Confidence-based laboratory test reduction recommendation algorithm

Abstract

AbstractUnnecessary laboratory tests present health risks and increase healthcare costs. We propose a new deep learning model to identify unnecessary hemoglobin (Hgb) tests for patients admitted to the hospital. Machine learning models might generate less reliable results due to noisy inputs containing low-quality information. We estimate prediction confidence to measure reliability of predicted results. Using a “select and predict” design philosophy, we aim to maximize prediction performance by selectively considering samples with high prediction confidence for recommendations. We use a conservative definition of unnecessary laboratory tests, which we define as stable and below the lower normal bound (LBNR). Our model accommodates irregularly sampled observational data to make full use of variable correlations (i.e., with other laboratory test values) and temporal dependencies (i.e., previous observations) in order to select candidates for training and prediction. Using data collected from a teaching hospital in Houston, our model achieves Hgb prediction performance with a normality AUC at 95.89% and a Hgb stability AUC at 95.94%, while recommending a reduction of 9.91% of Hgb tests that were deemed unnecessary.