Prediction models using artificial intelligence and longitudinal data from electronic health records: a systematic...

Area of Science:

• Medical Informatics
• Machine Learning
• Health Outcomes Research

Background:

• Longitudinal data from electronic health records (EHRs) offer valuable insights for predicting health outcomes.
• Artificial intelligence (AI) techniques are increasingly applied to analyze complex EHR data.

Approach:

• Systematic review of studies utilizing AI with longitudinal EHR data for health outcome prediction.
• Searched multiple databases (MEDLINE, Scopus, Web of Science, IEEE Xplorer) up to January 2022.
• Assessed study methodology, AI techniques, prediction tasks, validation, and reporting quality.

Key Points:

• Eighty-one studies were included, predominantly predicting disease development or events using diagnoses and drug treatments.
• Recurrent Neural Networks and transformer architectures were common, with increasing use of combined layers and attention mechanisms.
• Most studies had poor reporting quality, a high risk of bias, and inadequate performance assessment due to single train-test partitions.

Conclusions:

• AI models show promise in leveraging longitudinal EHR data for health predictions.
• Heterogeneity in reporting, lack of public EHR datasets, and limited code sharing impede reproducibility.
• Standardized reporting and data sharing are crucial for advancing AI in healthcare.