Poster

1
Table 1. Sample Selection PREDICTORS OF CHANGE IN ADHERENCE STATUS FROM ONE YEAR TO THE NEXT AMONG PATIENTS WITH TYPE 2 DIABETES MELLITUS ON ORAL ANTI- DIABETIC DRUGS S. Lane Slabaugh, PharmD, MBA 1 ; Xiaomei Peng, MD, PhD 2 ; Vishal Saundankar, MS 1 ; Anthony Louder, PhD, RPh 1 ; Haoda Fu, PhD 2 ; Peinie Young, PharmD 1 ; Angel Rodriguez, MD 2 ; Ayad K. Ali, RPh, PhD 2 ; Haya Ascher- Svanum, PhD 2 1. Comprehensive Health Insights, Humana, Louisville, KY, USA; 2. Eli Lilly and Company, Indianapolis, IN, USA Background Diabetes is a leading cause of morbidity, mortality and medical resource utilization in the United States 1 . Treatment is aimed at keeping blood glucose levels close to normal and preventing or delaying complications. However, it has been estimated that only 50% of US diabetes patients achieve an A1C level <7% 2 . Non-adherence to antidiabetic medications has been identified as one of the major factors related to poor glycemic control 3 . It has been estimated that increases in medication adherence of only 20% could reduce total health care spending by as much as $1074 on average per diabetes patient 4 . Although adherence to medication is typically conceptualized and studied as a dichotomy (adherent/nonadherent), most patients are likely partially adherent with their medication and show fluctuations in level of adherence over time 5 . Objective Assess adult type 2 diabetes patients whose adherence status to oral antidiabetics changed from one year to the next (adherent to nonadherent or nonadherent to adherent), and identify predictors of change in adherence status. Methods Study Design: A retrospective claims analysis of Medicare Advantage with prescription drug insurance (MAPD) patients. Data Source: Health insurance claims (both pharmacy and medical) from Humana, a large managed care organization in the United States. Inclusion and Exclusion Criteria: 1. Enrollment in a Humana MAPD plan for the duration of the years 2010, 2011, and 2012 2. At least 19 years old as of January 1, 2010 3. At least two prescription claims for medication(s) across any of the OAD classes during the each of the years within the study period. 4. Patients are only included in the PDC calculation if the first fill of their medication occurs at least 91 days before the end of the enrollment period in each year of 2010, 2011 and 2012. 5. Insulin users were excluded. Outcomes: - Adherence was evaluated dichotomously for each study period using and 80% threshold. Patients with a proportion of days covered (PDC) 80% during a given period were “Adherent” and those with a PDC <80% were “Non-adherent”. PDC was calculated as defined in the Medicare Health & Drug Plan Quality and Performance Ratings 2013 Part C & Part D Technical Notes. This measure is adapted from the Medication Adherence-Proportion of Days Covered measure which was developed and endorsed by the Pharmacy Quality Alliance (PQA). Limitations The results of this study were based on administrative claims data from a large national health plan. Retrospective database studies using administrative claims are prone to coding errors of omission and commission and incomplete claims information. The transferability of the predictive models developed as part of this study may be somewhat limited. The predictive models may not perform as well in other populations of diabetic patients when the available data is substantially different than that from the medical, pharmacy, and enrollment data used to develop these models. The models used a uniform profit/loss matrix. Future studies may demonstrate improved model performance if more accurate profit-loss values are incorporated. Conclusions Almost one-third of the OAD users demonstrated an adherence status change from one year to the next, with slightly more of those users being adherent in the baseline year and shifting to a nonadherent status in the follow-up year. This represents a opportunity for health plans to positively impact the medication utilization behavior of its membership prescribed chronic medications. This study demonstrated that predictive models may provide a valuable method of identifying those likely to have an adherence status change from a large pool of health plan members using OADs for the treatment of diabetes. The three models evaluated demonstrated a fair amount of consistency in the predictors identified as being important for identifying changes in adherence. These predictors are readily available in most payers data platforms and the use of these types of models could help health plans improve their allocation of resources used to address adherence measures which are incorporated in quality measures (i.e. CMS Star measures). References 1. Economic costs of diabetes in the U.S. In 2007. Diabetes Care, 2008. 31(3): p. 596-615. 2. CDC, National diabetes fact sheet: national estimates and general information on diabetes and prediabetes in the United States, 2011. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention. 2011. 3. Hepke, K.L., M.T. Martus, and D.A. Share, Costs and utilization associated with pharmaceutical adherence in a diabetic population. Am J Manag Care, 2004. 10(2 Pt 2): p. 144-51. 4. Sokol, M.C., et al., Impact of medication adherence on hospitalization risk and healthcare cost. Med Care, 2005. 43(6): p. 521-30. 5. Cramer, J.A., A systematic review of adherence with medications for diabetes. Diabetes Care, 2004. 27(5): p. 1218-24. Academy of Managed Care Pharmacy Nexus 2014| Boston, MA October 7-10, 2014 Results Jan 1, 2010 Dec 31, 2012 Jan 1, 2011 Study Period: Jan 1, 2010 to Dec 31, 2012 Baseline Adherence Status (2010 PDC) Jan 1, 2012 Training and Validation Models Test Models Figure 1. Study Design Model 1 Model 1A Model 1B Description Adherent/Nonadherent Baseline Adherence Status Change Adherent Baseline – Non- Adherent at Follow-up Nonadherent Baseline Adherence at Follow-up Key Predictors Variable 1 Total Number of Rxs Filled With 90 days Supply Total Number of Rxs Filled With 90 days Supply Diabetes-related Pill Burden in 2010 Variable 2 Diabetes-related Pill Burden in 2010 Diabetes-related Pill Burden in 2010 Longest Gap in OADs During 2010 Variable 3 Longest Gap in OADs During 2010 Longest Gap in OADs During 2010 Monthwise Patient Oscillation From Adherent to non-Adherent and Vice Versa- Year 2010 Variable 4 Monthwise Patient Oscillation From Adherent to non-Adherent and Vice Versa- Year 2010 Total Number of Anti-Diabetic Classes Filled Total Number of Rxs Filled With 90 days Supply Variable 5 Copay For The Last OAD Filled Copay For The Last OAD Filled Total Pill Burden in 2010 Key predictors were based on the importance scores calculated by the model. Table 3. Key Predictors of Adherence Status Change from Baseline Year 2010 to Follow-up Year 2011 Overall, 30% of the members had an adherence status change. Although patients with and without adherence status change differed significantly on each of the studied baseline characteristics, the groups’ differences were very small in terms of clinical significance. Table 4. Test Model Performance The negative predictive values (NPV) across the three models were high. These NPVs demonstrate that the models were able to identify patients who are not likely to change their adherence status from one year to the next with a high level of probability (71-86%). Research Collaboration Figure 2. Validation Model Sensitivity and Specificity Rates Model 1 Model 1A Model 1B Cutoff 0.32 0.20 0.47 Sensitivity 66.76% 79.68% 43.84% Specificity 47.49% 44.48% 76.83% Misclassification Rate 46.66% 46.00% 35.13% Positive Predictive Value 35.66% 34.71% 51.84% Negative Predictive Value 76.62% 85.53% 70.63% Table 2. Demographic and Clinical Characteristics 2010 2011 Criterion Population Excluded Population Remaining All patients with OADs during 2010/2011/2012 1,472,820 All patients with >= 2 OADs during 2010/2011/2012 552,238 920,582 Continuous Enrollment 584,524 336,058 Age > 19 198 335,860 No Socioeconomic Data Available* 97,458 238,402 FINAL SAMPLE 238,402 *Socioeconomic data (i.e. net worth, marital status, etc.) was provided by KBM, a third party consumer data vendor. 2012 Follow-up Adherence Status (2011 PDC) Follow-up Adherence Status (2012 PDC) Baseline Adherence Status (2011 PDC) Statistical Analyses: - Multivariable regression with gradient boosting trees were used were used to develop training and validation predictive models using baseline adherence status (2010 PDC) and demographic, clinical, and socioeconomic inputs to predict the follow-up period adherence status (2011 PDC). Three models were developed: Baseline Adherent and Nonadherent (Model 1), Baseline Adherent only (Model 1A), and Baseline Nonadherent only (Model 1B). - Models were selected based on the validation misclassification rate and their performance was evaluated by baseline adherence status (2011 PDC) to predict the follow-up period adherence status (2012 PDC). Methods – cont. 0 20 40 60 80 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Cutoff Model 1 True Positive Rate - Sensitivity True Negative Rate - Specificity 0 20 40 60 80 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Cutoff Model 1B True Positive Rate - Sensitivity True Negative Rate - Specificity 0 20 40 60 80 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Cutoff Model 1A True Positive Rate - Sensitivity True Negative Rate - Specificity Characteristic Status Change No Status Change P-value n (%) 70,604 (29.6) 167,798 (71.4) - Baseline Adherence Status – n (%) Adherent 36,948 (52.3) 112,042 (66.8) - Nonadherent 33,656 (47.7) 55,756 (32.2) - Age, years - mean (s.d.) 71.6 (8.9) 71.8 (9.0) <0.0001 Male – n (%) 32,429 (45.9) 76,129 (45.3) 0.0120 Race - n (%) White 57,083 (80.8) 133,124 (79.3) <0.0001 Black 7,276 (10.3) 19,470 (11.6) Hispanic 1,299 (1.8) 3,429 (2.0) Other 2,166 (3.0) 4,949 (2.9) Unknown 2,780 (3.9) 6,826 (4.0) Geographic Region - n (%) Northeast 3,983 (5.6) 9,452 (5.6) <0.0001 Midwest 18,889 (26.7) 43,563 (25.9) South 39,520 (55.9) 96,593 (57.5) West 8,209 (11.6) 18,178 (10.8) Unknown 3 (0.0) 12 (0.0) Population Density – n (%) Urban 41,222( 58.38) 97,412 (58.0) 0.0095 Suburban 18,253 (25.8) 43,231 (25.7) Rural 10,691 (15.1) 26,213 (15.6) Unknown 438 (0.6) 942 (0.5) Plan Characteristics - n (%) Low Income Subsidy (LIS) Status Only 5,131 (7.2) 14,139 (8.4) <0.0001 Dual Eligibility Only 202( 0.2) 502 (0.3) 0.5914 LIS Status and Dual Eligibility 6,195 (8.7) 17,606(10.4) <0.0001 Copay of last OAD during the baseline period – mean (s.d.) $14 ($48) $16 ($54) <0.0001 median, I.Q. range $3, $8 $4, $9 Number of antidiabetic classes used - mean (s.d.) 1.5 (0.6) 1.4 (0.6) <0.0001 Use of each antidiabetic class - n (%) Biguanides 55,068 (78.0) 131,636 (78.4) <0.0001 Sulfonylureas 34,836 (49.3) 93,500 (55.7) <0.0001 Thiazolidinediones 10,442 (14.7) 33,166 (19.7) <0.0001 Use mail-order pharmacy for antidiabetic medications - n (%) 45,691 (64.7) 87,816 (52.3) <0.0001 Claims for 90 days’ supply of antidiabetic meds - mean (s.d.) 2.1 (2.7) 2.4 (2.6) <0.0001 Pill Burden – mean (s.d.) Diabetes-related pill burden 1.1 (0.5) 0.9 (0.4) <0.0001 Total pill burden 5.2 (2.5) 5.0 (2.4) <0.0001 Longest Medication Gap [Days] - mean (s.d.) 22.6 (31.2) 26.5 (32.67) <0.0001

description

adsgfjhgmkl

Transcript of Poster

  • Table 1. Sample Selection

    PREDICTORS OF CHANGE IN ADHERENCE STATUS FROM ONE YEAR TO THE NEXT AMONG PATIENTS WITH TYPE 2 DIABETES MELLITUS ON ORAL ANTI-DIABETIC DRUGS

    S. Lane Slabaugh, PharmD, MBA1; Xiaomei Peng, MD, PhD2; Vishal Saundankar, MS1; Anthony Louder, PhD, RPh1; Haoda Fu, PhD2; Peinie Young, PharmD1; Angel Rodriguez, MD2; Ayad K. Ali, RPh, PhD2; Haya Ascher-Svanum, PhD2

    1. Comprehensive Health Insights, Humana, Louisville, KY, USA; 2. Eli Lilly and Company, Indianapolis, IN, USA

    Background Diabetes is a leading cause of morbidity, mortality and medical resource utilization in the United States1. Treatment is aimed at keeping blood glucose levels close to normal and preventing or delaying complications. However, it has been estimated that only 50% of US diabetes patients achieve an A1C level 19 198 335,860

    No Socioeconomic Data Available* 97,458 238,402 FINAL SAMPLE 238,402 *Socioeconomic data (i.e. net worth, marital status, etc.) was provided by KBM, a third party consumer data vendor.

    2012

    Follow-up Adherence Status (2011 PDC)

    Follow-up Adherence Status (2012 PDC)

    Baseline Adherence Status (2011 PDC)

    Statistical Analyses: - Multivariable regression with gradient boosting trees were used were used to develop training and validation predictive models

    using baseline adherence status (2010 PDC) and demographic, clinical, and socioeconomic inputs to predict the follow-up period adherence status (2011 PDC). Three models were developed: Baseline Adherent and Nonadherent (Model 1), Baseline Adherent only (Model 1A), and Baseline Nonadherent only (Model 1B).

    - Models were selected based on the validation misclassification rate and their performance was evaluated by baseline adherence status (2011 PDC) to predict the follow-up period adherence status (2012 PDC).

    Methods cont.

    0

    20

    40

    60

    80

    100

    0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

    Cutoff

    Model 1

    True Positive Rate - SensitivityTrue Negative Rate - Specificity

    0

    20

    40

    60

    80

    100

    0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

    Cutoff

    Model 1B

    True Positive Rate - SensitivityTrue Negative Rate - Specificity

    0

    20

    40

    60

    80

    100

    0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

    Cutoff

    Model 1A

    True Positive Rate - SensitivityTrue Negative Rate - Specificity

    Characteristic Status Change No Status Change P-value n (%) 70,604 (29.6) 167,798 (71.4) - Baseline Adherence Status n (%) Adherent 36,948 (52.3) 112,042 (66.8) - Nonadherent 33,656 (47.7) 55,756 (32.2) - Age, years - mean (s.d.) 71.6 (8.9) 71.8 (9.0)