Pneumonia is a leading cause of mortality in limited resource settings (LRS), which are common in low- and middle-income countries (LMICs). Accurate referrals can reduce the devastating impact of pneumonia, especially in LRS. Discriminating pneumonia from other respiratory conditions based only on symptoms is a major challenge. Machine learning has shown promise in overcoming the diagnostic difficulties of pneumonia (i.e., low specificity of symptoms, lack of accessible diagnostic tests and varied clinical presentation). Many scientific papers are now focusing on deep-learning methods applied to clinical images, which is unaffordable for initial patient referral in LMICs. The current study used a dataset of 4500 patients (1500 patients affected by bronchitis, 3000 by pneumonia) from a middle-income country, containing information on subject population characteristics, symptoms and laboratory test results. Manual feature selection was performed, focusing on clinical symptoms that are easily measurable in LRS and in community settings. Three common machine learning methods were tested and compared: logistic regression; decision tree and support vector machine. Models were developed through a holdout process of training-validation and testing. We focused on six clinically relevant, easily interpreted patient symptoms as best indicators for pneumonia. Our final model was a decision tree, achieving an AUC of 93%, with the advantage of being fully intelligible and easily interpreted. The performance achieved suggested that intelligible machine learning models can enhance symptom-based referral of pneumonia in LRS and in community settings.

A machine learning model for supporting symptom-based referral and diagnosis of bronchitis and pneumonia in limited resource settings

Pecchia L.
2021-01-01

Abstract

Pneumonia is a leading cause of mortality in limited resource settings (LRS), which are common in low- and middle-income countries (LMICs). Accurate referrals can reduce the devastating impact of pneumonia, especially in LRS. Discriminating pneumonia from other respiratory conditions based only on symptoms is a major challenge. Machine learning has shown promise in overcoming the diagnostic difficulties of pneumonia (i.e., low specificity of symptoms, lack of accessible diagnostic tests and varied clinical presentation). Many scientific papers are now focusing on deep-learning methods applied to clinical images, which is unaffordable for initial patient referral in LMICs. The current study used a dataset of 4500 patients (1500 patients affected by bronchitis, 3000 by pneumonia) from a middle-income country, containing information on subject population characteristics, symptoms and laboratory test results. Manual feature selection was performed, focusing on clinical symptoms that are easily measurable in LRS and in community settings. Three common machine learning methods were tested and compared: logistic regression; decision tree and support vector machine. Models were developed through a holdout process of training-validation and testing. We focused on six clinically relevant, easily interpreted patient symptoms as best indicators for pneumonia. Our final model was a decision tree, achieving an AUC of 93%, with the advantage of being fully intelligible and easily interpreted. The performance achieved suggested that intelligible machine learning models can enhance symptom-based referral of pneumonia in LRS and in community settings.
2021
Diagnostics
Healthcare
LMICs
Machine learning
Pneumonia
Screening tool
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12610/65976
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 56
  • ???jsp.display-item.citation.isi??? ND
social impact