A classification framework for exploiting sparse multi-variate temporal features with application to adverse drug event detection in medical records

Abstract

BackgroundAdverse drug events (ADEs) as well as other preventable adverse events in the hospital setting incur a yearly monetary cost of approximately $3.5 billion, in the United States alone. Therefore, it is of paramount importance to reduce the impact and prevalence of ADEs within the healthcare sector, not only since it will result in reducing human suffering, but also as a means to substantially reduce economical strains on the healthcare system. One approach to mitigate this problem is to employ predictive models. While existing methods have been focusing on the exploitation of static features, limited attention has been given to temporal features.MethodsIn this paper, we present a novel classification framework for detecting ADEs in complex Electronic health records (EHRs) by exploiting the temporality and sparsity of the underlying features. The proposed framework consists of three phases for transforming sparse and multi-variate time series features into a single-valued feature representation, which can then be used by any classifier. Moreover, we propose and evaluate three different strategies for leveraging feature sparsity by incorporating it into the new representation.ResultsA large-scale evaluation on 15 ADE datasets extracted from a real-world EHR system shows that the proposed framework achieves significantly improved predictive performance compared to state-of-the-art. Moreover, our framework can reveal features that are clinically consistent with medical findings on ADE detection.ConclusionsOur study and experimental findings demonstrate that temporal multi-variate features of variable length and with high sparsity can be effectively utilized to predict ADEs from EHRs. Two key advantages of our framework are that it is method agnostic, i.e., versatile, and of low computational cost, i.e., fast; hence providing an important building block for future exploitation within the domain of machine learning from EHRs.Electronic supplementary materialThe online version of this article (10.1186/s12911-018-0717-4) contains supplementary material, which is available to authorized users.

Identifying medication harm in hospitalised patients: a bimodal, targeted approach

“…A recent study described the development of multiple risk prediction models. 36 The authors utilised a de-identified dataset from a Swedish hospital. A series of ICD-10 codes related to the diagnosis of medication harm were used as the outcomes for models (e.g.…”

“…causality assessment to ensure codes were correctly allocated). 36 An Australian study also reported using ML and ICD codes to detect medication harm in a tertiary hospital. The automated algorithm demonstrated promising performance with an AUC of 0.803.…”

Time-ordered comorbidity correlations identify patients at risk of mis- and overdiagnosis

“…Diagnostic errors are common and it is estimated that everyone will experience at least one diagnostic error in their lifetime 1 . Although there is general agreement on co- or multi-morbidities complicating the diagnostic procedure resulting in higher risk for erroneous diagnoses, definitions for erroneous diagnoses lack consistency and several terms are used to refer to flawed diagnoses 2 – 5 . The terms misdiagnosis and overdiagnosis are often used interchangeably and they can be difficult to distinguish 3 .…”

“…Furthermore, to the best of our knowledge, no methods to identify overdiagnosis at the single patient level have been developed. Hence, there is a need to systematically identify patients at risk of mis- and overdiagnosis, leading to unnecessary harm from treatments and potentially missed underlying diseases 5 , 8 , 9 .…”

New and Increasing Rates of Adverse Events Can be Found in Unstructured Text in Electronic Health Records using the Shakespeare Method

“…Writers may explicitly state an apparent association between treatment and adverse outcome (“attributed”) or state the simple treatment and outcome without an association (“unattributed”). More critically, attributed and unattributed potential AEs (PAEs) may not necessarily be captured in structured data (e.g., diagnosis and procedure codes) [14, 23, 39].…”