RESUMO
We developed an inherently interpretable multilevel Bayesian framework for representing variation in regression coefficients that mimics the piecewise linearity of ReLU-activated deep neural networks. We used the framework to formulate a survival model for using medical claims to predict hospital readmission and death that focuses on discharge placement, adjusting for confounding in estimating causal local average treatment effects. We trained the model on a 5% sample of Medicare beneficiaries from 2008 and 2011, based on their 2009-2011 inpatient episodes (approximately 1.2 million), and then tested the model on 2012 episodes (approximately 400 thousand). The model scored an out-of-sample AUROC of approximately 0.75 on predicting all-cause readmissions-defined using official Centers for Medicare and Medicaid Services (CMS) methodology-or death within 30-days of discharge, being competitive against XGBoost and a Bayesian deep neural network, demonstrating that one need-not sacrifice interpretability for accuracy. Crucially, as a regression model, it provides what blackboxes cannot-its exact gold-standard global interpretation, explicitly defining how the model performs its internal "reasoning" for mapping the input data features to predictions. In doing so, we identify relative risk factors and quantify the effect of discharge placement. We also show that the posthoc explainer SHAP provides explanations that are inconsistent with the ground truth model reasoning that our model readily admits.
Assuntos
Teorema de Bayes , Medicare , Alta do Paciente , Readmissão do Paciente , Humanos , Readmissão do Paciente/estatística & dados numéricos , Alta do Paciente/estatística & dados numéricos , Estados Unidos/epidemiologia , Feminino , Idoso , Masculino , Redes Neurais de Computação , Idoso de 80 Anos ou maisRESUMO
Iodine deficiency in pregnancy is a common problem in the United States and parts of Europe, but whether iodine deficiency is associated with increased pregnancy loss has not been well studied. The LIFE study provided an excellent opportunity to examine the relationship between iodine status and pregnancy loss because women were monitored prospectively to ensure excellent ascertainment of conceptions. The LIFE study, a population-based prospective cohort study, monitored 501 women who had discontinued contraception within two months to become pregnant; 329 became pregnant, had urinary iodine concentrations measured on samples collected at enrollment, and were followed up to determine pregnancy outcomes. Of the 329, 196 had live births (59.5%), 92 (28.0%) had losses, and 41 (12.5%) withdrew or were lost to follow up. Urinary iodine concentrations were in the deficiency range in 59.6% of the participants. The risk of loss, however, was not elevated in the mildly deficient group (hazard ratio 0.69, 95% confidence interval 0.34, 1.38), the moderately deficient group (hazard ratio 0.81, 95% confidence interval 0.43, 1.51), or the severely deficient group (hazard ratio 0.69, 95% confidence interval 0.32, 1.50). Iodine deficiency, even when moderate to severe, was not associated with increased rates of pregnancy loss. This study provides some reassurance that iodine deficiency at levels seen in many developed countries does not increase the risk of pregnancy loss.