A Poisson binomial based statistical testing framework for comprehensive comorbidity discovery across massive Electronic Health Record datasets

Abstract

Discovery of comorbidities (the concomitant occurrence of distinct medical conditions in the same patient) is a prerequisite for creating forecasting tools for downstream outcomes research. Current comorbidity discovery applications are designed for small datasets and use stratification to control for confounding variables such as age, sex, or ancestry. Stratification lowers false positive rates, but reduces power, as the size of the study cohort is decreased. Here, we describe a Poisson Binomial based approach to comorbidity discovery (PBC) designed for big-data applications that circumvents the need for stratification. PBC adjusts for confounding demographic variables on a per-patient basis, and models temporal relationships. We benchmark PBC using two datasets, the publicly available MIMIC-IV; and the entire Electronic Health Record (EHR) corpus of the University of Utah Hospital System, encompassing over 1.6 million patients, to compute comorbidity statistics on 4,623,841 pairs of potentially comorbid medical terms. The results of this computation are provided as a searchable web resource. Compared to current methods, the PBC approach reduces false positive associations, while retaining statistical power to discover true comorbidities.

An Explainable Artificial Intelligence Approach for Predicting Cardiovascular Outcomes using Electronic Health Records

“…Unlike Neural nets, which are incredibly scalable, multimorbidity networks can model a maximum of only 30 or so variables at once 28,36,37 . It is therefore necessary to pre-identify high impact variables when modeling an outcome, a need fulfilled by PBC 10 . We argue that the ability to rigorously investigate interrelations among 30 or so primary determinants represents a giant step forward towards understanding cardiovascular disease outcomes.…”

“…By contrast, the PBC approach maintains power across all comparisons. For more on these points, see 10 .…”

“…We evaluated every pairwise combination of diagnoses, procedures, and medications mentioned in our EHR corpus for comorbid associations, using PBC 10 to adjust on a patient-by-patient basis for the potentially confounding demographic variables shown in Figure 1. Figure 2A summarizes the results of this computation as a patient disease network.…”

“…We used a massively-scalable comorbidity discovery method called Poisson Binomial based Comorbidity (PBC) discovery 10 to search Electronic Health Records (EHRs) from the University of Utah and Primary Children's Hospital (over 1.6 million patients and 77 million visits) for comorbid diagnoses, procedures, and medications. In this context, we refer to co-occurring medical diagnoses, procedures and medications using the single blanket term, comorbidity.…”