Identifying Existing Evidence to Potentially Develop a Machine Learning Diagnostic Algorithm for Cough in Primary Care Settings: Scoping Review

• DOI: 10.2196/46929
• Publication Date: 2023-12-14T15:21:45Z
• Authors: Julia Cummerow; Christin Wienecke; Nicola Engler; Philip Marahrens; Philipp Gruening; Jost Steinhäuser
• Keywords: PsycINFO; Odds

Background Primary care is known to be one of the most complex health settings because high number theoretically possible diagnoses. Therefore, process clinical decision-making in primary includes analytical and nonanalytical factors such as gut feelings dealing with uncertainties. Artificial intelligence also mandated offer support finding valid Nevertheless, translate some aspects what occurs during a consultation into machine-based diagnostic algorithm, probabilities for underlying diagnoses (odds ratios) need determined. Objective Cough common reasons general practice, core discipline care. The aim this scoping review was identify available data on cough predictor various encountered practice. In context an ongoing project, we reflect database basis algorithm. Furthermore, discuss applicability algorithm against background specifics Methods PubMed, Scopus, Web Science, Cochrane Library databases were searched defined search terms, supplemented by gray literature via German Journal Family Medicine until April 20, 2023. inclusion criterion explicit analysis any conceivable disease. Exclusion criteria articles that did not provide original study results, languages other than English or German, mention predictor. Results total, 1458 records identified screening, which 35 met our criteria. Most results (11/35, 31%) found chronic obstructive pulmonary others distributed among asthma unspecified airway disease, infectious diseases, bronchogenic carcinoma, dyspepsia gastroesophageal reflux adverse effects angiotensin-converting enzyme inhibitors. Positive odds ratios influenza, COVID-19 infections, bronchial whereas nonspecified diseases inconsistent. Conclusions Reliable practice are scarce. example does sufficient contribute machine learning–based meaningful way.