Reinventing Clinical Decision Support: Data Analytics, Artificial Intelligence, and Diagnostic Reasoning - Book Review

With a stock pile of excellent books on health information and technology on my desk and electronically, I plan to review one book per week.

First up is this book on Clinical Decision Support with a strong focus on AI written by John Halamka now at Mayo Clinic and Paul Cerrato. The authors bring together the current thinking and research about clinical decision support. This includes issues in diagnostic reasoning, such as, how diagnostic errors occur; the trends in data analytics including predictive analytics; and the pros and cons of artificial intelligence and machine learning as they will influence clinical decision support. Each chapter contains a rich literature review with clear conclusions on the current state of these technologies and techniques.

Clinical decision support has been traditionally based on medical training from experts and clinical practice guidelines. However, diagnostic errors, even though rare, are too common. Three new trends identified in this work should be highlighted:

• Artificial intelligence and machine learning - This technology has the potential for a major impact on decision support because of the availability of large data sets and the ability to analyze radiology and pathology images using faster computers and new algorithms. At the same time, the authors are clear about the fact that this technology is relatively new in healthcare and their are significant limitations and ethical issues.
• Predictive analytics - In a chapter on "Reengineering Data Analytics" the authors focus on Multiple Sclerosis disease prediction. The key to predictive models is determining which factors are included and the ability to stratify patients into high and low risk.
• Precision Medicine - the authors examined the role of genetic predisposition and the challenges of importing this data into the EMR and displaying it in a way that it can be easily interpreted. However, the suggest that the current potential for precision medicine is in pharmacogenomics - known sensitivity to medicines based on a genomic profile. This kind of data becomes actionable at the point of care as a physician can modify a prescription based on the data.

One of the concluding sections of the book which summarizes the state of CDS is: "Promising solutions, unrealistic expectations." The authors point out that the metrics used to measure the problem of diagnostic errors all have shortcomings and that cognitive errors from Type 1 and Type 2 reasoning need to be addressed. They also caution the use of AI/ML and point to the need to apply critical thinking to evaluate these emerging tools. While many are critical of these tools and reluctant to use them, the authors argue that these are not dissimilar from risk calculators commonly used in clinical practice. Data analytics using ML drive us to learn new tools, such as, random forest modeling, proportional hazards regression and logistic regression. Including these advanced data science tools need to be integrated into medical education for future care providers. Finally, they identify the need for systems biology and precision medicine to be integrated into CDS systems.

In conclusion, the authors state:

AI will never replace a competent physician. That said, there's little doubt that a competent physician who uses all the tools that AI has to offer will soon replace the competent physician who ignores these tools.