A healthcare system’s moral duty is to learn from its patients to improve future care. Models developed from patient data are now used to inform care and reduce medical costs by preventing unnecessary procedures.
Our previous discussions of the Severe COVID-19 Adaptive Risk Predictor (SCARP) model with Dr. Brian Garibaldi demonstrated the power of patient data to inform medical decisions through dynamic models. The outcomes are then fed back into the model to continue improving its accuracy. SCARP was developed within the Johns Hopkins inHealth Precision Medicine initiative. InHealth was established over a decade ago to improve health outcomes and reduce wasteful healthcare expenditures. It involves all areas of Johns Hopkins in improving patient care. Dr. Scott Zeger, a member of the initial group that designed inHealth, answers questions about the method. The former vice provost for research currently holds faculty positions across the schools of Medicine, Arts & Sciences, and Public Health, where he is primarily appointed in biostatistics.
We wanted to use Johns Hopkins Medicine as a laboratory to discover how to transform Johns Hopkins into a learning healthcare system. The idea is that each of the clinics at Johns Hopkins Medicine would operate in a virtuous learning cycle, where the healthcare they provided to the current patients would generate data, and that data would be analyzed to generate new knowledge, which would in turn improve the care for the next patients. Our approach was to identify mechanistically-anchored and clinically-relevant subgroups of patients to which we can tailor health interventions.
From the very beginning inHealth was conceived as a partnership of all the divisions of the university, the health system, and the Applied Physics Laboratory. All have contributed substantially to inHealth’s success. Health is more than the absence of disease. Essential expertise about people's well-being exists across the entire university from arts to social sciences to engineering. Since we want to not just promote health, but prevent specific diseases, the requisite expertise is the domain of many different disciplines in public health, nursing, and medicine.
inHealth also has methodologic needs. If you're going to learn from data in real time, you need people who are experts in learning, so that brings in the psychological and brain sciences. If you're going to acquire large amounts of data, manage it, learn from it, and then disseminate what you've learned, you need the data expertise we have across this university from applied math and statistics to computer science to biostatistics. Note, we have the country's top-rated Department of Biostatistics in the School of Public Health. The School of Medicine has medical informatics and biostatistics/bioinformatics specialists. All of these divisions had expertise that inHealth needs.
COVID-19 is just one example of the problem that we face in every clinic. In nearly all of medicine, clinicians seek better answers to three primary questions: What is this patient’s disease state? What is their disease trajectory? And, among the interventions available, what's the best one for this patient? Those questions are asked every day in every clinic and dynamic data modeling can help answer them.
One place where we've made a lot of progress is our scleroderma clinic. Scleroderma is an autoimmune disease that has long-term serious effects. Patients often experience life-threatening drops in lung function or heart function. It's complicated because clinicians have to ask those three questions about measures for five or six organs affected by scleroderma. They have to assemble all that information and try to piece together how this person's doing, where are they headed, and if they should change their treatments. We’ve now captured all that dynamic data, not only about the patient in hand, but for all the patients we have seen at the clinic. Because JHM has long-term follow up on nearly 4,000 patients, our scleroderma specialists can now characterize each patient’s trajectory by borrowing strength from the lessons learned from a large number of similar patients who have come before. They are discovering subgroups of patients whose dynamic disease processes differ and should therefore be treated accordingly.
The United States is spending roughly $3.3 trillion on healthcare this year. The second most expensive country per capita, the Netherlands, would spend just $2.1 trillion if it were the same size. We could still be number one in expenditures and spend $1.2 trillion less each year, which is 5% of GDP. A primary motivation for founding inHealth was to reduce wasteful, ineffective healthcare spending. We can reduce costs by avoiding things that don’t add value to the patient’s well-being. Careful analysis of healthcare data helps identify those opportunities.
One of inHealth’s first applications was in prostate cancer. If you have a low-risk cancer, Hopkins urologists encourage their patients to participate in active surveillance rather than pursuing major procedures like radiation therapy or removing the prostate. About 40-to-50% of all prostate diagnoses are in this initially low-risk category. Patients come in every 6 to 12 months, to be examined for signs and symptoms that the cancer is now more severe than was originally determined. A team of clinical and data scientists combined dynamic biomarkers, pathology and imaging data for 1,500 patients then built a statistical model to calculate the probability that a patient’s tumor remains indolent versus has become aggressive. Removing the prostate on average costs about $50,000 per case, and of all the prostatectomies done, 20-to-30% or more end up being indolent tumors once the removed tissue is further examined. This modeling helps identify patients who can avoid those expensive surgeries and the serious side effects they cause.
At some point the data-informed approach to medicine will be so routine, we will stop calling this inHealth Precision Medicine. It will just be Johns Hopkins Medicine committed to making precise measurements of every patient, learning from those measurements, and then continuously improving the care for the patients that come after. Every JHM clinic, not just the ones in which we're experimenting now, will capture all of the data from all of their patients, continuously analyze it, and continuously improve the care that they provide. Where we don’t know how to do well by patients in a particular disease, we're accelerating the research, focusing on research discoveries that subset the patients so that we can better tailor the treatments to those individuals.
A final reminder. There's this notion that if we collect enough data in the practice of medicine, and we have fast enough computers and smart enough data analysts, the truth will emerge. For a subset of problems these ingredients are sufficient, but for many others, we’ll need to innovate. It's not sufficient to only learn by carefully observing clinical practice, we also have to perform designed experiments — randomized clinical trials. I expect Johns Hopkins to become a ‘learning health system’ of the future, where we're not just learning from observations of our patients, but also by inviting all of our patients to participate in randomized trials. Where the existing evidence is clear, each patient will receive the treatment best for persons like them. Where the evidence is equivocal, we can give every patient the opportunity to access the best current or experimental treatments and to contribute precious unbiased evidence about what works best for future patients like them.