Pattern-based Fall Prediction using Hospital Clinical Notes
Yashodhya V. Wijesinghe, Yue Xu, Yuefeng Li, Qing Zhang · 2020
Falls are considered as the second most leading cause of lethal and non-lethal injuries among the aging population. The fall prediction enables to reduce the health care cost and the negative impacts related to fall such as loss of independence.The objective of this paper is to propose an approach to select discriminative patterns from clinical notes. The selected discriminative patterns are used as features to predict falls among elderly people based on their clinical data. A collection of 12,911 labeled clinical notes of patients aged ≥ 65 was selected from a medical database, named Medical Information Mart for Intensive Care (MIMICIII), to create the corpus. The classification algorithms, Logistic Regression, Support Vector Machine and Random Forest are used to evaluate the effectiveness of the selected discriminative patterns for predicting falls. The Support Vector Machine classification algorithm gave the best results. The prediction accuracy of using the proposed discriminative patterns is higher than the baseline approaches when predicting `Fall” using the clinical notes. The results of this research suggest that the proposed discriminative pattern mining approach was able to generate a set of interesting patterns that were able to distinguish between fall and not fall clinical notes.