AI Framework to Forecast Medication Usage with ICD-9/10 Code

Accurate forecasting of medication usage and ICD-9/10 code streams is critical for optimizing medical logistics, especially during periods of high demand, such as pandemics, disease outbreaks, wartime, or natural disasters. In this study, we develop a novel forecasting framework using unsupervised learning techniques and Natural Language Processing (NLP) methods to build vector representations of ICD-9/10 code stream and medication usage from Electronic Health Record (EHR) data. Multiple time-series forecasting models, including Vector Autoregression (VAR), Temporal Fusion Transformer (TFT), and Autoregressive Long Short-Term Memory (AR-LSTM) are trained, tested and evaluated. Finally multiple TFT and AR-LSTM models with different lookback horizons are trained and ensembled together to achieve better forecasting accuracy in near further. The AI framework is trained and tested with MIMIC-IV ER and MIMIC-III datasets, resulting in the average error as low as 5% for 5-th day forecasting and 18% for the 10-th day forecasting. The results demonstrate the ensemble model’s superior performance on near-future medication usage forecast as well as ICD code progression, offering valuable insights for healthcare logistics and decision-making. This adaptive forecasting approach provides a robust tool for managing healthcare logistics under extreme and fluctuating conditions, enhancing the ability to deliver timely care, ultimately saving lives and resources.