Improving length of stay prediction using a hidden Markov model

Area of Science:

• Medical Informatics
• Machine Learning
• Critical Care Medicine

Background:

• Accurate estimation of intensive care unit (ICU) length of stay (LOS) is essential for managing healthcare costs and optimizing resource allocation.
• Physician predictions of ICU LOS are often inaccurate, highlighting the need for improved forecasting methods.
• The increasing availability of electronic health records (EHRs) provides opportunities for data-driven approaches like machine learning to enhance LOS prediction.

Purpose of the Study:

• To develop and evaluate a novel machine learning framework for predicting ICU patient length of stay.
• To leverage physiological measurements from the initial 48 hours of ICU admission for accurate LOS prediction.
• To demonstrate the efficacy of the proposed model in capturing temporal patient data representations.

Main Methods:

• A hidden Markov model (HMM)-based framework was developed to predict ICU length of stay.
• The model utilizes physiological measurements collected within the first 48 hours of patient admission.
• The framework was validated using real-world intensive care unit patient data.

Main Results:

• The hidden Markov model framework effectively captures temporal patient representations.
• The proposed model demonstrated consistent performance in outperforming most existing baseline methods for ICU LOS prediction.
• The framework shows significant potential for improving the accuracy of length of stay estimations.

Conclusions:

• The developed hidden Markov model framework offers a promising approach for accurate ICU length of stay prediction.
• Utilizing early physiological data with advanced machine learning can significantly improve healthcare resource management and patient outcome prediction.
• This data-driven method has the potential to enhance clinical decision-making and reduce healthcare expenditures.