Combining models improves ICU vital sign forecasts for each variable
Which Model, For Which Variable? Per-Variable Forecast Combination in Sparse ICU Time Series
Machine Learning
Summary
Forecasting medical variables in intensive care units is tricky because no single prediction model works best for all measurements. The authors show that taking the best model for each variable and combining predictions improves accuracy, especially for data that is sparse or less frequently recorded. They tested this approach on five different ICU datasets and found consistent improvement over any single model. Their method also adapts to different data sizes and types, helping to choose the right model for each variable.
What this means in practice
- •For hospital data teams: Improve accuracy of ICU patient vital sign and lab value forecasts by using per-variable model combinations tailored to data sparsity and quality.
- •For clinical decision support developers: Enhance ICU forecasting tools by integrating variable-specific model weighting to better predict patient states across diverse ICU databases.$Commercial implications: Enables creation of improved patient monitoring software with customized forecast accuracy needed by hospitals and healthcare providers.
Authors
Marius Fracarolli, Stefan Riezler
Abstract
Clinical forecasting papers typically report a single best architecture. We show that on multivariate ICU physiology this question conceals variable-specific differences: no architecture is best on every variable, and which model wins changes from variable to variable. Across five ICU databases (MIMIC-III, MIMIC-IV, eICU, HiRID, and UMM (University Medicine Mannheim), a German single-center ICU) we forecast 94 channels over a 24-hour horizon with a pool of 14 models, and combine them with weights fitted per variable. On test, the combination beats the best single model on all five databases by 6.1-11.0% mean squared error (mean 8.7%). The gain is concentrated where clinical data is thinnest: the sparsest third of channels improves by 11.9% and laboratory values by 7.5%, against 3.1% for the densest third. Rank-based harmonic weighting of the best five models per variable, with no fitted weights at all, recovers about four-fifths of the gain. Extending the pool afterwards with a per-channel gradient-boosted tree, ARIMA and ETS makes the tree the validation-selected best single model on all five databases; per-variable combination still improves on it by 2.9-5.9% (significant on four of five), while equal weighting of the same members does not. As on univariate benchmarks, learned combination beats simple averaging - but only once the weights are constrained. We also report which model to prefer when, and show that the ordering changes with database size.