- Email: [email protected]
Use of rosiglitazone and pioglitazone immediately after the cardiovascular risk warnings
Available online at www.sciencedirect.com
Research in Social and Administrative Pharmacy 8 (2012) 47–59
Original Research
Use of rosiglitazone and pioglitazone immediately after the cardiovascular risk warnings Rahul Jain, Ph.D.a,*, C. Daniel Mullins, Ph.D.b, Helen Lee, Pharm.D., M.B.A.c, Winston Wong, Pharm.D.c a
Department of Clinical and Administrative Pharmacy, College of Pharmacy, University of Georgia, 250 Green Street, Athens, GA 30602, USA b Department of Pharmaceutical Health Services Research, University of Maryland School of Pharmacy, 220 Arch Street, 12th Floor, Baltimore, MD 21201, USA c Pharmacy Management, CareFirst BlueCross BlueShield, 1501 S. Clinton Street, 8th Floor, Mail Stop CT08-10, Baltimore, MD 21224, USA
Abstract Background: Meta-analyses of oral hypoglycemic agents (OHAs) revealed that rosiglitazone increased the risk of myocardial infarction (MI) and heart failure (HF) and that pioglitazone increased the risk of HF and decreased the risk of MI. Objective: To characterize the change in the pattern of use of OHAs immediately after the publication of these meta-analyses on May 21, 2007. Methods: Pharmacy and medical claims data for a managed care organization were analyzed for patients continuously enrolled from January 1, 2005, to November 30, 2007, with at least 1 pharmacy claim for OHA in the 13-month period between November 1, 2006, and November 30, 2007. A 5-month prepublication period (November 1, 2006, through March 31, 2007) was compared with a 5-month postpublication period (July 1, 2007, through November 30, 2007) using a differences-in-differences multinomial logistic regression. This regression explored discontinuation; continuation with monotherapy or adding another drug; and switching to a drug different from the index monotherapy drug after adjusting for gender, age, type of insurance, past 1-year history of MI or HF, and risk factors for MI and HF in the past 1 year. Results: The relative rate of switching to nonindex drug in the postpublication relative to prepublication was 2.64 (P ¼ .046) for monotherapy rosiglitazone users and 0.72 (P ¼ .583) for monotherapy pioglitazone users. The differences-in-differences estimate of the rate of switching to nonindex drugs for monotherapy rosiglitazone users was 3.64 (P ¼ .090) times higher relative to the estimate for monotherapy pioglitazone users. Conclusion: The pattern of use differed fundamentally between monotherapy rosiglitazone users and users of all other monotherapy OHAs in the postperiod. Not only were monotherapy rosiglitazone patients switching to non-rosiglitazone drugs at a higher rate, but the rate also was more than 3 times higher than similar switches among monotherapy pioglitazone users in the postperiod relative to the preperiod. This shows that the market response as observed by patient/prescriber decisions to the adverse news was interpreted narrowly to monotherapy rosiglitazone, and there is little or no spillover to the other drugs. * Corresponding author. Tel.: þ1 706 542 5327; fax: þ1 706 542 5228. E-mail address: [email protected] (R. Jain). 1551-7411/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.sapharm.2010.12.003
48
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Therefore, this study found that there was a differential effect of meta-analyses on the use of the 2 drugs. Ó 2012 Elsevier Inc. All rights reserved. Keywords: Diabetes; Pattern of use; Differences-in-differences
Introduction Thiazolidinediones (TZDs) were introduced in the 1990s as a second-line option after lifestyle modifications and maximum tolerated doses of metformin were not effective for the management of type 2 diabetes mellitus. As a class, TZDs have a unique mechanism of action that involves insulin sensitivity in patients with type 2 diabetes.1 TZDs are used as monotherapy as well as along with other oral hypoglycemic agents (OHAs) in combination therapy.2-4 Chemically, its members are derivatives of the parent compound TZD and include rosiglitazone (Avandia), pioglitazone (Actos), and troglitazone (Rezulin). Troglitazone was introduced in the late 1990s but was found to be associated with an idiosyncratic reaction, leading to drug-induced hepatitis. It was withdrawn from the U.S. market on March 21, 2000.5 Initial research had shown that rosiglitazone6 and pioglitazone7 may have potentially been beneficial to cardiovascular health. However, subsequent research raised concerns regarding the adverse cardiac effect of TZDs. A retrospective study published in 2003 found that TZDs might increase the risk of heart failure (HF).8 Another retrospective case-control study published in 2005 found an association between TZD therapy and hospitalization for HF within 60 days of the prescription date.9 On May 21, 2007, Nissen and Wolski10 publisheda their findings of a metaanalysis. This study found that for the rosiglitazone group, relative to the control group, the odds ratio (OR) for myocardial infarction (MI) was 1.43 (95% confidence interval [CI] ¼ 1.031.98) and the OR for death from cardiovascular causes was 1.64 (95% CI ¼ 0.98-2.74). Two other meta-analyses have been published since then, which investigated the risk of cardiovascular events associated with rosiglitazone11 and pioglitazone.12 The first study found that among patients with type 2 diabetes, rosiglitazone use for at least 12 months is associated with
a
a significantly increased risk of MI (RR ¼ 1.42, 95% CI ¼ 1.06-1.91) and HF (RR [relative risk] ¼ 2.09, 95% CI ¼ 1.52-2.88), without a significantly increased risk of cardiovascular mortality (RR ¼ 0.09, 95% CI ¼ 0.63-1.26). The second study found that pioglitazone use is associated with a significantly lower risk of death, MI, or stroke (hazards ratio [HR] ¼ 0.82, 95% CI ¼ 0.72-0.94, P ¼ .005) among a diverse population of patients with diabetes. Serious HF is increased (HR ¼ 1.41, 95% CI ¼ 1.14-1.76, P ¼ .002) by pioglitazone use, although without an associated increase in mortality. In response to these findings, on August 14, 2007, the Food and Drug Association (FDA) issued a black box warning13 for both pioglitazone and rosiglitazone about increased HF risks. On November 14, 2007, the black box warning14 for rosiglitazone was updated to include warnings that the drug may cause MI. No such warning was issued for pioglitazone. Intuitively, the market response as observed by patient/prescriber decisions to the risk warning or “adverse news” could be to discount it entirely, interpret it narrowly with respect to the specific drug involved for each adverse event, or interpret it broadly and assume that the safety problems apply to other drugs in the same therapeutic class.
Objectives The primary objective of this study was to characterize the change in the pattern of use of OHAs in general and the TZDs specifically immediately after the publication of these metaanalyses. The rest of the article characterizes changes in the patterns of use, switching, and discontinuation among pioglitazone, rosiglitazone, metformin, and sulfonylurea users before and after the publication of the meta-analyses. Subsequently, the change in pattern of use of rosiglitazone
This article was published at www.nejm.org on May 21, 2007. Subsequently, it was published in the Journal on June 14, 2007.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
relative to that of pioglitazone is presented to evaluate if changes in use had differential effect across TZDs. Methods Study sample: Inclusion and exclusion criteria The study population consisted of beneficiaries covered by a managed care organization with approximately 3.4 million members with medical benefits, of whom 1.2 million members had pharmacy benefits within the states of Maryland, District of Columbia, and Northern Virginia. Only adult patients (aged 18 years and older) who were continuously enrolled from January 1, 2005, through November 30, 2007, with both medical and pharmacy benefits and at least 1 pharmacy claim for an OHA between November 1, 2006, and November 30, 2007, were included. Patients were excluded if they had a pharmacy claim for insulinb or any injectable (eg, exenatide) or newer agentsc (eg, sitagliptin or any combination of sitagliptin) during January 1, 2005, through November 30, 2007. In addition to these, the analysis concentrates only on patients on monotherapy TZDs. This is done for multiple reasons. First, patients on monotherapy TZDs are likely to be those who are either contraindicated or intolerant to other OHAs and whose glycemic control is not so poor as to require the use of insulin. Therefore, these patients are likely to face cardiovascular risks because of the use of TZDs and not because of poor glycemic control. If there were an effect of the meta-analysis on prescription pattern, it would be the lower bound of the effect on all TZD users. Second, in our data, more than 70% of the patients had a monotherapy index drug. Third, because our study is based on claims data, the severity of diabetes is not known; therefore, by concentrating only on monotherapy TZD patients, the analysis eliminates the bias because of b
49
unobservable severity of diabetes (omitted variable bias) and limits the confounding because of concomitant diabetes medicines. Hence, we concentrated only on the monotherapy users. Study periods This study examines the change in pattern of use of OHAs before and after the publication of meta-analyses. The pre-publication period was defined as November 1, 2006, through March 31, 2007. Similarly, the post-publication period was defined as July 1, 2007, through November 30, 2007. Data for the period from April 1, 2007, to June 30, 2007, were excluded to allow for new prescriptions written after May 21, 2007, to trickle into the data. Fig. 1 shows the timeline for preperiod and postperiod diagrammatically. Because some patients may have filled OHA prescriptions just before the beginning of prepublication period (November 1, 2006), a 60-day window (September 2, 2006, through October 31, 2006) allowed patients to consume the prescriptions that may have been filled just before November 1, 2007. The index drug for the prepublication period was then defined as the first OHA or first combination of OHAs after the 60day window. Similarly, for the post-publication period, a similar 60-day windowd (May 2, 2007, through June 30, 2007) was allowed, and the index drug for the post-publication period was then defined as the first OHA or first combination of OHAs after the initial 60-day period. Definition of continuation, switching, and discontinuation This study was a retrospective longitudinal data analysis of medical and pharmacy claims of patients on OHA. After the initial prescription, each patient may have refilled the same drug, may have added another drug, switched to another drug or combination of drugs, or discontinued completely. Therefore, in both pre-publication and
Patients on insulin were excluded because patients on insulin may have more advanced diabetes or may have type 1 diabetes, which could not be identified in our data set. c The FDA approved sitagliptin on October 17, 2006. Because it is a newer agent, a very small number of patients were on this medication. Therefore, it was excluded from the analysis to eliminate any small sample biases in the estimate. d This study is analyzing the effect of the publication of a meta-analysis on the prescribing patterns of pioglitazone and rosiglitazone. Hence, this study does not require new users; it only requires new prescriptions during the study time frame. Because most prescriptions were for 30 days, a 60-day window was allowed for patients to consume the prescriptions that were filled just before the publication of the meta-analysis. The mail order prescriptions were dealt with exactly as retail prescriptions.
50
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Meta-analysis released on May/21/07
Apr/01/07
Nov/01/06
Mar/31/07 Pre-Publication Period
Jun/30/07 Time Nov/30/07
Jul/01/07
Post-Publication Period
Fig. 1. Timeline for change in pattern of use.
post-publication periods, the following definitions were used to classify discontinuation and switch. Discontinuation If there was no subsequent fill or the period between their previous prescription and a new prescription for the index drug was longer than the number of days of supply plus 30 days, then discontinuation of therapy was recorded. Switch A switch was recorded for those who had a prescription claim for an OHA different from the index drug within 30 days after the date of their last prescription plus the number of days of supply. The drugs were identified using the product names and are enumerated in Table 1. A distinction was made between adding another drug to the index drug (switching from monotherapy to combination therapy) and switching to a drug entirely different from the index drug. Patients were censored at the date of discontinuation, switching, or at the end of the follow-up period. Analytical approach Covariates The analysis included demographic variables (age and gender) and type of insurance (Health Maintenance Organization [HMO], Preferred Provider Organization [PPO], Point of Service Plans [POS], or other). It also included the history of observable risk factors for MI15 and HF16 (see Table 2 for International Classification of Diseases [ICD]-9 codes) and history of MI (ICD-9 code 410.xx) and HF (ICD-9 code 428.xx). The history of the risk factors for MI and HF and past history of MI and HF were measured as at
least 1 hospitalization/diagnosis or an emergency room visit with the above-stated ICD-9 codes listed in any diagnosis claim in the past 365 days. Therefore, for the pre-publication period, these variables were measured between November 1, 2005, and October 31, 2006, and for the postpublication period, they were measured between July 1, 2006, and June 30, 2007. Other risk factors for MI, such as tobacco use, physical inactivity, and obesity, were not observed in the data and, therefore, not included in the analysis. Statistical analysis The change in pattern of use compared the rates of switching, discontinuation, or continuation of the index drug before and after the publication of meta-analyses. This implied comparing 2 multinomial variables, one for the pre-publication period and the other for the post-publication period. Clearly, comparison based on binomial distribution is not suitable. Therefore, simultaneous CIs for the changes in multinomial proportions between pre-publication and post-publication samples17,18 were estimated. The aim of this study was to elicit the effect of the meta-analyses on the pattern of use of TZDs. The meta-analyses implied strong adverse news for rosiglitazone. Therefore, it is expected that the pattern of use of rosiglitazone users would be affected after the publication of the meta-analyses. However, some of the change could also be because of time trend or changes in other factors that were not linked to the publication of the meta-analyses. Therefore, a simple comparison in pattern of use of rosiglitazone patients before and after the publication is not enough to bring out the true effect of these studies.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Hence, to identify the true effect of the metaanalyses, a comparator group is needed, which was not affected by the meta-analyses. This comparator group is referred to as the control group in the differences-in-differences literature. The underlying assumption in the differences-in-differences approach is that in absence of the adverse news, the gap in pattern of use between rosiglitazone users and the comparator group would remain the same over time. Because the aim was to also test whether the “market” interpreted these publications narrowly or broadly and the meta-analyses did not imply strong adverse news for pioglitazone, the pioglitazone patients become ideal candidates for the comparator group. Therefore, the change in pattern of use of rosiglitazone relative to that of pioglitazone was estimated by estimating a differences-in-differences multinomial logistic regression. The dependent variable was defined as the categorical variable (see Equation 1). This variable indicated whether the patient continued their monotherapy or added another drug, discontinued, or switched to nonindex drug.
factors for MI. Also included in the regression were the indicator for index drug is rosiglitazone, indicator that the observation is from postpublication period, and an interaction term of post and rosiglitazone. The underlying assumption in the differences-in-differences approach is that in absence of the adverse news, the gap between the pattern of use of rosiglitazone users and pioglitazone users over time would remain the same. The results of this regression would provide the overall effect of the publication of meta-analyses. The coefficient of the interaction term would capture the effect of the publications because after covariate adjustment the only source attributing to variations in the rate difference between the preperiod and the postperiod is the publication itself. In a linear regression model of differencesin-differences, the coefficient of the interaction term is the parameter of interest; however, in this nonlinear specification of multinomial logistic regression, the coefficient of the interaction term is not the estimator of the true interaction effect.19-21 Therefore, the results are presented as
8 < 1 if continued their monotherapy or added another drug Yi ¼ 2 if discontinued : 3 if switched to nonindex drug
Impact of studies on Rosi ¼
Impact of studies on Pio ¼
Predicted prob of switching to non rosi in POST period Predicted prob of switching to non rosi in PRE period
Prob of switching to non pio in POST period Prob of switching to non pio in PRE period
The independent variables included age, gender, past 1 year’s history of MI, and risk factors for MI. Because history of HF is also a risk factor for HF, it was not included in the regression. Additionally, hypertension is a risk factor for both MI and HF; therefore, including it in both the risk factors will give rise to multicollinearity problem. Hence, it was included only in the risk
51
ð1Þ
ð2Þ
ð3Þ
relative risks (see Equation 2 and Equation 3) and as the ratio of relative risks (Equation 4) Diff ¼
Impact of studies on rosi Impact of studies on pio
ð4Þ
If the probability of rosiglitazone users switching to a non-rosiglitazone drug increases after the
52
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 1 Product names of the OHAs
Table 2 Risk factors for MI and HF and their ICD-9 codes
OHA
Risk factor
Product name
Monotherapy drug’s product names Sulfonylurea Amaryl Chlorpropamide Diabeta Diabinese Glimepiride Glipizide Glipizide ER Glipizide XL Glucotrol Glucotrol XL Glyburide Glyburide Micronized Glynase Pres-Tab Micronase Tolazamide Tolbutamide Metformin Metformin Fortamet Glucophage Glucophage XR Glumetza Metformin ER Metformin HCL Metformin Hydrochloride Riomet Pioglitazone Actos Rosiglitazone Avandia Combination therapy drug’s product names Sulf-Met Glipizide/Metformin Glipizide/Metformin HCL Glucovance Glyburide Micronized/ Metformin HCL Glyburide Micronized/ Metformin Hydrochloride Glyburide/Metformin HCL Metaglip Pio-Met Actoplus Met Rosi-Met Avandamet Pio-Sulf Duetact Rosi-Sulf Avandaryl
Risks factors for MI Hypertension Essential hypertension Chronic venous hypertension High blood cholesterol Pure hypercholesterolemia Pure hyperglyceridemia Mixed hyperlipidemia Hyperchylomicronemia Other and unspecified hyperlipidemia Risk factors for HF Hypertension Essential hypertension Chronic venous hypertension Aortic and mitral valve disease Diseases of aortic valve Diseases of mitral valve Diseases of mitral and aortic valves Aortic coarctation Cardiac dysrhythmias Pulmonary hypertension Tricuspid valve disease Pulmonary valve disease Ischemic heart disease Cardiomyopathy
ICD-9 code
401.xx 459.3x 272.0 272.1 272.2 272.3 272.4
401.xx 459.3x 395.xx 394.xx 396.xx 747.10 427.xx 416.0x 397.0x 746.xx 410-414.xx 425.xx
and number of patients included depend on the inclusion/exclusion criteria and not necessarily powered to detect differences. Sensitivity analysis The main outcomes of the study, namely switching and discontinuation, are defined in terms of the number of days of supply plus 30 days’ grace period. The results may be sensitive to this assumption of the 30-day grace period. Therefore, as sensitivity analysis, the differences-in-differences analysis was repeated with a grace period of 60 and 15 days, respectively. Results
publication of the meta-analyses, the “impact of studies on rosi” would be greater than 1; otherwise, it would be less than (or equal to) 1. Similar interpretation would hold for the “impact of studies on pio.” Therefore, the differencesin-differences estimate of the impact of the metaanalyses would be the ratio of these impacts. Standard deviation and CIs (P) were computed using the delta method.22 Because this study is based on claims data, the actual sample sizes
The study sample consisted of 15,779 adult patients with continuous enrollment for both pharmacy and medical insurance and at least 1 prescription for an OHA between January 1, 2005, and November 30, 2007. Of these, 2103 patients had at least 1 prescription of either insulin or an injectable drug (eg, exenatide) or a newer agent (eg, sitagliptin or any combination of sitagliptin) between January 1, 2005, and November 30, 2007, and were, therefore, excluded.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
The pre-publication period was defined as November 1, 2006, through March 31, 2007. During this period, a total of 8436 patients had at least 1 prescription for an OHA. Of these, 6199 patients (73.5%) were assigned a monotherapy OHA as the index drug. The postpublication period was defined as July 1, 2007, through November 30, 2007, and in this period, 7582 patients had at least 1 fill of OHA, of which 1653 were combination therapy users. Therefore, data on 5929 patients (78.2%) were used to analyze the patterns of use in the postpublication period. The differences-in-differences multinomial regression focused on those patients whose index drug was either monotherapy pioglitazone or
monotherapy rosiglitazone for the pre-publication or the post-publication period. Therefore, all the patients who were on monotherapy metformin or monotherapy sulfonylurea in the pre-publication and post-publication periods were eliminated. Consequently, 5359 patients in the pre-publication period and 5327 patients in the post-publication period were dropped from the regression analysis, leaving us with (840 þ 602) ¼ 1442 patients. Fig. 2 shows how study samples were derived. Changes in the patterns of use Table 3 details the unadjusted changes in the patterns of use of monotherapy OHA users between the pre-publication and post-publication
Adult patients with continuous enrollment to both pharmacy and medical coverage and at least one prescription for oral anti-diabetic drug between Jan 01, 2005 and Nov 30, 2007: 15,779
Patients with at least one prescription of insulin or newer drugs (exenatide and sitagliptin) between Jan 01, 2005 and Nov 30, 2007: 2,103 (minus) PRE PUBLICATION PERIOD
Cohort of patients with at least one prescription for oral anti-diabetic drug in the pre-publication period: 8,436
POST PUBLICATION PERIOD
Cohort of patients with at least one prescription for oral anti-diabetic drug in the post-publication period: 7,582
Patients with combination therapy index drug: 2237 (minus)
Patients with combination therapy index drug: 1,653 (minus)
COHORT OF PATIENTS USED TO CHARACTERIZE PATTERN OF USE IN THE PRE-PUB PERIOD: 6,199
COHORT OF PATIENTS USED TO CHARACTERIZE PATTERN OF USE IN THE POST-PUB PERIOD: 5,929
Patients with Metformin or Sulfonylurea as index drug: 5,359 (minus)
COHORT OF PATIENTS FROM PRE-PUB PERIOD FOR D-I-D: 840
53
Patients with Metformin or Sulfonylurea as index drug: 5,327 (minus)
COHORT OF PATIENTS FROM POST PUB PERIOD FOR D-I-D: 602
TOTAL NUMBER OF OBSERVATION FOR DIFFERENCES-INDIFFERENCES MULTINOMIAL REGRESSION: 840+602 = 1442
Fig. 2. Cohort of patients for analysis.
54
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 3 Change in pattern of use of monotherapy oral antidiabetic drug users between pre-publication and post-publication periods Monotherapy drugs
Pre-publication
Post-publication
n
n
%
Change (% points)
P value
%
Rosiglitazone Rosi Mono or Rosi Combo Discontinued Switch to other drug
368 270 90 8
73.37 24.46 2.17
170 127 33 10
74.71 19.41 5.88
1.34 5.04 3.71a
.378 .163 .066
Pioglitazone Pio Mono or Pio Combo Discontinued Switch to other drug
472 350 115 7
74.15 24.36 1.48
432 363 64 5
84.03 14.81 1.16
9.88b 9.55b 0.33
!.001 !.001 .363
Metformin Met Mono or Met Combo Discontinued Switch to other drug
4253 2717 1479 57
63.88 34.78 1.34
4368 3161 1166 41
72.37 26.69 0.94
8.48b 8.08b 0.40a
!.001 !.001 .086
Sulfonylurea Sulf Mono or Sulf Combo Discontinued Switch to other drug
1106 807 288 11
72.97 26.04 0.99
959 829 127 3
86.44 13.24 0.31
13.48b 12.80b 0.68a
!.001 !.001 .059
a b
Significant at the 10% level. Significant at the 5% level.
periods. The number of patients whose index drug was classified as monotherapy rosiglitazone users declined in the post-publication period to less than half (pre-publication n ¼ 368; postpublication n ¼ 170). However, the number of other monotherapy users remained nearly the same across the 2 periods. The discontinuation rates decreased for all monotherapy users in the post-publication period. The largest decrease was among sulfonylurea users (12.8% points, P ! .001). Sulfonylurea users also recorded the largest increase in either continuing their monotherapy or adding another drug to monotherapy sulfonylurea (13.5% points, P ! .001). The rates for switching to a nonindex drug remained nearly the same for pioglitazone (0.33% points, P ¼ .363) and decreased for metformin (0.4% points, P ¼ .086) and sulfonylurea (0.7% points, P ¼ .059) users; however, the switching rates for rosiglitazone users increased (3.7% points, P ¼ .066) in the post-publication period. Changes in the patterns of use: Rosiglitazone versus pioglitazone In particular, the rates of switching to a nonindex drug before and after the publications of the meta-analyses using multinomial logistic regression were examined. Continuing with the index drug or adding another drug was assigned as the
reference category (category 1). Table 4 enumerates the descriptive statistics. It shows that across index drugs and periods, the distribution of patients by the type of insurance type remained nearly same (HMO w35%, POS between 7% and 11%, PPO between 38% and 46%, and others between 11% and 14%). Men were slightly more than women (men between 52% and 57%). Average age of the patients was approximately 56 years. Patients with history of MI ranged between less than 0.01% to 1.69%, and patients with history of HF ranged between 1.27% and 2.94%. Patients with history of hypertension ranged between 52% and 57%. Patients with history of any risk factors for MI (other than hypertension) ranged from 52% to 60%, which went up to 70%-76% when hypertension was also included. Finally, patients with history of any risk factors for HF (other than hypertension) were approximately 23%. The estimates of the multinomial logistic regression are presented in Table 5. The results show that relative to continuing with the index drug or adding another drug (reference category), the likelihood of discontinuing or switching to nonindex drug decreases with age (OR for age variable for discontinuation ¼ 0.97 [P % .001] and for switching to nonindex drug ¼ 0.96 [P ¼ .052]). Additionally, relative to continuing with the index drug or adding another drug, patients with risk factors for
55
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 4 Descriptive statistics (n ¼ 840 patients in the pre-publication period and n ¼ 602 patients in the post-publication period) Variables
n Type of insurance ¼ HMO Type of insurance ¼ POS Type of insurance ¼ PPO Type of insurance ¼ other insurance Gender ¼ male Patients with history of MI Patients with history of HF Patients with history of hypertension Patients with history of any of the risk factors for MI (other than hypertension) Patients with history of any of the risk factors for HF (other than hypertension) Patients with history of any of the risk factors for MI (including hypertension) Age, in years (as of January 1, 2005), mean (standard deviation)
Rosiglitazone
Pioglitazone
Pre-publication
Post-publication
Pre-publication
Post-publication
n
%
n
%
n
n
368 131 29 162 46
35.60 7.88 44.02 12.50
170 61 20 65 24
472 35.88 155 11.76 40 38.24 220 14.12 57
432 32.84 164 8.47 45 46.61 173 12.08 50
37.96 10.42 40.05 11.57
199 2 9 201
54.08 0.54 2.45 54.62
94 0 5 96
55.29 275 0.00 8 2.94 6 56.47 247
58.26 243 1.69 3 1.27 9 52.33 248
56.25 0.69 2.08 57.41
195
52.99 102
60.00 246
52.12 252
58.33
23.37
38
22.35 110
23.31 111
25.69
70.11 128
75.29 333
70.55 332
76.85
86
258
56.98 (10.72) d
57.51 (11.17) d
MI (including hypertension) were less likely to discontinue (OR ¼ 0.68 [P ¼ .008]) but more likely to switch to nonindex drug (OR ¼ 3.40 [P ¼ .049]). Goodness-of-fit statistics of this regression were as follows. deviance value ¼ 1139.932 with 1184 degrees of freedom (P ¼ 1.00) and Pearson value ¼ 1625.463 with 1814 degrees of freedom (P ¼ .9994). In addition, the likelihood ratio test for all betas equal to zero was rejected at P ! .0001 (c2 ¼ 80.692, degrees of freedom ¼ 18). As discussed previously, the main results are presented as the adjusted risk ratio of switching to a non-rosiglitazone drug for rosiglitazone users in the post-publication versus the pre-publication period and a similar ratio for pioglitazone users. The results are documented in Table 6. The analysis found Impact of studies on Rosi Prob of switch to non rosi in POST period Prob of switch to non rosi in PRE period ¼ 2:64 ðP ¼ 0:046Þ ¼
In other words, there was a 2.64-fold (P ¼ .046) increase in switching to nonindex drug
%
56.20 (10.76) d
%
56.13 (10.40) d
among rosiglitazone users in the postperiod relative to preperiod. For switching to the nonindex drug, among pioglitazone users in the postperiod compared with the preperiod, there was no change: Impact of studies on Pio Prob of switch to non pio in POST period Prob of switch to non pio in PRE period ¼ 0:72 ðP ¼ 0:583Þ
¼
Thus, the results show that the trend of switching to the nonindex drug moved in the opposite direction, that is, rosiglitazone users were more likely to switch to non-rosiglitazone drug in postperiod, whereas pioglitazone users were less likely (although statistically insignificant) to switch to non-pioglitazone drug in postperiod. Finally, the differences-in-differences estimate of switching to a nonindex drug showed that in the post-publication period, the rate of switching for rosiglitazone users was higher by 3.64 (P ¼ .090) times relative to that for pioglitazone users. In addition to the main analysis, sensitivity analysis on the definition of the main outcome
56
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 5 ORs (with associated P values) of differences-in-differences multinomial regression Parameter
Discontinued POc
OR
P O c2
0.54 1.04 1.52b 0.97a 1.20
!.0001 .834 d !.0001 .203
0.64 1.45 4.03b 0.96c 0.72
.456 .477 d .052 .415
0.83 1.27 0.68a
.173 .726 .008
0.67 0.00 3.40a
.291 .982 .049
1.09
.617
0.72
.522
OR 1 if observation in the post period 1 if index drug is rosiglitazone Post rosiglitazone (interaction term) Age 1 if insurance type is HMO (base is all other types of insurance) 1 if gender is male (base is gender ¼ female) 1 if patient has a history of MI 1 if patient has any risk factors for MI (including hypertension) 1 if patient has any risk factors for HF (other than hypertension)
Switched to nonindex drug 2
a
a
Significant at the 5% level. In this nonlinear specification of multinomial logistic regression, the coefficient of the interaction term is not the estimator of the true interaction effect and the P value estimated by statistical packages is incorrect; therefore, we do not report them.19,21 c Significant at the 10% level. b
variables was also performed. Switching and discontinuation of the index drug were defined in terms of the number of days of supply plus 30 days’ grace period. As sensitivity analysis, the main analysis was repeated with a grace period of 60 and 15 days. For both the analyses, the estimated size of the effect was similar; however, with 15 days’ grace period, the differences-in-differences estimates were statistically insignificant. Discussion The main thrust of this study is to learn and document the effect of unanticipated new clinical information (the meta-analyses) on the change in the pattern of use of OHAs in general and TZDs in specific. The release of the meta-analyses is akin to an unanticipated change in the body of knowledge that was directed toward one of the OHAs, that is, rosiglitazone and hence may be assumed to be exogenous. Because both pioglitazone and rosiglitazone are from the TZD class of drugs, pioglitazone became a natural contender for the comparator group. If the market’s response is to discount this adverse news, then there should not be any difference in the pattern of use of OHAs in general and TZDs specifically. A retrospective analysis of pharmacy claims by Starner et al23 described the trends in use of rosiglitazone and pioglitazone from January 1, 2007, through May 31, 2008, the period covering the release of meta-analysis
and issuance of black box warnings.23 They calculated the average number of pharmacy claims per day per million members for rosiglitazone, pioglitazone, and a comparison drug, sitagliptin. They found that the aggregate pharmacy claims for rosiglitazone decreased by more than half between January 2007 and December 2007 and up to twothirds by May 2008. If prescribers reacted cautiously by assuming that the safety problems apply to all the drugs in the therapeutic class, then the prescriptions for both pioglitazone and rosiglitazone should have similar decline. However, Starner et al23 found that pharmacy claims per million members for pioglitazone remained steady over the same period. Therefore, it seems that the response to the adverse news was interpreted more narrowly to rosiglitazone. The main findings of this study are that (1) the number of patients on monotherapy rosiglitazone reduced by half in the post-publication period but remained stable for all the other monotherapy drugs over the same period, (2) there is a significant decrease in the discontinuation rates for all other drugs in the postperiod but an insignificant decrease in the discontinuation rates for monotherapy rosiglitazone patients over the same period, (3) there is a significant increase in the rates of either continuing or adding another drug for all the other drugs but a marginal and insignificant increase for monotherapy rosiglitazone patients in the postperiod, (4) there is a slight and insignificant decrease in switching to a nonindex drug for all
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
57
Table 6 Switching to nonindex drugs (post-publication vs pre-publication): rosiglitazone versus pioglitazone Adjusted relative risk: post versus pre (P value) Rosi (post vs pre)
Adjusted ratio of relative risk: rosi versus pio (P value)
2.64a (P ¼ .046) Rosi (post vs pre)/pio (post vs pre)
Pio (post vs pre) a b
0.72 (P ¼ .583)
Significant at the 5% level. Significant at the 10% level.
other drugs in the postperiod but a large increase in the rates of switching to nonindex drug for monotherapy rosiglitazone users, and (5) monotherapy rosiglitazone users were 3 times more likely to switch to a non-rosiglitazone drug in the postperiod relative to monotherapy pioglitazone users. This analysis extends the previous study of Starner et al23 by going a step further and analyzing how the patients assigned to each OHA change their pattern of use in the post-publication period relative to the pre-publication period. As the results illustrate, the pattern of use differed fundamentally between monotherapy rosiglitazone users and users of all the other monotherapy OHAs in the postpublication period. The most pronounced result was the differences in rates of switching to nonindex drugs. The results from the multinomial regression demonstrate that not only are monotherapy rosiglitazone patients switching to non-rosiglitazone drugs at a higher rate, but the rate also is more than 3 times higher than the similar switches among monotherapy pioglitazone users in the post-publication period relative to the pre-publication period. This shows that the market response, as observed by patient/prescriber decisions, to the adverse news was interpreted narrowly to rosiglitazone, and there is little or no spillover to the other drugs. Many researchers have analyzed the effect of risk warnings on the use of other drugs as well. For example, studies have been conducted on the effect of 2004 FDA warnings24-26 on the increased suicidal risks in antidepressant trials for the treatment of youth with major depressive disorder, and those studies, like this study, have found that the risk e
3.64b (P ¼ .090)
warnings lead to the decrease in the use of antidepressants in children and youth. More specifically, Kurian et al27 estimated a 33% decline in new users aged between 2 and 17 years after the FDA warnings among Tennessee Medicaid youth data. Libby et al28 found that antidepressant dispensing among depression-diagnosed youths (5-17 years of age) fell from 59.2% in the pre-warning period to 55% in the post-warning period between October 1998 and September 2005 in commercially insured population. In a follow-up article using the same data set, Libby et al29 showed sustained decline in antidepressant treatment up to 2007. Limitations The data analysis is based on a population from Maryland, District of Columbia, and Northern Virginia, and the results may differ for data from other states or commercial data sets. Additionally, medical and pharmacy claims data were used to extract information about history of MI and HF and risk factors for MI and HF, and therefore, risk factors without a claim-related diagnosis are not included in the analysis, such as tobacco use, physical inactivity, and obesity. Also, studies have shown that there are diagnostic coding errors in Medicaid claims data,30 and the same may be true for these data as well. As with all the studies using claims data only, there was no information on clinical indicators, such as blood pressure or laboratory results. Last, other comorbidities were not taken into account. Because the study is from a longitudinal data set, 440 patients are in both the pre-publication and post-publication periods.e This may cause the
Four hundred forty patients were in both the pre-publication and post-publication periods. Of these, 270 patients’ index drug was pioglitazone in both the periods. In other words, of those patients on pioglitazone in the pre-publication period, 57% were also on pioglitazone in the post-publication period. An additional 130 patients’ index drug was rosiglitazone in both the periods. This implies that of those patients on rosiglitazone in the pre-publication period, only 35% were also on the same drug in the post-publication period. The rest of the 40 patients’ index drug in the pre-publication period was rosiglitazone but in the post-publication period was pioglitazone.
58
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
problem of correlated responses. The analysis for change of pattern and differences-in-differences multinomial regression ignores this correlation. This study compares monotherapy rosiglitazone users with monotherapy pioglitazone users, and therefore, this study may not be generalizable to all patients with type 2 diabetes. Conclusion Comparing the pattern of use of drugs before and after the publication of meta analysis shows that the number of patients on monotherapy rosiglitazone reduced by half in the postpublication period but remained stable for all the other monotherapy drugs over the same period. It also showed that there was a significant decrease in the discontinuation rates for all other drugs in the post-publication period but an insignificant decrease in the discontinuation rates for monotherapy rosiglitazone patients over the same period. Additionally, there was a significant increase in the rates of either continuing or adding another drug for all the other drugs but a marginal and insignificant increase for monotherapy rosiglitazone patients in the post-publication period. Furthermore, there was a slight and insignificant decrease in switching to a nonindex drug for all other drugs in the post-publication period but a large increase in the rates of switching to a nonindex drug for monotherapy rosiglitazone users, and finally, monotherapy rosiglitazone users were 3 times more likely to switch to a non-rosiglitazone drug in the post-publication period relative to pioglitazone users. Therefore, our results show that the publications about safety risks had differential effect between the 2 drugs within the therapeutic class. Acknowledgments Funding for the first author’s postdoctoral fellowship was provided by an unrestricted grant from Takeda Pharmaceutical Company. Portions of this manuscript were presented as posters at the annual meeting of the American Association of Diabetes Education, August 5-8, 2009, in Atlanta, GA, and at the 12th Annual European Congress of the International Society for Pharmacoeconomics and Outcomes Research, October 24-27, 2009, in Paris, France. One of the authors reported receipt of consulting or honoraria from Amgen, Amylin, Bayer, BMS, Genentech, GlaxoSmithKline, Lilly Merck, Novartis, Pfizer,
and Sanofi-Aventis and grant support from GlaxoSmithKline, Novartis, Pfizer, and SanofiAventis.
References 1. Owens DR. Thiazolidinediones: a pharmacological overview. Clin Drug Invest 2002;22:485–505. 2. Meriden T. Progress with thiazolidinediones in the management of type 2 diabetes mellitus. Clin Ther 2004;26(2):177–190. 3. Bell DS. Type 2 diabetes mellitus: what is the optimal treatment regimen? Am J Med 2004;116(suppl 5A): 23S–29S. 4. Yanagawa T, Araki A, Sasamoto K, Shirabe S, Yamanouchi T. Effect of antidiabetic medications on microalbuminuria in patients with type 2 diabetes. Metabolism 2004;53(3):353–357. 5. Henney JE. Withdrawal of troglitazone and cisapride. JAMA 2000;283(17):2228. 6. Haffner SM, Greenberg AS, Weston WM, Chen H, Williams K, Freed MI. Effect of rosiglitazone treatment on nontraditional markers of cardiovascular disease in patients with type 2 diabetes mellitus. Circulation 2002;106(6):679–684. 7. Dormandy JA, Charbonnel B, Eckland DJ, et al. Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study (PROspective pioglitAzone Clinical Trial In macroVascular Events): a randomised controlled trial. Lancet 2005;366(9493):1279–1289. 8. Delea TE, Edelsberg JS, Hagiwara M, Oster G, Phillips LS. Use of thiazolidinediones and risk of heart failure in people with type 2 diabetes: a retrospective cohort study. Diabetes Care 2003;26(11): 2983–2989. 9. Hartung DM, Touchette DR, Bultemeier NC, Haxby DG. Risk of hospitalization for heart failure associated with thiazolidinedione therapy: a Medicaid claims-based case-control study. Pharmacotherapy 2005;25(10):1329–1336. 10. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular causes. N Engl J Med 2007;356(24): 2457–2471. 11. Singh S, Loke YK, Furberg CD. Long-term risk of cardiovascular events with rosiglitazone: a metaanalysis. JAMA 2007;298(10):1189–1195. 12. Lincoff AM, Wolski K, Nicholls SJ, Nissen SE. Pioglitazone and risk of cardiovascular events in patients with type 2 diabetes mellitus: a metaanalysis of randomized trials. JAMA 2007;298(10): 1180–1188. 13. News F. Manufacturers of some diabetes drugs to strengthen warning on heart failure risk. Available at: http://www.fda.gov/NewsEvents/ Newsroom/PressAnnouncements/2007/ucm108966. htm. Accessed 01.12.09.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59 14. FDA. FDA Adds boxed warning for heart-related risks to anti-diabetes drug Avandia. Available at: http://www.fda.gov/NewsEvents/Newsroom/Press Announcements/2007/ucm109026.htm. Accessed 01. 12.09. 15. American Heart Association. Risk factors and coronary heart disease. Available at: http://www. americanheart.org/presenter.jhtml?identifier¼4726. Accessed 12.12.08. 16. Hunt SA, Abraham WT, Chin MH, et al. ACC/AHA 2005 Guideline Update for the Diagnosis and Management of Chronic Heart Failure in the Adult: a report of the American College of Cardiology/ American Heart Association Task Force on Practice Guidelines (Writing Committee to Update the 2001 Guidelines for the Evaluation and Management of Heart Failure): developed in collaboration with the American College of Chest Physicians and the International Society for Heart and Lung Transplantation: endorsed by the Heart Rhythm Society. Circulation 2005;112(12):e154–e235. 17. Goodman LA. Simultaneous confidence intervals for contrasts among multinomial populations. Ann Math Stat 1964;35(2):716–725. 18. Goodman LA. On simultaneous confidence intervals for multinomial proportions. Technometrics 1965; 7(2):247–254. 19. Ai C, Norton EC. Interaction terms in logit and probit models. Econ Lett 2003;80(1):123–129. 20. Mullahy J. Interaction effects and difference-indifference estimation in loglinear models. NBER Technical Working Papers 1999;254. 21. Norton E, Wang H, Ai C. Computing interaction effects and standard errors in logit and probit models. Stata J 2004;4:103–116. 22. Greene W. Econometric Analysis. 5th ed. New Jersey, Prentice Hall: Macmillan; 2003.
59
23. Starner CI, Schafer JA, Heaton AH, Gleason PP. Rosiglitazone and pioglitazone utilization from January 2007 through May 2008 associated with five risk-warning events. J Manag Care Pharm 2008;14(6):523–531. 24. FDA. FDA public health advisory: worsening depression and suicidality in patients being treated with antidepressant. Available at: http://www. fda.gov/cder/drug/antidepressants/Antidepressanst PHA.htm. Accessed 01.12.2010. 25. FDA. FDA news: FDA launches a multi-pronged strategy to strengthen safeguards for children treated with antidepressant medications. Available at: http:// www.fda.gov/bbs/topics/news/2004/NEW01124. html. Accessed 01.12.2010. 26. FDA. FDA public health advisory: reports of suicidality in pediatric patients being treated with antidepressant medications for major depressive disorder (MDD). Available at: http://www.fda.gov/cder/ drug/advisory/mdd.htm. Accessed 01.12.2010. 27. Kurian BT, Ray WA, Arbogast PG, Fuchs DC, Dudley JA, Cooper WO. Effect of regulatory warnings on antidepressant prescribing for children and adolescents. Arch Pediatr Adolesc Med 2007;161: 690–696. 28. Libby AM, Brent DA, Morrato EH, Orton HD, Allen R, Valuck RJ. Decline in treatment of pediatric depression after FDA advisory on risk of suicidality with SSRIs. Am J Psychiatry 2007;164: 884–891. 29. Libby AM, Orton H, Valuck RJ. Persisting decline in depression treatment after FDA warnings. Arch Gen Psychiatry 2009;66:633–639. 30. Hennessy S, Bilker WB, Weber A, Strom BL. Descriptive analyses of the integrity of a US Medicaid claims database. Pharmacoepidemiol Drug Saf 2003;12(2):103–111.
Research in Social and Administrative Pharmacy 8 (2012) 47–59
Original Research
Use of rosiglitazone and pioglitazone immediately after the cardiovascular risk warnings Rahul Jain, Ph.D.a,*, C. Daniel Mullins, Ph.D.b, Helen Lee, Pharm.D., M.B.A.c, Winston Wong, Pharm.D.c a
Department of Clinical and Administrative Pharmacy, College of Pharmacy, University of Georgia, 250 Green Street, Athens, GA 30602, USA b Department of Pharmaceutical Health Services Research, University of Maryland School of Pharmacy, 220 Arch Street, 12th Floor, Baltimore, MD 21201, USA c Pharmacy Management, CareFirst BlueCross BlueShield, 1501 S. Clinton Street, 8th Floor, Mail Stop CT08-10, Baltimore, MD 21224, USA
Abstract Background: Meta-analyses of oral hypoglycemic agents (OHAs) revealed that rosiglitazone increased the risk of myocardial infarction (MI) and heart failure (HF) and that pioglitazone increased the risk of HF and decreased the risk of MI. Objective: To characterize the change in the pattern of use of OHAs immediately after the publication of these meta-analyses on May 21, 2007. Methods: Pharmacy and medical claims data for a managed care organization were analyzed for patients continuously enrolled from January 1, 2005, to November 30, 2007, with at least 1 pharmacy claim for OHA in the 13-month period between November 1, 2006, and November 30, 2007. A 5-month prepublication period (November 1, 2006, through March 31, 2007) was compared with a 5-month postpublication period (July 1, 2007, through November 30, 2007) using a differences-in-differences multinomial logistic regression. This regression explored discontinuation; continuation with monotherapy or adding another drug; and switching to a drug different from the index monotherapy drug after adjusting for gender, age, type of insurance, past 1-year history of MI or HF, and risk factors for MI and HF in the past 1 year. Results: The relative rate of switching to nonindex drug in the postpublication relative to prepublication was 2.64 (P ¼ .046) for monotherapy rosiglitazone users and 0.72 (P ¼ .583) for monotherapy pioglitazone users. The differences-in-differences estimate of the rate of switching to nonindex drugs for monotherapy rosiglitazone users was 3.64 (P ¼ .090) times higher relative to the estimate for monotherapy pioglitazone users. Conclusion: The pattern of use differed fundamentally between monotherapy rosiglitazone users and users of all other monotherapy OHAs in the postperiod. Not only were monotherapy rosiglitazone patients switching to non-rosiglitazone drugs at a higher rate, but the rate also was more than 3 times higher than similar switches among monotherapy pioglitazone users in the postperiod relative to the preperiod. This shows that the market response as observed by patient/prescriber decisions to the adverse news was interpreted narrowly to monotherapy rosiglitazone, and there is little or no spillover to the other drugs. * Corresponding author. Tel.: þ1 706 542 5327; fax: þ1 706 542 5228. E-mail address: [email protected] (R. Jain). 1551-7411/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.sapharm.2010.12.003
48
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Therefore, this study found that there was a differential effect of meta-analyses on the use of the 2 drugs. Ó 2012 Elsevier Inc. All rights reserved. Keywords: Diabetes; Pattern of use; Differences-in-differences
Introduction Thiazolidinediones (TZDs) were introduced in the 1990s as a second-line option after lifestyle modifications and maximum tolerated doses of metformin were not effective for the management of type 2 diabetes mellitus. As a class, TZDs have a unique mechanism of action that involves insulin sensitivity in patients with type 2 diabetes.1 TZDs are used as monotherapy as well as along with other oral hypoglycemic agents (OHAs) in combination therapy.2-4 Chemically, its members are derivatives of the parent compound TZD and include rosiglitazone (Avandia), pioglitazone (Actos), and troglitazone (Rezulin). Troglitazone was introduced in the late 1990s but was found to be associated with an idiosyncratic reaction, leading to drug-induced hepatitis. It was withdrawn from the U.S. market on March 21, 2000.5 Initial research had shown that rosiglitazone6 and pioglitazone7 may have potentially been beneficial to cardiovascular health. However, subsequent research raised concerns regarding the adverse cardiac effect of TZDs. A retrospective study published in 2003 found that TZDs might increase the risk of heart failure (HF).8 Another retrospective case-control study published in 2005 found an association between TZD therapy and hospitalization for HF within 60 days of the prescription date.9 On May 21, 2007, Nissen and Wolski10 publisheda their findings of a metaanalysis. This study found that for the rosiglitazone group, relative to the control group, the odds ratio (OR) for myocardial infarction (MI) was 1.43 (95% confidence interval [CI] ¼ 1.031.98) and the OR for death from cardiovascular causes was 1.64 (95% CI ¼ 0.98-2.74). Two other meta-analyses have been published since then, which investigated the risk of cardiovascular events associated with rosiglitazone11 and pioglitazone.12 The first study found that among patients with type 2 diabetes, rosiglitazone use for at least 12 months is associated with
a
a significantly increased risk of MI (RR ¼ 1.42, 95% CI ¼ 1.06-1.91) and HF (RR [relative risk] ¼ 2.09, 95% CI ¼ 1.52-2.88), without a significantly increased risk of cardiovascular mortality (RR ¼ 0.09, 95% CI ¼ 0.63-1.26). The second study found that pioglitazone use is associated with a significantly lower risk of death, MI, or stroke (hazards ratio [HR] ¼ 0.82, 95% CI ¼ 0.72-0.94, P ¼ .005) among a diverse population of patients with diabetes. Serious HF is increased (HR ¼ 1.41, 95% CI ¼ 1.14-1.76, P ¼ .002) by pioglitazone use, although without an associated increase in mortality. In response to these findings, on August 14, 2007, the Food and Drug Association (FDA) issued a black box warning13 for both pioglitazone and rosiglitazone about increased HF risks. On November 14, 2007, the black box warning14 for rosiglitazone was updated to include warnings that the drug may cause MI. No such warning was issued for pioglitazone. Intuitively, the market response as observed by patient/prescriber decisions to the risk warning or “adverse news” could be to discount it entirely, interpret it narrowly with respect to the specific drug involved for each adverse event, or interpret it broadly and assume that the safety problems apply to other drugs in the same therapeutic class.
Objectives The primary objective of this study was to characterize the change in the pattern of use of OHAs in general and the TZDs specifically immediately after the publication of these metaanalyses. The rest of the article characterizes changes in the patterns of use, switching, and discontinuation among pioglitazone, rosiglitazone, metformin, and sulfonylurea users before and after the publication of the meta-analyses. Subsequently, the change in pattern of use of rosiglitazone
This article was published at www.nejm.org on May 21, 2007. Subsequently, it was published in the Journal on June 14, 2007.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
relative to that of pioglitazone is presented to evaluate if changes in use had differential effect across TZDs. Methods Study sample: Inclusion and exclusion criteria The study population consisted of beneficiaries covered by a managed care organization with approximately 3.4 million members with medical benefits, of whom 1.2 million members had pharmacy benefits within the states of Maryland, District of Columbia, and Northern Virginia. Only adult patients (aged 18 years and older) who were continuously enrolled from January 1, 2005, through November 30, 2007, with both medical and pharmacy benefits and at least 1 pharmacy claim for an OHA between November 1, 2006, and November 30, 2007, were included. Patients were excluded if they had a pharmacy claim for insulinb or any injectable (eg, exenatide) or newer agentsc (eg, sitagliptin or any combination of sitagliptin) during January 1, 2005, through November 30, 2007. In addition to these, the analysis concentrates only on patients on monotherapy TZDs. This is done for multiple reasons. First, patients on monotherapy TZDs are likely to be those who are either contraindicated or intolerant to other OHAs and whose glycemic control is not so poor as to require the use of insulin. Therefore, these patients are likely to face cardiovascular risks because of the use of TZDs and not because of poor glycemic control. If there were an effect of the meta-analysis on prescription pattern, it would be the lower bound of the effect on all TZD users. Second, in our data, more than 70% of the patients had a monotherapy index drug. Third, because our study is based on claims data, the severity of diabetes is not known; therefore, by concentrating only on monotherapy TZD patients, the analysis eliminates the bias because of b
49
unobservable severity of diabetes (omitted variable bias) and limits the confounding because of concomitant diabetes medicines. Hence, we concentrated only on the monotherapy users. Study periods This study examines the change in pattern of use of OHAs before and after the publication of meta-analyses. The pre-publication period was defined as November 1, 2006, through March 31, 2007. Similarly, the post-publication period was defined as July 1, 2007, through November 30, 2007. Data for the period from April 1, 2007, to June 30, 2007, were excluded to allow for new prescriptions written after May 21, 2007, to trickle into the data. Fig. 1 shows the timeline for preperiod and postperiod diagrammatically. Because some patients may have filled OHA prescriptions just before the beginning of prepublication period (November 1, 2006), a 60-day window (September 2, 2006, through October 31, 2006) allowed patients to consume the prescriptions that may have been filled just before November 1, 2007. The index drug for the prepublication period was then defined as the first OHA or first combination of OHAs after the 60day window. Similarly, for the post-publication period, a similar 60-day windowd (May 2, 2007, through June 30, 2007) was allowed, and the index drug for the post-publication period was then defined as the first OHA or first combination of OHAs after the initial 60-day period. Definition of continuation, switching, and discontinuation This study was a retrospective longitudinal data analysis of medical and pharmacy claims of patients on OHA. After the initial prescription, each patient may have refilled the same drug, may have added another drug, switched to another drug or combination of drugs, or discontinued completely. Therefore, in both pre-publication and
Patients on insulin were excluded because patients on insulin may have more advanced diabetes or may have type 1 diabetes, which could not be identified in our data set. c The FDA approved sitagliptin on October 17, 2006. Because it is a newer agent, a very small number of patients were on this medication. Therefore, it was excluded from the analysis to eliminate any small sample biases in the estimate. d This study is analyzing the effect of the publication of a meta-analysis on the prescribing patterns of pioglitazone and rosiglitazone. Hence, this study does not require new users; it only requires new prescriptions during the study time frame. Because most prescriptions were for 30 days, a 60-day window was allowed for patients to consume the prescriptions that were filled just before the publication of the meta-analysis. The mail order prescriptions were dealt with exactly as retail prescriptions.
50
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Meta-analysis released on May/21/07
Apr/01/07
Nov/01/06
Mar/31/07 Pre-Publication Period
Jun/30/07 Time Nov/30/07
Jul/01/07
Post-Publication Period
Fig. 1. Timeline for change in pattern of use.
post-publication periods, the following definitions were used to classify discontinuation and switch. Discontinuation If there was no subsequent fill or the period between their previous prescription and a new prescription for the index drug was longer than the number of days of supply plus 30 days, then discontinuation of therapy was recorded. Switch A switch was recorded for those who had a prescription claim for an OHA different from the index drug within 30 days after the date of their last prescription plus the number of days of supply. The drugs were identified using the product names and are enumerated in Table 1. A distinction was made between adding another drug to the index drug (switching from monotherapy to combination therapy) and switching to a drug entirely different from the index drug. Patients were censored at the date of discontinuation, switching, or at the end of the follow-up period. Analytical approach Covariates The analysis included demographic variables (age and gender) and type of insurance (Health Maintenance Organization [HMO], Preferred Provider Organization [PPO], Point of Service Plans [POS], or other). It also included the history of observable risk factors for MI15 and HF16 (see Table 2 for International Classification of Diseases [ICD]-9 codes) and history of MI (ICD-9 code 410.xx) and HF (ICD-9 code 428.xx). The history of the risk factors for MI and HF and past history of MI and HF were measured as at
least 1 hospitalization/diagnosis or an emergency room visit with the above-stated ICD-9 codes listed in any diagnosis claim in the past 365 days. Therefore, for the pre-publication period, these variables were measured between November 1, 2005, and October 31, 2006, and for the postpublication period, they were measured between July 1, 2006, and June 30, 2007. Other risk factors for MI, such as tobacco use, physical inactivity, and obesity, were not observed in the data and, therefore, not included in the analysis. Statistical analysis The change in pattern of use compared the rates of switching, discontinuation, or continuation of the index drug before and after the publication of meta-analyses. This implied comparing 2 multinomial variables, one for the pre-publication period and the other for the post-publication period. Clearly, comparison based on binomial distribution is not suitable. Therefore, simultaneous CIs for the changes in multinomial proportions between pre-publication and post-publication samples17,18 were estimated. The aim of this study was to elicit the effect of the meta-analyses on the pattern of use of TZDs. The meta-analyses implied strong adverse news for rosiglitazone. Therefore, it is expected that the pattern of use of rosiglitazone users would be affected after the publication of the meta-analyses. However, some of the change could also be because of time trend or changes in other factors that were not linked to the publication of the meta-analyses. Therefore, a simple comparison in pattern of use of rosiglitazone patients before and after the publication is not enough to bring out the true effect of these studies.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Hence, to identify the true effect of the metaanalyses, a comparator group is needed, which was not affected by the meta-analyses. This comparator group is referred to as the control group in the differences-in-differences literature. The underlying assumption in the differences-in-differences approach is that in absence of the adverse news, the gap in pattern of use between rosiglitazone users and the comparator group would remain the same over time. Because the aim was to also test whether the “market” interpreted these publications narrowly or broadly and the meta-analyses did not imply strong adverse news for pioglitazone, the pioglitazone patients become ideal candidates for the comparator group. Therefore, the change in pattern of use of rosiglitazone relative to that of pioglitazone was estimated by estimating a differences-in-differences multinomial logistic regression. The dependent variable was defined as the categorical variable (see Equation 1). This variable indicated whether the patient continued their monotherapy or added another drug, discontinued, or switched to nonindex drug.
factors for MI. Also included in the regression were the indicator for index drug is rosiglitazone, indicator that the observation is from postpublication period, and an interaction term of post and rosiglitazone. The underlying assumption in the differences-in-differences approach is that in absence of the adverse news, the gap between the pattern of use of rosiglitazone users and pioglitazone users over time would remain the same. The results of this regression would provide the overall effect of the publication of meta-analyses. The coefficient of the interaction term would capture the effect of the publications because after covariate adjustment the only source attributing to variations in the rate difference between the preperiod and the postperiod is the publication itself. In a linear regression model of differencesin-differences, the coefficient of the interaction term is the parameter of interest; however, in this nonlinear specification of multinomial logistic regression, the coefficient of the interaction term is not the estimator of the true interaction effect.19-21 Therefore, the results are presented as
8 < 1 if continued their monotherapy or added another drug Yi ¼ 2 if discontinued : 3 if switched to nonindex drug
Impact of studies on Rosi ¼
Impact of studies on Pio ¼
Predicted prob of switching to non rosi in POST period Predicted prob of switching to non rosi in PRE period
Prob of switching to non pio in POST period Prob of switching to non pio in PRE period
The independent variables included age, gender, past 1 year’s history of MI, and risk factors for MI. Because history of HF is also a risk factor for HF, it was not included in the regression. Additionally, hypertension is a risk factor for both MI and HF; therefore, including it in both the risk factors will give rise to multicollinearity problem. Hence, it was included only in the risk
51
ð1Þ
ð2Þ
ð3Þ
relative risks (see Equation 2 and Equation 3) and as the ratio of relative risks (Equation 4) Diff ¼
Impact of studies on rosi Impact of studies on pio
ð4Þ
If the probability of rosiglitazone users switching to a non-rosiglitazone drug increases after the
52
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 1 Product names of the OHAs
Table 2 Risk factors for MI and HF and their ICD-9 codes
OHA
Risk factor
Product name
Monotherapy drug’s product names Sulfonylurea Amaryl Chlorpropamide Diabeta Diabinese Glimepiride Glipizide Glipizide ER Glipizide XL Glucotrol Glucotrol XL Glyburide Glyburide Micronized Glynase Pres-Tab Micronase Tolazamide Tolbutamide Metformin Metformin Fortamet Glucophage Glucophage XR Glumetza Metformin ER Metformin HCL Metformin Hydrochloride Riomet Pioglitazone Actos Rosiglitazone Avandia Combination therapy drug’s product names Sulf-Met Glipizide/Metformin Glipizide/Metformin HCL Glucovance Glyburide Micronized/ Metformin HCL Glyburide Micronized/ Metformin Hydrochloride Glyburide/Metformin HCL Metaglip Pio-Met Actoplus Met Rosi-Met Avandamet Pio-Sulf Duetact Rosi-Sulf Avandaryl
Risks factors for MI Hypertension Essential hypertension Chronic venous hypertension High blood cholesterol Pure hypercholesterolemia Pure hyperglyceridemia Mixed hyperlipidemia Hyperchylomicronemia Other and unspecified hyperlipidemia Risk factors for HF Hypertension Essential hypertension Chronic venous hypertension Aortic and mitral valve disease Diseases of aortic valve Diseases of mitral valve Diseases of mitral and aortic valves Aortic coarctation Cardiac dysrhythmias Pulmonary hypertension Tricuspid valve disease Pulmonary valve disease Ischemic heart disease Cardiomyopathy
ICD-9 code
401.xx 459.3x 272.0 272.1 272.2 272.3 272.4
401.xx 459.3x 395.xx 394.xx 396.xx 747.10 427.xx 416.0x 397.0x 746.xx 410-414.xx 425.xx
and number of patients included depend on the inclusion/exclusion criteria and not necessarily powered to detect differences. Sensitivity analysis The main outcomes of the study, namely switching and discontinuation, are defined in terms of the number of days of supply plus 30 days’ grace period. The results may be sensitive to this assumption of the 30-day grace period. Therefore, as sensitivity analysis, the differences-in-differences analysis was repeated with a grace period of 60 and 15 days, respectively. Results
publication of the meta-analyses, the “impact of studies on rosi” would be greater than 1; otherwise, it would be less than (or equal to) 1. Similar interpretation would hold for the “impact of studies on pio.” Therefore, the differencesin-differences estimate of the impact of the metaanalyses would be the ratio of these impacts. Standard deviation and CIs (P) were computed using the delta method.22 Because this study is based on claims data, the actual sample sizes
The study sample consisted of 15,779 adult patients with continuous enrollment for both pharmacy and medical insurance and at least 1 prescription for an OHA between January 1, 2005, and November 30, 2007. Of these, 2103 patients had at least 1 prescription of either insulin or an injectable drug (eg, exenatide) or a newer agent (eg, sitagliptin or any combination of sitagliptin) between January 1, 2005, and November 30, 2007, and were, therefore, excluded.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
The pre-publication period was defined as November 1, 2006, through March 31, 2007. During this period, a total of 8436 patients had at least 1 prescription for an OHA. Of these, 6199 patients (73.5%) were assigned a monotherapy OHA as the index drug. The postpublication period was defined as July 1, 2007, through November 30, 2007, and in this period, 7582 patients had at least 1 fill of OHA, of which 1653 were combination therapy users. Therefore, data on 5929 patients (78.2%) were used to analyze the patterns of use in the postpublication period. The differences-in-differences multinomial regression focused on those patients whose index drug was either monotherapy pioglitazone or
monotherapy rosiglitazone for the pre-publication or the post-publication period. Therefore, all the patients who were on monotherapy metformin or monotherapy sulfonylurea in the pre-publication and post-publication periods were eliminated. Consequently, 5359 patients in the pre-publication period and 5327 patients in the post-publication period were dropped from the regression analysis, leaving us with (840 þ 602) ¼ 1442 patients. Fig. 2 shows how study samples were derived. Changes in the patterns of use Table 3 details the unadjusted changes in the patterns of use of monotherapy OHA users between the pre-publication and post-publication
Adult patients with continuous enrollment to both pharmacy and medical coverage and at least one prescription for oral anti-diabetic drug between Jan 01, 2005 and Nov 30, 2007: 15,779
Patients with at least one prescription of insulin or newer drugs (exenatide and sitagliptin) between Jan 01, 2005 and Nov 30, 2007: 2,103 (minus) PRE PUBLICATION PERIOD
Cohort of patients with at least one prescription for oral anti-diabetic drug in the pre-publication period: 8,436
POST PUBLICATION PERIOD
Cohort of patients with at least one prescription for oral anti-diabetic drug in the post-publication period: 7,582
Patients with combination therapy index drug: 2237 (minus)
Patients with combination therapy index drug: 1,653 (minus)
COHORT OF PATIENTS USED TO CHARACTERIZE PATTERN OF USE IN THE PRE-PUB PERIOD: 6,199
COHORT OF PATIENTS USED TO CHARACTERIZE PATTERN OF USE IN THE POST-PUB PERIOD: 5,929
Patients with Metformin or Sulfonylurea as index drug: 5,359 (minus)
COHORT OF PATIENTS FROM PRE-PUB PERIOD FOR D-I-D: 840
53
Patients with Metformin or Sulfonylurea as index drug: 5,327 (minus)
COHORT OF PATIENTS FROM POST PUB PERIOD FOR D-I-D: 602
TOTAL NUMBER OF OBSERVATION FOR DIFFERENCES-INDIFFERENCES MULTINOMIAL REGRESSION: 840+602 = 1442
Fig. 2. Cohort of patients for analysis.
54
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 3 Change in pattern of use of monotherapy oral antidiabetic drug users between pre-publication and post-publication periods Monotherapy drugs
Pre-publication
Post-publication
n
n
%
Change (% points)
P value
%
Rosiglitazone Rosi Mono or Rosi Combo Discontinued Switch to other drug
368 270 90 8
73.37 24.46 2.17
170 127 33 10
74.71 19.41 5.88
1.34 5.04 3.71a
.378 .163 .066
Pioglitazone Pio Mono or Pio Combo Discontinued Switch to other drug
472 350 115 7
74.15 24.36 1.48
432 363 64 5
84.03 14.81 1.16
9.88b 9.55b 0.33
!.001 !.001 .363
Metformin Met Mono or Met Combo Discontinued Switch to other drug
4253 2717 1479 57
63.88 34.78 1.34
4368 3161 1166 41
72.37 26.69 0.94
8.48b 8.08b 0.40a
!.001 !.001 .086
Sulfonylurea Sulf Mono or Sulf Combo Discontinued Switch to other drug
1106 807 288 11
72.97 26.04 0.99
959 829 127 3
86.44 13.24 0.31
13.48b 12.80b 0.68a
!.001 !.001 .059
a b
Significant at the 10% level. Significant at the 5% level.
periods. The number of patients whose index drug was classified as monotherapy rosiglitazone users declined in the post-publication period to less than half (pre-publication n ¼ 368; postpublication n ¼ 170). However, the number of other monotherapy users remained nearly the same across the 2 periods. The discontinuation rates decreased for all monotherapy users in the post-publication period. The largest decrease was among sulfonylurea users (12.8% points, P ! .001). Sulfonylurea users also recorded the largest increase in either continuing their monotherapy or adding another drug to monotherapy sulfonylurea (13.5% points, P ! .001). The rates for switching to a nonindex drug remained nearly the same for pioglitazone (0.33% points, P ¼ .363) and decreased for metformin (0.4% points, P ¼ .086) and sulfonylurea (0.7% points, P ¼ .059) users; however, the switching rates for rosiglitazone users increased (3.7% points, P ¼ .066) in the post-publication period. Changes in the patterns of use: Rosiglitazone versus pioglitazone In particular, the rates of switching to a nonindex drug before and after the publications of the meta-analyses using multinomial logistic regression were examined. Continuing with the index drug or adding another drug was assigned as the
reference category (category 1). Table 4 enumerates the descriptive statistics. It shows that across index drugs and periods, the distribution of patients by the type of insurance type remained nearly same (HMO w35%, POS between 7% and 11%, PPO between 38% and 46%, and others between 11% and 14%). Men were slightly more than women (men between 52% and 57%). Average age of the patients was approximately 56 years. Patients with history of MI ranged between less than 0.01% to 1.69%, and patients with history of HF ranged between 1.27% and 2.94%. Patients with history of hypertension ranged between 52% and 57%. Patients with history of any risk factors for MI (other than hypertension) ranged from 52% to 60%, which went up to 70%-76% when hypertension was also included. Finally, patients with history of any risk factors for HF (other than hypertension) were approximately 23%. The estimates of the multinomial logistic regression are presented in Table 5. The results show that relative to continuing with the index drug or adding another drug (reference category), the likelihood of discontinuing or switching to nonindex drug decreases with age (OR for age variable for discontinuation ¼ 0.97 [P % .001] and for switching to nonindex drug ¼ 0.96 [P ¼ .052]). Additionally, relative to continuing with the index drug or adding another drug, patients with risk factors for
55
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 4 Descriptive statistics (n ¼ 840 patients in the pre-publication period and n ¼ 602 patients in the post-publication period) Variables
n Type of insurance ¼ HMO Type of insurance ¼ POS Type of insurance ¼ PPO Type of insurance ¼ other insurance Gender ¼ male Patients with history of MI Patients with history of HF Patients with history of hypertension Patients with history of any of the risk factors for MI (other than hypertension) Patients with history of any of the risk factors for HF (other than hypertension) Patients with history of any of the risk factors for MI (including hypertension) Age, in years (as of January 1, 2005), mean (standard deviation)
Rosiglitazone
Pioglitazone
Pre-publication
Post-publication
Pre-publication
Post-publication
n
%
n
%
n
n
368 131 29 162 46
35.60 7.88 44.02 12.50
170 61 20 65 24
472 35.88 155 11.76 40 38.24 220 14.12 57
432 32.84 164 8.47 45 46.61 173 12.08 50
37.96 10.42 40.05 11.57
199 2 9 201
54.08 0.54 2.45 54.62
94 0 5 96
55.29 275 0.00 8 2.94 6 56.47 247
58.26 243 1.69 3 1.27 9 52.33 248
56.25 0.69 2.08 57.41
195
52.99 102
60.00 246
52.12 252
58.33
23.37
38
22.35 110
23.31 111
25.69
70.11 128
75.29 333
70.55 332
76.85
86
258
56.98 (10.72) d
57.51 (11.17) d
MI (including hypertension) were less likely to discontinue (OR ¼ 0.68 [P ¼ .008]) but more likely to switch to nonindex drug (OR ¼ 3.40 [P ¼ .049]). Goodness-of-fit statistics of this regression were as follows. deviance value ¼ 1139.932 with 1184 degrees of freedom (P ¼ 1.00) and Pearson value ¼ 1625.463 with 1814 degrees of freedom (P ¼ .9994). In addition, the likelihood ratio test for all betas equal to zero was rejected at P ! .0001 (c2 ¼ 80.692, degrees of freedom ¼ 18). As discussed previously, the main results are presented as the adjusted risk ratio of switching to a non-rosiglitazone drug for rosiglitazone users in the post-publication versus the pre-publication period and a similar ratio for pioglitazone users. The results are documented in Table 6. The analysis found Impact of studies on Rosi Prob of switch to non rosi in POST period Prob of switch to non rosi in PRE period ¼ 2:64 ðP ¼ 0:046Þ ¼
In other words, there was a 2.64-fold (P ¼ .046) increase in switching to nonindex drug
%
56.20 (10.76) d
%
56.13 (10.40) d
among rosiglitazone users in the postperiod relative to preperiod. For switching to the nonindex drug, among pioglitazone users in the postperiod compared with the preperiod, there was no change: Impact of studies on Pio Prob of switch to non pio in POST period Prob of switch to non pio in PRE period ¼ 0:72 ðP ¼ 0:583Þ
¼
Thus, the results show that the trend of switching to the nonindex drug moved in the opposite direction, that is, rosiglitazone users were more likely to switch to non-rosiglitazone drug in postperiod, whereas pioglitazone users were less likely (although statistically insignificant) to switch to non-pioglitazone drug in postperiod. Finally, the differences-in-differences estimate of switching to a nonindex drug showed that in the post-publication period, the rate of switching for rosiglitazone users was higher by 3.64 (P ¼ .090) times relative to that for pioglitazone users. In addition to the main analysis, sensitivity analysis on the definition of the main outcome
56
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
Table 5 ORs (with associated P values) of differences-in-differences multinomial regression Parameter
Discontinued POc
OR
P O c2
0.54 1.04 1.52b 0.97a 1.20
!.0001 .834 d !.0001 .203
0.64 1.45 4.03b 0.96c 0.72
.456 .477 d .052 .415
0.83 1.27 0.68a
.173 .726 .008
0.67 0.00 3.40a
.291 .982 .049
1.09
.617
0.72
.522
OR 1 if observation in the post period 1 if index drug is rosiglitazone Post rosiglitazone (interaction term) Age 1 if insurance type is HMO (base is all other types of insurance) 1 if gender is male (base is gender ¼ female) 1 if patient has a history of MI 1 if patient has any risk factors for MI (including hypertension) 1 if patient has any risk factors for HF (other than hypertension)
Switched to nonindex drug 2
a
a
Significant at the 5% level. In this nonlinear specification of multinomial logistic regression, the coefficient of the interaction term is not the estimator of the true interaction effect and the P value estimated by statistical packages is incorrect; therefore, we do not report them.19,21 c Significant at the 10% level. b
variables was also performed. Switching and discontinuation of the index drug were defined in terms of the number of days of supply plus 30 days’ grace period. As sensitivity analysis, the main analysis was repeated with a grace period of 60 and 15 days. For both the analyses, the estimated size of the effect was similar; however, with 15 days’ grace period, the differences-in-differences estimates were statistically insignificant. Discussion The main thrust of this study is to learn and document the effect of unanticipated new clinical information (the meta-analyses) on the change in the pattern of use of OHAs in general and TZDs in specific. The release of the meta-analyses is akin to an unanticipated change in the body of knowledge that was directed toward one of the OHAs, that is, rosiglitazone and hence may be assumed to be exogenous. Because both pioglitazone and rosiglitazone are from the TZD class of drugs, pioglitazone became a natural contender for the comparator group. If the market’s response is to discount this adverse news, then there should not be any difference in the pattern of use of OHAs in general and TZDs specifically. A retrospective analysis of pharmacy claims by Starner et al23 described the trends in use of rosiglitazone and pioglitazone from January 1, 2007, through May 31, 2008, the period covering the release of meta-analysis
and issuance of black box warnings.23 They calculated the average number of pharmacy claims per day per million members for rosiglitazone, pioglitazone, and a comparison drug, sitagliptin. They found that the aggregate pharmacy claims for rosiglitazone decreased by more than half between January 2007 and December 2007 and up to twothirds by May 2008. If prescribers reacted cautiously by assuming that the safety problems apply to all the drugs in the therapeutic class, then the prescriptions for both pioglitazone and rosiglitazone should have similar decline. However, Starner et al23 found that pharmacy claims per million members for pioglitazone remained steady over the same period. Therefore, it seems that the response to the adverse news was interpreted more narrowly to rosiglitazone. The main findings of this study are that (1) the number of patients on monotherapy rosiglitazone reduced by half in the post-publication period but remained stable for all the other monotherapy drugs over the same period, (2) there is a significant decrease in the discontinuation rates for all other drugs in the postperiod but an insignificant decrease in the discontinuation rates for monotherapy rosiglitazone patients over the same period, (3) there is a significant increase in the rates of either continuing or adding another drug for all the other drugs but a marginal and insignificant increase for monotherapy rosiglitazone patients in the postperiod, (4) there is a slight and insignificant decrease in switching to a nonindex drug for all
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
57
Table 6 Switching to nonindex drugs (post-publication vs pre-publication): rosiglitazone versus pioglitazone Adjusted relative risk: post versus pre (P value) Rosi (post vs pre)
Adjusted ratio of relative risk: rosi versus pio (P value)
2.64a (P ¼ .046) Rosi (post vs pre)/pio (post vs pre)
Pio (post vs pre) a b
0.72 (P ¼ .583)
Significant at the 5% level. Significant at the 10% level.
other drugs in the postperiod but a large increase in the rates of switching to nonindex drug for monotherapy rosiglitazone users, and (5) monotherapy rosiglitazone users were 3 times more likely to switch to a non-rosiglitazone drug in the postperiod relative to monotherapy pioglitazone users. This analysis extends the previous study of Starner et al23 by going a step further and analyzing how the patients assigned to each OHA change their pattern of use in the post-publication period relative to the pre-publication period. As the results illustrate, the pattern of use differed fundamentally between monotherapy rosiglitazone users and users of all the other monotherapy OHAs in the postpublication period. The most pronounced result was the differences in rates of switching to nonindex drugs. The results from the multinomial regression demonstrate that not only are monotherapy rosiglitazone patients switching to non-rosiglitazone drugs at a higher rate, but the rate also is more than 3 times higher than the similar switches among monotherapy pioglitazone users in the post-publication period relative to the pre-publication period. This shows that the market response, as observed by patient/prescriber decisions, to the adverse news was interpreted narrowly to rosiglitazone, and there is little or no spillover to the other drugs. Many researchers have analyzed the effect of risk warnings on the use of other drugs as well. For example, studies have been conducted on the effect of 2004 FDA warnings24-26 on the increased suicidal risks in antidepressant trials for the treatment of youth with major depressive disorder, and those studies, like this study, have found that the risk e
3.64b (P ¼ .090)
warnings lead to the decrease in the use of antidepressants in children and youth. More specifically, Kurian et al27 estimated a 33% decline in new users aged between 2 and 17 years after the FDA warnings among Tennessee Medicaid youth data. Libby et al28 found that antidepressant dispensing among depression-diagnosed youths (5-17 years of age) fell from 59.2% in the pre-warning period to 55% in the post-warning period between October 1998 and September 2005 in commercially insured population. In a follow-up article using the same data set, Libby et al29 showed sustained decline in antidepressant treatment up to 2007. Limitations The data analysis is based on a population from Maryland, District of Columbia, and Northern Virginia, and the results may differ for data from other states or commercial data sets. Additionally, medical and pharmacy claims data were used to extract information about history of MI and HF and risk factors for MI and HF, and therefore, risk factors without a claim-related diagnosis are not included in the analysis, such as tobacco use, physical inactivity, and obesity. Also, studies have shown that there are diagnostic coding errors in Medicaid claims data,30 and the same may be true for these data as well. As with all the studies using claims data only, there was no information on clinical indicators, such as blood pressure or laboratory results. Last, other comorbidities were not taken into account. Because the study is from a longitudinal data set, 440 patients are in both the pre-publication and post-publication periods.e This may cause the
Four hundred forty patients were in both the pre-publication and post-publication periods. Of these, 270 patients’ index drug was pioglitazone in both the periods. In other words, of those patients on pioglitazone in the pre-publication period, 57% were also on pioglitazone in the post-publication period. An additional 130 patients’ index drug was rosiglitazone in both the periods. This implies that of those patients on rosiglitazone in the pre-publication period, only 35% were also on the same drug in the post-publication period. The rest of the 40 patients’ index drug in the pre-publication period was rosiglitazone but in the post-publication period was pioglitazone.
58
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59
problem of correlated responses. The analysis for change of pattern and differences-in-differences multinomial regression ignores this correlation. This study compares monotherapy rosiglitazone users with monotherapy pioglitazone users, and therefore, this study may not be generalizable to all patients with type 2 diabetes. Conclusion Comparing the pattern of use of drugs before and after the publication of meta analysis shows that the number of patients on monotherapy rosiglitazone reduced by half in the postpublication period but remained stable for all the other monotherapy drugs over the same period. It also showed that there was a significant decrease in the discontinuation rates for all other drugs in the post-publication period but an insignificant decrease in the discontinuation rates for monotherapy rosiglitazone patients over the same period. Additionally, there was a significant increase in the rates of either continuing or adding another drug for all the other drugs but a marginal and insignificant increase for monotherapy rosiglitazone patients in the post-publication period. Furthermore, there was a slight and insignificant decrease in switching to a nonindex drug for all other drugs in the post-publication period but a large increase in the rates of switching to a nonindex drug for monotherapy rosiglitazone users, and finally, monotherapy rosiglitazone users were 3 times more likely to switch to a non-rosiglitazone drug in the post-publication period relative to pioglitazone users. Therefore, our results show that the publications about safety risks had differential effect between the 2 drugs within the therapeutic class. Acknowledgments Funding for the first author’s postdoctoral fellowship was provided by an unrestricted grant from Takeda Pharmaceutical Company. Portions of this manuscript were presented as posters at the annual meeting of the American Association of Diabetes Education, August 5-8, 2009, in Atlanta, GA, and at the 12th Annual European Congress of the International Society for Pharmacoeconomics and Outcomes Research, October 24-27, 2009, in Paris, France. One of the authors reported receipt of consulting or honoraria from Amgen, Amylin, Bayer, BMS, Genentech, GlaxoSmithKline, Lilly Merck, Novartis, Pfizer,
and Sanofi-Aventis and grant support from GlaxoSmithKline, Novartis, Pfizer, and SanofiAventis.
References 1. Owens DR. Thiazolidinediones: a pharmacological overview. Clin Drug Invest 2002;22:485–505. 2. Meriden T. Progress with thiazolidinediones in the management of type 2 diabetes mellitus. Clin Ther 2004;26(2):177–190. 3. Bell DS. Type 2 diabetes mellitus: what is the optimal treatment regimen? Am J Med 2004;116(suppl 5A): 23S–29S. 4. Yanagawa T, Araki A, Sasamoto K, Shirabe S, Yamanouchi T. Effect of antidiabetic medications on microalbuminuria in patients with type 2 diabetes. Metabolism 2004;53(3):353–357. 5. Henney JE. Withdrawal of troglitazone and cisapride. JAMA 2000;283(17):2228. 6. Haffner SM, Greenberg AS, Weston WM, Chen H, Williams K, Freed MI. Effect of rosiglitazone treatment on nontraditional markers of cardiovascular disease in patients with type 2 diabetes mellitus. Circulation 2002;106(6):679–684. 7. Dormandy JA, Charbonnel B, Eckland DJ, et al. Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study (PROspective pioglitAzone Clinical Trial In macroVascular Events): a randomised controlled trial. Lancet 2005;366(9493):1279–1289. 8. Delea TE, Edelsberg JS, Hagiwara M, Oster G, Phillips LS. Use of thiazolidinediones and risk of heart failure in people with type 2 diabetes: a retrospective cohort study. Diabetes Care 2003;26(11): 2983–2989. 9. Hartung DM, Touchette DR, Bultemeier NC, Haxby DG. Risk of hospitalization for heart failure associated with thiazolidinedione therapy: a Medicaid claims-based case-control study. Pharmacotherapy 2005;25(10):1329–1336. 10. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular causes. N Engl J Med 2007;356(24): 2457–2471. 11. Singh S, Loke YK, Furberg CD. Long-term risk of cardiovascular events with rosiglitazone: a metaanalysis. JAMA 2007;298(10):1189–1195. 12. Lincoff AM, Wolski K, Nicholls SJ, Nissen SE. Pioglitazone and risk of cardiovascular events in patients with type 2 diabetes mellitus: a metaanalysis of randomized trials. JAMA 2007;298(10): 1180–1188. 13. News F. Manufacturers of some diabetes drugs to strengthen warning on heart failure risk. Available at: http://www.fda.gov/NewsEvents/ Newsroom/PressAnnouncements/2007/ucm108966. htm. Accessed 01.12.09.
Jain et al./Research in Social and Administrative Pharmacy 8 (2012) 47–59 14. FDA. FDA Adds boxed warning for heart-related risks to anti-diabetes drug Avandia. Available at: http://www.fda.gov/NewsEvents/Newsroom/Press Announcements/2007/ucm109026.htm. Accessed 01. 12.09. 15. American Heart Association. Risk factors and coronary heart disease. Available at: http://www. americanheart.org/presenter.jhtml?identifier¼4726. Accessed 12.12.08. 16. Hunt SA, Abraham WT, Chin MH, et al. ACC/AHA 2005 Guideline Update for the Diagnosis and Management of Chronic Heart Failure in the Adult: a report of the American College of Cardiology/ American Heart Association Task Force on Practice Guidelines (Writing Committee to Update the 2001 Guidelines for the Evaluation and Management of Heart Failure): developed in collaboration with the American College of Chest Physicians and the International Society for Heart and Lung Transplantation: endorsed by the Heart Rhythm Society. Circulation 2005;112(12):e154–e235. 17. Goodman LA. Simultaneous confidence intervals for contrasts among multinomial populations. Ann Math Stat 1964;35(2):716–725. 18. Goodman LA. On simultaneous confidence intervals for multinomial proportions. Technometrics 1965; 7(2):247–254. 19. Ai C, Norton EC. Interaction terms in logit and probit models. Econ Lett 2003;80(1):123–129. 20. Mullahy J. Interaction effects and difference-indifference estimation in loglinear models. NBER Technical Working Papers 1999;254. 21. Norton E, Wang H, Ai C. Computing interaction effects and standard errors in logit and probit models. Stata J 2004;4:103–116. 22. Greene W. Econometric Analysis. 5th ed. New Jersey, Prentice Hall: Macmillan; 2003.
59
23. Starner CI, Schafer JA, Heaton AH, Gleason PP. Rosiglitazone and pioglitazone utilization from January 2007 through May 2008 associated with five risk-warning events. J Manag Care Pharm 2008;14(6):523–531. 24. FDA. FDA public health advisory: worsening depression and suicidality in patients being treated with antidepressant. Available at: http://www. fda.gov/cder/drug/antidepressants/Antidepressanst PHA.htm. Accessed 01.12.2010. 25. FDA. FDA news: FDA launches a multi-pronged strategy to strengthen safeguards for children treated with antidepressant medications. Available at: http:// www.fda.gov/bbs/topics/news/2004/NEW01124. html. Accessed 01.12.2010. 26. FDA. FDA public health advisory: reports of suicidality in pediatric patients being treated with antidepressant medications for major depressive disorder (MDD). Available at: http://www.fda.gov/cder/ drug/advisory/mdd.htm. Accessed 01.12.2010. 27. Kurian BT, Ray WA, Arbogast PG, Fuchs DC, Dudley JA, Cooper WO. Effect of regulatory warnings on antidepressant prescribing for children and adolescents. Arch Pediatr Adolesc Med 2007;161: 690–696. 28. Libby AM, Brent DA, Morrato EH, Orton HD, Allen R, Valuck RJ. Decline in treatment of pediatric depression after FDA advisory on risk of suicidality with SSRIs. Am J Psychiatry 2007;164: 884–891. 29. Libby AM, Orton H, Valuck RJ. Persisting decline in depression treatment after FDA warnings. Arch Gen Psychiatry 2009;66:633–639. 30. Hennessy S, Bilker WB, Weber A, Strom BL. Descriptive analyses of the integrity of a US Medicaid claims database. Pharmacoepidemiol Drug Saf 2003;12(2):103–111.