Dataset Information

Can supervised deep learning architecture outperform autoencoders in building propensity score models for matching?

ABSTRACT:

Purpose

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.

Methods

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.

Results

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.

Conclusion

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.

SUBMITTER: Karim ME

PROVIDER: S-EPMC11295454 | biostudies-literature | 2024 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Can supervised deep learning architecture outperform autoencoders in building propensity score models for matching?

Karim Mohammad Ehsanul ME

BMC medical research methodology 20240802 1

<h4>Purpose</h4>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.<h4>Methods</h4>Utilizing a plasmode simulation based on the Right Heart Catheterization dataset, under a variety of sett ...[more]

PMID: 39095707

Similar Datasets

Project description:BackgroundThe diagnostic process is a key element of medicine but it is complex and prone to errors. Infectious diseases are one of the three categories of diseases in which diagnostic errors can be most harmful to patients. In this study we aimed to estimate the effect of initial misdiagnosis of the source of infection in patients with bacteraemia on 14 day mortality using propensity score methods to adjust for confounding.MethodsData from a previously described longitudinal cohort of patients diagnosed with monobacterial bloodstream infection (BSI) at the Leiden University Medical Centre (LUMC) between 2013 and 2015 were used. Propensity score matching and inversed probability of treatment weighting (IPTW) were applied to correct for confounding. The average treatment effect on the treated (ATT), which in this study was the average effect of initial misdiagnosis on the misdiagnosed (AEMM), was estimated. Methodological issues that were encountered when applying propensity score methods were addressed by performing additional sensitivity analyses. Sensitivity analyses consisted of varying caliper in propensity score matching and using different truncated weights in inversed probability of treatment weighting.ResultsData of 887 patients were included in the study. Propensity scores ranged between 0.015 and 0.999 and 80 patients (9.9%) had a propensity score > 0.95. In the matched analyses, 35 of the 171 misdiagnosed patients died within 14 days (20.5%), versus 10 of the 171 correctly diagnosed patients (5.8%), yielding a difference of 14.6% (7.6%; 21.6%). In the total group of patients, the observed percentage of patients with an incorrect initial diagnosis that died within 14 days was 19.8% while propensity score reweighting estimated that their probability of dying would have been 6.5%, if they had been correctly diagnosed (difference 13.3% (95% CI 6.9%;19.6%)). After adjustment for all variables that showed disbalance in the propensity score a difference of 13.7% (7.4%; 19.9%) was estimated. Sensitivity analyses yielded similar results. However, performing weighted analyses without truncation yielded unstable results.ConclusionThus, we observed a substantial increase of 14 day mortality in initially misdiagnosed patients. Furthermore, several patients received propensity scores extremely close to one and were almost sure to be initially misdiagnosed.

Project description:It is increasingly important to accurately and comprehensively estimate the effects of particular clinical treatments. Although randomization is the current gold standard, randomized controlled trials (RCTs) are often limited in practice due to ethical and cost issues. Observational studies have also attracted a great deal of attention as, quite often, large historical datasets are available for these kinds of studies. However, observational studies also have their drawbacks, mainly including the systematic differences in baseline covariates, which relate to outcomes between treatment and control groups that can potentially bias results. Propensity score methods, which are a series of balancing methods in these studies, have become increasingly popular by virtue of the two major advantages of dimension reduction and design separation. Within this approach, propensity score matching (PSM) has been empirically proven, with outstanding performances across observational datasets. While PSM tutorials are available in the literature, there is still room for improvement. Some PSM tutorials provide step-by-step guidance, but only one or two packages have been covered, thereby limiting their scope and practicality. Several articles and books have expounded upon propensity scores in detail, exploring statistical principles and theories; however, the lack of explanations on function usage in programming language has made it difficult for researchers to understand and follow these materials. To this end, this tutorial was developed with a six-step PSM framework, in which we summarize the recent updates and provide step-by-step guidance to the R programming language. This tutorial offers researchers with a broad survey of PSM, ranging from data preprocessing to estimations of propensity scores, and from matching to analyses. We also explain generalized propensity scoring for multiple or continuous treatments, as well as time-dependent PSM. Lastly, we discuss the advantages and disadvantages of propensity score methods.

Project description:ObjectivesAt present, there are no established clinical guidelines for radiofrequency ablation (RFA) of peribiliary hepatocellular carcinoma (HCC). Therefore, the aim of this study was to compare the long-term outcomes of RFA for peribiliary vs. non-peribiliary HCC.MethodsThis retrospective study included 282 patients with peribiliary HCC (n = 109) or non-peribiliary HCC (n = 173) who received RFA between February 2013 and May 2021. Local tumor progression (LTP), overall survival (OS), disease-free survival (DFS), and complications were compared before and after propensity score matching (PSM).ResultsBefore PSM, there were no significant differences in 5-year LTP rates (26.3% vs. 23.6%, p = 0.602), OS rates (56.6% vs. 68.0%, p = 0.586), or DFS rates (22.9% vs. 25.7%, p = 0.239) between the peribiliary and non-peribiliary groups. After PSM, there were no significant differences in the 1-, 3-, and 5-year LTP rates (13.0%, 23.1%, and 26.3% vs. 12.1%, 25.1%, and 28.2%, respectively, p = 0.857), OS rates (97.2%, 73.5%, and 56.6% vs. 95.3%, 79.5%, and 70.6%, p = 0.727), or DFS rates (59.4%, 29.4%, and 22.9% vs. 64.2%, 33.1%, and 23.8%, p = 0.568) between the peribiliary non-peribiliary groups. Peribiliary location was not a significant prognostic factor for LTP (p = 0.622) or OS (p = 0.587). In addition, mild intrahepatic bile duct dilatation was more frequent in the peribiliary group (9.2% vs. 2.8%, p = 0.045).ConclusionLong-term outcomes of RFA were similar for peribiliary and non-peribiliary HCC. RFA is a viable alternative for treatment of peribiliary HCC.Critical relevance statementThe local tumor progression (LTP), overall survival (OS), and disease-free survival (DFS) rates after radiofrequency ablation (RFA) were similar for peribiliary and non-peribiliary hepatocellular carcinoma (HCC).Key pointsThere are currently no clinical guidelines for radiofrequency ablation (RFA) of peribiliary hepatocellular carcinoma (HCC). Local tumor progression, overall survival, and disease-free survival after RFA were similar for peribiliary and non-peribiliary HCC. RFA is a viable alternative for the treatment of peribiliary HCC.

Project description:BackgroundLoneliness and unemployment are each detrimental to health and well-being. Recent evidence suggests a potential bidirectional relationship between loneliness and unemployment in working age individuals. As most existing research focuses on the outcomes of unemployment, this paper seeks to understand the impact of loneliness on unemployment, potential interaction with physical health, and assess bidirectionality in the working age population.MethodsThis study utilised data from waves 9 (2017-19) and 10 (2018-2020) of the Understanding Society UK Household Longitudinal Study. Nearest-neighbour probit propensity score matching with at least one match was used to infer causality by mimicking randomisation. Analysis was conducted in three steps: propensity score estimation; matching; and stratification. Propensity scores were estimated controlling for age, gender, ethnicity, education, marital status, household composition, number of own children in household and region. Findings were confirmed in panel data random effect models, and heterogeneous treatment effects assessed by the matching-smoothing method.ResultsExperience of loneliness in at least one wave increased the probability of being unemployed in wave 10 by 17.5 [95%CI: 14.8, 20.2] percentage points. Subgroup analysis revealed a greater effect from sustained than transitory loneliness. Further exploratory analysis identified a positive average treatment effect, of smaller magnitude, for unemployment on loneliness suggesting bidirectionality in the relationship. The impact of loneliness on unemployment was further exacerbated by interaction with physical health.ConclusionsThis is the first study to directly consider the potentially bidirectional relationship between loneliness and unemployment through analysis of longitudinal data from a representative sample of the working age population. Findings reinforce the need for greater recognition of wider societal impacts of loneliness. Given the persisting and potentially scarring effects of both loneliness and unemployment on health and the economy, prevention of both experiences is key. Decreased loneliness could mitigate unemployment, and employment abate loneliness, which may in turn relate positively to other factors including health and quality of life. Thus, particular attention should be paid to loneliness with additional support from employers and government to improve health and well-being.

Dataset Information

Can supervised deep learning architecture outperform autoencoders in building propensity score models for matching?

Purpose

Methods

Results

Conclusion

Publications

Can supervised deep learning architecture outperform autoencoders in building propensity score models for matching?

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets