Structured learning in time-dependent Cox models.

Cox models with time-dependent coefficients and covariates are widely used in survival analysis. In high-dimensional settings, sparse regularization techniques are employed for variable selection, but existing methods for time-dependent Cox models lack flexibility in enforcing specific sparsity patterns (ie, covariate structures). We propose a flexible framework for variable selection in time-dependent Cox models, accommodating complex selection rules. Our method can adapt to arbitrary grouping structures, including interaction selection, temporal, spatial, tree, and directed acyclic graph structures. It achieves accurate estimation with low false alarm rates. We develop the sox package, implementing a network flow algorithm for efficiently solving models with complex covariate structures. sox offers a user-friendly interface for specifying grouping structures and delivers fast computation. Through examples, including a case study on identifying predictors of time to all-cause death in atrial fibrillation patients, we demonstrate the practical application of our method with specific selection rules.

Effect of statin use for the primary prevention of cardiovascular disease among older adults: a cautionary tale concerning target trials emulation.

OBJECTIVES: Evidence concerning the effect of statins in primary prevention of cardiovascular disease (CVD) among older adults is lacking. Using Quebec population-wide administrative data, we emulated a hypothetical randomized trial including older adults >65 years on April 1, 2013, with no CVD history and no statin use in the previous year. STUDY DESIGN AND SETTING: We included individuals who initiated statins and classified them as exposed if they were using statin at least 3 months after initiation and nonexposed otherwise. We followed them until March 31, 2018. The primary outcome was the composite endpoint of coronary events (myocardial infarction, coronary bypass, and percutaneous coronary intervention), stroke, and all-cause mortality. The intention-to-treat (ITT) effect was estimated with adjusted Cox models and per-protocol effect with inverse probability of censoring weighting. RESULTS: A total of 65,096 individuals were included (mean age = 71.0 ± 5.5, female = 55.0%) and 93.7% were exposed. Whereas we observed a reduction in the composite outcome (ITT-hazard ratio (HR) = 0.75; 95% CI: 0.68-0.83) and mortality (ITT-HR = 0.69; 95% CI: 0.61-0.77) among exposed, coronary events increased (ITT-HR = 1.46; 95% CI: 1.09-1.94). All multibias E-values were low indicating that the results were not robust to unmeasured confounding, selection, and misclassification biases simultaneously. CONCLUSION: We cannot conclude on the effectiveness of statins in primary prevention of CVD among older adults. We caution that an in-depth reflection on sources of biases and careful interpretation of results are always required in observational studies.

Test negative design for vaccine effectiveness estimation in the context of the COVID-19 pandemic: A systematic methodology review.

BACKGROUND: During the height of the global COVID-19 pandemic, the test-negative design (TND) was extensively used in many countries to evaluate COVID-19 vaccine effectiveness (VE). Typically, the TND involves the recruitment of care-seeking individuals who meet a common clinical case definition. All participants are then tested for an infection of interest. OBJECTIVES: To review and describe the variation in TND methodology, and disclosure of potential biases, as applied to the evaluation of COVID-19 VE during the early vaccination phase of the pandemic. METHODS: We conducted a systematic review by searching four biomedical databases using defined keywords to identify peer-reviewed articles published between January 1, 2020, and January 25, 2022. We included only original articles that employed a TND to estimate VE of COVID-19 vaccines in which cases and controls were evaluated based on SARS-CoV-2 laboratory test results. RESULTS: We identified 96 studies, 35 of which met the defined criteria. Most studies were from North America (16 studies) and targeted the general population (28 studies). Outcome case definitions were based primarily on COVID-19-like symptoms; however, several papers did not consider or specify symptoms. Cases and controls had the same inclusion criteria in only half of the studies. Most studies relied upon administrative or hospital databases assembled for a different (non-evaluation) clinical purpose. Potential unmeasured confounding (20 studies), misclassification of current SARS-CoV-2 infection (16 studies) and selection bias (10 studies) were disclosed as limitations by some studies. CONCLUSION: We observed potentially meaningful deviations from the validated design in the application of the TND during the COVID-19 pandemic.

Two-stage targeted maximum likelihood estimation for mixed aggregate and individual participant data analysis with an application to multidrug resistant tuberculosis.

In this study, we develop a new method for the meta-analysis of mixed aggregate data (AD) and individual participant data (IPD). The method is an adaptation of inverse probability weighted targeted maximum likelihood estimation (IPW-TMLE), which was initially proposed for two-stage sampled data. Our methods are motivated by a systematic review investigating treatment effectiveness for multidrug resistant tuberculosis (MDR-TB) where the available data include IPD from some studies but only AD from others. One complication in this application is that participants with MDR-TB are typically treated with multiple antimicrobial agents where many such medications were not observed in all studies considered in the meta-analysis. We focus here on the estimation of the expected potential outcome while intervening on a specific medication but not intervening on any others. Our method involves the implementation of a TMLE that transports the estimation from studies where the treatment is observed to the full target population. A second weighting component adjusts for the studies with missing (inaccessible) IPD. We demonstrate the properties of the proposed method and contrast it with alternative approaches in a simulation study. We finally apply this method to estimate treatment effectiveness in the MDR-TB case study.

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.

Identifying patients with multidrug-resistant tuberculosis who may benefit from shorter durations of treatment.

OBJECTIVE: Studying treatment duration for rifampicin-resistant and multidrug-resistant tuberculosis (MDR/RR-TB) using observational data is methodologically challenging. We aim to present a hypothesis generating approach to identify factors associated with shorter duration of treatment. STUDY DESIGN AND SETTING: We conducted an individual patient data meta-analysis among MDR/RR-TB patients restricted to only those with successful treatment outcomes. Using multivariable linear regression, we estimated associations and their 95% confidence intervals (CI) between the outcome of individual deviation in treatment duration (in months) from the mean duration of their treatment site and patient characteristics, drug resistance, and treatments used. RESULTS: Overall, 6702 patients with successful treatment outcomes from 84 treatment sites were included. We found that factors commonly associated with poor treatment outcomes were also associated with longer treatment durations, relative to the site mean duration. Use of bedaquiline was associated with a 0.51 (95% CI: 0.15, 0.87) month decrease in duration of treatment, which was consistent across subgroups, while MDR/RR-TB with fluoroquinolone resistance was associated with 0.78 (95% CI: 0.36, 1.21) months increase. CONCLUSION: We describe a method to assess associations between clinical factors and treatment duration in observational studies of MDR/RR-TB patients, that may help identify patients who can benefit from shorter treatment.

Marginal structural models with latent class growth analysis of treatment trajectories: Statins for primary prevention among older adults.

Latent class growth analysis is increasingly proposed as a solution to summarize the observed longitudinal treatment into a few distinct groups. When latent class growth analysis is combined with standard approaches like Cox proportional hazards models, confounding bias is not properly addressed because of time-varying covariates that have a double role of confounders and mediators. We propose to use latent class growth analysis to classify individuals into a few latent classes based on their medication adherence pattern, then choose a working marginal structural model that relates the outcome to these groups. The parameter of interest is defined as a projection of the true marginal structural model onto the chosen working model. Simulation studies are used to illustrate our approach and compare it with unadjusted, baseline covariates adjusted, time-varying covariates adjusted, and inverse probability of trajectory groups weighted adjusted models. Our proposed approach yielded estimators with little or no bias and appropriate coverage of confidence intervals in these simulations. We applied our latent class growth analysis and marginal structural model approach to a database comprising information on 52,790 individuals from the province of Quebec, Canada, aged more than 65 and who were statin initiators to estimate the effect of statin-usage trajectories on a first cardiovascular event.

Polypharmacy and risk of mortality among patients with heart failure following hospitalization: a nested case-control study.

Heart failure (HF) is associated with morbidity, rehospitalization and polypharmacy. The incidence rate of mortality in HF patients with polypharmacy is poorly studied. We examine the association of polypharmacy with mortality risk in incident hospitalized HF patients with a primary diagnosis after discharge from the hospital using Quebec administrative databases, Canada from 1999 to 2015. Polypharmacy, cardiovascular (CV) polypharmacy and non-CV polypharmacy were respectively defined as exposure to ≥ 10 drugs, ≥ 5 CV drugs and ≥ 5 non-CV drugs within three months prior to the case or the control selection date. We conducted a nested case-control study to estimate rate ratios (RR) of all-cause mortality using a multivariate conditional logistic regression during one-year of follow-up. We identified 12,242 HF patients with a mean age of 81.6 years. Neither CV polypharmacy (RR 0.97, 95%CI 0.82-1.15) nor non-CV polypharmacy (RR 0.93, 95%CI 0.77-1.12) were associated with lower mortality risk. However, all polypharmacy (RR 1.31, 95%CI 1.07-1.61) showed an association with mortality risk. Myocardial infarction, valvular disease, peripheral artery disease, diabetes, major bleeding, chronic kidney disease, high comorbidity score, high Frailty score, hydralazine and spironolactone users were associated with increasing mortality risk, ranging from 15 to 61%, while use of angiotensin II inhibitors, beta-blockers, statins, anticoagulant, and antiplatelets were associated with lower risk, ranging from 23 to 32%.

Identifying asthma patients at high risk of exacerbation in a routine visit: A machine learning model.

BACKGROUND: Tools capable of predicting the risk of asthma exacerbations can facilitate asthma management in clinical practice. However, existing tools require additional data from patients beyond electronic medical records. OBJECTIVE: To predict asthma exacerbation in an upcoming year using electronically accessible data conditional on past adherence to asthma medications. METHODS: This retrospective cohort study included patients with ≥1 hospitalization or ≥2 medical claims for asthma within 2 consecutive years between 2002 and 2015 in Quebec administrative databases. Cohort entry (CE) was defined as the date of the first asthma-related ambulatory visit on or after meeting the operational definition of asthma. Adherence to each controller medication and use of each rescue medication was measured in the year prior to CE. Elastic-net regularized logistic regression was applied. RESULTS: Among 98,823 patients, the mean age was 55.9 years and 36.2% were men. The area under the curve for prediction was 0.708. In the model, the use of long-acting anticholinergic or long-acting ß2-agonists in the year prior to CE increased the odds of exacerbation by 24% and 21%, respectively. Among patients who received rescue medication, low and high adherence to controller medications increased the odds by 2%-5% compared with patients with medium adherence. Patients with a predicted risk of ≥0.20 were more likely to develop future exacerbation. CONCLUSION: This risk prediction indicated that asthma-related medication use increased the risk of asthma exacerbation. A potential U-shaped relationship between adherence to controller medications and the risk of exacerbation was identified among users of rescue medications.

Estimands and Estimation of COVID-19 Vaccine Effectiveness Under the Test-Negative Design: Connections to Causal Inference.

The test-negative design is routinely used for the monitoring of seasonal flu vaccine effectiveness. More recently, it has become integral to the estimation of COVID-19 vaccine effectiveness, in particular for more severe disease outcomes. Because the design has many important advantages and is becoming a mainstay for monitoring postlicensure vaccine effectiveness, epidemiologists and biostatisticians may be interested in further understanding the effect measures being estimated in these studies and connections to causal effects. Logistic regression is typically applied to estimate the conditional risk ratio but relies on correct outcome model specification and may be biased in the presence of effect modification by a confounder. We give and justify an inverse probability of treatment weighting (IPTW) estimator for the marginal risk ratio, which is valid under effect modification. We use causal directed acyclic graphs, and counterfactual arguments under assumptions about no interference and partial interference to illustrate the connection between these statistical estimands and causal quantities. We conduct a simulation study to illustrate and confirm our derivations and to evaluate the performance of the estimators. We find that if the effectiveness of the vaccine varies across patient subgroups, the logistic regression can lead to misleading estimates, but the IPTW estimator can produce unbiased estimates. We also find that in the presence of partial interference both estimators can produce misleading estimates.

Clinical Correlates Identify ProBDNF and Thrombo-Inflammatory Markers as Key Predictors of Circulating p75NTR Extracellular Domain Levels in Older Adults.

The p75NTR receptor binds all neurotrophins and is mostly known for its role in neuronal survival and apoptosis. Recently, the extracellular domain (ECD) of p75NTR has been reported in plasma, its levels being dysregulated in numerous neurological diseases. However, the factors associated with p75NTR ECD levels remain unknown. We investigated clinical correlates of plasma p75NTR ECD levels in older adults without clinically manifested neurological disorders. Circulating p75NTR levels were measured by enzyme-linked immunosorbent assay in plasma obtained from participants in the BEL-AGE cohort (n = 1,280). Determinants of plasma p75NTR ECD levels were explored using linear and non-linear statistical models. Plasma p75NTR ECD levels were higher in male participants; were positively correlated with circulating concentrations of pro-brain-derived neurotrophic factor, and inflammatory markers interleukin-6 and CD40 Ligand; and were negatively correlated with the platelet activation marker P-selectin. While most individuals had p75NTR levels ranging from 43 to 358 pg/ml, high p75NTR levels reaching up to 9,000 pg/ml were detectable in a subgroup representing 15% of the individuals studied. In this cohort of older adults without clinically manifested neurological disorders, there was no association between plasma p75NTR ECD levels and cognitive performance, as assessed by the Montreal Cognitive Assessment score. The physiological relevance of high p75NTR ECD levels in plasma warrants further investigation. Further research assessing the source of circulating p75NTR is needed for a deeper understanding of the direction of effect, and to investigate whether high p75NTR ECD levels are predictive biomarkers or consequences of neuropathology.

Doubly robust adaptive LASSO for effect modifier discovery.

Effect modification occurs when the effect of a treatment on an outcome differsaccording to the level of some pre-treatment variable (the effect modifier). Assessing an effect modifier is not a straight-forward task even for a subject matter expert. In this paper, we propose a two-stageprocedure to automatically selecteffect modifying variables in a Marginal Structural Model (MSM) with a single time point exposure based on the two nuisance quantities (the conditionaloutcome expectation and propensity score). We highlight the performance of our proposal in a simulation study. Finally, to illustrate tractability of our proposed methods, we apply them to analyze a set of pregnancy data. We estimate the conditional expected difference in the counterfactual birth weight if all women were exposed to inhaled corticosteroids during pregnancy versus the counterfactual birthweight if all women were not, using data from asthma medications during pregnancy.

A potential outcomes approach to defining and estimating gestational age-specific exposure effects during pregnancy.

Many studies seek to evaluate the effects of potentially harmful pregnancy exposures during specific gestational periods. We consider an observational pregnancy cohort where pregnant individuals can initiate medication usage or become exposed to a drug at various times during their pregnancy. An important statistical challenge involves how to define and estimate exposure effects when pregnancy loss or delivery can occur over time. Without proper consideration, the results of standard analysis may be vulnerable to selection bias, immortal time-bias, and time-dependent confounding. In this study, we apply the "target trials" framework of Hernán and Robins in order to define effects based on the counterfactual approach often used in causal inference. This effect is defined relative to a hypothetical randomized trial of timed pregnancy exposures where delivery may precede and thus potentially interrupt exposure initiation. We describe specific implementations of inverse probability weighting, G-computation, and Targeted Maximum Likelihood Estimation to estimate the effects of interest. We demonstrate the performance of all estimators using simulated data and show that a standard implementation of inverse probability weighting is biased. We then apply our proposed methods to a pharmacoepidemiology study to evaluate the potentially time-dependent effect of exposure to inhaled corticosteroids on birthweight in pregnant people with mild asthma.

Predictive Factors of Detectable Viral Load in HIV-Infected Patients.

Despite availability of effective antiretroviral therapy (ART), many HIV patients still have a detectable viral load (VL). Predictive factors of detectable VL are not well documented. This study was done at two large multidisciplinary HIV outpatient clinics at the Centre hospitalier de l'Université de Montréal (CHUM) and the McGill University Health Centre (MUHC). This is a retrospective case-control study of patients treated between 2016 and 2018. Cases had a VL ≥50 copies/mL in 2018. Controls had an undetectable VL from 2016 to 2018. Matching was based on gender and year of HIV diagnosis. Primary objective was to identify predictive factors of detectable VL. Secondary objectives included to identify predictive factors of virologic failure, low persistent viremia, and viral blip. A forward stepwise model selection by the Akaike Information Criterion of the conditional logistic regression was used to identify predictive factors. Two hundred cases were identified and matched with 200 controls. The cohort was mostly male (68.0%) with a median age of 54 years (21-83 years). Among cases, viral blip was the most common type of detectable VL (43.0%). The strong predictive factors for a detectable VL were adherence to ART and seeking health care services. Asylum seekers were less at risk of detectable VL. Adherence to ART was the only strong predictive factor for virologic failure. Three main predictive factors of detectable VL were identified in two ambulatory clinic hospitals in Montreal. Ascertaining these factors will allow for identification of patients more at risk of detectable VL.

Proton pump inhibitors and risk of gastric cancer: population-based cohort study.

OBJECTIVE: To determine whether new users of proton pump inhibitors (PPIs) are at an increased risk of gastric cancer compared with new users of histamine-2 receptor antagonists (H2RAs). DESIGN: Using the UK Clinical Practice Research Datalink, we conducted a population-based cohort study using a new-user active comparator design. From 1 January 1990 to 30 April 2018, we identified 973 281 new users of PPIs and 193 306 new users of H2RAs. Cox proportional hazards models were fit to estimate HRs and 95% CIs of gastric cancer, and the number needed to harm was estimated using the Kaplan-Meier method. The models were weighted using standardised mortality ratio weights using calendar time-specific propensity scores. Secondary analyses assessed duration and dose-response associations. RESULTS: After a median follow-up of 5.0 years, the use of PPIs was associated with a 45% increased risk of gastric cancer compared with the use of H2RAs (HR 1.45, 95% CI 1.06 to 1.98). The number needed to harm was 2121 and 1191 for five and 10 years after treatment initiation, respectively. The HRs increased with cumulative duration, cumulative omeprazole equivalents and time since treatment initiation. The results were consistent across several sensitivity analyses. CONCLUSION: The findings of this large population-based cohort study indicate that the use of PPIs is associated with an increased risk of gastric cancer compared with the use of H2RAs, although the absolute risk remains low.

Proton pump inhibitors and risk of colorectal cancer.

OBJECTIVE: To determine whether proton pump inhibitors (PPIs) are associated with an increased risk of colorectal cancer, compared with histamine-2 receptor antagonists (H2RAs). DESIGN: The United Kingdom Clinical Practice Research Datalink was used to identify initiators of PPIs and H2RA from 1990 to 2018, with follow-up until 2019. Cox proportional hazards models were fit to estimate marginal HRs and 95% CIs of colorectal cancer. The models were weighted using standardised mortality ratio weights using calendar time-specific propensity scores. Prespecified secondary analyses assessed associations with cumulative duration, cumulative dose and time since treatment initiation. The number needed to harm was calculated at five and 10 years of follow-up. RESULTS: The cohort included 1 293 749 and 292 387 initiators of PPIs and H2RAs, respectively, followed for a median duration of 4.9 years. While the use of PPIs was not associated with an overall increased risk of colorectal cancer (HR: 1.02, 95% CI 0.92 to 1.14), HRs increased with cumulative duration of PPI use (<2 years, HR: 0.93, 95% CI 0.83 to 1.04; 2-4 years, HR: 1.45, 95% CI 1.28 to 1.60; ≥4 years, HR: 1.60, 95% CI 1.42 to 1.80). Similar patterns were observed with cumulative dose and time since treatment initiation. The number needed to harm was 5343 and 792 for five and 10 years of follow-up, respectively. CONCLUSION: While any use of PPIs was not associated with an increased risk of colorectal cancer compared with H2RAs, prolonged use may be associated with a modest increased risk of this malignancy.

Modeling treatment effect modification in multidrug-resistant tuberculosis in an individual patientdata meta-analysis.

Effect modification occurs while the effect of the treatment is not homogeneous across the different strata of patient characteristics. When the effect of treatment may vary from individual to individual, precision medicine can be improved by identifying patient covariates to estimate the size and direction of the effect at the individual level. However, this task is statistically challenging and typically requires large amounts of data. Investigators may be interested in using the individual patient data from multiple studies to estimate these treatment effect models. Our data arise from a systematic review of observational studies contrasting different treatments for multidrug-resistant tuberculosis, where multiple antimicrobial agents are taken concurrently to cure the infection. We propose a marginal structural model for effect modification by different patient characteristics and co-medications in a meta-analysis of observational individual patient data. We develop, evaluate, and apply a targeted maximum likelihood estimator for the doubly robust estimation of the parameters of the proposed marginal structural model in this context. In particular, we allow for differential availability of treatments across studies, measured confounding within and across studies, and random effects by study.

A comparison of confounder selection and adjustment methods for estimating causal effects using large healthcare databases.

PURPOSE: Confounding adjustment is required to estimate the effect of an exposure on an outcome in observational studies. However, variable selection and unmeasured confounding are particularly challenging when analyzing large healthcare data. Machine learning methods may help address these challenges. The objective was to evaluate the capacity of such methods to select confounders and reduce unmeasured confounding bias. METHODS: A simulation study with known true effects was conducted. Completely synthetic and partially synthetic data incorporating real large healthcare data were generated. We compared Bayesian adjustment for confounding (BAC), generalized Bayesian causal effect estimation (GBCEE), Group Lasso and Doubly robust estimation, high-dimensional propensity score (hdPS), and scalable collaborative targeted maximum likelihood algorithms. For the hdPS, two adjustment approaches targeting the effect in the whole population were considered: Full matching and inverse probability weighting. RESULTS: In scenarios without hidden confounders, most methods were essentially unbiased. The bias and variance of the hdPS varied considerably according to the number of variables selected by the algorithm. In scenarios with hidden confounders, substantial bias reduction was achieved by using machine-learning methods to identify proxies as compared to adjusting only by observed confounders. hdPS and Group Lasso performed poorly in the partially synthetic simulation. BAC, GBCEE, and scalable collaborative-targeted maximum likelihood algorithms performed particularly well. CONCLUSIONS: Machine learning can help to identify measured confounders in large healthcare databases. They can also capitalize on proxies of unmeasured confounders to substantially reduce residual confounding bias.