Modelling EHR timeseries by restricting feature interaction
This work addresses the challenge of handling missing and noisy data in EHR time-series for healthcare professionals, but it is incremental as it builds on existing recurrent models with a specific modification.
The paper tackled the problem of predicting clinical outcomes like mortality and acute kidney injury from noisy and missing time-series data in electronic health records by proposing a recurrent neural network model that restricts feature interactions to reduce overfitting, resulting in improvements of 1.1% to 2.2% in AU-ROC compared to existing state-of-the-art models.
Time series data are prevalent in electronic health records, mostly in the form of physiological parameters such as vital signs and lab tests. The patterns of these values may be significant indicators of patients' clinical states and there might be patterns that are unknown to clinicians but are highly predictive of some outcomes. Many of these values are also missing which makes it difficult to apply existing methods like decision trees. We propose a recurrent neural network model that reduces overfitting to noisy observations by limiting interactions between features. We analyze its performance on mortality, ICD-9 and AKI prediction from observational values on the Medical Information Mart for Intensive Care III (MIMIC-III) dataset. Our models result in an improvement of 1.1% [p<0.01] in AU-ROC for mortality prediction under the MetaVision subset and 1.0% and 2.2% [p<0.01] respectively for mortality and AKI under the full MIMIC-III dataset compared to existing state-of-the-art interpolation, embedding and decay-based recurrent models.