RNN-BOF: A Multivariate Global Recurrent Neural Network for Binary Outcome Forecasting of Inpatient Aggression (2312.01029v1)

Abstract: Psychometric assessment instruments aid clinicians by providing methods of assessing the future risk of adverse events such as aggression. Existing machine learning approaches have treated this as a classification problem, predicting the probability of an adverse event in a fixed future time period from the scores produced by both psychometric instruments and clinical and demographic covariates. We instead propose modelling a patient's future risk using a time series methodology that learns from longitudinal data and produces a probabilistic binary forecast that indicates the presence of the adverse event in the next time period. Based on the recent success of Deep Neural Nets for globally forecasting across many time series, we introduce a global multivariate Recurrent Neural Network for Binary Outcome Forecasting, that trains from and for a population of patient time series to produce individual probabilistic risk assessments. We use a moving window training scheme on a real world dataset of 83 patients, where the main binary time series represents the presence of aggressive events and covariate time series represent clinical or demographic features and psychometric measures. On this dataset our approach was capable of a significant performance increase against both benchmark psychometric instruments and previously used machine learning methodologies.

• The paper presents RNN-BOF, a global RNN model that uses LSTM cells to predict binary outcomes of inpatient aggression.
• The study applies a holistic training approach integrating continuous, categorical, and binary patient data for improved accuracy.
• The paper demonstrates that RNN-BOF outperforms traditional psychometric instruments, as evidenced by a higher AUC-PRG in forecasting aggression.

Recurrent Neural Networks Enhance Prediction of Inpatient Aggression

Inpatient aggression poses significant challenges in psychiatric settings, prompting the need for reliable methods to predict such events. Traditional tools, like psychometric assessment instruments, offer structured measures to assess future risks. However, they lack the ability to harness temporal relationships in recurrent patient assessments. A paper explored an alternative approach, presenting a novel model, named Recurrent Neural Network for Binary Outcome Forecasting (RNN-BOF), that leverages the power of artificial intelligence to predict binary outcomes like inpatient aggression.

The paper harnessed a recurrent neural network (RNN), a type of deep learning model adept at handling sequential data, to provide probabilistic forecasts of whether an adverse event, such as patient aggression, would occur in the following period. RNN-BOF stands out for its global training approach, which means it learns from the data of a whole population of patients simultaneously. This not only allows the model to gather richer insights by identifying patterns across different patients but also circumvents the need for every patient’s data to follow similar trends.

The researchers constructed the RNN-BOF architecture using long short-term memory (LSTM) cells. Unlike other time series forecasters, RNN-BOF is designed to predict the likelihood of the next time interval being an 'event' (aggressive incident) or 'non-event', providing a probability score associated with this outcome. Particularly innovative is the model’s ability to include various predictors in its consideration: it can process binary time series data indicating aggressive incidents, as well as continuous and categorical patient data like clinical features and demographic information.

The effectiveness of RNN-BOF was tested on a real-world dataset from Thomas Embling Hospital, which included daily records of aggression incidents alongside psychometric evaluations and other patient features. The model was benchmarked against conventional psychometric instruments and several state-of-the-art machine learning classifiers. The results were promising: RNN-BOF demonstrated a significant increase in performance over both the traditional instruments and other machine learning approaches.

What makes the performance comparison compelling is the use of the Precision Recall Gain (PRG) curve, which is regarded as better suited for datasets with imbalanced outcomes, such as the one used in the paper where actual aggressive incidents were rare compared to non-incidents. The increased area under the PRG curve (AUC-PRG) for RNN-BOF indicated its superior ability to distinguish between true instances of aggression and false alarms—a crucial factor in clinical settings where false positives could drain resources and cause unnecessary distress.

Given these findings, the potential applications for RNN-BOF and similar AI-driven models are broad, extending beyond the psychiatric inpatient setting. For example, this forecasting approach could aid in anticipating various binary outcomes in different medical fields or in predicting events within law enforcement contexts.

In summary, the paper positions RNN-based models as promising instruments in binary outcome forecasting tasks, offering more nuanced and dynamic risk assessments than many current methods. As data collection in clinical environments becomes increasingly rich and complex, the application of such advanced AI models may well be a significant step forward in proactive patient care and resource management.