Research Abstract: Free Paper - Oral Presentation Australian and New Zealand Society for Geriatric Medicine Annual Scientific Meeting 2023

Navigating the machine learning pipeline: a scoping review of inpatient delirium prediction models. (#39)

Tom R Strating 1 , Leila S Hanjani 1 , Ida Tornvall 1 , Ruth E Hubbard 1 , Ian Scott 2
  1. Centre for Health Services Research, University of Queensland, Milton, QLD, Australia
  2. Greater Brisbane Clinical School, Brisbane

Research Abstract

Title: Navigating the machine learning pipeline: a scoping review of inpatient delirium prediction models.

Aims

Early identification of inpatients at risk of developing delirium and implementing preventive measures could avoid up to 40% of delirium cases. Predictive machine learning (ML) applications promise to accelerate the diagnosis and treatment of delirium, but the extent of current research into such applications is unclear, hence the need for a scoping review.

 

Methods

We searched ten databases up to June 2022 for studies of delirium prediction models based on ML methods. Articles were excluded if there was insufficient information regarding methods used to develop, validate, and deploy the model or evaluate its performance. Included models were categorised by their stage of development and assessed for their performance and utility in clinical practice.

 

Results

Among 921 screened studies, 39 met the eligibility criteria, all published since 2017. In-silico predictive performance was consistently high; however, only six articles (15.4%) externally validated their findings on a different population. Three studies (7.7%) deployed their model in clinical workflows and demonstrated clinically useful predictive accuracy associated with high user acceptance.

 

Conclusions

This scoping review indicates rapidly growing research into ML delirium prediction models for hospital settings. Our findings show ML prediction models have the potential to identify inpatients at risk of developing delirium before symptom onset. However, few models were subject to external validation, which, when performed, revealed degraded performance. While few studies underwent prospective evaluation in clinical settings, performance and user acceptance seemed promising in those that did. Our findings have implications for data scientists and clinicians interested in implementing ML to predict and prevent delirium.