++
This JAMA Guide to Statistics and Methods explains the use of regression discontinuity analysis on observational data—the difference in effect size estimate between regression analyses using an exposure variable above and beneath a threshold of interest—to distinguish changes associated with an intervention from background ecological or secular changes.
++
In a report published in JAMA Network Open in 2019, Sukul and colleagues1 used a regression discontinuity design (RDD) study to determine if a 2011 change in Medicare policy, expanding the number of secondary diagnostic codes allowed in Medicare billing from 9 to 24, was associated with a change in the observed disease severity of Medicare patients, independent of any real change in these patients’ medical condition.
++
The apparent disease severity, based on secondary diagnostic codes, might increase because of real changes in the overall disease burden over time, as an artifact from the listing of additional codes due to the greater number allowed, or some combination these factors. The RDD is a method for separating the contribution of the abrupt change in policy—discontinuity—to observed increases in outcomes from the contribution of other factors that may produce more gradual changes.2
+++
USE OF THE RDD METHOD
+++
Why Is an RDD Study Used?
++
The RDD method is used to study the outcomes related to an abrupt change when it is not possible to randomly assign patients to the conditions before and after the change. A goal of an RDD is to minimize the potential influence of confounding on the estimated effect of a policy or treatment change.2 Unobserved confounding is particularly concerning in nonrandomized studies because this bias cannot be completely removed using conventional statistical methods, such as regression or propensity score–based analysis.3 An RDD attempts to minimize the risk for unobserved confounding when generating the association between an exposure and the change in the outcome of interest.4
+++
Description of the RDD Method
++
The RDD approach relies on having a continuous variable, the “running variable,” for which the levels above a certain cutoff abruptly change the probability of receiving one treatment or the other, resulting in a “natural experiment.” The 2011 change in Medicare policy for secondary diagnosis codes represents just such a natural experiment, and calendar time represents the running variable. Because the presence of potentially confounding factors should be the same immediately around the date when the policy change occurred, estimates of the effect of the change at the cutoff point should not be substantially confounded.5
++
To evaluate the relationship of the change with the outcome of interest, the outcome is first estimated for patients above the cutoff as the value of the running variable approaches the cutoff from above. The same outcome is then estimated for patients below the cutoff as the value of the running ...