Predictive Modelling for Length of Stay with the MIMIC-III Critical Care Database

As the world's population grows and ages, the importance of well-organized and efficient healthcare facilities that cater to a wider community has become increasingly important. As a result, the optimal use of resources in these facilities is critical. Digitization of most healthcare facilities has led to more and more data being electronically collected, opening up opportunities to use data science techniques for optimization. Efficient and effective discharge decision making plays a vital role in this regard. In this study, the MIMIC-III Critical Care Database was used, to analyze and understand factors that affect the length of stay leading to the development of a novel composite super learner model to predict the same at the point of admission. This would be a vital stepping stone towards improving discharge planning at hospitals. The model developed in this paper is able to predict the length of stay with significant improved performance (1.99 days RMSE) compared to previous work (9.5 days RMSE) on the problem. This paper provides evidence that performance can be improved through conceptual modelling with a richer and wider feature set, without having to restrict the focus on to a specific disease or diagnosis, or having to compensate on informational value by opting for classification.