Observational studies of treatment effectiveness in neurology.

The capacity and power of data from cohorts, registries and randomized trials to provide answers to contemporary clinical questions in neurology has increased considerably over the past two decades. Novel sophisticated statistical methods are enabling us to harness these data to guide treatment decisions, but their complexity is making appraisal of clinical evidence increasingly demanding. In this review, we discuss several methodological aspects of contemporary research of treatment effectiveness in observational data in neurology, aimed at academic neurologists and analysts specializing in outcomes research. The review discusses specifics of the sources of observational data and their key features. It focuses on the limitations of observational data and study design, as well as statistical approaches aimed to overcome these limitations. Among the examples of leading clinical themes typically studied with analyses of observational data, the review discusses methodological approaches to comparative treatment effectiveness, development of diagnostic criteria and definitions of clinical outcomes. Finally, this review provides a brief summary of key points that will help clinical audience critically evaluate design and analytical aspects of studies of disease outcomes using observational data.

The application of target trials with longitudinal targeted maximum likelihood estimation to assess the effect of alcohol consumption in adolescence on depressive symptoms in adulthood.

Time-varying confounding is a common challenge for causal inference in observational studies with time-varying treatments, long follow-up periods, and participant dropout. Confounder adjustment using traditional approaches can be limited by data sparsity, weight instability and computational issues. The Nicotine Dependence in Teens (NDIT) study is a prospective cohort study involving 24 data collection cycles from 1999 to date, among 1,294 students recruited from 10 high schools in Montreal, Canada, including follow-up into adulthood. Our aim is to estimate associations between the timing of alcohol initiation and the cumulative duration of alcohol use on depression symptoms in adulthood. Based on the target trials framework, we define intention-to-treat and as-treated parameters in a marginal structural model with sex as a potential effect-modifier. We then use the observational data to emulate the trials. For estimation, we use pooled longitudinal target maximum likelihood estimation (LTMLE), a plug-in estimator with double robust and local efficiency properties. We describe strategies for dealing with high-dimensional potential drinking patterns and practical positivity violations due to a long follow-up time, including modifying the effect of interest by removing sparsely observed drinking patterns from the loss function and applying longitudinal modified treatment policies to represent the effect of discouraging drinking.

A marginal structural model to estimate the effect of antidepressant medication treatment on major cardiovascular events among people with post-traumatic stress disorder.

BACKGROUND: Previous evidence on antidepressant medication and cardiovascular disease (CVD) among patients with posttraumatic stress disorder (PTSD) has been inconclusive. We estimated the association between antidepressant medication and CVD by applying a marginal structural model. METHODS: We analyzed medical utilization records of 27 170 people with PTSD without prior major cardiovascular events in the Korean National Health Insurance Database (NHID). PTSD and CVD were defined in accordance with the recorded ICD-10 diagnostic codes. We acquired information on antidepressant use from the NHID and categorized them by medication type. A composite major adverse cardiovascular events (MACE) outcome was defined as coronary artery disease with revascularization, ischaemic stroke, and/or haemorrhagic stroke. We used inverse probability of treatment weighting to estimate the parameters of a marginal structural discrete-time survival analysis regression model, comparing the resulting estimates to those derived from traditional time-fixed and time-varying Cox proportional hazards regression. We calculated cumulative daily defined doses to test for a dose-response relationship. RESULTS: People exposed to antidepressants showed a higher hazard of MACE [hazard ratio (HR) 1.34, 95% confidence interval (CI) 1.18-1.53]. The estimated effects were strongest for selective serotonin reuptake inhibitors (HR 1.24, 95% CI 1.08-1.44) and TCAs (HR 1.33, 95% CI 1.13-1.56). Exposure to serotonin-norepinephrine reuptake inhibitors did not appear to increase the risk of MACE. People exposed to higher doses of antidepressants showed higher risk of MACE. CONCLUSIONS: In a national cohort of people with PTSD, exposure to antidepressant medications increased the risk of MACE in a dose-response fashion.

Instrumental variable estimation of the causal hazard ratio.

Cox's proportional hazards model is one of the most popular statistical models to evaluate associations of exposure with a censored failure time outcome. When confounding factors are not fully observed, the exposure hazard ratio estimated using a Cox model is subject to unmeasured confounding bias. To address this, we propose a novel approach for the identification and estimation of the causal hazard ratio in the presence of unmeasured confounding factors. Our approach is based on a binary instrumental variable, and an additional no-interaction assumption in a first-stage regression of the treatment on the IV and unmeasured confounders. We propose, to the best of our knowledge, the first consistent estimator of the (population) causal hazard ratio within an instrumental variable framework. A version of our estimator admits a closed-form representation. We derive the asymptotic distribution of our estimator and provide a consistent estimator for its asymptotic variance. Our approach is illustrated via simulation studies and a data application.

Consistent inverse probability of treatment weighted estimation of the average treatment effect with mismeasured time-dependent confounders.

In longitudinal studies, the inverse probability of treatment weighted (IPTW) method is commonly employed to estimate the effect of time-dependent treatments on an outcome of interest. However, it has been documented that when the confounders are subject to measurement error, the naive IPTW method which simply ignores measurement error leads to biased treatment effect estimation. In the existing literature, there is a lack of measurement error correction methods that fully remove measurement error effect and produce consistent treatment effect estimation. In this article, we develop a novel consistent IPTW estimation procedure for longitudinal studies. The key step of the proposed method is to use the observed data to construct a corrected function that is unbiased of the unknown IPTW function. Simulation studies reveal that the proposed method outperforms the existing consistent and approximate measurement error correction methods for IPTW estimation of the average treatment effect. Finally, we apply the proposed method to analyze a real dataset.

Causal inference in survival analysis using longitudinal observational data: Sequential trials and marginal structural models.

Longitudinal observational data on patients can be used to investigate causal effects of time-varying treatments on time-to-event outcomes. Several methods have been developed for estimating such effects by controlling for the time-dependent confounding that typically occurs. The most commonly used is marginal structural models (MSM) estimated using inverse probability of treatment weights (IPTW) (MSM-IPTW). An alternative, the sequential trials approach, is increasingly popular, and involves creating a sequence of "trials" from new time origins and comparing treatment initiators and non-initiators. Individuals are censored when they deviate from their treatment assignment at the start of each "trial" (initiator or noninitiator), which is accounted for using inverse probability of censoring weights. The analysis uses data combined across trials. We show that the sequential trials approach can estimate the parameters of a particular MSM. The causal estimand that we focus on is the marginal risk difference between the sustained treatment strategies of "always treat" vs "never treat." We compare how the sequential trials approach and MSM-IPTW estimate this estimand, and discuss their assumptions and how data are used differently. The performance of the two approaches is compared in a simulation study. The sequential trials approach, which tends to involve less extreme weights than MSM-IPTW, results in greater efficiency for estimating the marginal risk difference at most follow-up times, but this can, in certain scenarios, be reversed at later time points and relies on modelling assumptions. We apply the methods to longitudinal observational data from the UK Cystic Fibrosis Registry to estimate the effect of dornase alfa on survival.

Variability of the response to immunotherapy among subgroups of patients with multiple sclerosis.

BACKGROUND AND PURPOSE: This study assessed the effect of patient characteristics on the response to disease-modifying therapy (DMT) in multiple sclerosis (MS). METHODS: We extracted data from 61,810 patients from 135 centers across 35 countries from the MSBase registry. The selection criteria were: clinically isolated syndrome or definite MS, follow-up ≥ 1 year, and Expanded Disability Status Scale (EDSS) score ≥ 3, with ≥1 score recorded per year. Marginal structural models with interaction terms were used to compare the hazards of 12-month confirmed worsening and improvement of disability, and the incidence of relapses between treated and untreated patients stratified by their characteristics. RESULTS: Among 24,344 patients with relapsing MS, those on DMTs experienced 48% reduction in relapse incidence (hazard ratio [HR] = 0.52, 95% confidence interval [CI] = 0.45-0.60), 46% lower risk of disability worsening (HR = 0.54, 95% CI = 0.41-0.71), and 32% greater chance of disability improvement (HR = 1.32, 95% CI = 1.09-1.59). The effect of DMTs on EDSS worsening and improvement and the risk of relapses was attenuated with more severe disability. The magnitude of the effect of DMT on suppressing relapses declined with higher prior relapse rate and prior cerebral magnetic resonance imaging activity. We did not find any evidence for the effect of age on the effectiveness of DMT. After inclusion of 1985 participants with progressive MS, the effect of DMT on disability mostly depended on MS phenotype, whereas its effect on relapses was driven mainly by prior relapse activity. CONCLUSIONS: DMT is generally most effective among patients with lower disability and in relapsing MS phenotypes. There is no evidence of attenuation of the effect of DMT with age.

Distinct hyperuricemia trajectories are associated with different risks of incident diabetes: A prospective cohort study.

BACKGROUND AND AIM: Conflicting results suggest a link between serum uric acid and diabetes and previous studies ignored the effect of continuous exposure of serum uric acid on diabetes risk. This study aims to characterize hyperuricemia trajectories in middle-aged adults and to examine its potential impact on diabetes risk, considering the role of obesity, dyslipidemia, and hypertension. METHODS AND RESULTS: The cohort included 9192 participants who were free of diabetes before 2013. The hyperuricemia trajectories during 2009-2013 were identified by latent class growth models. Incident diabetes during 2014-2018 was used as the outcome. Modified Poisson regression models were used to assess the association of trajectories with diabetes. Furthermore, marginal structural models were used to estimate the mediating effects of the relationship between hyperuricemia trajectories and diabetes. We identified three discrete hyperuricemia trajectories: high-increasing (n = 5794), moderate-stable (n = 2049), and low-stable (n = 1349). During 5 years of follow-up, we documented 379 incident diabetes cases. Compared with the low-stable pattern, the high-increasing pattern had a higher risk of developing diabetes (RR, 1.42; 95% CI: 1.09-1.84). In addition, the percentages of total effect between the high-increasing hyperuricemia pattern and diabetes mediated by obesity, dyslipidemia, and hypertension were 24.41%, 18.26%, and 6.29%. However, the moderate-stable pattern was not associated with an increased risk of diabetes. CONCLUSIONS: These results indicate that the high-increasing hyperuricemia trajectory is significantly associated with an increased risk of diabetes. Furthermore, obesity, dyslipidemia, and hypertension play mediating roles in the relationship between the high-increasing hyperuricemia pattern and increased diabetes risk.

Directed Acyclic Graph Assisted Method For Estimating Average Treatment Effect.

Observational data, such as electronic clinical records and claims data, can prove invaluable for evaluating the Average Treatment Effect (ATE) and supporting decision-making, provided they are employed correctly. The Inverse Probability of Treatment Weighting (IPTW) method, based on propensity scores, has demonstrated remarkable efficacy in estimating ATE, assuming that the assumptions of exchangeability, consistency, and positivity are met. Directed Acyclic Graphs (DAGs) offer a practical approach to assess the exchangeability assumption, which asserts that treatment assignment and potential outcomes are independent given a set of confounding variables that block all backdoor paths from treatment assignment to potential outcomes. To ensure a consistent ATE estimator, one can adjust for a minimally sufficient adjustment set of confounding variables that block all backdoor paths from treatment assignment to the outcome. To enhance the efficiency of ATE estimators, our proposal involves incorporating both the minimally sufficient adjustment set of confounding variables and predictors into the propensity score model. Extensive simulations were conducted to evaluate the performance of propensity score-based IPTW methods in estimating ATE when different sets of covariates were included in the propensity score models. The simulation results underscored the significance of including the minimally sufficient adjustment set of confounding variables along with predictors in the propensity score models to obtain a consistent and efficient ATE estimator. We applied this proposed method to investigate whether tracheostomy was causally associated with in-hospital infant mortality, utilizing the 2016 Healthcare Cost and Utilization Project Kids' Inpatient Database. The estimated ATE was found to be approximately 2.30%-2.46% with p-value >0.05.

Association of Statins and 28-Day Mortality Rates in Patients Hospitalized With Severe Acute Respiratory Syndrome Coronavirus 2 Infection.

BACKGROUND: Statins may be protective in severe acute respiratory syndrome coronavirus 2 SARS-CoV-2 infection. The aim of the current study was to evaluate the effect of in-hospital statin use on 28-day mortality rates and intensive care unit (ICU) admission among patients with SARS-CoV-2, stratified into 4 groups: those who used statins before hospitalization (treatment continued or discontinued in the hospital) and those who did not (treatment newly initiated in the hospital or never initiated). METHODS: In a cohort study of 1179 patients with SARS-CoV-2, record review was used to assess demographics, laboratory measurements, comorbid conditions, and time from admission to death, ICU admission, or discharge. Using marginal structural Cox models, we estimated hazard ratios (HRs) for death and ICU admission. RESULTS: Among 1179 patients, 676 (57%) were male, 443 (37%) were >65 years old, and 493 (46%) had a body mass index ≥30 (calculated as weight in kilograms divided by height in meters squared). Inpatient statin use reduced the hazard of death (HR, 0.566; P=.008). This association held among patients who did and those who did not use statins before hospitalization (HR, 0.270 [P=.003] and 0.493 [P=.04], respectively). Statin use was associated with improved time to death for patients aged >65 years but not for those ≤65 years old. CONCLUSION: Statin use during hospitalization for SARS-CoV-2 infection was associated with reduced 28-day mortality rates. Well-designed randomized control trials are needed to better define this relationship.

Sunscreens With High Versus Low Sun Protection Factor and Cutaneous Squamous Cell Carcinoma Risk: A Population-Based Cohort Study.

Evidence on sunscreen use and cutaneous squamous cell carcinoma (cSCC) risk is limited. Most studies have not taken sun protection factor (SPF) into consideration and used nonusers of sunscreen as the reference group. Nonusers are likely a priori at lower cSCC risk than users. No study has investigated the effect of high- versus low-SPF sunscreens on cSCC, appropriately adjusting for time-varying confounding. Using data from the Norwegian Women and Cancer Study (1991-2016), we investigated whether use of SPF ≥15 versus SPF <15 sunscreens reduces cSCC risk. We used a marginal structural Cox proportional hazards model with inverse probability of treatment and censoring weights to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). During follow-up of 148,781 women (mean follow-up, 14.3 years), 653 women were diagnosed with cSCC. The effect on cSCC risk of sunscreens with SPF ≥15 versus SPF <15 was close to the null when used at any latitudes (HR = 1.02, 95% CI: 0.82, 1.27) and when used in lower-latitude settings (HR = 1.05, 95% CI: 0.84, 1.32). In conclusion, we found no indication that sunscreens with SPF ≥15 reduced Norwegian women's cSCC risk more than sunscreens with SPF <15, suggesting that either there is no difference in their effects long-term or the difference is diluted by incorrect application.

Association of Time-Updated Anion Gap With Risk of Kidney Failure in Advanced CKD: A Cohort Study.

RATIONALE & OBJECTIVE: High anion gap acidosis frequently develops in patients with advanced chronic kidney disease (CKD) and might be involved in kidney injury. Its impact on kidney outcomes, however, has not been well studied. We sought to examine the association between time-updated anion gap and the risk of kidney failure with replacement therapy (KFRT) among patients with advanced CKD. STUDY DESIGN: Retrospective cohort study. SETTING & PARTICIPANTS: 1,168 patients with CKD glomerular filtration rate categories 3b-5 (G3b-G5) who had available data on anion gap. EXPOSURE: High time-updated anion gap defined as values ≥ 9.2 (top 25th percentile). OUTCOME: KFRT and death. ANALYTICAL APPROACH: Marginal structural models were fit to characterize the association between anion gap and study outcomes while accounting for potential time-dependent confounding. RESULTS: The mean baseline estimated glomerular filtration rate (eGFR) of the study participants was 28 mL/min/1.73 m2. Over a median follow-up period of 3.1 years, 317 patients progressed to KFRT (7.5 per 100 patient-years), and 146 died (3.5 per 100 patient-years). In the marginal structural models, a high anion gap was associated with a higher rate of KFRT (HR, 3.04 [95% CI, 1.94-4.75]; P < 0.001). This association was stronger in patients with a baseline eGFR of <30 mL/min/1.73 m2 (P for interaction = 0.05). High anion gap was also associated with a higher mortality rate (HR, 5.56 [95% CI, 2.95-10.5]; P < 0.001). Sensitivity analyses with different definitions of high anion gap showed similar results. LIMITATIONS: Observational study design and selection bias due clinical indications for measuring anion gap. CONCLUSIONS: Among patients with advanced CKD, high anion gap was associated with an increased risk of progression to KFRT and death.

Comparative Effectiveness of Dialysis Modality on Laboratory Parameters of Mineral Metabolism.

INTRODUCTION: Chronic kidney disease-mineral and bone disorders (CKD-MBD) are prevalent in patients undergoing maintenance dialysis. Yet, there are limited and mixed evidence on the effects of different dialysis modalities involving longer treatment times or higher frequencies on CKD-MBD markers. METHODS: This cohort study used data from 132,523 incident dialysis patients treated with any of the following modalities: conventional thrice-weekly in-center hemodialysis, nocturnal in-center hemodialysis (NICHD), home hemodialysis (HHD), or peritoneal dialysis (PD) from 2007 to 2011. We used marginal structural models fitted with inverse probability weights to adjust for fixed and time-varying confounding and informative censoring. We estimated the average effects of treatments with different dialysis modalities on time-varying serum concentrations of CKD-MBD markers: albumin-corrected calcium, phosphate, parathyroid hormone (PTH), and alkaline phosphatase (ALP) using pooled linear regression. RESULTS: Most of the cohort were exclusively treated with conventional in-center hemodialysis, while few were ever treated with NICHD or HHD. At the baseline, PD patients had the lowest mean and median values of PTH, while NICHD patients had the highest median values. During follow-up, compared to hemodialysis patients, patients treated with NICHD had lower mean serum PTH (19.8 pg/mL [95% confidence interval: 2.8, 36.8] lower), whereas PD and HHD patients had higher mean PTH (39.7 pg/mL [31.6, 47.8] and 51.2 pg/mL [33.0, 69.3] higher, respectively). Compared to hemodialysis patients, phosphate levels were lower for patients treated with NICHD (0.44 mg/dL [0.37, 0.52] lower), PD (0.15 mg/dL [0.12, 0.19] lower), or HHD (0.33 mg/dL [0.27, 0.40] lower). There were no clinically meaningful associations between dialysis modalities and concentrations of calcium or ALP. CONCLUSION: In incident dialysis patients, compared to treatment with conventional in-center hemodialysis, treatments with other dialysis modalities with longer treatment times or higher frequency were associated with different patterns of serum phosphate and PTH. Given the recent growth in the use of dialysis modalities other than hemodialysis, the associations between the treatment and the CKD-MBD markers warrant additional study.

Estimating the causal effect of transient anemia status on renal and cardiovascular outcomes in community-dwelling patients in Japan at the beginning of impaired renal function using marginal structural modeling.

BACKGROUND: Anemia status may be transient. Causal associations between changes in anemia status over time and adverse outcome development are not well characterized in community-dwelling subjects at the beginning of impaired kidney function. METHODS: This retrospective cohort study used annual health checkup and medical and pharmacy claims data from the JMDC between January 2005 and June 2019. Community-dwelling subjects in Japan with a pre-index estimated glomerular filtration rate (eGFR) ≥ 60 mL/min/1.73 m2 followed by a subsequent eGFR < 60 mL/min/1.73 m2 (index) were included. The composite renal outcome was ≥ 30% eGFR reduction over 3 years from baseline, serum creatinine doubling, progression to chronic dialysis, kidney transplantation, or eGFR < 15 mL/min/1.73 m2. The composite cardiovascular outcome was fatal and non-fatal unstable angina, myocardial infarction, heart failure, or cerebrovascular event. Time-dependent anemia risk was evaluated using Breslow's estimator and marginal structural Cox models (MSM). RESULTS: In 32,870 included subjects, 1,396 had anemia at baseline. Adverse outcome incidence was higher in the baseline anemic group, but absolute differences in renal and cardiovascular outcomes between groups were diminished after adjusting for baseline characteristics. In MSM, time-dependent anemia status was associated with higher risk of renal (hazard ratio [95% confidence interval]; 2.6 [1.7-3.8]) and cardiovascular (1.6 [1.2-2.2]) outcomes and mortality (2.8 [1.8-4.3]). Absolute differences in survival probabilities were retained over time but were clinically marginal (1.1-2.7% over 6 years). CONCLUSIONS: Even in subjects at the very early stage of impaired kidney function, early detection and treatment of anemia may help reduce the development of negative sequelae.

Childhood stunting and subsequent educational outcomes: a marginal structural model analysis from a South African longitudinal study.

OBJECTIVE: To examine the association between childhood stunting and grade completion (as educational outcome) in South Africa. DESIGN: Longitudinal study. Data were obtained using the National Income Dynamics Study over five waves (2008 to 2017). Children were tracked at wave 1 in 2008 until wave 5 in 2017 to determine their total years of schooling. We controlled for time-variant and time-varying confounding with a marginal structural model to estimate the associations between childhood stunting and subsequent grade completion. SETTING: Nationally representative study of South African households. PARTICIPANTS: A total of 2629 children aged 2 and 3 years in 2008. RESULTS: We observed a substantial decrease in the prevalence of stunting between wave 1 (28·2 %) and wave 4 (8·6 %). Our marginal structural model results suggest that childhood stunting was significantly associated with decreased odds (22 % less likely) of grade completion (OR = 0·78; 95 % CI: 0·40, 0·86; P = 0·015), while those who were only stunted during early childhood had a 29 % reduction in the odds of grade completion (OR = 0·71; 95 % CI: 0·51, 0·82; P = 0·020). CONCLUSION: These findings underscore the fact that stunting is a significant predictor of academic achievement, whose effects might be long-lasting.

Dealing With Treatment-Confounder Feedback and Sparse Follow-up in Longitudinal Studies: Application of a Marginal Structural Model in a Multiple Sclerosis Cohort.

The beta-interferons are widely prescribed platform therapies for patients with multiple sclerosis (MS). We accessed a cohort of patients with relapsing-onset MS from British Columbia, Canada (1995-2013), to examine the potential survival advantage associated with beta-interferon exposure using a marginal structural model. Accounting for potential treatment-confounder feedback between comorbidity, MS disease progression, and beta-interferon exposure, we found an association between beta-interferon exposure of at least 6 contiguous months and improved survival (hazard ratio (HR) = 0.63, 95% confidence interval 0.47, 0.86). We also assessed potential effect modifications by sex, baseline age, or baseline disease duration, and found these factors to be important effect modifiers. Sparse follow-up due to variability in patient contact with the health system is one of the biggest challenges in longitudinal analyses. We considered several single-level and multilevel multiple imputation approaches to deal with sparse follow-up and disease progression information; both types of approach produced similar estimates. Compared to ad hoc imputation approaches, such as linear interpolation (HR = 0.63), and last observation carried forward (HR = 0.65), all multiple imputation approaches produced a smaller hazard ratio (HR = 0.53), although the direction of effect and conclusions drawn concerning the survival advantage remained the same.

Using generalized linear models to implement g-estimation for survival data with time-varying confounding.

Using data from observational studies to estimate the causal effect of a time-varying exposure, repeatedly measured over time, on an outcome of interest requires careful adjustment for confounding. Standard regression adjustment for observed time-varying confounders is unsuitable, as it can eliminate part of the causal effect and induce bias. Inverse probability weighting, g-computation, and g-estimation have been proposed as being more suitable methods. G-estimation has some advantages over the other two methods, but until recently there has been a lack of flexible g-estimation methods for a survival time outcome. The recently proposed Structural Nested Cumulative Survival Time Model (SNCSTM) is such a method. Efficient estimation of the parameters of this model required bespoke software. In this article we show how the SNCSTM can be fitted efficiently via g-estimation using standard software for fitting generalised linear models. The ability to implement g-estimation for a survival outcome using standard statistical software greatly increases the potential uptake of this method. We illustrate the use of this method of fitting the SNCSTM by reanalyzing data from the UK Cystic Fibrosis Registry, and provide example R code to facilitate the use of this approach by other researchers.

Simulating longitudinal data from marginal structural models using the additive hazard model.

Observational longitudinal data on treatments and covariates are increasingly used to investigate treatment effects, but are often subject to time-dependent confounding. Marginal structural models (MSMs), estimated using inverse probability of treatment weighting or the g-formula, are popular for handling this problem. With increasing development of advanced causal inference methods, it is important to be able to assess their performance in different scenarios to guide their application. Simulation studies are a key tool for this, but their use to evaluate causal inference methods has been limited. This paper focuses on the use of simulations for evaluations involving MSMs in studies with a time-to-event outcome. In a simulation, it is important to be able to generate the data in such a way that the correct forms of any models to be fitted to those data are known. However, this is not straightforward in the longitudinal setting because it is natural for data to be generated in a sequential conditional manner, whereas MSMs involve fitting marginal rather than conditional hazard models. We provide general results that enable the form of the correctly specified MSM to be derived based on a conditional data generating procedure, and show how the results can be applied when the conditional hazard model is an Aalen additive hazard or Cox model. Using conditional additive hazard models is advantageous because they imply additive MSMs that can be fitted using standard software. We describe and illustrate a simulation algorithm. Our results will help researchers to effectively evaluate causal inference methods via simulation.

Data-adaptive longitudinal model selection in causal inference with collaborative targeted minimum loss-based estimation.

Causal inference methods have been developed for longitudinal observational study designs where confounding is thought to occur over time. In particular, one may estimate and contrast the population mean counterfactual outcome under specific exposure patterns. In such contexts, confounders of the longitudinal treatment-outcome association are generally identified using domain-specific knowledge. However, this may leave an analyst with a large set of potential confounders that may hinder estimation. Previous approaches to data-adaptive model selection for this type of causal parameter were limited to the single time-point setting. We develop a longitudinal extension of a collaborative targeted minimum loss-based estimation (C-TMLE) algorithm that can be applied to perform variable selection in the models for the probability of treatment with the goal of improving the estimation of the population mean counterfactual outcome under a fixed exposure pattern. We investigate the properties of this method through a simulation study, comparing it to G-Computation and inverse probability of treatment weighting. We then apply the method in a real-data example to evaluate the safety of trimester-specific exposure to inhaled corticosteroids during pregnancy in women with mild asthma. The data for this study were obtained from the linkage of electronic health databases in the province of Quebec, Canada. The C-TMLE covariate selection approach allowed for a reduction of the set of potential confounders, which included baseline and longitudinal variables.

Adjusting for time-varying confounders in survival analysis using structural nested cumulative survival time models.

Accounting for time-varying confounding when assessing the causal effects of time-varying exposures on survival time is challenging. Standard survival methods that incorporate time-varying confounders as covariates generally yield biased effect estimates. Estimators using weighting by inverse probability of exposure can be unstable when confounders are highly predictive of exposure or the exposure is continuous. Structural nested accelerated failure time models (AFTMs) require artificial recensoring, which can cause estimation difficulties. Here, we introduce the structural nested cumulative survival time model (SNCSTM). This model assumes that intervening to set exposure at time t to zero has an additive effect on the subsequent conditional hazard given exposure and confounder histories when all subsequent exposures have already been set to zero. We show how to fit it using standard software for generalized linear models and describe two more efficient, double robust, closed-form estimators. All three estimators avoid the artificial recensoring of AFTMs and the instability of estimators that use weighting by the inverse probability of exposure. We examine the performance of our estimators using a simulation study and illustrate their use on data from the UK Cystic Fibrosis Registry. The SNCSTM is compared with a recently proposed structural nested cumulative failure time model, and several advantages of the former are identified.