Research can tell us about the efficacy of a new drug or medical intervention. But when it’s applied in the real world, the complexities of medical practice and behaviors of patients and their clinicians often mediate its effectiveness. A patient may struggle to remember to take a new medicine, for instance, while lack of fidelity to a proven care model may produce suboptimal results. Thus, there is a need for ongoing research into what does and does not work outside of ideal research environments.
Tapping into new sources of data—including biometric readings from wireless devices, patterns of health service use and costs from electronic health records and medical claims, and insights on what people think and do drawn from their social media and internet use—can help bridge the divide between what’s been shown to work in randomized controlled trials and what actually produces value as measured by patients, providers, and payers. Making sense of all this information is an enormous challenge: the amount of health care data grows by nearly 50 percent a year. It requires a radically different set of analytic methods based on predictive analytics, machine learning, and artificial intelligence.
Bridging the Old and the New
Already, some health care providers—especially accountable care organizations that assume risk for the total costs of care for designated patients—are using these approaches to identify those most in need of help, make predictions about the efficacy of different treatments and the potential for adverse events, and give recommendations accordingly—moving us from predictive to prescriptive analytics. For an example of how health systems are tapping into new sources of data to alter clinical practice patterns, see “Reframing Analytics: Transforming Insights into Action,” Catalyst, June 29, 2017.
New analytic approaches offer the opportunities to learn about much larger populations than a particular pool of patients attributed to a single provider or health system. By gathering data on the real-world experiences of large populations of patients through program support platforms, digital devices, and mobile apps, the pharmaceutical industry can, for example, study how certain groups who are underrepresented in clinical trials—older adults with multiple chronic conditions and minorities among them—fare on new medications and find novel indications for drugs that have been proven safe. This also supports industry requirements to report the results of safety surveillance among large populations to the FDA.
Broader and Deeper Understanding of Patients’ Behaviors
New data sources and research methods can also shed light on what fuels behavior and why people do or do not play active roles in managing their health and care. Such insights are vital to engaging the growing number of people living with chronic conditions whose needs may evolve over time as they gain experience and improve their self care.
Technology allows us to personalize interventions and continually refine our approaches using the principles of machine learning. It can be used to detect the underlying reasons a patient may skip his or her medication—perhaps forgetfulness one day and concern about side effects another—and alter supportive messages accordingly, leveraging what’s been proven to be effective for the individual in the past. “Smart” interventions like these enable us to customize our approaches, rather than simply designing interventions that work for most, but not all, patients.
Working in collaboration with Google Analytics, Sanofi is leveraging public and proprietary data in an effort to understand how the attitudes and preferences among those with diabetes affect whether they pursue flu vaccination, which has been shown to significantly reduce their rates of hospital admission and mortality. Sanofi is also collaborating with the Duke Clinical Research Institute and the Center for Assessment Technology and Continuous Health (CATCH) at Massachusetts General Hospital to mine anonymized data to predict and mitigate problems with medication adherence among those with diabetes. These projects seek to create behavioral phenotypes that can be used to tailor interventions to accommodate patients’ diverse social and behavioral characteristics.
Recommendations to Health Care Providers, Payers, and Policymakers
To realize the benefits of these new data sources and methods we need to become much more nimble. Health care leaders, as well as researchers, need to build up their analytic capabilities and make greater use of learning systems. Our efforts must be grounded in what matters most to patients, not to health care payers and providers. And we should not be afraid of experimentation, of learning from low-cost simulations as well as failures.
Ultimately, our success will depend on collaboration among all parties: patients, providers, payers, and yes, the pharmaceutical industry. Contributions from each sector will create the most comprehensive views of patients’ experiences and offer the most salient lessons for promoting better health outcomes and lower costs.
To read the first blog in this two-part series, visit: From Randomized Controlled Trials to Real World Evidence: Finding the Balance.