Machine learning-based spirometry reference values for the Iranian population: a cross-sectional study from the...

Area of Science:

• Pulmonary Medicine
• Biostatistics
• Machine Learning

Background:

• Spirometry is crucial for assessing lung function.
• Existing spirometry norm values may not accurately represent diverse populations like Iranians.
• Accurate reference values are needed for reliable diagnosis and management of respiratory diseases.

Purpose of the Study:

• To determine spirometric norm values for healthy Iranian adults.
• To compare the accuracy of KNN regression with MLR and LMS models for predicting spirometric parameters.
• To evaluate the performance of the derived values against established GLI-Caucasian and Iranian equations.

Main Methods:

• Spirometric data from 998 healthy Iranian adults (PERSIAN study) were analyzed.
• K-Nearest Neighbors (KNN) regression was employed to derive reference values for FEV1, FVC, FEV1/FVC, and FEF25-75%.
• Model performance was validated using 5-fold cross-validation and Mean Squared Error (MSE), comparing KNN regression against Multiple Linear Regression (MLR) and Lambda-Mu-Sigma (LMS) models.

Main Results:

• KNN regression demonstrated superior accuracy in predicting spirometric parameters compared to MLR and LMS models.
• The derived spirometry values showed lower MSE than the GLI-Caucasian and existing Iranian equations, indicating better prediction for the Iranian population.
• Specifically, the MSE for predicted FVC in females was 0.159 using KNN regression, significantly lower than Iranian (0.344) and GLI-Caucasian (0.397) equations.

Conclusions:

• Machine learning, specifically KNN regression, provides a flexible and accurate method for establishing population-specific spirometry reference values.
• The developed spirometry reference values are tailored for the Iranian adult population.
• An Excel calculator based on these findings can aid healthcare professionals in precise lung function assessment for Iranian adults.