Not all patients will react to a medical treatment in the same way. A treatment that works well for some patients may be ineffective, or even harmful, for others. And that often makes it difficult for patients and clinicians to make informed health and healthcare decisions.

One reason for this is that medical research often provides information only about the average effect of a particular treatment approach, not how an individual patient with a specific set of characteristics will react. But patients want to know, “What is likely to happen to someone like me?”

Answering that question is at the heart of PCORI’s work. In fact, our authorizing legislation requires us to study the effects of health interventions for different patient subpopulations. This is known as research on the heterogeneity of treatment effects (HTE), and we consider it an essential component of the patient-centered comparative clinical effectiveness research (CER) we support.

The results of HTE analyses can be used to develop care programs designed to provide the treatment most likely to be effective for a particular patient—a step toward personalized medicine.

Seeking Better Methods for Analyzing Complex Data

At PCORI, we often say that in research, methods matter. That’s certainly true when it comes to the quality of the evidence produced by HTE analyses—and used by clinicians and patients. Traditional approaches, such as estimating treatment effects for subgroups of patients based on single risk factors, can lead researchers to detect HTE where it doesn’t exist and fail to detect true HTE where it does.

We’re addressing this issue in a variety of ways. For example, PCORI’s Methodology Standards include guidance on conducting appropriate HTE analyses to identify the risks and benefits of an intervention for particular subgroups of patients. Several of the research projects we fund focus on HTE as well.

One study, the subject of a recent paper in a major scientific journal, clearly illustrates how HTE analysis can help patients and doctors make better-informed decisions. This work was funded in part by PCORI to explore variation in treatment effects in randomized controlled trials. Re-examining data from a previous trial of a diabetes prevention program, the researchers used 17 factors to assign patients to diabetes-risk groups. In the study, some patients took metformin, a drug that helps to control blood sugar levels and is recommended to prevent diabetes. The research team found that among patients in the highest-risk group (the top quartile), those who took metformin were less likely to develop diabetes during the next three years, but the rest of the study population showed little benefit. In fact, those at lowest risk for diabetes (in the bottom quartile) showed no benefit at all from the medication.

This HTE analysis suggests a major, positive health impact from prescribing metformin to people at high risk for diabetes—but no benefit for patients at lower risk. However, all patients may experience any of a long list of negative side effects, as well as higher out-of-pocket expenses. In this example, methods research led to findings that can help patients and clinicians make better-informed choices about care.

Taking a Wide Variety of Approaches

Our Methods and Pilot Projects programs have funded additional studies that take different approaches to improving the usefulness of HTE analysis.

• A team at the Johns Hopkins University is developing recommendations and software to help researchers employ HTE analysis in patient-centered outcomes research. The team is using Bayesian statistical methods, which build on information about how treatment effects vary across populations.
• A study at the University of South Carolina is investigating statistical methods for examining large sets of observational data, collected as a treatment has been used in practice. Such observational studies can be more practical than randomized clinical trials in cases where an intervention not only affects individual patients differently but also can have multiple effects within a single patient (for example, a medicine might lower blood pressure, prevent kidney problems, and cause headache and upset stomach).
• Researchers at the University of Miami School of Medicine are developing an improvement on a machine learning technique to identify subgroups with different treatment responses.
• A project at Brigham and Women's Hospital aims to develop and test a tool for predicting whether a patient will take a medication as prescribed. Many patients do not take their prescribed medications, which leads to extra hospitalizations and costs; it also makes medical studies difficult to interpret.
• To improve the value of CER based on observational data, a study at the University of North Carolina at Chapel Hill is assessing specific methods, including HTE analysis, in a comparison of treatments administered to older adults after heart attacks.

Research on methods for HTE analyses can improve the evidence available for clinicians, patients, and other stakeholders to use in determining—on an individual level—whether a treatment is likely to be effective. We welcome your thoughts, questions, and suggestions on our work to apply HTE analyses to improve health outcomes.