Improving Methods for Studying Chronic Disease

For patients with chronic medical conditions, physicians often need to repeatedly adjust medication doses up or down based on the patient’s response to the prior dose. When researchers use data from medical records to compare the outcomes of various treatment regimens, all those dose changes make it difficult to tease out the effects and determine the best way to prescribe the medications.

Statisticians have developed analytic methods over the years to address this problem, but each one has limitations. A new PCORI-funded project is creating an open-source educational toolkit that includes two sophisticated statistical applications to adjust for complications introduced by dose changes. The web-based, user-friendly package will guide clinical researchers step by step using statistical tools that had previously been available only to highly trained statisticians at a few universities.

“The goal is to encourage applied researchers to use advanced techniques, with greatly reduced difficulty,” says project leader Yi Zhang, PhD, MS, of the Medical Technology and Practice Patterns Institute in Bethesda, Maryland. This nonprofit organization conducts research on the clinical and economic implications of healthcare technologies.

“Although patients are not directly involved in the toolkit project, it will certainly have an impact on them,” says Julia M. Kim, MD, MPH, of the Department of Pediatrics, Division of Quality and Safety, at Johns Hopkins University. As a clinician, she is one of two stakeholder representatives serving on the project’s advisory board.

“The research methods applied with this toolkit will provide reliable data for people to make informed decisions about their health,” says Kim.

A Clinical Scenario

To provide examples for toolkit users, Zhang and her team will use the statistical methods to conduct studies on three clinical scenarios of patients with chronic conditions. One of these conditions is kidney failure, an illness Zhang has been studying for several years. Even when receiving dialysis, patients are at increased risk for anemia—low levels of red blood cells that carry oxygen throughout the body—which makes patients feel tired and may lead to heart problems.

To counteract anemia, doctors prescribe a hormone called erythropoietin (EPO), which is normally produced by the kidneys and stimulates the bone marrow to produce red blood cells. Despite many studies over the years, there isn’t agreement on when to start a patient on EPO, how much to give, or the optimal concentration of red blood cells.

One reason this condition is so hard to study is that doctors adjust the dose of EPO frequently. Patients typically receive three treatment sessions per week, and the EPO dose may change monthly depending on a patient’s red blood cell concentrations. A patient who hasn’t responded well would get a dose increase, and a patient who has responded well would get a decreased dose, or EPO might be discontinued.

“Controversy continues, however, regarding the optimal anemia correction for dialysis patients,” Zhang notes.

Calculating the Results of Different Strategies

Ideally, researchers would conduct a clinical trial where patients are randomly assigned to treatment regimens that differ by variables such as dose and treatment targets. However, the large number of potential EPO dosing strategies makes randomized trials impractical, if not impossible.

Instead, researchers, when studying chronic disease treatments, often turn to large electronic databases of patient records, such as those of Medicare, Medicaid, or health maintenance organizations, to see—after the fact—how patients were treated and how they fared.

Even though it’s more feasible than a randomized, controlled clinical trial, this type of observational study often leaves questions: How can researchers be sure of the effect a treatment is having when the dose and the patient’s response keep changing? Would the patient’s condition have changed similarly without the treatment, or with a higher or lower dose?

Zhang’s software toolkit incorporates two recently developed advanced statistical techniques, called IP weighting and g-formula. Both techniques manipulate the data to mimic the setup of a randomized, controlled clinical trial. “We’re calculating what would have happened if the patient had received a different treatment strategy,” Zhang explains.

This toolkit should help researchers determine the best use of EPO for patients with kidney failure. Those results would then be expected to influence policy regarding public and private reimbursement for this expensive drug. Policy has shifted over the years, but currently, Medicare includes EPO expense as part of overall reimbursement for kidney failure patients.

Zhang and her team anticipate that the toolkit will be useful for observational studies of many common chronic conditions, including heart disease, diabetes, and obesity. According to Kim, “There are so many changes going on in medicine. We want to make sure that new studies are being done in a methodologically robust way to help inform practice guidelines.”