Rajan M (East Orange REAP), Pogach LM
(East Orange REAP), Tseng CL
(East Orange REAP), Litaker DG
(Cleveland VAMC), Aron DC
(Cleveland VAMC)
Objectives:
NCQA recently proposed restricting a < 7% A1C measure to individuals < 65 years without serious comorbidities. This measure may still result in incorrect quality inferences based upon facility differences in important patient-level factors, including insulin treatment, that pose significant barriers to achieving < 7%. Furthermore, a threshold measure is relatively insensitive to significant clinical improvements that do not reach “goal.” Our objective was to compare facility rankings for veterans younger than 65 years, with and without insulin treatment, using both a threshold and continuous measure.
Methods:
We identified 352,003 veterans with diabetes < 65 years from 128 facilities in FY2003 and eliminated 127, 583 subjects with comorbid conditions considered contraindications to tight control as previously described (AmJManCare 2007;13:133). The remaining 224,420 subjects were stratified into those receiving (22%) or not receiving insulin (78%) in 2003. The mean number of veterans per facility was 1753 (range 200-5000). Using the last A1c within year, we evaluated the percent of veterans < 7%A1c, and the proportion of maximal quality adjusted life years (QALYs) earned within a continuous 7.9% to 7.0% range (Diabetes Care 2006;29:241). Facilities were ranked within the insulin user and non-user groups using Spearman Rank Coefficient and decile changes of the top and bottom (10%) performers.
Results:
The mean A1c and % < 7% were 7.5, 34.5% overall; 8.4, 17.5% for the insulin group; and 7.3, 39.3% for the non-insulin group. The Spearman rank correlation of the insulin to non-insulin group was 0.72 and 0.76 using < 7% and QALYs, respectively. Of the 13 highest ranked facilities for non-insulin group, 7 changed > = one decile when using QALY’s, and 10 changed using the < 7% threshold. There was no change in the bottom ranked facilities using QALYs; 2 facilities changed using thresholds.
Implications:
A1c outcomes and identification of “highest” performing facilities were markedly different when stratified by insulin usage. Compared to a threshold measure, the continuous measure was somewhat more consistent in identifying outliers.
Impacts:
Providing results separately for insulin users and non-users may prioritize quality improvement efforts and more fairly evaluate performance. A continuous measure may be preferable to threshold measures based upon face-validity considerations of assessing incremental improvement.