Predictive analytics in health care: how can we know it works?

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Biostatistics

Background:

• Growing recognition of the need for transparency in research methodology and findings.
• Increasing use of artificial intelligence (AI) for developing predictive algorithms in healthcare.
• Ethical considerations surrounding proprietary algorithms and their impact on patient care.

Purpose of the Study:

• To advocate for the public availability of algorithms used in predictive modeling.
• To highlight the importance of transparency for validation, performance assessment, and algorithm refinement.
• To address the ethical implications of withholding algorithms for commercial purposes.

Main Methods:

• Argumentative approach based on principles of scientific transparency and ethical research conduct.
• Discussion of the necessity for open-source software for 'black box' machine learning algorithms.
• Emphasis on the role of journals, funders, and clinical guidelines in promoting algorithm transparency.

Main Results:

• Public availability of algorithms enables independent external validation.
• Transparency facilitates assessment of performance heterogeneity across different settings and over time.
• Open access to algorithms is essential for refinement, updating, and ensuring accountability.

Conclusions:

• It is paramount to make AI-driven predictive algorithms publicly available for ethical and scientific reasons.
• Journals and funders should mandate transparency for publications involving predictive algorithms.
• Clinical guidelines should prioritize the recommendation of publicly accessible algorithms to ensure trust and efficacy.