Objective To define measures of Medicare diagnosis coding intensity that capture the dynamics of changes in coding prevalence.

Study setting and design Retrospective analysis of coding practices for risk adjustment using observational claims data from Medicare beneficiaries assigned to Accountable Care Organizations (ACOs) in 2017 and 2018.

Data sources Enrollment and claims data from 2017-2018 of a random 20 percent sample of Medicare beneficiaries were subset to include those assigned to an ACO in 2018. Beneficiaries were included in analyses if they were continuously enrolled in Medicare Parts A and B in 2017 and 2018 and aged 65 years or older during that time. Beneficiaries were excluded if they were originally enrolled in Medicare due to end-stage renal disease or were residing outside of the United States.

Principal findings Borrowing terminology from epidemiology, prevalence of a diagnosis code can be decomposed into incidence (the proportion of beneficiaries that newly have the code) and persistence (the proportion of beneficiaries who previously had the code and continue to do so). Together, these two measures define what we call steady-state prevalence, the hypothetical long-run prevalence implied by no changes in current rates of incidence and persistence of coding. The concept of steady-state prevalence can help explain why observed prevalence tends to grow over time, apart from continued behavioral change. We demonstrate this and also illustrate the application of our measures for the most expensive diagnoses in Medicare’s risk-adjustment payment model for ACOs.

Conclusions Diagnostic coding is a highly heterogeneous behavior that can vary in response to changes in incentives. Researchers and policymakers can better monitor such behavior by measuring the incidence and persistence of a given code separately. In addition, changes in coding practices can take years to be fully reflected in current coding prevalence even in the absence of further behavioral change.

What is known on this topic:

Medicare incentivizes providers and insurers to code for as many diagnoses per beneficiary as possible, contributing to billions of dollars of Medicare program spending.

Empirical research indicates that the frequency of diagnostic coding in Medicare has increased steadily in recent years.

Current measures of coding practices for a given diagnosis largely count the prevalence, or the proportion of beneficiaries with a specified code, in one time period.

What this study adds:

This study proposes several new measures of coding practices decomposing diagnosis code prevalence into incident and persistent codes over two years, which policymakers can use to monitor coding behaviors.

This study defines steady-state prevalence as the ultimate prevalence of a code implied by current coding patterns hypothetically remaining unchanged.

Applying our measures to Medicare Accountable Care Organization coding finds that at current rates of incidence and persistence, observed prevalence of coding will increase even without further behavioral change.

TGM, OME, TN, and SR report no conflicts of interest. MC reports being the Chair of MedPAC and sits on the board of Massachusetts Health Connector and previously the HCCI, serving on advisory boards and panels for NIHCM, CBO, BCBSA, BHI, Aledade Inc., and equity in Waymark Inc. JMM reports research support from the National Institute on Aging, Agency for Healthcare Research and Quality, Arnold Ventures, and the Commonwealth Fund, and from an institutional gift for research and education from Humana to Harvard Medical School; personal income from Oak Ridge Associated Universities for services as a senior adviser to the Center for Medicare and Medicaid Innovation; consulting fees from RTI International, MITRE, Analysis Group, Phillips and Cohen, Action Now Initiative, and the West Health Policy Center; speaker honorarium from America's Physician Groups; and service as a member of the board of directors for the Institute for Accountable Care. The content of this article is solely the responsibility of the authors and does not necessarily reflect the views of any organization with which they are affiliated.

TGM, MC, JMM, and TN were funded by the National Institute on Aging of the National Institutes of Health under award number P01-AG032952. OME was funded by the National Library of Medicine grant 5T15LM007033- 40 and the Stanford Interdisciplinary Graduate Fellowship. SR was funded by the NIH Director's New Innovator Award DP2-MD012722.

I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.

Yes

The details of the IRB/oversight body that provided approval or exemption for the research described are given below:

This project was approved by the Centers for Medicare and Medicaid Services' privacy board and the Harvard Medical School Institutional Review Committee, which also waived the requirement for obtaining informed consent because all administrative data were deidentified.

I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.

Yes

I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).

Yes

I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.

Yes

No data are available. Under the requisite data use agreements, data for the analyses are not publicly available.