The Secondary Use of Electronic Health Records for Data Mining: Data Characteristics and Challenges

Abstract

The primary objective of implementing Electronic Health Records (EHRs) is to improve the management of patients’ health-related information. However, these records have also been extensively used for the secondary purpose of clinical research and to improve healthcare practice. EHRs provide a rich set of information that includes demographics, medical history, medications, laboratory test results, and diagnosis. Data mining and analytics techniques have extensively exploited EHR information to study patient cohorts for various clinical and research applications, such as phenotype extraction, precision medicine, intervention evaluation, disease prediction, detection, and progression. But the presence of diverse data types and associated characteristics poses many challenges to the use of EHR data. In this article, we provide an overview of information found in EHR systems and their characteristics that could be utilized for secondary applications. We first discuss the different types of data stored in EHRs, followed by the data transformations necessary for data analysis and mining. Later, we discuss the data quality issues and characteristics of the EHRs along with the relevant methods used to address them. Moreover, this survey also highlights the usage of various data types for different applications. Hence, this article can serve as a primer for researchers to understand the use of EHRs for data mining and analytics purposes.

Learning the progression patterns of treatments using a probabilistic generative model

“…The model can be also regarded as a data augmentation tool when little information is available, for example, for rare diseases. In addition to this, since healthcare datasets are frequently incomplete and the removal of missing values may result in a dataset that is too small or induce statistical bias [1], the model has the ability to impute such missing values in the trajectories of patients or reconstruct incomplete sequences of actions. In terms of interpretability, our model provides easier comprehension and explanation for end-users than other approaches developed in the healthcare setting [22,23,24,25,26,27].…”

“…These repositories contain systematized collections of patient data, including demographics, procedures, diagnosis, medications, costs, medical service providers and so on. The order of occurrence of these medical events in EHRs provides valuable information on the treatment trajectory of a patient which could improve the understanding of the disease [1]. Therefore, we represent each patient using chronologically ordered sequences of medical actions.…”

“…synthetic datasets of different sizes to evaluate the behavior of the learning algorithm by comparing the learned models with the original generative model underlying the data. The corresponding source code is publicly available1 . Secondly, we apply the model on real-world EHRs about breast cancer disease to classify the sequences of actions and segment them in different progression stages.3.1.…”

Predicting Alzheimer’s Disease with Interpretable Machine Learning

“…With the generation of huge volume of medical and health data, it has significantly stimulated the rapid development and application of data mining techniques in the field of healthcare [4]. Studies on data-driven diagnostic or prognostic models for diseases are on the rise, evolving from classical statistical methods to machine learning models [5].…”

Characterizing suicidal ideation, suicidal behaviors, and service utilization among unhoused individuals using a health information exchange

“…Suicide and homelessness share many risk factors, but few have studied the temporal relationship between homelessness and SI/SB, assessed patterns of suicide‐related service utilization around the onset of homelessness, or distinguished between risk factors of SI versus SB. Data from health information exchanges (HIEs) are a potentially valuable, complementary resource for filling these knowledge gaps (Sarkar, 2022; Sarwar et al, 2022).…”