Change in Claims-Based Diabetes Medications Is a Diabetes Improvement Indicator

Background:

Because laboratory test results are less available to researchers than claims data, a claims-based indicator of diabetes improvement would be valuable.

Objectives:

To determine whether a decrease in medication use for diabetes parallels clinical improvement in glycemic control.

Study Design:

This was a retrospective cohort study using up to 3.5 years of pharmacy and laboratory data from 1 private insurer. Data included 104 patients with diabetes who underwent bariatric surgery and had at least 1 glycated hemoglobin (A1C) test before and after surgery.

Methods:

We assigned each A1C test to a 90-day interval before or after surgery. Medication availability was noted for the midpoint of the interval (on insulin, on oral medications, count of medications). Each subject could contribute 1 presurgery and up to 3 postsurgery observations. We recorded the changes in A1C test results and medication use from the presurgery to the postsurgery period. Using the A1C test as the reference standard, positive and negative predictive values of medication-based indicators were calculated.

Results:

After bariatric surgery, A1C test values decreased by more than 1% and the count of unique medications decreased by 0.6. All 3 medication-based indicators had high positive predictive values (0.85) and low negative predictive values (0.20), and count of medications had better performance than the other indicators.

Conclusions:

Without clinical information, a decrease in use of medications can serve as a proxy for clinical improvement. Validation of results in other settings is needed.

Am J Manag Care. 2013;19(7):572-578

Diabetes is a major health concern in the United States given its high prevalence and significant morbidity. Many interventions have been implemented to improve glycemic control, especially those targeting weight loss.^1-5 Glycated hemoglobin (A1C) is the standard method of monitoring diabetes mellitus control, including responses to treatment.^6-8 In health services research, clinical data often are not available to researchers, particularly in studies that use administrative claims for outcomes evaluation.^9,10

Claims data have been used to describe the outcomes of interventions or as a method for risk stratification among individuals with diabetes.^11-15 However, it is hard to know whether glycemic control has improved without access to laboratory test data. Therefore, in this study we determined whether a decrease in the use of medications for diabetes parallels clinical improvement in glycemic control among individuals with substantial weight loss. We used individuals with diabetes who had undergone bariatric surgery, as they are expected to have improvement in A1C after surgery.^16-19 We hypothesized that a decrease in medication use from the presurgical to the postsurgical period would parallel an improvement in A1C values over the same period.

MATERIALS AND METHODS

This was a retrospective cohort study using up to 3.5 years of pharmacy and laboratory data (up to 1 year before and 2.5 years after bariatric surgery).

Data

We accessed claims and laboratory data from 1 Blue Cross Blue Shield plan, as part of a large ongoing collaborative effort. We received data on all members who met at least 1 of the following inclusion criteria at any point during 2002 through 2006: (1) completed a health risk assessment that captured member height and body weight; (2) had an obesity diagnosis reported in aclaim; (3) had a paid or denied claim for bariatric surgery; (4) had a paid or denied claim for a medication for promoting weight loss; or (5) were more than 12 years of age and had a diagnosis of hyperlipidemia, type 2 diabetes, sleep apnea, gall bladder disease or surgery, or metabolic syndrome. These diagnoses were identified in the claims by Current Procedural Terminology codes, International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) codes, and National Drug Codes or Diagnosis-Related Group codes.

We acquired (1) enrollment files from administrative data; (2) benefits information to determine medical and pharmacy coverage; (3) inpatient, outpatient, and pharmacy claims records containing ICD-9-CM diagnosis, Current Procedural Terminology codes, prescription National Drug Codes, and the starting date and length of prescriptions; and (4) laboratory records containing titles of test, test dates, and results in a subset of the population.

Sample Selection

We restricted analyses to enrollees with bariatric surgery and diabetes. We considered patients to have had bariatric surgery if they had appropriate ICD-9-CM procedure codes or Current Procedural Terminology codes.²⁰ For 4 of the ICD-9-CM codes, we required that a code for morbid obesity (ICD-9-CM 287.1) be present concurrently with the procedure code. We defined individuals as having type 2 diabetes if they had 1 relevant inpatient code or 2 outpatient codes separated by at least 30 days. The relevant codes were 250.xx, 648.0 (diabetes mellitus with pregnancy), 362.0 (diabetic retinopathy), or 266.41 (diabetic cataract). Individuals with only 250.x3 (type 1 diabetes) were not included. Additionally, any individual filling a prescription for a medication for treatment of hyperglycemia was included. If the prescription was for metformin alone, the individual was also required to have an ICD-9-CM code for diabetes for inclusion in this group.

A1C Test Value Calculation

We classified the dates of the laboratory tests into a 90-day period before surgery and a 90-day period after surgery (time zero). We required an individual to have pharmacy coverage throughout the entire 90-day period for his/her results to be evaluated in that period. We had only the month of the laboratory test result, so we arbitrarily assigned the middle of the month (day 15) to each record and then calculated the mean test result for each period. We limited observations to 360 days before and 900 daysafter surgery, and to enrollees with at least 1 presurgery and 1 postsurgery A1C value. We created pre-post test value pairs by matching any presurgery observation with any postsurgery observation from the same individual. We further restricted analyses to the latest presurgery observation and up to the 3 earliest postsurgery observations from the same individual.

Calculation of Drug in Hand

For a 90-day period, we used pharmacy claims to determine whether an enrollee had diabetes drugs in hand on the median day in that period. Diabetes drugs were classified into the following categories: alpha-glycosidase inhibitors, dipeptidyl peptidase inhibitors, exenatide, metformin, nateglinide, sulfonylureas, thiazolidinediones, and insulin. We calculated the number of unique medications as the count of medications from these categories that were in hand on the median day in a period.

Definition of the Change From Presurgery to Postsurgery Periods

Table 1

We defined change as a 3-level categorical variable: “decrease,” “same,” and “increase” (). For example, an enrollee would be assigned to the decrease category if he/she switched from having any diabetes medication in the presurgery period to no medication in the postsurgery period, or had a drop in the count of unique medications, or had a postsurgery A1C test value lower than the presurgery value by at least 0.4%.

We further collapsed these 3 categories into a binary variable: “favorable” or “unfavorable” (Table 1). Observations in the decrease category were considered favorable while those in the increase category were unfavorable. For observations classified into the “same” category, the use of no diabetes medications in both periods was classified as favorable, whereas others were unfavorable. For “same” observations of A1C test values, postsurgery test values lower than the presurgery values,by any amount, were considered favorable; the rest were unfavorable.

Statistical Methods

Using the binary change in A1C test results as a gold standard, we calculated the positive predictive value and negative predictive value for 3 binary medication-based indicators. We also calculated the Spearman’s correlation coefficient for 2 continuous variables: change in the count of unique medications and change in the test value (a simple difference between the presurgery and postsurgery values).

Sensitivity Analyses

eAppendix A

eAppendix B

eAppendix C

Given the high prevalence of being a nonuser of oral medication/insulin or having a zero count of medications in both periods, we tested the results while restricting the analyses to enrollees who were active users in at least 1 period to identify whether these medication-based indicators performed better among active users (restricted set; , available at www.ajmc.com). In addition, we also tested the outcomes with the sample limited to 1 pre-post observation pair (the latest presurgery observation and the earliest postsurgery observation; 104 pairs in total; , available at www.ajmc.com). Furthermore, when calculating drug in hand, instead of using the median day only, we also investigated whether a longer period (15 days; from a week before to a week after the median day in a period) would result in differences in the performance of medication-based indicators; the results are shown in (available at www.ajmc.com).

The data were de-identified in accordance with the Health Insurance Portability and Accountability Act’s definition of a limited data set. The Johns Hopkins University Office of Research Subjects deemed the study to be exempt from federal regulations because the research activities were considered to be of minimal risk to subjects, since the subjects were not identifiable.

RESULTS

Figure

The shows the flow chart for sample selection.

Characteristics of the Sample

Table 2

There were 208 paired presurgery and postsurgery observations (A1C and medication use) from 104 enrollees when including up to the 3 earliest postsurgery observations (). The mean age of the 104 eligible enrollees was 47.9 years, and females accounted for 80% of the sample. Of 208 pairs in the presurgery period, the mean A1C test value was 7.5%. About half of the individuals used oral medications, 12% used insulin, the average number of unique medications was 1.1, and 41% were nonusers. We observed reductions in medication use and A1C in the postsurgery period: the mean A1C value decreased to 6.1%, only 19% had oral medications, only 3% had insulin, the average number of unique medications was 0.4, and close to 80% were nonusers.

Distribution of the Change in Test Indicators

Table 3

For the 3 medication-based indicators, the majority of pairs were assigned to the same level categorically (54%, 90%, and 47% for the change in oral medication, insulin, and count of unique medications, respectively) (). Among the rest, more pairs were in the decrease category than in the increase category (39%, 10%, and 48% for the change in oral medication, insulin, and count of unique medications, respectively). For the change in A1C test value, about three-fourths were in the decrease category; only 7% were in the increase category.

As for binary indicators of change, more than 80% of 208 pairs had favorable change across all 4 measures (81%, 97%, 84%, and 87% for the change in oral medication, insulin, count of unique medications, and A1C value, respectively).

Performance of the Medication-Based Indicators

Table 4

All 3 medication-based indicators had high predictive positive values and low negative predictive values under most circumstances (). The positive predictive values were similar across all 3 indicators, with more variation in the negative predictive values. The predictive positive values of all 3 indicators were about 0.87. In analyses restricted to patients with any medication use, the positive predictive values ranged from 0.80 (insulin) to 0.91 (oral medication) (eAppendix A). The negative predictive values ranged from 0.18 (oral medication) to 0.33 (insulin); however, the 95% confidence intervals associated with insulin were substantially wider than others. When we restricted the analyses to a single postsurgery observation for each individual, results were minimally changed.

When using a 15-day period to calculate drug in hand, we found that the actual counts of medication in hand from the same period increased compared with those calculated by using just the median day; for example, the proportion of subjects on oral medication in the presurgery period among these 208 pairs increased from 50.5% to 59.1% (Table 1 in eAppendix C), which could be expected because the length of time used to determine drug in hand was much longer. However, when it came to the distribution and performance of the medication-based indicators, both sets generated very similar results (Tables 2 and 3 in eAppendix C).

All 3 medication-based indicators had very high sensitivity (>0.85) and very low specificity (<0.25). There was a modest but statistically significant correlation between the change in count of unique medications and the change in A1C test value. Spearman’s correlation coefficient was 0.24 without restriction (Table 4) and 0.33 after restriction (Table 2 in eAppendix A).

DISCUSSION

As expected, after bariatric surgery, patients had a decrease in both their A1C test value (a reduction of >1%) and medication use (a mean reduction of >0.6 in the count of unique medications). This favorable change across 3 medication-based indicators was highly predictive of the favorable change in the A1C test value; in contrast, unfavorable change was not a good predictor of higher A1C test values. The count of unique medications was the best indicator, given its relatively high and stable predictive values.

In our study, patients were taking oral medication in only half of the 208 observations in the presurgical period. Although not taking medication may indicate mild disease, it also may indicate poor adherence, which is prevalent among individuals with diabetes. A systematic review of 15 retrospective studies found that the adherence to oral hypoglycemic agent therapy ranged from 36% to 93% among patients on treatment from 6 to 24 months; insulin adherence was 62% to 64%.²¹ A different meta-analysis showed that the average adherence to medication in patients with diabetes is 67.5%.²² Therefore, even though he 50% adherence in our study did not seem high, it is still comparable to what was found in other studies.

We focused on the pre-post change after bariatric surgery, which led to the high prevalence of favorable change in A1C results (the gold standard in this study). This focus contributed to the high positive predictive value and the low negative predictive value observed, because the positive predictive value is higher for a given test sensitivity if the prevalence of the outcome is higher. The positive and negative predictive values would be different in a different patient population. Therefore, results from this study may not be directly applied to other populations, and the validity of our findings under different settings should be tested.²³ We chose not to include patients who did not have bariatric surgery because the relationship between the counts of medications and A1C would be in the opposite direction. Individuals with diabetes rarely have improvement in A1C without intensification of medication; only significant weight loss or major lifestyle changes would allow this improvement to happen.

This study should be replicated with a larger data set so that the point estimates can be more precise and stable, especially among those with an unfavorable change in medication-basedindicators. Furthermore, various types of weight-loss interventions might have different impacts on the degree of the change in the medication use; while subgroup analysis was not possible in the current study given the small sample size, it can be investigated further in a future study. An additional limitation of this study is that we did not account for intensification of medication that may be accomplished by changing the dose strength or the frequency of dosing; thus, we may have misclassified individuals as having a stable dose of medication when they actually increased their daily dose. In addition, we did not closely examine changes in medication class. Instead, we took only a simple look at this topic (eg, changing from insulin/oral medication to no medication); further study is necessary to address this issue.

CONCLUSIONS

This study shows that medication-based indicators may serve as a proxy to laboratory measures for interventions that target weight loss—especially when there is significant weight loss, such as after bariatric surgery.^24,25 However, when an unfavorable result is observed, further validation is necessary. This validation study contributes to the literature on the use of medication-based indicators as outcomes for research purposes.Author Affiliations: From Department of Health Policy and Management (H-YC, TMR, JPW), Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD; Department of Medicine (JMC, JBS), School of Medicine, Johns Hopkins University, Baltimore, MD; Department of Epidemiology (JMC, JBS), Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD.

Funding Source: None.

Author Disclosures: The authors (H-YC, TMR, JMC, JPW, JBS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (H-YC, JBS); acquisition of data (TMR, JMC, JPW, JBS); analysis and interpretation of data (H-YC, TMR, JMC, JBS); drafting of the manuscript (H-YC); critical revision of the manuscript for important intellectual content (TMR, JMC, JPW, JBS); statistical analysis (HYC, TMR); obtaining funding (JMC, JPW, JBS); administrative, technical, or logistic support (TMR); and supervision (JBS).

Address correspondence to: Hsien-Yen Chang, PhD, Department of Health Policy and Management, Bloomberg School of Public Health, Johns Hopkins University, 624 N Broadway, Baltimore, MD 21205. E-mail: hchang2@jhsph.edu.1. Anderson JW, Conley SB, Nicholas AS. One hundred pound weight losses with an intensive behavioral program: changes in risk factors in 118 patients with long-term follow-up. Am J Clin Nutr. 2007;86(2): 301-307.

2. Fernandez A, Schillinger D, Warton EM, et al. Language barriers, physician-patient language concordance, and glycemic control among insured Latinos with diabetes: the Diabetes Study of Northern California (DISTANCE). J Gen Intern Med. 2011;26(2):170-176.

3. Murata GH, Duckworth WC, Shah JH, Wendel CS, Mohler MJ, Hoffman RM. Blood glucose monitoring is associated with better glycemic control in type 2 diabetes: a database study. J Gen Intern Med. 2009;24(1):48-52.

4. Blonde L. Improving care for patients with type 2 diabetes: applying management guidelines and algorithms, and a review of new evidence for incretin agents and lifestyle intervention. Am J Manag Care. 2011;17(suppl 14):S368-S376.

5. Samsa GP, Kolotkin RL, Williams GR, Nguyen MH, Mendel CM. Effect of moderate weight loss on health-related quality of life: n analysis of combined data from 4 randomized trials of sibutramine vs placebo.Am J Manag Care. 2001;7(9):875-883.

6. Walker EA, Shmukler C, Ullman R, et al. Results of a successful telephonic intervention to improve diabetes control in urban adults: a randomized trial. Diabetes Care. 2011;34(1):2-7.

7. Chen HS, Wu TE, Jap TS, Chen RL, Lin HD. Effects of health educationon glycemic control during holiday time in patients with type 2 diabetes mellitus. Am J Manag Care. 2008;14(1):45-51.

8. Coberley C, Hamar B, Gandy B, et al. Impact of telephonic interventions on glycosylated hemoglobin and low-density lipoprotein cholesterol testing. Am J Manag Care. 2007;13(4):188-192.

9. Chang HY, Clark JM, Weiner JP. Morbidity trajectories as predictors of utilization: multi-year disease patterns in Taiwan’s National Health Insurance Program. Med Care. 2011;49(10):918-923.

10. Chang HY, Weiner JP. An in-depth assessment of a diagnosis-based risk adjustment model based on national health nsurance claims: the application of the Johns Hopkins Adjusted Clinical Group case-mix system in Taiwan. BMC Med. 2010;8:7.

11. Chang HY, Weiner JP, Richards TM, Bleich SN, Segal JB. Validating the adapted Diabetes Complications Severity Index in claims data. Am J Manag Care. 2012;18(11):721-726.

12. Pladevall M, Williams LK, Potts LA, Divine G, Xi H, Lafata JE. Clinical outcomes and adherence to medications measured by claims data in patients with diabetes. Diabetes Care. 2004;27(12):2800-2805.

13. Nrishnarajah G, Bhosle M, Chapman R. Health care costs associated with treatment modification in type 2 diabetes mellitus patients taking oral anti-diabetic drugs. Manag Care. 2011;20(6):42-48.

14. Hong JS, Kang HC. Relationship between oral antihyperglycemic medication adherence and hospitalization, mortality, and healthcare costs in adult ambulatory care patients with type 2 diabetes in South Korea. Med Care. 2011;49(4):378-384.

15. Chang HY, Weiner JP, Richards TM, Bleich SN, Segal JB. Predicting costs with diabetes complications severity index in claims data. Am J Manag Care. 2012;18(4):213-219.

16. Nocca D, Guillaume F, Noel P, et al. Impact of laparoscopic sleeve gastrectomy and laparoscopic gastric bypass on HbA1c blood level and pharmacological treatment of type 2 diabetes mellitus in severe or morbidly obese patients: results of a multicenter prospective study at 1 year. Obes Surg. 2011;21(6):738-743.

17. Lee WJ, Chong K, Ser KH, et al. Gastric bypass vs sleeve gastrectomy for type 2 diabetes mellitus: a randomized controlled trial. Arch Surg. 2011;146(2):143-148.

18. Kim S, Richards WO. Long-term follow-up of the metabolic profiles in obese patients with type 2 diabetes mellitus after Roux-en-Y gastric bypass. Ann Surg. 2010;251(6):1049-1055.

19. Padwal R, Klarenbach S, Wiebe N, et al. Bariatric surgery: a systematic review of the clinical and economic evidence. J Gen Intern Med. 2011;26(10):1183-1194.

20. Steele KE, Prokopowicz GP, Chang HY, et al. Risk of complications after bariatric surgery among individuals with and without type 2 diabetes mellitus. Surg Obes Relat Dis. 2012;8(3):305-330.

21. Cramer JA. A systematic review of adherence with medications for diabetes. Diabetes Care. 2004;27(5):1218-1224.

22. DiMatteo MR. Variations in patients’ adherence to medical recommendations: a quantitative review of 50 years of research. Med Care.2004;42(3):200-209.

23. Irwig L, Bossuyt P, Glasziou P, Gatsonis C, Lijmer J. Designing studies to ensure that estimates of test accuracy are transferable. BMJ. 2002;324(7338):669-671.

24. Adams TD, Davidson LE, Litwin SE, et al. Health benefits of gastric bypass surgery after 6 years. JAMA. 2012;308:1122-1131.

25. Himpens J, Verbrugghe A, Cadiere GB, et al. Long-term results of laparoscopic Roux-en-Y gastric bypass: evaluation after 9 years. Obes Surg. 2012;22(11):1586-1593.