Deep Learning on Electronic Health Records to Improve Disease Coding Accuracy.

Characterization of a patient's clinical phenotype is central to biomedical informatics. ICD codes, assigned to inpatient encounters by coders, is important for population health and cohort discovery when clinical information is limited. While ICD codes are assigned to patients by professionals trained and certified in coding there is substantial variability in coding. We present a methodology that uses deep learning methods to model coder decision making and that predicts ICD codes. Our approach predicts codes based on demographics, lab results, and medications, as well as codes from previous encounters. We are able to predict existing codes with high accuracy for all three of the test cases we investigated: diabetes, acute renal failure, and chronic kidney disease. We employed a panel of clinicians, in a blinded manner, to assess ground truth and compared the predictions of coders, model and clinicians. When disparities between the model prediction and coder assigned codes were reviewed, our model outperformed coder assigned ICD codes.

Geographic, Temporal, and Sociodemographic Differences in Opioid Poisoning.

INTRODUCTION: Not enough is known about the epidemiology of opioid poisoning to tailor interventions to help address the growing opioid crisis in the U.S. The objective of this study is to expand the current understanding of opioid poisoning through the use of data analytics to evaluate geographic, temporal, and sociodemographic differences of opioid poisoning- related hospital visits in a region of New York State with high opioid poisoning rates. METHODS: This retrospective cohort study utilized patient-level New York State all-payer hospital data (2010-2016) combined with Census data to evaluate geographic, patient, and community factors for 9,714 Long Island residents with an opioid poisoning-related inpatient or outpatient hospital facility discharge. Temporal, 7-year opioid poisoning rates and trends were evaluated, and geographic maps were generated. Overall, significance tests and tests for linear trend were based upon logistic regression. Analyses were completed between 2017 and 2018. RESULTS: Since 2010, Long Island and New York State opioid poisoning hospital visit rates have increased 2.5- to 2.7-fold (p<0.001). Opioid poisoning hospital visit rates decreased for men, white patients, and self-payers (p<0.001) and increased for Medicare payers (p<0.001). Communities with high opioid poisoning rates had lower median home values, higher percentages of high school graduates, were younger, and more often white patients (p<0.01). Maps displayed geographic patterns of communities with high opioid poisoning rates overall and by age group. CONCLUSIONS: Findings highlight the changing demographics of the opioid poisoning epidemic and utility of data analytics tools to identify regions and patient populations to focus interventions. These population identification techniques can be applied in other communities and interventions.