Using a neural network to screen a population for asthma.

Area of Science:

• Computational biology
• Respiratory medicine
• Epidemiology

Background:

• Limited resources hinder comprehensive asthma diagnosis in large populations.
• Prioritization is crucial for efficient screening and timely clinical review.
• Asthma diagnosis requires full clinical assessment, often not feasible for mass screening.

Purpose of the Study:

• To develop and apply a neural network for ranking individuals based on their likelihood of asthma.
• To utilize respiratory questionnaire responses for predicting asthma probability.
• To enable resource-efficient prioritization for asthma screening.

Main Methods:

• A neural network was trained using questionnaire responses and expert-assigned asthma probability labels.
• A stratified random sample of 6825 community survey respondents was used for training and validation.
• The trained network ranked the entire population to identify individuals needing further assessment.

Main Results:

• Setting the screening threshold for the top 10% (683 individuals) identified 239 patients without a prior diagnosis needing assessment.
• Among these 239 patients, 74% were predicted to have a confirmed asthma diagnosis.
• The model provided a ranked order of asthma likelihood across the entire population.

Conclusions:

• This neural network approach facilitates population prioritization for asthma screening.
• It optimizes resource allocation by focusing clinical assessments on high-likelihood individuals.
• The method supports efficient identification of undiagnosed asthma in resource-constrained settings.