Conventional and hybrid time series models for forecasting medication dispensing and errors integration in automated dispensing cabinets

Résumé fourni par la source

Automated dispensing cabinets (ADCs) represent a critical innovation in modern healthcare, revolutionizing medication management by improving efficiency, accuracy, and security. With the increasing reliance on these technologies, optimizing their performance is paramount. This study aims to apply conventional, hybrid time series, and machine learning models to forecast three key performance indicators of ADCs: items dispensation, override occurrences, and error integration. Using monthly data from the MICU at Almoosa Hospital between January 2023 and December 2024, we employed both traditional linear time series models (e.g., autoregressive models, simple exponential smoothing, autoregressive moving average, and theta models) and advanced non-linear machine learning models (e.g., NPAR, Artificial Neural Networks (ANN)) in various hybrid configurations (ARIMA-ANN, EMS-ANN, NPAR-ANN). Model accuracy was assessed using key metrics such as RMSE, MAE, MAPE, and RMSLE, with a bootstrap 95% CI to ensure the best performance for predicting future trends. The study demonstrated that the NPAR-ANN, a hybrid model combining nonparametric ARIMA and artificial neural networks, showed superior performance. The model demonstrated the best performance, achieving the lowest RMSE values. Specifically, for the number of items issued, it attained an RMSE of 71.50, for overrides an RMSE of 15.43, while for error integration, and an RMSE of 20.92 lowest among all competing models. This novel study modeled key parameters of ADCs, providing data-driven insights that can inform hospital decision-making and optimize medication management. The study showcased the application of hybrid machine learning models in forecasting critical ADC parameters, offering valuable data-driven insights for hospital administrators.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.