Effects of intensified care management activities and diabetes medication copayment reduction on medication adherence and health care costs

Diabetes mellitus (hereinafter referred to as diabetes) is a serious health concern affecting the daily lives of many Americans in both clinical and financial aspects. Diabetes affects approximately 26 million people of all ages in the United States, and the total estimated medical costs of diagnosed diabetes in the U.S. in 2012 exceeded $200 billion. To obtain better health outcomes, prevent complications, and reduce unnecessary costs, some health insurance plans encourage patients to enroll in diabetes management programs that monitor patient's health status more closely and assist in the adoption of healthy behaviors and habits. The aims of this study are to compare medication adherence rates and total healthcare cost among patients participating in a Diabetes Care Management Incentive Program offered by a commercial health insurer with usual care. This study was performed using a retrospective cohort study design; subjects were insurance plan members with diabetes using metformin-containing medications. Logistic regression analyses were performed to measure the degree of association between intervention status (i.e. participation in the diabetes incentive program) and adherence rates. The adjusted odds ratio with 95% confidence intervals were reported as the measure of effect. For the total healthcare cost analysis, the Mann-Whitney U test was utilized to evaluate differences between the median intervention and non-intervention cost values. Odds ratios for rates of achieving medication possession ratio (MPR) of 0.80 or greater among the intervention group as compared with the nonintervention groups were 0.966 (95% CI: 0.739 1.264) in the bivariate logistic regression model, 0.995 (95% CI: 0.755 -1.312) in the full logistic regression model, and 1.008 (95% CI: 0.765 1.328) in the fitted logistic regression model. Additionally, the mean annual total healthcare cost was $8,827.01 ($735.58 per month) in the intervention group and $10,096.53 ($841.38 per month) in the nonintervention group), yet the difference was not statistically significant (p = 0.2327). Study results indicate that the medication adherence rates among patients using metformin-containing medications were similar between members who were enrolled in the diabetes management program and members who were not enrolled in the program. However, members participating in the program incurred approximately $2,200 less in annual total healthcare cost.

This study will help in fulfilling pharmacoeconomic-research demands about diabetes management programs and contribute to the development of more economically efficient diabetes management programs. The study's hypothesis is that the members enrolled in the diabetes management program will achieve higher medication adherence rate than those who are not enrolled in the diabetes management program. A secondary hypothesis is that the members enrolled in the diabetes management program will experience reduced total healthcare expenditure compared to the members nor enrolled in the diabetes management program.

METHOD
This study was a retrospective cohort study, assessing the association between a diabetes management program and its effects on participants' medication adherence and medical costs. The data were provided by a commercial health insurer, and included members' medical diagnoses, basic demographics, medical and pharmacy spending information, and health care procedures. compared between intervention and non-intervention groups.
The distribution of costs was analyzed prior to determining the appropriate statistical test for evaluating group differences in healthcare spending. Mean costs by group were reported, and where data were skewed, a log transformation was performed. The student's t-test was used to assess mean differences in cost, and given the skewed nature of the data, median-based tests were also performed to assess the statistical significance of these cost differences.
These costs analyses included the same independent variables from the adherence analysis: age, gender, comorbidities, insulin usage, and total number of medications dispensed. To evaluate the statistical significance of differences in these baseline characteristics and health expenditure, the students t-test was utilized. The Mann-Whitney U test was also utilized to evaluate differences between the median intervention and non-intervention cost values.
Data analysis was performed using SAS (version 9.3).

RESULTS
A total of 284 intervention members and 5,528 non-intervention members met all the cohort selection criteria (see Figure A). The intervention group had a mean age of 54.06 years (SD=9.50) and the non-intervention group had a mean age of 54.59 years (SD=8.59) (see Table 1).About one-quarter in the intervention group (26.06%, n=74) and the non-intervention group (25.92%, n=1,433) were between 18-49 years of age. The frequency of the population aged50-64 years was 61.97% in the intervention group and 64.36%, while the frequency of population who is 65 years and above was 11.97% in the intervention group and 9.71% in the non-intervention group. Percent differences across age groups were not statistically significant (p=0.4396).
The frequency of males was 61.27% in the intervention group and 61.09% in the non-intervention group, and the frequency of females was 38.73% in the intervention group and 38.91% in the non-intervention group. Although it was not a statistically significant difference, the frequency of insulin use was 39.79% in the intervention group and 35.42% in the non-intervention group (p=0.952).
Differences in the prevalence of the comorbidities of respiratory disease and cardiovascular disease were statistically significant between the two groups (p=0.0409 Respiratory disease, p=0.0026 Cardiovascular disease); 7.04% of the intervention group and 10.89% of the non-intervention group had a respiratory disease, while 7.04% of the intervention group and 13.17% of the non-intervention group had a documentation of cardiovascular disease, respectively. Additionally, 15.49% of the intervention group and 14.35% of the non-intervention group had a mental health disorder, although this difference was not statistically significant (p=0.5920).The frequency of thoseusing1 to 7 total medications was 25.00% in the intervention group and 20.98% in the non-intervention group, and the frequency of those using 8 to 13 total medications was 29.58% in the intervention group and 31.35% in the non-intervention group. The frequency of those using 14 to 17 total medications was 16.90% in the intervention group and 16.75% in the nonintervention group, and the frequency of those using more than 18 total medications was 28.52% in the intervention group and 30.92% in the nonintervention group. The frequency based on the number of total medications was not statistically significant, and both intervention and non-intervention patients appeared to utilize a similar number of medications during the measurement period.

Results (Medication adherence)
Result assessing medication adherence revealed that 72.89% of the intervention group and 73.55% of the non-intervention group achieved MPR of 0.80 or greater, although this difference was not statistically significant (p=0.8042).
Mean ages of the subgroups who achieved MPR greater than or equal to 0.80 and who did not achieve MPR of 0.80 or higher were 55.24 and 52.64, respectively (p=<0.0001, SD=8.34 and 9.15) (see Table 2).Adherence rates increased with age: Being enrolled in the intervention was not associated with higher medication adherence rate in the bivariate logistic regression analysis(see Table 3). Based on the saturated logistic regression model, the impact of intervention group status was also not significantly associated with medication adherence (see Table 4). In the fitted multiple logistic regression model, the impact of intervention group status was not statistically significant (see Table 5). The mean total healthcare costs of male and female patients in the intervention group were $7,761.1 and $10,513.0 respectively, and the mean total healthcare costs of males and females in the non-intervention group were $9,785.9 and $10,583.1, respectively(see Table 7). Analysis of age groups revealed a mean total healthcare cost of intervention and non-intervention population aged 18 to 49 years to be $7,015.2 and $8,244.1, respectively. For those who were 50 to 64 years old, intervention group had the mean total healthcare costs of $9,117.0, and nonintervention group had the mean total healthcare costs of $10,551.7. The mean total healthcare costs of intervention and non-intervention population who are 65 years old and above were $11,269.4 and $12,028.0, respectively. All the mean total healthcare cost results based on different age groups were not statistically significant. For the total medication counts, the mean total healthcare costs of intervention and non-intervention members who are taking 1 to 7 total medications were $4,495.2 and $6,261.6, respectively. For those who are taking8 to 13 total medications, intervention group had the mean total healthcare costs of $6,836.4, and non-intervention group had the mean total healthcare costs of $7,661.0. The mean total healthcare costs of intervention and non-intervention members who are taking 14 to 17 total medications were $7,689.9 and $9,380.5, respectively. The mean total healthcare costs of intervention and non-intervention members who are taking 18 or more total medications were $15,362.3 and $15,548.9, respectively.

DISCUSSION
Diabetes mellitus is a serious obstacle for the United States health care system in both clinical and financial terms. Affecting approximately 8.3% of the U.S. population, it is the seventh leading cause of death, and is a major cause of cardiovascular diseases, the first leading cause of death in the U.S. 1,22 Discussion (Medication adherence) The first goal of this study was to determine whether the participation in a  Grenard et al (2011) found that depressed patients are 1.76 times more likely to be non-adherent to their medications compared to patients without depression. 23 Therefore, the diabetes management program should specifically target depressed patients to prevent patients' disengagement and improve program's performance.
Other independent characteristics including gender were not associated with medication adherence. The subgroup that used insulin was less likely to achieve MPR of 0.80 or greater, although this association was not consistent across the three logistic regression models. In the fitted logistic regression model, the odds ratio for insulin use was 0.874 (95% CI: 0.769 -0.993), (p=0.0379). This result is maybe due to the relationship between diabetes severity and insulin usage.
If diabetes worsens, patients tend to switch to insulin therapy from oral diabetes medications including metformin.
The results of this study indicate that an incentive-based diabetes management program did not yield increased rates of medication adherence among participants when compared with rates among members with diabetes not participating in the diabetes management program. Medication adherence rates were similar between the two groups in all three logistic regression statistical tests, which indicates that the program likely did not yield clinical benefits as a consequence of more consistent medication taking. However, medication adherence rates were already fairly high among all patients included in this study, suggesting that the opportunity for improvement was limited.
Discussion (Healthcare cost) The second aim of this study was to determine whether the participation of Total healthcare costs varied across several of the independent variables evaluated. As expected, older patients were more likely to have higher mean total healthcare cost than younger patients in both the intervention group and nonintervention group. Female patients were more likely to have higher mean total healthcare cost than male patients in both the intervention group and nonintervention group. Patients using insulin were more likely to have higher mean total healthcare cost than population not using any insulin in both intervention group and non-intervention group, reflecting the progressed disease status among these patients. Patients using more medications were more likely to have higher mean total healthcare cost than population using less numbers of medications in both intervention group and non-intervention group. Analysis of the relationship between patient age and total medication dispensings suggested that these two continuous variables were not highly correlated (Spearman r = 0.179), suggesting that both increasing age and a greater number of medications used were independently associated with higher health care expenditure.
As expected, patients with comorbidities (respiratory disease, mental health disorder, and cardiovascular disease) were more likely to have a higher mean total healthcare cost than population without comorbidities. When examining costs across comorbidity categories, the mean total healthcare costs of intervention and non-intervention population with respiratory diseases were $23,337.4 and $17,937.7, respectively; the mean total healthcare costs of intervention and nonintervention population with cardiovascular diseases were $24,627.6 and $19,503.9, respectively. These results reveal that older members with more comorbidities and more medications prescribed incurred greater total healthcare costs. Based on this finding, diabetes management program should consider focusing on older members with comorbidities and use of a greater number of medications to reduce total healthcare costs through the programs components.

Discussion (Limitations)
There were several limitations in this study. First, intervention and nonintervention groups were fundamentally different, and this difference prevented us to confirm that the diabetes management program solely contributed to intervention group's lower total healthcare costs. In the intervention group, subgroups with respiratory diseases and cardiovascular diseases were 7.04% and 7.04%, and in non-intervention group, subgroups with respiratory diseases and cardiovascular diseases were 10.89% and 13.17% (p=0.0409 and 0.0026). The intervention group was generally healthier than the non-intervention group.
Additionally, there were no comparisons to medication adherence rates or cost in prior months. Without having access to previous data, the study was unable to determine if the intervention was progressively improving members' medication adherence and costs. It is possible that the intervention may have provided greater gains or losses in medication adherence and costs from the previous year than the non-intervention group, yet we were not able to evaluate the progressive impacts on the diabetes management program. Moreover, the study's 12-month evaluation period may have been too limited to measure the intervention's impact on medication adherence and total healthcare costs.
Furthermore, the study was unable to determine the temporal relationships between examined variables due to the short time period for follow up. In this case, there is a possibility that some members could have been diagnosed with a cardiovascular disease on the last day of their enrollment periods of 12 months and would have been labeled as cardiovascular patients for the entire study period. This limitation could falsely increase cost burdens of participants with some comorbidities. An additional limitation is that the study only considered metformin and metformin-containing medications for adherence to diabetes medication. As metformin is usually the first line oral therapy agent for diabetes mellitus and nonmetformin medications are add-on therapies. Users of other oral diabetes medications such as sulfonylureas, DPP-4 inhibitors, or thiazolidinediones were excluded in this study. The medication adherence rates of this study, therefore, could over-represent healthier populations, and overall oral diabetes medication adherence rates may be different from the result.
Additionally, another study limitation is that the administrative data source only included information about paid claims and excluded any procedure or medication that were paid out-of-pocket. The study, also, assumed that members consumed the dispensed medications, but compliance to dispensed medications was not measured or assessed. There is a possibility that the study misclassified members that did not take medications that had been dispensed as adherent to their medications. Also, the data source did not include information about patient race, ethnicity, and socioeconomic factors, which may have been associated with both the independent variables studied and the outcomes of medication adherence, and total healthcare cost.
There were cost outliers, affecting average medical, pharmacy, and total healthcare cost values. The highest cost for the non-intervention group was $345,862.33 and for the intervention group was $166,091.36. To evaluate the two groups not affected by the outliers, a median-based test was performed in the statistical analysis. Due to these outliers in the cost analysis, standard deviations became higher than mean values, and analysis results turned out to be statistically not significant.
In addition, the lack of randomization may not avoid impacts from unidentified or unseen biases or confounders. Also, members deciding to enter the diabetes management program could have been more careful about their own health, and this may have possibly led to a selection bias.