Effect of Drug Therapy on HEDIS Measurements of HbA1c

Document Sample
Effect of Drug Therapy on HEDIS Measurements of HbA1c Powered By Docstoc
					     Effect of Drug Therapy on HEDIS
     Measurements of HbA1c Control
     In Diabetes Patients
     Gary Bazalo, MS, MBA1; Richard Weiss, MS1; Nathaniel Clark, MD2; Berhanu Alemayehu, PhD3; Felicia Forma3;
     Garrett Ingram ME3
      Managed Solutions LLC, Mount Freedom, N.J.; 2Senior Medical Adviser for Diabetes, Novo Nordisk Inc.,
     Princeton, N.J. ; 3Health Economics & Outcomes Research, Novo Nordisk Inc., Princeton, N.J.

        The National Committee for Qual-        ABSTRACT
     ity Assurance (NCQA 2008) collects            The purpose of this study was to corroborate an earlier study that ex-
     data from over 90 percent of the           plored the relationship between a health plan’s Health Plan Employer
     health plans in the United States to       Data and Information Set (HEDIS) score for glycolated hemoglobin (HbA1c)
     evaluate and report health plan per-       control in diabetes patients and its utilization of insulin and oral diabetes
     formance with respect to quality of        products. Prescription volumes were tracked for four categories of dia-
     care and service.The tool that NCQA        betes drug therapy: analog insulin, human insulin, single-source brand
     uses to collect this data is the Health    oral products, and multisource generic oral products, for calendar years
     Plan Employer Data and Informa-            2005 and 2006. The prescription shares of each of the four drug categories
     tion Set (HEDIS) (NCQA 2007,               for each health plan were matched to the health plan’s HEDIS measure-
     NCQA 2005). HEDIS addresses a              ments of HbA1c control for each year. Univariate and multivariate regres-
     wide range of important health care        sion analysis was performed between the health plan’s HbA1c -based
     states, including comprehensive dia-       HEDIS score and its prescription share of each drug category. A favorable
     betes care. Comprehensive diabetes         and statistically significant (p < 0.01) relationship was found between plan
     care data collected by HEDIS and re-       HbA1c HEDIS score and plan prescription share of analog insulin in both
     ported annually by NCQA are used           2005 and 2006. The correlation between HEDIS scores and human insulin
     by employers, providers, and con-          was not statistically significant. Unfavorable relationships were found be-
     sumers to compare health plans’ care       tween HEDIS scores and both the single-source brand (statistically signifi-
                                                cant) and the multisource generic oral category prescription shares (not
         Key Words: HEDIS, diabetes care,       significant). These results corroborate the relationships found in our ear-
         analog insulin, human insulin          lier study, although a cause and effect relationship cannot be confirmed.

                                               for their diabetic patient populations.   Control” will report the percentage of
         Disclosure                               Although results for individual        patients with diabetes with HbA1c <7
         Gary Bazalo and Richard Weiss dis-    plans vary greatly, nationally, nearly    percent. The most recent NCQA re-
         close receiving honoraria or other    30 percent of patients with diabetes      port stated that 41.8 percent of the
         financial benefit from Novo Nordisk   who are enrolled in commercial            population with diabetes is reported
         relevant to the products or pro-      health plans have poor control of         to be within good control as defined
         grams mentioned in this article.      their disease, as defined by glycolated   by this new measure (NCQA 2007).
                                               hemoglobin (HbA1c) levels >9 per-            Attempts have been made to cor-
                                               cent (NCQA 2007). There has been a        relate disease management, adher-
         Address for correspondence            continual improvement in the per-         ence, and preventative care programs
         Gary Bazalo, MS, MBA                  centages of patients in poor diabetic     with improvements in diabetes
         Managed Solutions, LLC                control from a high of 42.5 percent in    HEDIS scores. While some of these
         PO Box 526                            2000 to 29.6 percent in 2006, the most    studies have shown an improvement
         Mt. Freedom, NJ 07970-0526            recent year for which data is avail-      in HEDIS scores, others have demon-
         E-mail:                               able. Recently, NCQA has added a          strated inconsistent results (Sidorov
           gary@managedsolutionsllc.com        new measure of performance with           2002, Bramley 2006, Quenan 2000).
                                               respect to diabetes. “Good HbA 1c            In 2007, we reported a correlation

40         MANAGED CARE / FEBRUARY 2009
between high use of certain medi-          generic oral products consisted of all                 year. The total diabetes prescription
cations with better diabetes HEDIS         single source and multisource oral                     volume for eligible plans was about
scores. For the first time, the rela-      products respectively captured in the                  21.2 million in 2005 and about 24.5
tionship between utilization of spe-       IMS prescription database. The pre-                    million in 2006. Combined branded
cific classes of medications and de-       scription share of each of the four                    and multisource generic oral prod-
clining rates of patients in poor          drug classes was calculated based on                   ucts represented slightly more than
diabetes control was demonstrated          all prescription diabetes products in                  two thirds of total diabetes prescrip-
(Weiss 2007). That study examined          the PlanTrak database. Therefore, the                  tions (68 percent and 68.1 percent
the relationship among health plans        prescription shares of the four classes                for 2005 and 2006, respectively).Ana-
of HEDIS-measured HbA1c control            within each health plan totaled 100                    log insulin prescription share in-
and the prescription share of four         percent.                                               creased from 20.7 percent in 2005 to
categories of diabetes products: ana-                                                             23 percent in 2006, while human in-
log insulin, human insulin, branded        ANALYSIS                                               sulin share decreased from 11.3 per-
oral products, and generic oral prod-         Univariate (simple) and multi-                      cent to 8.9 percent during the same
ucts. We reported a strong positive        variate regression analysis was per-                   time period. The oral multisource
relationship between better HEDIS          formed between the health plan’s                       generic market share increased
scores and higher analog insulin pre-      (HbA1c)-based HEDIS score and its                      slightly from 48.1 percent in 2005 to
scription share (P<0.0001).                prescription share of each drug cate-                  49.4 percent in 2006, while branded
   The purpose of the current study        gory. The effect of plan size and re-                  oral market share decreased from
was to investigate whether these re-       gion of the United States was also ex-                 19.8 percent in 2005 to 18.6 percent
sults could be corroborated with a         plored via multivariate regression. In                 in 2006. The mean HEDIS score for
more robust and recent data set.           order to assess the effect of copay-                   percentage of patients with diabetes
                                           ments on HEDIS score we matched                        enrolled in eligible plans who were
METHODS                                    HEDIS score to the percentage of di-                   not in control improved from 29.6
   To identify the level of HbA1c con-     abetes prescriptions in the lowest co-                 percent ± 7.7 in 2005 to 28.8 percent
trol for a given health plan, this study   payment tier in each plan and per-                     ± 7.13 in 2006 (P=NS).
used the HEDIS measurement “per-           formed a linear regression across                         A statistically significant correla-
cent of diabetes patients having poor      health plans to determine the rela-                    tion (p = 0.002 and <0.001, respec-
HbA1c control” as reported in the          tionship between HEDIS score and                       tively) was seen between higher ana-
NCQA Quality Compass product               percentage of diabetes prescriptions                   log insulin prescription shares and
(NCQA 2008). These measurements            with the lowest copayment.                             better (i.e., lower percentage of pa-
were matched to the plan’s prescrip-                                                              tients with HbA1c level > 9 percent)
tion share (based on prescription vol-     RESULTS                                                HEDIS scores among health plans in
ume) in each calendar year for four          A total of 194 eligible health plans                 both 2005 and 2006 (Table 2). That is,
groupings of antidiabetic products:        were identified for 2005, and 211                      a higher analog insulin prescription
(1) analog insulin, (2) human insulin,     health plans were identified for 2006.                 share was associated with a lower
(3) single-source branded oral prod-       Table 1 summarizes the prescription                    (fewer patients with HbA1c level > 9
ucts, and (4) multisource generic oral     volume from eligible plans for each                    percent) HEDIS score. The slope
products. A health plan was included
in the study for 2004 or 2005, pro-
                                             TABLE 1
vided the plan had a valid HbA 1c
“poor control” HEDIS measurement             Summary of prescription volume and share for eligible plans
and at least 2,000 total diabetes pre-                                             2005                                        2006
scriptions in that year.                                                       N = 194 plans                               N = 211 plans
   Prescription volume for each prod-
                                                                       TRx volume           TRx share           TRx volume        TRx share
uct group was reported by IMS
HealthCare (Plymouth Meeting, Pa.)           Analog insulin             4,404,688            20.7%               5,651,117         23.0%
for individual health plans in its           Human insulin              2,395,031            11.3%               2,182,978          8.9%
PlanTrak report. Analog insulins in-         Total insulin              6,799,719            32.0%               7,834,095         31.9%
cluded the insulin lispro, insulin as-       Brand orals                4,211,152            19.8%               4,573,738         18.6%
part, and insulin glargine product           Generic orals             10,223,185            48.1%              12,127,876         49.4%
lines. Human insulin included the in-        Total orals               14,434,337            68.0%              16,701,614         68.1%
sulin isophane (Humulin) and                 Total                    21,234,056            100.0%             24,535,709         100.0%
human insulin isophane suspension            TRx = Total prescriptions (new and refill); N = number of plans in analysis
(Novolin) product lines.Branded and

                                                                                          FEBRUARY 2009 / MANAGED CARE                        41
       TABLE 2
       Simple linear regression models of HEDIS score with analog insulin, human insulin, branded oral,
       and generic oral prescription share
                                      Correlation                                                    Correlation
                                          with                                                           with
                                      HEDIS 2005              Model                                  HEDIS 2006              Model
       Drug Class                      (N = 194)*           coefficient            P value†           (N = 211)*           coefficient            P value†
       Analog insulin share              −0.22                 -0.319                 0.002              −0.30                −0.424                < .001
       Human insulin share               −0.09                 -0.218                 0.197              −0.10                −0.257                0.128
       OAD share (brand )                 0.22                  0.378                 0.002               0.326                0.573                < .001
       OAD share (generic)                0.07                  0.910                 0.321               0.12                 0.180                0.090
       *A negative correlation value indicates a favorable correlation between HbA1c control and prescription share (i.e., higher prescription share is associated
       with a lower HbA1c “poor control” score)
        t test        OAD = oral antidiabetic drug

     (model coefficient in Table 3) of the                 tive correlation between poorer (i.e.,                   the effect of analog and human in-
     linear regression line was –0.319 in                  higher percentage of patients with                       sulin prescription share, with higher
     2005 and –0.424 in 2006.The slope in                  HbA1c level > 9 percent) HEDIS scores                    analog insulin utilization continuing
     2006 indicates that a 10-share-point                  and higher oral product prescription                     to be statistically significantly associ-
     increase in analog insulin utilization                share. This unfavorable correlation                      ated in both 2005 and 2006 (p = 0.004
     was associated with a 4.2 percentage                  was statistically significant in the case                and p = 0.013, respectively) with bet-
     point decline in the “poor control”                   of branded oral product prescription                     ter HEDIS scores. Higher human in-
     HEDIS score (i.e., is associated with                 share in 2005 and 2006 (p = 0.002 and                    sulin utilization was not associated
     an improvement in HbA1c control).                     p<0.001, respectively). The unfavor-                     with better HEDIS scores. Branded
     Linear regression analysis using                      able correlation was not significant                     oral share continued to be associated
     human insulin share as the inde-                      for generic oral agents in either year.                  with poorer HEDIS scores at a statis-
     pendent variable showed a favorable                      To further compare the effect of                      tically significant level (p = 0.011 and
     correlation between human insulin                     product group prescription share on                      p = 0.002 in 2005 and 2006, respec-
     share and improvement of “poor con-                   HEDIS score, a multivariate regres-                      tively).
     trol” in both years. However, the re-                 sion was conducted for each year with                       Evaluation of plan enrollment and
     lationship was not as strong as the                   analog insulin share, human insulin                      location in the multiple regression
     analog insulin relationship (model                    share, and branded oral share as in-                     analysis revealed that only two U.S.
     coefficient = –0.218 in 2005 and                      dependent variables (Table 3). Since                     regions reached statistical significance
     –0.257 in 2006) and not statistically                 the four categories total 100 percent,                   (South Atlantic/Central and Moun-
     significant in either year.                           the generic oral value is completely                     tain) with both regions unfavorably
        Linear regression analysis using ei-               dependent on the values of the other                     correlated to HEDIS scores (Table 4).
     ther branded oral product share or                    three shares and was dropped from                        However the effects were small com-
     multisource generic oral share as the                 the model. The multivariate regres-                      pared to the product prescription
     independent variable showed a posi-                   sion clearly discriminated between                       shares. The inclusion of plan enroll-

       TABLE 3
       Multivariate regression analysis of HEDIS score with analog insulin, human insulin, and branded
       oral prescription share
                                             HEDIS 2005                                                      HEDIS 2006
                                           model coefficient                                               model coefficient
       Drug class                             (N = 194)*                          P value†                    (N = 211)*                         P value†
       Analog insulin share                       −0.302                           0.004                          −0.258                           0.013
       Human insulin share                         0.010                           0.961                           0.156                           0.427
       OAD share (brand )                          0.363                           0.011                           0.492                           0.002
       *A negative model coefficient indicates a favorable correlation between HbA1c control and prescription share. Model statistics: r2=0.091, p<0.001, F-test
       (year 2005); r2=0.141, p<0.001, F-test (year 2006).
        t test                  OAD = oral antidiabetic drug

  Multivariate regression analysis of HEDIS score with analog insulin, human insulin, and brand oral
  prescription share; plan enrollment; and plan location
                                    Correlation with                                                  Correlation with
  Drug class,                         HEDIS 2005                                                        HEDIS 2006
  enrollment, location                 (N = 194)*                         P value†                       (N = 211)*                          P value†
  Analog insulin share                    −0.263                              0.02                            −0.288                             0.01
  Human insulin share                      0.148                              0.48                             0.298                             0.15
  OAD share (brand )                       0.429                              0.01                             0.491                             0.00
  Small enrollment                        −0.010                              0.50                            −0.003                             0.82
  Medium enrollment                       −0.002                              0.91                            −0.011                             0.39
  Large enrollment                        −0.013                              0.47                            −0.006                             0.71
  East north central                      −0.007                              0.65                            −0.006                             0.64
  South Atlantic/Central                   0.035                              0.02                             0.032                             0.01
  Mountain                                 0.040                              0.02                             0.036                             0.02
  Pacific                                  0.011                              0.67                             0.002                             0.94
  Small enrollment = 100,000 to 249,999; medium enrollment = 250,000 to 749,999; large enrollment ≥750,000
  Enrollments under 100,000, northeast and mid-Atlantic used as reference and not included in model
  *A negative model coefficient indicates a favorable correlation between HbA1c control and prescription share. Model statistics: r2=0.161, p<0.001, F-test
  (year 2005); r2=0.207, p<0.001, F-test (year 2006).
   t test            OAD = oral antidiabetic drug

ment and location in the multiple re-                 DISCUSSION                                              with human insulin showing a slight
gression model increased the unfa-                       These results of this study corrob-                  (nonsignificant) unfavorable effect
vorable effect of human insulin with                  orate those reported in the earlier                     on HEDIS scores in both years. The
respect to analog insulin.                            study (years 2004 and 2005), which                      inclusion of plan enrollment and lo-
   The linear regression with per-                    used more limited prescription vol-                     cation in the multiple regression
centage of diabetes prescriptions on                  ume information (Weiss 2007). A                         analysis resulted in a further differ-
the lowest copayment tier as the in-                  strong relationship between better                      entiation between analog and human
dependent variable showed a small                     (lower) HEDIS scores for HbA1c con-                     insulin, with human insulin showing
nonsignificant effect (Table 5). For                  trol and higher analog insulin pre-                     a greater unfavorable impact, al-
2005, a higher percentage of pre-                     scription share was demonstrated by                     though still not reaching statistical
scriptions in the lowest copay tier was               both the univariate and multivariate                    significance. Both the univariate and
associated with a better (lower)                      regression analyses for 2005 and                        multivariate regressions demon-
HEDIS score. The slope was –11.3,                     2006. Human insulin share showed a                      strated an unfavorable relationship
indicating a 1.1 percentage point                     favorable relationship to HEDIS                         between oral product shares and
drop in HEDIS score for a 10-share-                   scores in the univariate analysis in                    HEDIS scores. The relationship was
point increase in prescriptions in the                both years but without statistical sig-                 much more pronounced for the
lowest copay tier (p = 0.06). In 2006,                nificance. The multivariate analysis                    brand oral products, with higher
this relationship diminished to a                     attributed the improvement in                           brand oral market shares in health
slope of –7.4.                                        HEDIS scores to analog insulin only,                    plans associated with statistically sig-

  Results of univariate (simple) regression between HEDIS score and percentage of diabetes
  prescriptions in lowest copayment tier
                   Mean percent of diabetes prescriptions by copayment tier
                          Tier 1                        Tier 2                   Tier 3                           Slope                        p-value
  2005                    68.4%                        30.7%                      0.9%                           -11.3                           0.06
  2006                    70.5%                        28.2%                      1.3%                            -7.4                           0.18
  Percentages are based on prescriptions for products with known copayment tiers

                                                                                                     FEBRUARY 2009 / MANAGED CARE                             43
     nificantly poorer HbA1c control. One       tions in the portion of Type 1 and         HbA1c levels. Studies should be con-
     interpretation for this result is that     Type 2 patients among health plans.        ducted to confirm these findings with
     patients and physicians may be at-         Therefore, the proportion of Type 1        more data points, compare them with
     tempting to avoid insulin therapy by       patients would not be expected to be       new standards-of-care measures, and
     using a variety of newer oral agents.      a significant factor in these results.     longitudinally track the effect of new
         The lack of a relationship between        No data were available in this study    medications on diabetes care as meas-
     HEDIS score and percentage of dia-         to determine concomitant use of in-        ured by HEDIS.
     betes prescriptions in the lowest          sulin and oral products, or among
     copay tier may be due to the fact that     products within a drug class.The pre-      REFERENCES
     the copay tier status was missing for      scription share analysis assumes pa-       Bramley TJ, Gerbino PP, Nightingale BS,
     many products. The initial assump-         tients are on monotherapy, although             et al. Relationship of blood pressure
                                                                                                control to adherence with antihyper-
     tion was that a higher percentage of       concomitant use would not invali-               tensive monotherapy in 13 managed
     prescriptions in the lowest copay tier     date the findings regarding the rela-           care organizations. J Manag Care
     might result in better patient com-        tionship between HEDIS scores and               Pharm 2006;12:239–245.
     pliance and therefore better HbA1c         analog insulin prescription share.         National Committee for Quality Assur-
                                                                                                ance. The State of Health Care Quality
     control. Although this proved to be           The new NCQA measure of per-                 2005: Industry Trends and Analysis.
     the case directionally, the results did    centage of patients with diabetes with          Washington, DC, 2005. Available at:
     not reach statistical significance. We     good HbA1c control (<7 percent) will            http://www.ahrq.gov/qual/nhqr05/n
     calculated the percentage of pre-          provide the first data point to cor-            hqr05.htm. Accessed Jan. 14, 2009.
     scriptions in the lowest copay tier        roborate the results of the current        National Committee for Quality Assur-
                                                                                                ance “New HEDIS measures allow
     based only on the product prescrip-        and prior studies using a direct meas-          purchasers to compare health plan’s
     tions for which a copay tier was           ure of good control (NCQA 2006). A              resource use in addition to quality
     known.                                     future study might perform analyses             (press release).” Feb. 22, 2006. Avail-
         Although the results clearly indi-     similar to those conducted here, but            able at:
     cate that higher utilization of analog     using both performance measures                 http://www.ncqa.org/tabid/362/De-
                                                                                                fault.aspx. Accessed Jan. 14, 2009.
     insulin is associated with greater         over time.                                 National Committee for Quality Assur-
     HbA1c control in diabetes patients in         As new medications to treat dia-             ance. The State of Health Care Qual-
     commercial health plans, an associa-       betes enter the market, future studies          ity 2007. Washington, DC, 2007.
     tion does not imply a cause-and-           must account for their effect on dia-           Available at:
     effect relationship. Drug therapy is       betes performance measures. Thus, a
     only one factor in the management of       future study might correlate only the           SOHC_07.pdf. Accessed Jan. 14,
     diabetes. Comprehensive diabetes           newer agents with HEDIS results.                2009.
     care may also entail enrollment in            Future studies might also explore       National Committee for Quality Assur-
     disease management programs, im-           the causes of the correlations between          ance. Quality Compass. National
                                                                                                Committee for Quality Assurance.
     proved adherence to published treat-       better NCQA/HEDIS performance                   Available at:
     ment guidelines, and other factors         scores in diabetes and specific drug            http://www.ncqa.org/tabid/177/De-
     such as diet and exercise.                 classes and agents. Areas to explore            fault.aspx. Accessed Jan. 14, 2009.
         A potential confounding factor         might include disease severity, patient    About NCQA. National Committee for
     could be the proportion of Type 1 di-      population statistics, and plan inter-          Quality Assurance. Available at:
     abetes patients in a health plan.These     ventions such as disease and medi-              fault.aspx. Accessed Jan. 14, 2009.
     patients would raise the ratio of in-      cation management programs.                Quenan L, Remington P, Gohre F, et al.
     sulin to oral medications, but not                                                         The Wisconsin Collaborative Dia-
     necessarily the analog portion. While      CONCLUSIONS                                     betes Quality Improvement Project.
     Type 1 patients would be expected to          Data from this study corroborate             WMJ. 2000;99(3):48–52.
                                                                                           Sidorov J, Shull R, Tomcavage J, et al. Does
     be more adherent to their regimens,        an earlier study performed by the au-           diabetes disease management save
     they might not show any better con-        thors that indicate a statistically sig-        money and improve outcomes? Dia-
     trol than Type 2 patients because of       nificant correlation between higher             betes Care. 2002;25:684–689.
     the volatility of their glucose levels.    utilization of analog insulins and         Weiss R, Bazalo G, Clark N, Forma F, In-
                                                                                                gram G, Alemayehu B. Correlation of
     Type 1 patients represent only 5 per-      lower percentages of health plan                HEDIS Diabetes Health Plan Score
     cent to 10 percent of diabetes pa-         members with diabetes who are not               With Utilization of Diabetes Medi-
     tients, and therefore the impact on        in glycemic control, as defined by              cations. Managed Care Interface.
     the correlation between insulin uti-       HbA1c levels >9 percent. Human in-              2007; 10: 14–17.
     lization and improved HEDIS scores         sulin showed either a weaker positive
     is likely to be small. Also, there is no   effect (univariate analysis) or a neg-
     reason to expect significant fluctua-      ative effect (multivariate analysis) on