Personalizing Medication Recommendation with a Graph-Based Approach

The broad adoption of electronic health records (EHRs) has led to vast amounts of data being accumulated on a patient’s history, diagnosis, prescriptions, and lab tests. Advances in recommender technologies have the potential to utilize this information to help doctors personalize the prescribed medications. However, existing medication recommendation systems have yet to make use of all these information sources in a seamless manner, and they do not provide a justification on why a particular medication is recommended. In this work, we design a two-stage personalized medication recommender system called PREMIER that incorporates information from the EHR. We utilize the various weights in the system to compute the contributions from the information sources for the recommended medications. Our system models the drug interaction from an external drug database and the drug co-occurrence from the EHR as graphs. Experiment results on MIMIC-III and a proprietary outpatient dataset show that PREMIER outperforms state-of-the-art medication recommendation systems while achieving the best tradeoff between accuracy and drug-drug interaction. Case studies demonstrate that the justifications provided by PREMIER are appropriate and aligned to clinical practices.

Assessment of Natural Language Processing Methods for Ascertaining the Expanded Disability Status Scale Score From the Electronic Health Records of Patients With Multiple Sclerosis: Algorithm Development and Validation Study

Background The Expanded Disability Status Scale (EDSS) score is a widely used measure to monitor disability progression in people with multiple sclerosis (MS). However, extracting and deriving the EDSS score from unstructured electronic health records can be time-consuming. Objective We aimed to compare rule-based and deep learning natural language processing algorithms for detecting and predicting the total EDSS score and EDSS functional system subscores from the electronic health records of patients with MS. Methods We studied 17,452 electronic health records of 4906 MS patients followed at one of Canada’s largest MS clinics between June 2015 and July 2019. We randomly divided the records into training (80%) and test (20%) data sets, and compared the performance characteristics of 3 natural language processing models. First, we applied a rule-based approach, extracting the EDSS score from sentences containing the keyword “EDSS.” Next, we trained a convolutional neural network (CNN) model to predict the 19 half-step increments of the EDSS score. Finally, we used a combined rule-based–CNN model. For each approach, we determined the accuracy, precision, recall, and F-score compared with the reference standard, which was manually labeled EDSS scores in the clinic database. Results Overall, the combined keyword-CNN model demonstrated the best performance, with accuracy, precision, recall, and an F-score of 0.90, 0.83, 0.83, and 0.83 respectively. Respective figures for the rule-based and CNN models individually were 0.57, 0.91, 0.65, and 0.70, and 0.86, 0.70, 0.70, and 0.70. Because of missing data, the model performance for EDSS subscores was lower than that for the total EDSS score. Performance improved when considering notes with known values of the EDSS subscores. Conclusions A combined keyword-CNN natural language processing model can extract and accurately predict EDSS scores from patient records. This approach can be automated for efficient information extraction in clinical and research settings.

Linking electronic health records and in-depth interviews to inform efforts to integrate social determinants of health into health care delivery: Protocol for a qualitative research study (Preprint)

BACKGROUND Health systems are attempting to capture social determinants of health (SDoH) in electronic health records (EHR) and use these data to adjust care plans. To date, however, methods for identifying social needs, which are the SDoH prioritized by patients, have been underexplored, and there is little guidance as to how clinicians should act on SDoH data when caring for patients. Moreover, the unintended consequences of collecting and responding to SDoH are poorly understood. OBJECTIVE The objective of this study is to use two data sources, EHR data and patient interviews, to describe divergences between the EHR and patient experiences that could help identify gaps in documentation of SDoH in the EHR; highlight potential missed opportunities for addressing social needs; and identify unintended consequences of efforts to integrate SDoH into clinical care. METHODS We are conducting a qualitative study that merges discrete and free-text data from EHRs with in-depth interviews with women residing in rural, socio-economically deprived communities in the Mid-Atlantic region of the United States. Participants had to confirm that they had at least one visit with the large health system that serves the region. Interviews with the women included questions regarding health, interaction with the health system, and social needs. Next, with consent, for each participant we extracted discrete data (e.g., diagnoses; medication orders) and free-text clinician notes from this health system’s EHRs between 1996 and the year of the interview. We used a standardized protocol to create an EHR narrative, a free-text summary of the EHR data. We used NVivo to identify themes in the interviews and the EHR narratives. RESULTS To date, we have interviewed 88 women, including 51 White women, 19 Black women, 14 Latina women, 2 mixed Black and Latina women, and 2 Asian Pacific women. We have completed the EHR narratives on 66 women. The women range in age from 18 to 90. We found corresponding EHR data on all but 4 of the interview participants. Participants had contact with a wide range of clinical departments (e.g., psychiatry, neurology, infectious disease) and received care in various clinical settings (e.g., primary care clinics, emergency departments, inpatient hospitalizations). A preliminary review of the EHR narratives revealed that the clinician notes were a source of data on a range of SDoH, but did not always reflect the social needs that participants described in the interviews. CONCLUSIONS This study will provide unique insight into the demands and consequences of integrating SDoH into clinical care. This work comes at a pivotal point in time, as health systems, payors, and policy makers accelerate attempts to deliver care within the context of social needs.

Machine-learning based prediction of hypo- and hyperglycemia from electronic health records (Preprint)

BACKGROUND The increasing need for blood glucose (BG) management in hospitalized patients poses high demands on clinical staff and health care systems alike. Acute decompensations of BG levels (hypo- and hyperglycemia) adversely affect patient outcomes and safety. OBJECTIVE Acute BG decompensations pose a frequent and significant risk for inpatients. Ideally, proactive measures are taken before BG levels derail. We have generated a broadly applicable multiclass classification model for predicting decompensation events from patients’ electronic health records to indicate where adjustments of patient monitoring and/or therapeutic interventions are required. METHODS A retrospective cohort study was conducted of patients hospitalized at a tertiary hospital in Bern, Switzerland. Using patient details and routine data from electronic health records (EHRs), a multiclass prediction model for BG decompensation events (< 3.9 mmol/L (hypoglycemia), or > 10, > 13.9, or > 16.7 mmol/L (representing different degrees of hyperglycemia)) was generated, based on a second-level ensemble of gradient-boosted binary trees. RESULTS 63’579 hospital admissions of 33’212 patients were included in this study. The multiclass prediction model reached a specificity of 93.0%, 98.5%, and 93.6% and a sensitivity of 69.6%, 63.0%, and 65.5%, for the main categories of interest. i.e., non-decompensated cases, hypo- or hyperglycemia, respectively. The median prediction horizon was seven and four hours for hypo- and hyperglycemia, respectively. CONCLUSIONS EHRs hold the potential to reliably predict all kinds of BG decompensations. Readily available patient details and routine laboratory data can support the decisions for proactive interventions and thus help to reduce the detrimental health effects of hypo- and hyperglycemia.