Comparison of deep learning with traditional models to predict preventable acute care use and spending among heart failure patients.

Recent health reforms have created incentives for cardiologists and accountable care organizations to participate in value-based care models for heart failure (HF). Accurate risk stratification of HF patients is critical to efficiently deploy interventions aimed at reducing preventable utilization. The goal of this paper was to compare deep learning approaches with traditional logistic regression (LR) to predict preventable utilization among HF patients. We conducted a prognostic study using data on 93,260 HF patients continuously enrolled for 2-years in a large U.S. commercial insurer to develop and validate prediction models for three outcomes of interest: preventable hospitalizations, preventable emergency department (ED) visits, and preventable costs. Patients were split into training, validation, and testing samples. Outcomes were modeled using traditional and enhanced LR and compared to gradient boosting model and deep learning models using sequential and non-sequential inputs. Evaluation metrics included precision (positive predictive value) at k, cost capture, and Area Under the Receiver operating characteristic (AUROC). Deep learning models consistently outperformed LR for all three outcomes with respect to the chosen evaluation metrics. Precision at 1% for preventable hospitalizations was 43% for deep learning compared to 30% for enhanced LR. Precision at 1% for preventable ED visits was 39% for deep learning compared to 33% for enhanced LR. For preventable cost, cost capture at 1% was 30% for sequential deep learning, compared to 18% for enhanced LR. The highest AUROCs for deep learning were 0.778, 0.681 and 0.727, respectively. These results offer a promising approach to identify patients for targeted interventions.

Development of a machine learning algorithm for early detection of opioid use disorder.

BACKGROUND: Opioid use disorder (OUD) affects an estimated 16 million people worldwide. The diagnosis of OUD is commonly delayed or missed altogether. We aimed to test the utility of machine learning in creating a prediction model and algorithm for early diagnosis of OUD. SUBJECTS AND METHODS: We analyzed data gathered in a commercial claim database from January 1, 2006, to December 31, 2018 of 10 million medical insurance claims from 550 000 patient records. We compiled 436 predictor candidates, divided to six feature groups - demographics, chronic conditions, diagnosis and procedures features, medication features, medical costs, and episode counts. We employed the Word2Vec algorithm and the Gradient Boosting trees algorithm for the analysis. RESULTS: The c-statistic for the model was 0.959, with a sensitivity of 0.85 and specificity of 0.882. Positive Predictive Value (PPV) was 0.362 and Negative Predictive Value (NPV) was 0.998. Significant differences between positive OUD- and negative OUD- controls were in the mean annual amount of opioid use days, number of overlaps in opioid prescriptions per year, mean annual opioid prescriptions, and annual benzodiazepine and muscle relaxant prescriptions. Notable differences were the count of intervertebral disc disorder-related complaints per year, post laminectomy syndrome diagnosed per year, and pain disorders diagnosis per year. Significant differences were also found in the episodes and costs categories. CONCLUSIONS: The new algorithm offers a mean 14.4 months reduction in time to diagnosis of OUD, at potential saving in further morbidity, medical cost, addictions and mortality.