Dataset Information

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

ABSTRACT:

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.

SUBMITTER: Benasseur I

PROVIDER: S-EPMC9304306 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BACKGROUND:When trials are subject to departures from randomised treatment, simple statistical methods that aim to estimate treatment efficacy, such as per protocol or as treated analyses, typically introduce selection bias. More appropriate methods to adjust for departure from randomised treatment are rarely employed, primarily due to their complexity and unfamiliarity. We demonstrate the use of causal methodologies for the production of estimands with valid causal interpretation for time-to-event outcomes in the analysis of a complex epilepsy trial, as an example to guide non-specialist analysts undertaking similar analyses. METHODS:Two causal methods, the structural failure time model and inverse probability of censoring weighting, are adapted to allow for skewed time-varying confounders, competing reasons for treatment changes and a complicated time to remission outcome. We demonstrate the impact of various factors: choice of method (structural failure time model versus inverse probability of censoring weighting), model for inverse probability of censoring weighting (pooled logistic regression versus Cox models), time interval (for creating panel data for time-varying confounders and outcome), choice of confounders and (in pooled logistic regression) use of splines to estimate underlying risk. RESULTS:The structural failure time model could adjust for switches between trial treatments but had limited ability to adjust for the other treatment changes that occurred in this epilepsy trial. Inverse probability of censoring weighting was able to adjust for all treatment changes and demonstrated very similar results with Cox and pooled logistic regression models. Accounting for increasing numbers of time-varying confounders and reasons for treatment change suggested a more pronounced advantage of the control treatment than that obtained using intention to treat. CONCLUSIONS:In a complex trial featuring a remission outcome, underlying assumptions of the structural failure time model are likely to be violated, and inverse probability of censoring weighting may provide the most useful option, assuming availability of appropriate data and sufficient sample sizes. Recommendations are provided for analysts when considering which of these methods should be applied in a given trial setting.