Extraction of Clinical Phenotypic Information from Online Heterogeneous Healthcare Networks

Millions of patients are affected by adverse drug reactions (ADRs) every year. It represents a substantial burden on healthcare resources. Pharmacovigilance using text and data analytics has drawn substantial attention in the recent years. These techniques are mainly extracting the associations between drugs and ADRs using data sources such as spontaneous reporting systems, electronic health records, medical literature, and pharmacological databases. In this work, we are not only interested in extracting the associations between drugs and ADRs but also the associations between diseases and ADRs. There is an association between a disease and an ADR when the drugs treating the disease are associated with the same ADR, which means there might be an underlying mechanism-of-action (MOA) between the disease and the ADR [1]. The ADR can be considered as a clinical phenotypic biomarker for the disease. In addition, we are adopting the social media data as the data source in analytics. The social media provides timely and large volume of health consumer contributed information that overcomes the limitations the traditional data sources. We propose to construct a heterogeneous healthcare network from social media data and develop three path-mining techniques to the clinical phenotypic information. The experiments results demonstrate that the proposed method is effective in detecting significant and novel ADR-disease associations. Case study shows that many of the association can be supported by existing academic literatures.

I. Introduction

Clinical phenotypic information comes from actual patient data, which mimics a phenotypic “screen” of the drug effects on human [1]. In this work, we focus on the extraction of the associations between adverse drug reactions (ADRs) and diseases in the clinical phenotypic information contributed by patients themselves in social media. ADR is defined as an appreciably harmful or unpleasant reaction, which results from an intervention related to the use of a medicinal product, predicts hazard from future administration, and warrants prevention or specific treatment, or alteration of the dosage regimen, or withdrawal of the product [2]. ADR represents a significant health problem worldwide and it has been becoming the leading cause of death as well as diseases in United States [3]. A nationwide study in Netherlands demonstrated that 1.83% of hospital admissions are ADR related during 2001 [4]. Therefore, research about ADR is an important topic in healthcare informatics.