk difference: - 3.4%; 95% CI: - 4.6% to - 2.1%), ICU admissions (- 3.3%; - 4.5% to - 2.1%), and intubation (- 1.9%; - 2.8% to - 1.0%) but comparable inpatient deaths (0.6%; - 1.3% to 0.1%).

After controlling for confounding using doubly robust methods, antecedent statin use was associated with minimally lower risk of severe COVID-19-related outcomes, ICU admission and intubation, however, we were not able to corroborate a statin-associated mortality benefit.

In causal studies, the near-violation of the positivity may occur by chance, because of sample-to-sample fluctuation despite the theoretical veracity of the positivity assumption in the population. It may mostly happen when the exposure prevalence is low or when the sample size is small. We aimed to compare the robustness of g-computation (GC), inverse probability weighting (IPW), truncated IPW, targeted maximum likelihood estimation (TMLE), and truncated TMLE in this situation, using simulations and one real application. We also tested different extrapolation situations for the sub-group with a positivity violation. The results illustrated that the near-violation of the positivity impacted all methods. We demonstrated the robustness of GC and TMLE-based methods. Truncation helped in limiting the bias in near-violation situations, but at the cost of bias in normal conditions. The application illustrated the variability of the results between the methods and the importance of choosing the most appropriate one. In conclusion, compared to propensity score-based methods, methods based on outcome regression should be preferred when suspecting near-violation of the positivity assumption.

Existing studies have suggested superior performance of nonparametric machine learning over logistic regression for propensity score estimation. However, it is unclear whether the advantages of nonparametric propensity score modeling are carried to settings where there is clustering of individuals, especially when there is unmeasured cluster-level confounding. In this work we examined the performance of logistic regression (all main effects), Bayesian additive regression trees and generalized boosted modeling for propensity score weighting in clustered settings, with the clustering being accounted for by including either cluster indicators or random intercepts. We simulated data for three hypothetical observational studies of varying sample and cluster sizes. Confounders were generated at both levels, including a cluster-level confounder that is unobserved in the analyses. A binary treatment and a continuous outcome were generated based on seven scenarios with varying relationships between the treatment and confounders (linear and additive, nonlinear/nonadditive, nonadditive with the unobserved cluster-level confounder). Results suggest that when the sample and cluster sizes are large, nonparametric propensity score estimation may provide better covariate balance, bias reduction, and 95% confidence interval coverage, regardless of the degree of nonlinearity or nonadditivity in the true propensity score model. When the sample or cluster sizes are small, however, nonparametric approaches may become more vulnerable to unmeasured cluster-level confounding and thus may not be a better alternative to multilevel logistic regression. We applied the methods to the National Longitudinal Study of Adolescent to Adult Health data, estimating the effect of team sports participation during adolescence on adulthood depressive symptoms.

Post-discharge opioid consumption is a crucial patient-reported outcome informing opioid prescribing guidelines, but its collection is resource-intensive and vulnerable to inaccuracy due to nonresponse bias.

We developed a post-discharge text message-to-web survey system for efficient collection of patient-reported pain outcomes. We prospectively recruited surgical patients at Beth Israel Deaconess Medical Center in Boston, Massachusetts from March 2019 through October 2020, sending an SMS link to a secure web survey to quantify opioids consumed after discharge from hospitalization. Patient factors extracted from the electronic health record were tested for nonresponse bias and observable confounding. Following targeted learning-based nonresponse adjustment, procedure-specific opioid consumption quantiles (medians and 75th percentiles) were estimated and compared to a previous telephone-based reference survey.

6553 patients were included. Opioid consumption was measured in 44% of patients (2868), including 21% (1342) through survey response. Characteristics associated with inability to measure opioid consumption included age, tobacco use, and prescribed opioid dose. Among the 10 most common procedures, median consumption was only 36% of the median prescription size; 64% of prescribed opioids were not consumed. Among those procedures, nonresponse adjustment corrected the median opioid consumption by an average of 37% (IQR: 7, 65%) compared to unadjusted estimates, and corrected the 75th percentile by an average of 5% (IQR: 0, 12%). This brought median estimates for 5/10 procedures closer to telephone survey-based consumption estimates, and 75th percentile estimates for 2/10 procedures closer to telephone survey-based estimates.

SMS-recruited online surveying can generate reliable opioid consumption estimates after nonresponse adjustment using patient factors recorded in the electronic health record, protecting patients from the risk of inaccurate prescription guidelines.

Children may be exposed to numerous in-home environmental exposures (IHEE) that trigger asthma exacerbations. Spatially linking social and environmental exposures to electronic health records (EHR) can aid exposure assessment, epidemiology, and clinical treatment, but EHR data on exposures are missing for many children with asthma. To address the issue, we predicted presence of indoor asthma trigger allergens, and estimated effects of their key geospatial predictors.

Our study samples were comprised of children with asthma who provided self-reported IHEE data in EHR at a safety-net hospital in New England during 2004-2015. We used an ensemble machine learning algorithm and 86 multilevel features (e.g., individual, housing, neighborhood) to predict presence of cockroaches, rodents (mice or rats), mold, and bedroom carpeting/rugs in homes. We reduced dimensionality via elastic net regression and estimated effects by the G-computation causal inference method.

Our models reasonably predicted presence of cockroaches (area under receiver operating curves [AUC] = 0.65), rodents (AUC = 0.64), and bedroom carpeting/rugs (AUC = 0.64), but not mold (AUC = 0.54). In models adjusted for confounders, higher average household sizes in census tracts were associated with more reports of pests (cockroaches and rodents). Tax-exempt parcels were associated with more reports of cockroaches in homes. Living in a White-segregated neighborhood was linked with lower reported rodent presence, and mixed residential/commercial housing and newer buildings were associated with more reports of bedroom carpeting/rugs in bedrooms.

We innovatively applied a machine learning and causal inference mixture methodology to detail IHEE among children with asthma using EHR and geospatial data, which could have wide applicability and utility.

Controlling for large numbers of variables that collectively serve as 'proxies' for unmeasured factors can often improve confounding control in pharmacoepidemiologic studies utilizing administrative healthcare databases. There is a growing body of evidence showing that data-driven machine learning algorithms for high-dimensional proxy confounder adjustment can supplement investigator-specified variables to improve confounding control compared to adjustment based on investigator-specified variables alone. Consequently, there has been a recent focus on the development of data-driven methods for high-dimensional proxy confounder adjustment. In this paper, we discuss the considerations underpinning three areas for data-driven high-dimensional proxy confounder adjustment: 1) feature generation-transforming raw data into covariates (or features) to be used for proxy adjustment; 2) covariate prioritization, selection and adjustment; and 3) diagnostic assessment. We survey current approaches and recent advancements within each area, including the most widely used approach to proxy confounder adjustment in healthcare database studies (the high-dimensional propensity score or hdPS). We also discuss limitations of the hdPS and outline recent advancements that incorporate the principles of proxy adjustment with machine learning extensions to improve performance. We further discuss challenges and avenues of future development within each area. This manuscript is endorsed by the International Society for Pharmacoepidemiology (ISPE). This article is protected by copyright. All rights reserved.

Randomized controlled trials are the gold standard for measuring causal effects. However, they are often not always feasible, and causal treatment effects must be estimated from observational data. Observational studies do not allow robust conclusions about causal relationships unless statistical techniques account for the imbalance of pretreatment confounders across groups and key assumptions hold. Propensity score and balance weighting (PSBW) are useful techniques that aim to reduce the observed imbalances between treatment groups by weighting the groups to look alike on the observed confounders. Notably, there are many methods available to estimate PSBW. However, it is unclear a priori which will achieve the best trade-off between covariate balance and effective sample size for a given application. Moreover, it is critical to assess the validity of key assumptions required for robust estimation of the needed treatment effects, including the overlap and no unmeasured confounding assumptions. We present a step-by-step guide to the use of PSBW for estimation of causal treatment effects that includes steps on how to evaluate overlap before the analysis, obtain estimates of PSBW using multiple methods and select the optimal one, check for covariate balance on multiple metrics, and assess sensitivity of findings (both the estimated treatment effect and statistical significance) to unobserved confounding. We illustrate the key steps using a case study examining the relative effectiveness of substance use treatment programs and provide a user-friendly Shiny application that can implement the proposed steps for any application with binary treatments.

Robust estimation of exposure response analysis relies on correct specification of the model structure with traditional parametric approach. However, the assumptions of the handcrafted model may not always hold or verifiable. Here, we conducted a simulation study to assess the performance of machine learning-based techniques in exposure-response (E-R) analysis where data were generated by a complicated nonlinear system under one dose level. Two analysis options involving machine learning were evaluated. The first option was based on marginal structural model with inverse probability weighting, where machine learning (ML) was employed to improve the performance of propensity score estimation. The simulation results showed that propensity score predicted by ML was more robust than traditional multinomial logistic regression in terms of adjusting the confounding effects and unbiasedly estimating the E-R relationship. The second option estimated the E-R relationship by employing artificial neural network as a universal function approximator to the data generating mechanism, without the requirement of accurately hand-crafting the whole simulation system. The results demonstrated that the trained network was able to correctly predict the treatment effects across a certain range of adjacent dose levels. In contrast, traditional regression provided biased predictions, even when all confounders were included in the model. Our study demonstrated that ML may serve as a powerful tool for pharmacometrics analysis with its prediction flexibility in a nonlinear system and its capacity of approximating the ground truth.

Purpose: This study aims to describe health inequities experienced by transgender Hispanic (TH) individuals in the United States. Methods: This retrospective case-control study used the Behavioral Risk Factor Surveillance System (BRFSS) data from 2014 to 2018. Propensity score matching and logistic and negative binomial regression were used to compare TH survey respondents with other relevant populations across the following outcomes: health care access, health risk factors, self-reported chronic conditions, and perceived health status. Results: Relative to transgender White (TW) respondents, TH respondents (n=414) were less likely to report having health insurance (odds ratio [OR]: 0.35, p<0.001), a regular provider (OR=0.40, p<0.001), and were more likely to report cost barriers to care (OR=1.85, p<0.001) and HIV risk factors (OR=2.41, p<0.001). Similar results were found when comparing outcomes with cisgender White respondents. TH respondents reported fewer days of poor health (rate ratio [RR]=0.67, p<0.001), activity limited days (RR=0.64, p=0.011), and were less likely to report depression (OR=0.44, p<0.001) than TW respondents. Relative to cisgender Hispanic (CH) respondents, TH respondents experienced more cost barriers (OR=1.56, p=0.003), higher HIV risk (OR=3.38, p<0.001), and more activity limited days (RR=2.93, p<0.001). Conclusion: Our results demonstrate that TH individuals may be less likely to have access to health care and have poorer health-related quality-of-life when compared with either CH or TW individuals. It is vital that additional research further elucidate the challenges faced by this multiply marginalized population including racism and transphobia. Further health care solutions should be responsive to the unique challenges of the TH population at the individual and institutional level.

The purpose of this study was to compare hospitalization outcomes among US inpatients with brain metastases who received stereotactic radiosurgery (SRS) and/or non-SRS radiation therapies without neurosurgical intervention. A cross-sectional study was conducted whereby existing data on 35,199 hospitalization records (non-SRS alone: 32,981; SRS alone: 1035; SRS + non-SRS: 1183) from 2005 to 2014 Nationwide Inpatient Sample were analyzed. Targeted maximum likelihood estimation and Super Learner algorithms were applied to estimate average treatment effects (ATE), marginal odds ratios (MOR) and causal risk ratio (CRR) for three distinct types of radiation therapy in relation to hospitalization outcomes, including length of stay (' ≥ 7 days' vs. ' < 7 days') and discharge destination ('non-routine' vs. 'routine'), controlling for patient and hospital characteristics. Recipients of SRS alone (ATE = - 0.071, CRR = 0.88, MOR = 0.75) or SRS + non-SRS (ATE = - 0.17, CRR = 0.70, MOR = 0.50) had shorter hospitalizations as compared to recipients of non-SRS alone. Recipients of SRS alone (ATE = - 0.13, CRR = 0.78, MOR = 0.59) or SRS + non-SRS (ATE = - 0.17, CRR = 0.72, MOR = 0.51) had reduced risks of non-routine discharge as compared to recipients of non-SRS alone. Similar analyses suggested recipients of SRS alone had shorter hospitalizations and similar risk of non-routine discharge when compared to recipients of SRS + non-SRS radiation therapies. SRS alone or in combination with non-SRS therapies may reduce the risks of prolonged hospitalization and non-routine discharge among hospitalized US patients with brain metastases who underwent radiation therapy without neurosurgical intervention.

Methods for estimating heterogeneous treatment effect in observational data have largely focused on continuous or binary outcomes, and have been relatively less vetted with survival outcomes. Using flexible machine learning methods in the counterfactual framework is a promising approach to address challenges due to complex individual characteristics, to which treatments need to be tailored. To evaluate the operating characteristics of recent survival machine learning methods for the estimation of treatment effect heterogeneity and inform better practice, we carry out a comprehensive simulation study presenting a wide range of settings describing confounded heterogeneous survival treatment effects and varying degrees of covariate overlap. Our results suggest that the nonparametric Bayesian Additive Regression Trees within the framework of accelerated failure time model (AFT-BART-NP) consistently yields the best performance, in terms of bias, precision, and expected regret. Moreover, the credible interval estimators from AFT-BART-NP provide close to nominal frequentist coverage for the individual survival treatment effect when the covariate overlap is at least moderate. Including a nonparametrically estimated propensity score as an additional fixed covariate in the AFT-BART-NP model formulation can further improve its efficiency and frequentist coverage. Finally, we demonstrate the application of flexible causal machine learning estimators through a comprehensive case study examining the heterogeneous survival effects of two radiotherapy approaches for localized high-risk prostate cancer.