Predictive machine-learning model for screening iron deficiency without anaemia: a retrospective cohort study

Area of Science:

• Biomedical informatics
• Clinical data science
• Predictive modeling in healthcare

Background:

• Iron deficiency without anaemia (IDWA) is a common condition that can precede iron deficiency anaemia.
• Early identification of IDWA is crucial for timely intervention and prevention of complications.
• Utilizing routinely collected electronic health record (EHR) data offers a scalable approach for population-level screening.

Purpose of the Study:

• To develop and validate a machine-learning (ML) model for predicting IDWA using EHR data.
• To assess the accuracy of the ML model in identifying low ferritin levels (<30 ng/mL) in non-anaemic patients.
• To compare the ML model's performance against traditional methods and evaluate its utility in clinical settings.

Main Methods:

• A retrospective cohort study was conducted using EHR data from 211,486 adult patients.
• An extreme gradient-boosted decision tree ML algorithm was employed to predict low ferritin levels.
• The model utilized demographic data, complete blood count indices, and chemistry results.

Main Results:

• The ML model achieved an area under the curve (AUC) of 0.814, outperforming a model based solely on complete blood count indices (AUC 0.741).
• The model demonstrated a sensitivity of 70% with a specificity of 75.85%, a positive predictive value of 37.6%, and a negative predictive value of 92.41%.
• Performance varied across subgroups, with higher accuracy observed in men and postmenopausal women compared to premenopausal women.

Conclusions:

• The developed ML model effectively screens for IDWA using readily available EHR data.
• Implementation of this ML tool in clinical practice can facilitate the early diagnosis of IDWA.
• This approach holds promise for improving the management of iron deficiency in non-anaemic individuals.