Paul Cerrato and I have created a new book, Reinventing Clinical Decision Support
, our first to be published about Platform thinking. Although it is being published during my tenure at Mayo Clinic, it is not endorsed by Mayo Clinic and represents the personal opinions of Paul and me. Below is the preface.
In our last book, on mobile health(1), we wrote about the power of words such as cynicism, optimism, and transformation. Another word with powerful connotations is misdiagnosis. To a patient whose condition remains undetected, it is a source of frustration and anger. To a physician or nurse who has become a defendant in a malpractice lawsuit, it can likewise generate frustration and anger as they try to demonstrate that they did everything humanly possible to uncover the source of their patient’s symptoms.
The National Academy of Medicine’s report Improving Diagnosis in Health Care explains: “It is estimated that 5 percent of U.S. adults who seek outpatient care each year experience a diagnostic error. Postmortem examination research spanning decades has shown that diagnostic errors contribute to approximately 10 percent of patient deaths, and medical record reviews suggest that they account for 6 to 17 percent of adverse events in hospitals.”(2) An earlier report from the same group, To Err Is Human, came to a similar disturbing conclusion. The message between the lines of both reports is straightforward: Medical errors, including misdiagnoses, are often the consequences of being human. That same reality also comes across in a recent New England Journal of Medicine editorial: “The complexity of medicine now exceeds the capacity of the human mind.”(3)
Such complexity fosters humility—or at least it should. It requires humility for clinicians with years of experience successfully diagnosing patients’ ills to admit that they may be missing as many disorders as they catch. And the way the healthcare system is currently designed, that is a distinct possibility. When a patient is misdiagnosed by Dr. Jones, he often never goes back to him to say: You made a mistake, please try again. He is just as likely to move on to Dr. Smith in the hope that her diagnostic skills are more finely tuned. Humility is also required of clinicians to admit that the quantity of new research coming out in each specialty each year is so massive that it is virtually impossible for any one person to stay abreast of it. By one estimate, a new medical journal article is published once every 26 seconds, which translates to about 5,000 articles per day.(4)
Many diagnostic aids are now available to help address the epidemic of diagnostic errors we now face. Clinical decision support (CDS) systems, for example,are designed to help practitioners stay up to date on new developments without requiring them to spend their entire day reading the medical literature. Some CDS systems also offer symptom finders, decision trees, and other advanced features. But today’s digital tools only scratch the surface. Incorporating newly developed algorithms that take advantage of machine learning, neural networks, and a variety of other types of artificial intelligence (AI) can help address many of the shortcomings of human intelligence.
Fatima Paruk, MD, MPH, the chief medical officer at Allscripts, said it best: “[W]ith machine learning, clinical decision support can do so much more. We can transform systems laden with meaningless alerts to intelligent workflows and best practices driven by relevant patient history . . . Machine learning can enable clinical decision support based on multi-system analysis to understand which patients are at highest risk of a negative outcome, or to optimize treatment in real-time . . . Algorithms can parse available historical and current information to inform clinicians which patients are at risk for specific outcomes or deliver personalized treatment plans for patients with chronic conditions.”(5)
When will this next generation of CDS tools be available for clinicians in the trenches? When will we reinvent CDS? As the 8 chapters of this book point out, these tools are already emerging. Ignoring their value puts both clinicians and patients at risk.
This book begins with an examination of the diagnostic reasoning process itself, which includes how diagnostic errors are measured, what technological and cognitive errors are to blame, and what solutions are most likely to improve the process. It explores Type 1 and Type 2 reasoning methods, cognitive mistakes such as availability bias, affective bias, and anchoring, and potential solutions such as the Human Diagnosis Project.
AI and machine learning are explored at length, with plain clinical English explanations of convolutional neural networks, back propagation, and digital image analysis. Real-world examples of how these tools are being employed are also discussed, including the landmark Google study that demonstrated the value of deep learning in diagnosing diabetic retinopathy. Machine learning–enabled neural networks are also helping to detect melanoma, breast cancer, cancer metastasis, and colorectal cancer, and to manage severe sepsis. AI is even helping to address the opioid epidemic by reducing the number of pills being prescribed postoperatively. Each of these topics includes detailed references to the peer-reviewed medical literature.
With all the enthusiasm in the healthcare community about the role of AI and machine learning, it was also necessary to outline some of the criticisms, obstacles, and limitations of these new tools. Among the criticisms discussed is the relative lack of hard scientific evidence supporting some of the latest algorithms and the “explainability” dilemma. Most machine learning systems are based on advanced statistics and mind-bending mathematical equations, which have made many clinicians skeptical about their worth. We address the so called black box problem, along with potential solutions, including educational tutorials that open up the black box.
This book devotes an entire chapter to commercial CDS systems, comparing legacy products to the latest software platforms. The evidence to show that these are having an impact on patient outcomes is mixed—an issue explored in depth in this book. On a more positive note, this chapter explores many of the innovative developments being launched by vendors such as DynaMed (EBSCO), VisualDX, UpToDate Advanced, and Isabel Healthcare.
The chapter on data analytics does a deep dive into new ways to conduct subgroup analysis and how it is forcing healthcare executives to rethink the way they apply the results of large clinical trials to everyday medical practice. This re-evaluation is slowly affecting the way diabetes, heart disease, hypertension, and cancer are treated. The research discussed also suggests that data analytics will impact emergency medicine, medication management, and healthcare costs.
Any attempt to reinvent CDS also needs to tackle the outdated paradigm that still serves as the underpinning for most patient care. This reductionistic mindset insists that most diseases have a single cause. The latest developments in systems biology indicate otherwise and point to an ensemble of interacting contributing causes for most degenerative disorders. The new paradigm, which is being assisted by advances in AI, has spawned a new specialty called network medicine, which is poised to transform patient care at its roots.
Similarly, the current medical model relies too heavily on a population-based approach to medicine. This one-size-fits-all model is being replaced by a precision medicine approach that takes into account a long list of risk factors. And once again, this new paradigm is being supported by new technologies that help clinicians combine a patient’s genomic data, including pharmacogenomic test results, with the more traditional markers available in their electronic health record (EHR).
All these new developments would be useless if they could not be implemented in the real world. The final chapter outlines many of the use cases that have been put in place at Beth Israel Deaconess Medical Center (Boston) and elsewhere. These new programs are helping to improve the scheduling of 41 operating rooms, streamline the processing of patient consent forms before surgery, and much more.
Despite all these positive developments, it is important to emphasize that AI and machine learning will not solve all of healthcare’s problems. That will require an artful blend of artificial and human intelligence, as well as a healthy dose of emotional intelligence.
Finally, our enthusiastic take on digital innovation should not give readers the impression that AI will ever replace a competent physician. That said, there is 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.
Paul Cerrato, MA
John Halamka, MD, MS
1. Cerrato, P. and Halamka, J. (2019). Th e Transformative Power of Mobile Medicine.
Cambridge (MA): Academic Press/Elsevier.
2. Balogh, E., Miller, B. T., and Ball, J. R. (Eds.). (2015). Improving Diagnosis in
Health Care. Institute of Medicine, National Academies Press.
3. Obermeyer, Z. and Lee, T. H. (2017). Lost in Th ought: Th e Limits of the Human
Mind and the Future of Medicine. New England Journal of Medicine, vol. 377,
4. Garba, S., Ahmed, A., Mai, A., Makama, G., and Odigie, V. (2010). Proliferations of Scientifi c Medical Journals: A Burden or a Blessing. Oman Medical Journal, vol. 25, pp. 311–314.
5. Paruk, F. (2018, December 4). HIT Th ink 4 Keys to Success with AI and Machine
Learning. HealthData Management. Accessed on December 18, 2018, from https://