Potential applications and performance of machine learning techniques and algorithms in clinical practice: A...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Machine learning (ML) algorithms offer potential solutions for diverse clinical challenges, including diagnosis, prognosis, and treatment recommendations.
• The integration of ML into clinical practice is rapidly advancing, necessitating a comprehensive review of its current performance and implementation status.

Purpose of the Study:

• To systematically review and evaluate the performance of machine learning algorithms in clinical practice.
• To assess the progress and identify key ML algorithms and application areas in healthcare.

Main Methods:

• A systematic literature search was conducted across major databases (PubMed, MEDLINE, Scopus, Google Scholar, Cochrane Library, WHO Covid-19) for articles published between January 2011 and October 2021.
• Studies involving human subjects, ML techniques in clinical practice, and reporting performance metrics were included.

Main Results:

• Out of 873 unique articles, 36 were included. XGBoost (extreme gradient boosting) demonstrated high potential (7 studies), followed by random forest, logistic regression, and support vector machines (5 studies each).
• Outcome prediction was the most common application (33 studies), particularly for inflammatory diseases (7 studies), cancer, and neuropsychiatric disorders (5 studies each).
• While most studies met quality criteria (TRIPOD checklist), none achieved a 'low' bias rating (PROBAST checklist), and only three demonstrated actual clinical deployment.

Conclusions:

• Machine learning (ML) shows significant potential as a clinical decision support tool.
• Further research is required to validate ML algorithms for clinical use, focusing on enhancing quality, transparency, and interpretability to facilitate broader acceptance.