Changes in Transfusion Coding Among Hospitalized Medicare Beneficiaries after Implementation of ICD-10

The standardized transfusion ratio (STrR), a claims-based quality metric, was added to the ESKD Quality Incentive Program (QIP) in performance year 2016. The Centers for Medicare and Medicaid Services (CMS) recently proposed that the STrR constitute 22% of the total performance score for each dialysis facility in performance year 2019; currently, the STrR constitutes only 8.25% of the total performance score (1).

Almost 80% of blood transfusions among patients on dialysis are administered in inpatient settings (2). On October 1, 2015, the CMS mandated immediate migration from the International Classification of Diseases, Ninth Revision (ICD-9) taxonomy to the International Classification of Diseases, Tenth Revision (ICD-10) taxonomy to document diagnoses and procedures in inpatient settings. ICD-10 includes 256 codes for documentation of transfusions. We assessed whether hospitals consistently recorded blood transfusions before and after the advent of ICD-10.

Per a data use agreement with the CMS, we analyzed all Medicare fee-for-service claims by inpatient hospital providers in 2014–2016. We identified hospitalizations in short-term and critical access hospitals in the last four quarters of ICD-9 (October 2014 to September 2015) and the first four quarters of ICD-10 (October 2015 to September 2016). For each hospitalization, we assessed whether a blood transfusion was administered according to codes specified by the CMS (ICD-9 procedure codes: 99.03 and 99.04; ICD-10 procedure codes: 302xyz1 [x=3, 4, 5, or 6; y=0 or 3; z=H, N, or P] excluding 30233P1; and value code: 37). For each hospital, we tallied (1) the number of hospitalizations and (2) the number of hospitalizations with a blood transfusion during each quarter in the study era. Finally, we fit a Bayesian logistic regression model of the probability of blood transfusion during hospitalization, with a normally distributed random intercept and a normally distributed random effect for taxonomy (i.e., ICD-10 versus ICD-9) within each hospital; in this model, each observation was a hospital-quarter with s transfusions during t hospitalizations. To estimate the posterior distribution of the model, we used Gibbs sampling; we drew 5000 samples from each of three parallel chains with 1000-sample burn-in periods (3). For each hospital, we estimated the posterior distribution of the odds ratio (OR) of blood transfusion during hospitalization in the first year of ICD-10 versus the last year of ICD-9. We geocoded hospitals for which the median posterior OR of blood transfusion was <0.20 (i.e., >80% reduction in blood transfusion). We used SAS (Cary, NC) to process claims and R (Vienna, Austria) to model data.

During the study era, there were 20,485,461 hospitalizations among 4541 hospitals. In the last year of ICD-9 and the first year of ICD-10, blood transfusions were documented during 8.7% and 6.7% of hospitalizations, respectively. The median posterior OR of blood transfusion during hospitalization in the first year of ICD-10 versus the last year of ICD-9 was <0.20 in 545 hospitals (12.0%), 0.20–0.39 in 257 hospitals (5.7%), 0.40–0.79 in 1349 hospitals (29.7%), 0.80–1.19 in 1703 hospitals (37.5%), and ≥1.20 in 687 hospitals (15.1%). Hospitals with a median posterior OR <0.20 are mapped in Figure 1. With posterior probability >97.5%, the OR of blood transfusion during hospitalization in the first year of ICD-10 versus the last year of ICD-9 was <0.80 in 795 hospitals (17.5%), including 708 short-term hospitals (21.7%) and 87 critical access hospitals (6.8%).

Figure 1.

Hospitals with a large reduction in transfusion coding among hospitalized Medicare beneficiaries after implementation of ICD-10 are widely distributed across the US. Each dot indicates the location of a hospital with a median posterior odds ratio of blood transfusion during hospitalization in the first year of ICD-10 versus the last year of ICD-9 that was <0.20.

The unbiasedness of the STrR depends on accurate ascertainment of blood transfusions in Medicare fee-for-service claims and appropriate risk adjustment. The transition from ICD-9 to ICD-10 seems to have altered documentation of blood transfusions in the inpatient setting. Hundreds of hospitals sharply curtailed documentation of blood transfusions after the advent of ICD-10. According to survey data, units of red blood cells that were transfused in the United States decreased 14% between 2013 and 2015 (4). From this perspective, apparent year-over-year decreases in blood transfusion utilization exceeding 20% (if not 80%) strain credulity. Poorer quality of data is a more logical explanation. Hospitals generally lack a financial incentive to document blood transfusions; the transfusion of only bone marrow affects Medicare payment.

To maximize sample size, we analyzed documentation of blood transfusions among all Medicare fee-for-service claims by inpatient hospital providers. Changes in blood transfusion coding among hospitalized patients on dialysis were possibly less pronounced. Further analysis of this issue is needed, especially after claims data from calendar year 2017 are released. However, in the face of uncertainty about the validity of the STrR, the proposal to attribute 22% of the total performance score for each dialysis facility to the STrR seems ill timed. How the CMS should finalize the ESKD QIP for performance year 2019 is unclear. One straightforward modification is to decrease the weight assigned to the STrR and increase weights assigned to other clinical care measures (i.e., dialysis adequacy, hypercalcemia, ultrafiltration rate reporting, and vascular access type). In the long run, the CMS must ensure that the STrR is a function of valid data.

Disclosures

Both authors are employees of NxStage Medical.