QuBBD: Estimating drug-drug and drug-disease interactions for nursing homes residents

Brown University, Providence RI

Investigators

Abstract

Drug-drug interaction (DDI) is a situation in which one drug affects the activity of another. Drug-disease interaction (DDSI) is a situation where prescribed medication has the potential to exacerbate a preexisting disease. Both DDI and DDSI are major causes of adverse drug effects (ADEs) and represent a considerable threat to public health. Frail elderly patients have a higher prevalence of multiple morbidities and they are often prescribed a wide range of drugs to treat these morbidities. Thus, they are at increased risk for adverse effects from DDI and DDSI. Other than hospitalization, re-hospitalization and emergency room visits, little is known about the contribution of DDI and DDSI to other important outcomes, such as functional status, institutionalization, and mortality. In addition, some DDIs and DDSIs are unknown and require evidence-based updates. This award supports initiation of a collaborative research project with the long-term goal of developing methods that efficiently identify ADEs caused by DDIs and DDSIs by combining electronic health records with precise clinical data (such as gene sequencing). These findings will enable the updating of current guidelines, generate alerts, identify possible chemical compounds that may interact and result in ADEs, and propose new genetic drug pathways. The current objective, which is a step toward the long-term goal, is composed of two aims. The first aim is to combine current methods in machine learning and statistical causal inference to accurately identify DDI and DDSI. The second aim is to develop statistically valid methods that will impute missing patient level data in large detailed datasets. The project will exploit the availability of current datasets that could be augmented in the future to include precise clinical data. The Minimum Data Set is part of the U.S. federally mandated process for clinical assessment of all residents in Medicare or Medicaid certified nursing homes. It includes data on cognitive and functional abilities, disease diagnosis, psychological well-being, and medication use. The dataset comprises millions of patients, thus enabling the discovery of DDIs and DDSIs that occur in a small fraction of the population. The first aim comprises a combination of state-of-the-art supervised learning approaches such as LASSO, random forests, SVM, and Ensemble methods with efficient statistical causal inference approaches such as optimal matching, weighting, targeted learning, and multiple imputation. The approach is innovative because it combines causal inference methods with supervised learning methods to generate accurate and precise methods that have statistically valid operating characteristics when providing estimates for a specific patient. The second aim includes development of methods that rely on dimension reduction of the observed covariates to impute missing data. When imputing missing data, the rule of thumb is to use all of the available data, so that the missing at random assumption (MAR) is plausible. Dimension reduction aims to reduce the number of covariates that are used for prediction. Thus, development of an imputation algorithm with many missing covariates requires striking a balance between dimension reduction and plausibility of the MAR assumption. This award is supported by the National Institutes of Health Big Data to Knowledge (BD2K) Initiative in partnership with the National Science Foundation Division of Mathematical Sciences.