Identifying Existing Evidence to Potentially Develop a Machine Learning Diagnostic Algorithm for Cough in Primary...

Primary care consultations often involve complex decision-making due to a wide range of possible diagnoses. Clinical reasoning in this setting combines analytical and intuitive factors. Artificial intelligence has potential to assist in diagnostic processes. However, translating clinical consultations into machine-based algorithms requires data on symptom-diagnosis relationships. Cough is a frequent reason for primary care visits. Prior research has shown that cough is a common symptom across multiple conditions. No prior work had resolved how cough functions as a diagnostic predictor across general practice. This gap motivated a review of existing literature. That uncertainty drove the need to assess if sufficient data exist to support machine learning diagnostic tools. The lack of comprehensive data on cough as a predictor remains a challenge. This review aimed to clarify what is known about cough's role in diagnosis.

The goal was to evaluate the availability of data on cough as a diagnostic indicator in primary care. The specific problem was the lack of a clear database linking cough to various diagnoses. This review aimed to determine if existing literature could support the development of a machine learning diagnostic tool. The motivation stemmed from the need to improve diagnostic accuracy in primary care. The study focused on cough due to its high prevalence in consultations. The researchers sought to identify how cough functions as a diagnostic predictor. They examined if current evidence is sufficient for algorithm development. This work contributes to understanding the feasibility of machine learning in primary care diagnostics.

The researchers conducted a scoping review using multiple databases. They searched PubMed, Scopus, Web of Science, and the Cochrane Library. Gray literature was also included via the German Journal of Family Medicine. Search terms were defined to capture cough-related diagnostic studies. Inclusion criteria required explicit analysis of cough as a predictor. Excluded were non-English/German articles and those without original data. A total of 1458 records were identified for screening. After applying inclusion criteria, 35 articles were selected for analysis. The study focused on the frequency and strength of cough as a diagnostic indicator.

The most frequently studied diagnosis was chronic obstructive pulmonary disease. Eleven of 35 articles focused on this condition. Other diagnoses included asthma, infectious diseases, and bronchogenic carcinoma. Positive odds ratios were found for COPD, influenza, and bronchial carcinoma. Cough as a predictor for asthma showed inconsistent results. The data on cough as a diagnostic indicator remain limited. Only 31% of relevant studies focused on COPD. No strong evidence was found for cough predicting dyspepsia or GERD. The overall database lacked sufficient breadth and consistency.

The authors suggest that current evidence on cough as a diagnostic predictor is insufficient. They propose that the available data do not support meaningful machine learning algorithms. The researchers note that diagnostic specificity in general practice remains challenging. They emphasize the need for more comprehensive data on symptom-diagnosis relationships. The study highlights that cough's diagnostic value varies across conditions. The authors suggest that future work should focus on expanding the evidence base. They caution against overestimating the current utility of cough data in algorithm development. The findings suggest that further research is needed before implementing machine learning tools.