pharmacoepidemiology and drug safety 2014; 23: 348–356 Published online 23 December 2013 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3558
ORIGINAL REPORT
Cardiac mortality in users of olmesartan, other angiotensin-receptor blockers and angiotensin-converting enzyme inhibitors Alexander M. Walker1*, Caihua Liang2, C. Robin Clifford2, Crawford Parker3 and Allen Feldman3 1
World Health Information Science Consultants, Newton, MA, USA OptumInsight, Waltham, MA, USA 3 Daiichi Sankyo Pharma Development, Edison, NJ, USA 2
ABSTRACT Purpose Clinical trials of olmesartan for prevention of progression of renal disease in patients with diabetes showed renal protection but an unexpected imbalance in cardiac deaths. The US Food and Drug Administration requested from the manufacturer a cohort study of olmesartan, other angiotensin-receptor blockers (ARBs), and angiotensin-converting enzyme (ACE) inhibitors in a large population. Methods A retrospective cohort study was conducted with the cooperation of a US health insurer. Subject entry and follow-up ran from 2002 through 2009. In propensity-matched cohorts, the primary analysis considered continuous current users. Endpoints were sudden cardiac death (SCD) and all-cause mortality, identified through the US National Death Index, supplemented by insurance and hospital discharge data. Statistical estimation was based on proportional hazards analyses with 95% confidence intervals. Power calculations had shown that 25 000 olmesartan initiators would be required to detect relative risks (RRs) of SCD of twofold or greater. Results A total of 57 123 initiators of olmesartan were matched 1:2 to initiators of other ARBs and 41 801 to initiators of ACE inhibitors. Average follow-up time ranged from 8 to 9 months. Olmesartan initiators and comparators experienced similar patterns of both outcomes, with RRs ≤1.0 and upper confidence bounds ≤1.6. Among persons with prior use of hypoglycemic agents, in comparison with other ARBs, the RR of SCD for olmesartan users was 0.8, with an upper confidence bound of 2.2. Conclusion The results of this well-powered study do not raise concerns for the risk of SCD or death from all causes among olmesartan users in comparison with users of other ARBs or ACE inhibitors. Copyright © 2013 John Wiley & Sons, Ltd. key words—olmesartan; ARBs; ACE inhibitors; sudden cardiac death; mortality; safety; pharmacoepidemiology Received 26 July 2013; Revised 13 November 2013; Accepted 19 November 2013
BACKGROUND Two randomized trials that were intended to address olmesartan’s ability to prevent or delay the progression of renal disease in patients with diabetes showed unexpected elevations of cardiovascular mortality in comparison with placebo.1,2 In the ROADMAP study, the time to first occurrence of microalbuminuria was 23% longer in 2232 patients randomized to olmesartan 40 mg once daily than in 2215 patients randomized to placebo.1 There was a rough equality of combined cardiovascular endpoints between the olmesartan (46 events, 2.1%) and
*Correspondence to: A. M. Walker, World Health Information Science Consultants, 275 Grove Street, Suite 2-400, Newton, MA 02466, USA. Email: Alec. [email protected]
Copyright © 2013 John Wiley & Sons, Ltd.
placebo (37 events, 1.7%) arms, for a relative risk (RR) of 1.2 (95% confidence interval [CI]: 0.8–1.9). There was however a notable imbalance in cardiovascular mortality (15 deaths versus 3, RR: 5.0, 95%CI: 1.4–17.1). The second trial, called ORIENT, enrolled type 2 diabetic patients with nephropathy.2 A combined cardiovascular endpoint of cardiovascular death, stroke or nonfatal myocardial infarction occurred in 18 of 282 (6.4%) patients receiving olmesartan in a variable dose that was titratable from 10 to 40 mg daily and in 21 of 284 (7.4%) patients on placebo (RR: 0.9, 95%CI: 0.5–1.6). Cardiovascular mortality considered in isolation was higher in the olmesartan group than in the placebo group (10 deaths versus 3, RR: 3.4, 95% CI: 0.9–12.3). Aware of findings from ROADMAP and ORIENT, the US Food and Drug Administration (FDA) requested that the manufacturer of olmesartan, Daiichi
cardiac mortality in users of antihypertensive agents
Sankyo, Incorporated, “conduct an epidemiologic study using claims or electronic health records data to evaluate the comparative incidence of sudden cardiac death, in-hospital fatal myocardial infarction and total mortality in olmesartan users vs. users of other angiotensin-receptor blockers and in olmesartan users vs. ACE inhibitor users.”† The goal of the present study was to meet this requirement and to provide scientific evidence on the incidence of cardiac mortality in users of olmesartan, other angiotensin-receptor blockers (ARBs) and angiotensin-converting enzyme (ACE) inhibitors. METHODS Sponsorship and approvals The study was conducted by World Health Information Sciences Consultants (WHISCON) in cooperation with Optum, a member of the United Health Group companies, under a research contract with Daiichi Sankyo, Incorporated. The protocol was provided in draft to the FDA, modified in response to FDA comments, and resubmitted to the FDA before initiation, after no further suggestions were received. The original protocol as implemented had as its primary analyses a comparison of olmesartan initiators to initiators of other ARBs and of ACE inhibitors among persons who had not previously used either other ARBs or ACE inhibitors. Secondary analyses included a subgroup of persons with prior use of hypoglycemic agents, a comparison with ARBs initiators among persons with prior ACE inhibitor use, comparisons of olmesartan with specific ARBs (they are not reported here because they are not different from the overall comparison) and sensitivity analyses, these being intent-to-treat analyses in the primary cohort and a more stringent definition of continuous use in the as-treated analyses. (The sensitivity analyses are also not reported here, as they gave essentially the same results as those reported in the succeeding sections.) After seeing the study results, which included a nonsignificant elevation in risk of about twofold among olmesartan users in the primary cohort among persons with prior use of hypoglycemic agents, the FDA requested additional subgroup analyses in persons with prior hypoglycemic use and prior use of ACE inhibitors, persons whom the primary cohort had omitted.
349
Source population A large US commercial health insurance company provided insurance claims and enrollment data for this work. The insurance included drug benefits along with coverage for services. Although people of all ages were represented, the proportion of subjects aged 65 years and older (ages at which most Americans have health insurance through Medicare) was smaller than in the USA overall. For 2009, data relating to approximately 13.3 million individuals with both medical and pharmacy benefit coverage were available. Persons in the study were those who at age 20 years or older received a first dispensing for olmesartan, another ARB or an ACE inhibitor, the “study drugs,” between 2002 and 2009. For inclusion, an individual’s first dispensing of a study drug had to come after a period of at least 183 days during which the subject was eligible for insurance benefits. The individual had to have had no prior dispensing of any ARB. Individuals remained under observation until they terminated treatment with the initiating product, left insurance coverage, entered long-term care, died or reached the close of observation at the end of 2009. Termination of treatment was defined as the first occurrence of a gap of 30 days beyond the treatment that would be ascribable to a dispensing (calculated as dispensing date plus days of medication supplied), with no new dispensing. Additionally, patients who initiated an ARB while a member of the olmesartan or ACE inhibitor cohorts were censored at the first dispensing of an ARB, and patients who initiated olmesartan while a member of the other ARB or ACE inhibitor cohorts were censored at the first dispensing of olmesartan. Persons who initiated treatment with two different study drugs on the same day were not included. The end of insurance coverage was taken to have occurred 32 days following an administrative record of discontinuation of eligibility without an indication of reinstatement. If an individual terminated enrollment with the insurer and then enrolled again at a later date, each enrollment period was considered separately, so that a person could potentially leave and reenter the cohort through discontinuation and reenrollment. (Because the outcome under study was death, no person could suffer an event in two different enrolment periods.) All the cohort entry criteria had to be met in each enrollment period. Covariates
† Personal communication to Allen Feldman, Daiichi Sankyo, Incorporated, 3 June 2011.
Copyright © 2013 John Wiley & Sons, Ltd.
Definitions of baseline characteristics were keyed to the date of first treatment with a study drug. In addition Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
350
a. m. walker et al.
to age and sex, we noted health-care utilization and drugs dispensed in the 183 days prior to treatment initiation. Drugs were considered both in narrow therapeutic classes and as summary counts of the numbers of distinct drugs dispensed. Diagnoses associated with insurance claims for physician and hospital care were identified in the 30 days before treatment initiation and during the preceding 153 days.
Comparison groups Comparison groups were built without knowledge of outcome status, as the cohorts needed to be constructed in order to know which records to submit to the US National Death Index (NDI) for linkage. Within each of the eight calendar years of cohort accrual from 2002 through 2009 and within strata defined by the prior use of ACE inhibitors or not, we formed cohorts of initiators of the study drugs, with covariate balance achieved through matching initiators of other ARBs or ACE inhibitors to initiators of olmesartan, using a propensity score particular to that calendar year and prior-use stratum. In strata without prior use of ACE inhibitors, initiators of olmesartan could be compared both with initiators of other ARBs and of ACE inhibitors. In strata with prior use of ACE inhibitors, only the contrast of olmesartan initiators versus initiators of other ARBs was available. The propensity score models were initially derived by backward elimination in a logistic regression that modeled the choice of olmesartan versus either other ARBs or ACE inhibitors as a function of all the covariates. All retained predictors in the model had a p-value of 0.1 or less. After propensity score models had been estimated for each comparison (olmesartan versus other ARBs and olmesartan versus ACE inhibitors) in each stratum, we used the union of all retained predictor variables from all comparisons to fit a final common logistic model that had the same covariates for every stratum. Olmesartan patients were matched individually to two comparators from other ARB and ACE inhibitor initiators in the stratum with no prior ACE inhibitor use in the same calendar year, except in 2009 where each olmesartan patient was matched to one ARB patient, as there were not enough other ARB initiators for a 2:1 match. In the strata with prior ACE inhibitor use, olmesartan initiators were matched to two ARB initiators. Matching used a nearestneighbor algorithm with a caliper of 0.1 on the logit propensity scale, a value that represented approximately 0.2 times the average standard deviation of the year-specific logit propensity scores.3,4 Copyright © 2013 John Wiley & Sons, Ltd.
Outcomes The outcomes were sudden cardiac death (SCD) and death from any cause. Another planned outcome of in-hospital deaths from myocardial infarction yielded fewer than two additional deaths in any cohort, did not change any qualitative finding and is not reported here. In order to ascertain deaths, a business affiliate of the insurer (Optum) submitted personal identifying information of all 178 003 cohort members who terminated participation in the insurance plan from 2002 to 2010 to the NDI for linkage to the NDI records. If different enrollment periods for an individual were associated with different data (e.g., different family names or dates of birth), records corresponding to each data variant were submitted to the NDI. Potential matches returned by the NDI were accepted according to the NDI’s suggested guidelines. SCD was classified according to an algorithm proposed by Chung et al.5 This involved an NDI match to one of the submitted records with an underlying cause of death from Table S1. The match had to occur on or before termination of enrollment, with the additional requirement that there be no evidence of a terminal hospital stay. For 70 subjects for whom there were insurance claims recorded more than 30 days after an NDIdetermined death, or for whom the NDI date of death differed by more than seven days from an apparent death recorded in the insurance claims, two adjudicators examined the chronological insurance claims profile and the NDI record without knowledge of the individual’s exposure status. When the later claims reflected billings that could recur automatically (such as durable equipment rentals) or medical–legal services, the deaths were retained (32 occurrences). Individuals whose claims showed an uninterrupted sequence of billings spanning the putative date of death from the NDI were classed as not having died during study follow-up (38 occurrences). Thirty-three of the latter cases were persons for whom there was no Social Security number in the submitted identifying data from the insurer, so that the rejected, apparently false-positive NDI matches had been based only on similarity of names and dates of birth. There were in addition 10 persons for whom Social Security number was unavailable in the insurance files and who had no match in the NDI, but for whom there was an International Classification of Diseases, 9th Revision (ICD-9) code indicating death in the insurance claims, with no insurance claims more than seven Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
cardiac mortality in users of antihypertensive agents
days after. These events were retained in the all-cause mortality analyses but did not qualify as SCD cases. Analysis For each of the study endpoints, we undertook a common proportional hazards analysis over strata defined by calendar year and prior use of ACE inhibitors. In persons without prior use of ACE inhibitors, initiators of olmesartan were compared with initiators of other ARBs and with initiators of ACE inhibitors in separate analyses. In cohort members with prior use of ACE inhibitors, it was only possible to compare initiators of olmesartan and other ARBs. There was a prespecified subgroup analysis in persons with prior use of hypoglycemic agents and prior use of ACE inhibitors. The original protocol specified that the primary analysis would be in persons who had not used ACE inhibitors previously and that the combined results (including those with and without prior ACE inhibitor use) would be a secondary comparison. After it had reviewed (i) the overall combined findings and (ii) the subgroup analysis in persons who had not previously used ACE inhibitors and who had used hypoglycemic agents, the FDA requested that the analysis be expanded so that the subgroup of persons with prior hypoglycemic use would additionally include prior users of ACE inhibitors, permitting a combined analysis. To check for differential censoring by treatment group in the compared cohorts, we conducted proportional hazards analyses with censoring for any reason taken as the dependent variable and study drug group (olmesartan, other ARBs, and ACE inhibitors) as predictors. The hazard ratios for termination of follow-up, comparing olmesartan with comparators, were in both cases equal to 0.99. For the analyses in prior users of hypoglycemic agents, we checked the overall covariate distributions between olmesartan and each of the comparators. We added to the outcome model new terms that summarized covariates that were out of balance in this subgroup. In a combined as-treated analysis stratified by prior ACE inhibitor use, a subject could be an initiator of an ACE inhibitor in the primary cohort in one year, and then in a subsequent year start olmesartan or another ARB, qualifying as an olmesartan (or a matching other-ARB) initiator in the later year. We accounted for the second cohort membership by generating a new follow-up record for the second exposure and invoking a robust variance estimator. Copyright © 2013 John Wiley & Sons, Ltd.
351
Power calculations indicated that sizes for the olmesartan cohort above 25 000, with two comparators for each olmesartan initiator, would have a greater than 90% power to detect RRs of 2 or greater for SCD. RESULTS Characteristics of the source population and cohorts From 2002 through 2009, there were 42 410 olmesartan initiators who had not previously used ACE inhibitors and 21 930 who had. The corresponding counts for initiators of other ARBs (cohort candidates) with no prior ACE inhibitor use and initiators of ARBs with prior ACE inhibitor use were 130 225 and 68 439, respectively. Over the same time, there were 392 543 initiators of ACE inhibitors. The propensity models retained 205 predictor variables. Matching on propensity score yielded cohorts with close balance on demographics, diagnoses, and drug classes dispensed, as well as on a variety of health-care utilization measures. For no baseline characteristic was the prevalence difference between cohorts (summed over all the years together) more than 0.5%. The final matched cohort sizes for the comparison between olmesartan and other ARBs were 38 750 and 72 326, respectively, for patients with no prior use of ACE inhibitors. (Recall that in 2009, only a 1:1 match was possible, so the overall ratio fell short of 2:1.) These cohorts had average observation times of 8.3 and 8.7 months, respectively. The corresponding matched cohort sizes for the olmesartan (with no prior use of ACE inhibitors) versus ACE inhibitor comparisons were 41 801 and 83 602 (8.4 months average observation time in each). Among those with prior use of ACE inhibitors, there were 18 373 initiators of olmesartan (9.3 months average observation time) and 36 746 matched initiators of other ARBs (9.5 months), making for a total of 57 123 olmesartan initiators matched to 109 072 initiators of ARBs including groups with and without prior use of ACE inhibitors. All cohorts in this commercially insured population were predominantly middle-aged, approximately 52% being in the range 45–64 years. All cohorts were either 53 or 54% men. Table 1a through 1d shows demographic characteristics and summary measures of health-care utilization for cohort members without and with prior use of ACE inhibitors, and for the subsets of each of those groups who received hypoglycemic agents before cohort entry. The users of hypoglycemic agents were about 7% of the olmesartan users and Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
352
a. m. walker et al.
Table 1a. Age, sex, and health-care utilization in the period 183 days prior to cohort entry: persons without prior ACEI use All patients Drug N
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Olm
ARB
ACEI
Olm
ARB
Olm
ACEI
42 410
130 225
392 543
38 750
72 326
41 801
83 602
%
%
%
%
%
%
%
8.2 22.0 35.5 27.0 5.6 1.6
6.8 19.9 33.6 29.6 7.3 2.7
8.1 20.6 33.8 29.2 6.4 1.9
7.6 21.5 35.1 28.0 6.0 1.8
7.6 21.6 35.1 28.0 6.0 1.7
8.2 22.0 35.4 27.1 5.7 1.6
8.0 21.7 35.5 27.4 5.7 1.7
53.9 46.1 Mean 7.7 3.7 3.9 0.06 2956 510 1734 712
51.5 48.5 Mean 7.9 3.8 3.9 0.08 3560 551 1838 1172
58.2 41.8 Mean 7.5 3.7 3.3 0.14 4955 531 1809 2614
53.1 46.9 Mean 7.6 3.6 3.8 0.06 2982 502 1734 747
53.3 46.7 Mean 7.6 3.6 3.8 0.06 3126 513 1778 836
54.1 45.9 Mean 7.6 3.6 3.8 0.06 2940 501 1718 721
54.2 45.8 Mean 7.7 3.6 3.8 0.06 3172 526 1838 808
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
Table 1b. Age, sex and health-care utilization 183 days prior to cohort entry: persons with prior ACEI use All Drug N
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Olm
ARB
Olm
ARB
21 930
68 349
18 373
36 746
%
%
%
%
3.8 16.2 34.6 36.3 7.6 1.5
3.6 15.0 33.1 37.7 8.6 2.0
3.9 15.7 33.9 37.0 7.9 1.7
3.8 15.9 33.4 36.9 8.2 1.8
55.2 44.8 Mean 9.1 6.3 4.7 0.08 4321 936 2113 1272
53.2 46.8 Mean 9.8 6.8 5.0 0.14 5753 1035 2451 2268
54.1 45.9 Mean 9.1 6.3 4.6 0.09 4474 941 2140 1393
54.0 46.1 Mean 9.1 6.3 4.6 0.09 4593 953 2181 1459
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
(before matching) over 10% of the users of other ARBs and ACE inhibitors. All the tables give the distributions of characteristics in the general population Copyright © 2013 John Wiley & Sons, Ltd.
(before matching) and in the matched populations that form the basis of the analysis. Persons who filled prescriptions for hypoglycemic agents before they Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
353
cardiac mortality in users of antihypertensive agents
Table 1c. Age, sex, and health-care utilization in the period 183 days prior to cohort entry: persons without prior ACEI use, subset with prior use of hypoglycemic agents All patients
Matched
Drug
Olm
ARB
ACEI
Olm
ARB
Olm
ACEI
N
2991
13 765
56 421
2892
5536
2982
6403
%
%
%
%
%
%
%
5.6 16.5 33.5 33.4 8.9 2.2
4.8 15.2 31.3 36.4 9.5 2.8
7.2 18.4 32.7 32.4 7.7 1.8
5.3 16.0 33.3 34.0 9.1 2.3
5.9 17.4 33.0 34.4 7.4 2.0
5.6 16.5 33.6 33.3 8.9 2.3
6.5 19.5 34.6 30.3 7.4 1.8
51.4 48.7 Mean 9.5 6.6 4.7 0.09 4489 1045 2178 1267
52.2 47.8 Mean 9.7 6.8 4.8 0.13 5436 1138 2334 1963
57.8 42.2 Mean 8.9 6.3 4.2 0.15 5757 1037 2116 2604
51.3 48.7 Mean 9.5 6.5 4.7 0.10 4527 1049 2193 1285
52.5 47.5 Mean 9.0 6.4 4.4 0.09 4544 1026 2208 1311
51.4 48.6 Mean 9.5 6.6 4.7 0.09 4484 1037 2177 1271
53.3 46.7 Mean 8.9 6.1 4.4 0.08 4168 988 2175 1005
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
Table 1d. Age, sex and health-care utilization 183 days prior to cohort entry: persons with prior ACEI use, subset with prior use of hypoglycemic agents All
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Drugs
Olm
ARB
Olm
ARB
N
4398
17 566
4037
8211
%
%
%
%
2.2 12.9 32.6 41.0 9.7 1.6
2.6 11.9 30.7 42.5 10.4 2.0
2.3 12.7 32.2 41.2 10.0 1.7
2.7 12.8 31.3 41.1 10.2 1.8
55.7 44.3 Mean 10.6 9.1 5.4 0.12 6038 1622 2473 1942
55.1 45.0 Mean 11.1 9.3 5.6 0.19 7498 1632 2888 2978
55.5 44.5 Mean 10.6 9.1 5.3 0.13 6129 1620 2477 2032
56.2 43.8 Mean 10.2 8.7 5.1 0.12 5828 1564 2471 1794
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensinconverting enzyme inhibitor.
entered the cohort were older, carried more diagnoses and were heavier consumers of health care. Table S2a through S2d shows the 15 diagnoses most frequently Copyright © 2013 John Wiley & Sons, Ltd.
seen in each of the groups during the period from 183 through 31 days prior to cohort entry. For all groups, there was substantial balance in demographics, health-care utilization, and diagnoses after matching. As detailed below, a number of characteristics were nonetheless sufficiently out of balance in the subsets with prior hypoglycemic agent use to warrant further covariate control.
Mortality in comparison with initiators of other angiotensin-receptor blockers and angiotensin-converting enzyme inhibitors The analyses for SCD in olmesartan versus other ARBs and olmesartan versus ACE inhibitors are given in Table 2. Note that the comparison under the heading “All initiators” in the table includes persons with prior ACE inhibitor use and is necessarily restricted to ARBs. For both comparisons, the mortality from SCD was lower in the olmesartan group than in the comparator, and for both, the CIs included an RR of 1.0. All-cause mortality is shown in Table 3. The rates were lower in the olmesartan initiators than in the comparators, with upper confidence bounds for the RR at or just below 1.0. Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
354
a. m. walker et al.
Table 2. Mortality from sudden cardiac death Mortality Deaths
Person-years
Rate
95%CI
Ratio
95%CI
Olmesartan Other ARBs
14 32
Among persons without prior use of ACEIs 26 774.0 0.5 52 189.2 0.6
0.3–0.9 0.4–0.9
0.9
0.5–1.6
Olmesartan ACEIs
13 42
29 116.4 58 240.1
0.2–0.8 0.5–1.0
0.6
0.3–1.1
Olmesartan Other ARBs
24 60
0.8
0.5–1.3
0.4 0.7
All initiators (with or without prior use of ACEIs) 41 080.5 0.6 0.4–0.9 81 414.9 0.7 0.6–0.9
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and (for “All initiators”) by prior use of ACEIs. ARBs, angiotensin-receptor blockers; ACEIs, angiotensin-converting enzyme inhibitors; CI, confidence interval.
Table 3. All-cause mortality Mortality Deaths
Person-years
Rate
95%CI
Ratio
95%CI
Olmesartan Other ARBs
49 133
Among persons without prior use of ACEIs 26 774.0 1.8 52 189.2 2.5
1.4–2.4 2.1–3.0
0.7
0.5–1.0
Olmesartan ACEIs
51 150
29 116.4 58 240.1
1.3–2.3 2.2–3.0
0.7
0.5–0.9
Olmesartan Other ARBs
101 253
0.8
0.6–1.0
1.8 2.6
All initiators (with or without prior use of ACEIs) 41 080.5 2.5 2.0–3.0 81 414.9 3.1 2.7–3.5
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and by prior ACEI use. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
Users of hypoglycemic agents
DISCUSSION
The rate of SCD was slightly lower among olmesartan initiators who had used hypoglycemic agents previously than it was for similar ARB initiators overall. The overall RR, compared with other ARB users both with and without prior use of ACE inhibitors, was 0.8 with an upper confidence bound of 2.2 (Table 4). In the persons with prior use of hypoglycemic agents who moreover had not previously used ACE inhibitors, the SCD rate in olmesartan users exceeded that in comparators by twofold. This estimate is based on three deaths only in each group and has confidence bounds that run from 0.4 to greater than 10-fold. Although the point estimates in strata with and without prior use of ACE inhibitors were very different, the very small number of events meant that the differences were compatible with chance ( p = 0.18). Table 5 presents the all-cause mortality for the same comparisons. Mortality rates were essentially the same in each instance.
The size of the source data and the opportunity for extensive confounder control mean that these results provide reassurance about the relative safety of olmesartan as it has been used in the USA since market introduction in 2002. Because it reflects actual use, the population studied here differs from the diabetic olmesartan users in the ROADMAP and ORIENT studies. The overall finding of lower cardiac mortality in olmesartan users versus comparators effectively rules out an important mortality-increasing effect of olmesartan in general. Prior users of hypoglycemic agents proved to be a small subgroup, with large variations in point estimates of RR, but no overall indication of excess risk. Nonetheless, for statistical reasons, the study provides only modest assurance about the relative safety of olmesartan in diabetic populations. The signals that gave rise to this study arose in two trials of persons with type II diabetes, and so the
Copyright © 2013 John Wiley & Sons, Ltd.
Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
355
cardiac mortality in users of antihypertensive agents Table 4. Mortality from sudden cardiac death in prior users of hypoglycemic agents Mortality Deaths
Person-years
Rate
95%CI
Olmesartan Other ARBs
3 3
Among persons without prior use of ACEIs 1 881.6 1.6 0.3–4.7 3770.0 0.8 0.2–2.3
Olmesartan ACEIs
3 3
1951.3 4207.3
Olmesartan Other ARBs
6 15
1.5 0.7
Adjusted rate ratio
95%CI
2.1
0.4–11.7
2.0
0.4–10.7
0.8
0.3–2.2
0.3–4.5 0.1–2.1
All initiators (with or without prior use of ACEIs) 4811.8 1.2 0.5–2.7 10 101.7 1.5 0.8–2.4
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and with adjustment for covariates imbalanced because of restriction to a subset of the balanced cohort. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
Table 5. All-cause mortality in prior users of hypoglycemic agents Mortality Deaths
Person-years
Rate
95%CI
Olmesartan Other ARBs
8 16
Among persons without prior use of ACEIs 1881.6 4.3 1.8–8.4 3770.0 4.2 2.4–6.9
Olmesartan ACEIs
8 16
1951.3 4207.3
Olmesartan Other ARBs
25 53
4.1 3.8
1.8–8.1 2.2–6.2
All initiators (with or without prior use of ACEIs) 4811.8 5.2 3.4–7.7 10 101.7 5.2 3.9–6.9
Adjusted rate ratio
95%CI
1.0
0.4–2.4
0.9
0.4–2.1
1.0
0.6–1.6
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and with adjustment for covariates imbalanced because of restriction to a subset of the balanced cohort. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
subgroup with prior hypoglycemic use deserves special scrutiny. ROADMAP excluded persons who had used ACE inhibitors (or other ARBs) in the 6 months before study entry.1 In ORIENT, by contrast, approximately 73% of the patients had used ACE inhibitors prior to study entry, and participants were required as part of the study protocol to continue treatment.2 In the present study, the primary cohort without prior ACE inhibitor use corresponds therefore to the ROADMAP population. The combined results respond more nearly to the two external signals that prompted the study. Both ROADMAP and ORIENT were trials of olmesartan as a preventive therapy in a setting in which placebo was an appropriate comparator. Any risk excess associated with treatment could be seen as an absolute risk of therapy. In testing whether the concerns raised by ROADMAP and ORIENT might affect people treated for hypertension, we and the regulatory agency that requested this study moved into a related question of comparative safety. Nontreatment Copyright © 2013 John Wiley & Sons, Ltd.
of people with hypertension is not a therapeutic option, and so the question became: does olmesartan differ from other treatments, in particular other ARBs or ACE inhibitors, with respect to the risk of cardiac death? This study cannot answer the question of absolute risk in persons with hypertension that might be approached in a hypothetical (as well as unethical and clinically irrelevant) comparison with no treatment at all. In addition to shedding light on therapeutic alternatives, an observational study with active comparators rather than nontreatment as a reference group runs a smaller risk of bias. By comparison with expected medical differences between treated and untreated people, there is an anticipated rough similarity in characteristics between people receiving alternative therapies. A starting covariate balance between compared therapies substantially reduces the scope for confounding by indication, the broad category of biases introduced by the reasons for treatment, which often involve patient characteristics that are difficult to measure or to quantify.6 Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
356
a. m. walker et al.
Fully known covariates and realistic parameterization are assumed in statistical adjustment and matching. The effect of deviations from this ideal state is small when the poorly measured (or unknown) covariates have similar distributions in the compared groups. More than 90% of the olmesartan initiators in the database could be matched to initiators of other ARBs and of ACE inhibitors with very similar patterns of diagnoses associated with insurance claims, drugs dispensed, health-care utilization and demographic characteristics. The average follow-up times of 8 to 9 months from treatment initiation were adequate to evaluate effects of the study drugs over a corresponding term, but were too short to permit a duration–response analysis and they do not rule out effects that would occur with longer follow-up. (Both ROADMAP and ORIENT had average durations of follow-up of about 3 years.) Nor was the study size adequate to permit a dose–response analysis. Because the cohorts had to be formed in order to know which records to submit for NDI matching, it was not possible to screen for potential instruments among the 205 retained predictors. Given the small numbers of events, there is no post hoc technique that could accurately designate instruments. The possibility remains therefore that the analysis may have amplified the impact of unmeasured confounders through what has been called “Z-bias” or “bias amplification.”7,8 The restriction of the initial analysis to persons without prior use of ACE inhibitors had the effect of focusing on a population of treatment initiators who may have been substantially healthier than those with prior ACE inhibitor use. Accordingly, the analyses most representative of the experience of olmesartan initiators generally may be the combined analyses that include both persons who had and who had not used ACE inhibitors before cohort entry. CONFLICT OF INTEREST This work was funded by research contracts between Daiichi Sankyo, Incorporated, and World Health Information Science Consultants, LLC (WHISCON), and between WHISCON and Optum. Crawford Parker and Allen Feldman are employees of Daiichi Sankyo, Incorporated. KEY POINTS • Persons prescribed olmesartan, other ARBs and ACE inhibitors for the first time had similar rates of sudden cardiac death and all-cause mortality during an average of eight months of follow-up. • The presence or absence of diabetes mellitus and of prior treatment with ACE inhibitors did not affect the relative risks. Copyright © 2013 John Wiley & Sons, Ltd.
ETHICS STATEMENT The original protocol was completed on 15 September 2011, and the revision was finalized on 6 March 2012. The New England Institutional Review Board considered and approved the protocol and its revision on 7 November 2011 and 9 April 2012. The FDA requested supplementary analyses in September 2012, and these are reported here, along with additional results requested by reviewers. REFERENCES 1. Haller H, Ito S, Izzo JL, et al. Olmesartan for the delay or prevention of microalbuminuria in type 2 diabetes. N Engl J Med 2011; 364: 907–917. 2. Imai E, Chan JCN, Ito S, et al. Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study. Diabetologia 2011; 54: 2978–2986. 3. Rassen JA, Doherty M, Huang W, Schneeweiss S. Pharmacoepidemiology Toolbox. Boston, MA. http://www.hdpharmacoepi.org. 4. Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 2011; 10: 150–151. 5. Chung CP, Murray KT, Stein M, Hall K, Ray WA. A computer case definition for sudden cardiac death. Pharmacoepidemiol Drug Saf 2010; 19: 563–572. 6. Walker AM. Confounding by indication. (Editorial) Epidemiology 1996; 7: 335–336. 7. Brookhart MA, Stürmer T, Glynn RJ, Rassen J, Schneeweiss S. Confounding control in healthcare database research: challenges and potential approaches. Med Care 2010; 48(6 Suppl): S114–S120. 8. Pearl J. Invited commentary: understanding bias amplification. Am J Epidemiol 2011; 174(11): 1223–1227.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article: Table S1. ICD-10 codes for underlying cause of death that define sudden cardiac death in non-hospitalized patients. Adapted from Chung et al. (2010) Table S2a. 15 most common diagnoses in the period 31–-183 days prior to cohort entry: persons without prior ACEI use Table S2b. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons with prior ACEI use Table S2c. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons without prior ACEI use, subset with prior hypoglycemic agent use Table S2d. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons with prior ACEI use, subset with prior use of hypoglycemic agents
Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
ORIGINAL REPORT
Cardiac mortality in users of olmesartan, other angiotensin-receptor blockers and angiotensin-converting enzyme inhibitors Alexander M. Walker1*, Caihua Liang2, C. Robin Clifford2, Crawford Parker3 and Allen Feldman3 1
World Health Information Science Consultants, Newton, MA, USA OptumInsight, Waltham, MA, USA 3 Daiichi Sankyo Pharma Development, Edison, NJ, USA 2
ABSTRACT Purpose Clinical trials of olmesartan for prevention of progression of renal disease in patients with diabetes showed renal protection but an unexpected imbalance in cardiac deaths. The US Food and Drug Administration requested from the manufacturer a cohort study of olmesartan, other angiotensin-receptor blockers (ARBs), and angiotensin-converting enzyme (ACE) inhibitors in a large population. Methods A retrospective cohort study was conducted with the cooperation of a US health insurer. Subject entry and follow-up ran from 2002 through 2009. In propensity-matched cohorts, the primary analysis considered continuous current users. Endpoints were sudden cardiac death (SCD) and all-cause mortality, identified through the US National Death Index, supplemented by insurance and hospital discharge data. Statistical estimation was based on proportional hazards analyses with 95% confidence intervals. Power calculations had shown that 25 000 olmesartan initiators would be required to detect relative risks (RRs) of SCD of twofold or greater. Results A total of 57 123 initiators of olmesartan were matched 1:2 to initiators of other ARBs and 41 801 to initiators of ACE inhibitors. Average follow-up time ranged from 8 to 9 months. Olmesartan initiators and comparators experienced similar patterns of both outcomes, with RRs ≤1.0 and upper confidence bounds ≤1.6. Among persons with prior use of hypoglycemic agents, in comparison with other ARBs, the RR of SCD for olmesartan users was 0.8, with an upper confidence bound of 2.2. Conclusion The results of this well-powered study do not raise concerns for the risk of SCD or death from all causes among olmesartan users in comparison with users of other ARBs or ACE inhibitors. Copyright © 2013 John Wiley & Sons, Ltd. key words—olmesartan; ARBs; ACE inhibitors; sudden cardiac death; mortality; safety; pharmacoepidemiology Received 26 July 2013; Revised 13 November 2013; Accepted 19 November 2013
BACKGROUND Two randomized trials that were intended to address olmesartan’s ability to prevent or delay the progression of renal disease in patients with diabetes showed unexpected elevations of cardiovascular mortality in comparison with placebo.1,2 In the ROADMAP study, the time to first occurrence of microalbuminuria was 23% longer in 2232 patients randomized to olmesartan 40 mg once daily than in 2215 patients randomized to placebo.1 There was a rough equality of combined cardiovascular endpoints between the olmesartan (46 events, 2.1%) and
*Correspondence to: A. M. Walker, World Health Information Science Consultants, 275 Grove Street, Suite 2-400, Newton, MA 02466, USA. Email: Alec. [email protected]
Copyright © 2013 John Wiley & Sons, Ltd.
placebo (37 events, 1.7%) arms, for a relative risk (RR) of 1.2 (95% confidence interval [CI]: 0.8–1.9). There was however a notable imbalance in cardiovascular mortality (15 deaths versus 3, RR: 5.0, 95%CI: 1.4–17.1). The second trial, called ORIENT, enrolled type 2 diabetic patients with nephropathy.2 A combined cardiovascular endpoint of cardiovascular death, stroke or nonfatal myocardial infarction occurred in 18 of 282 (6.4%) patients receiving olmesartan in a variable dose that was titratable from 10 to 40 mg daily and in 21 of 284 (7.4%) patients on placebo (RR: 0.9, 95%CI: 0.5–1.6). Cardiovascular mortality considered in isolation was higher in the olmesartan group than in the placebo group (10 deaths versus 3, RR: 3.4, 95% CI: 0.9–12.3). Aware of findings from ROADMAP and ORIENT, the US Food and Drug Administration (FDA) requested that the manufacturer of olmesartan, Daiichi
cardiac mortality in users of antihypertensive agents
Sankyo, Incorporated, “conduct an epidemiologic study using claims or electronic health records data to evaluate the comparative incidence of sudden cardiac death, in-hospital fatal myocardial infarction and total mortality in olmesartan users vs. users of other angiotensin-receptor blockers and in olmesartan users vs. ACE inhibitor users.”† The goal of the present study was to meet this requirement and to provide scientific evidence on the incidence of cardiac mortality in users of olmesartan, other angiotensin-receptor blockers (ARBs) and angiotensin-converting enzyme (ACE) inhibitors. METHODS Sponsorship and approvals The study was conducted by World Health Information Sciences Consultants (WHISCON) in cooperation with Optum, a member of the United Health Group companies, under a research contract with Daiichi Sankyo, Incorporated. The protocol was provided in draft to the FDA, modified in response to FDA comments, and resubmitted to the FDA before initiation, after no further suggestions were received. The original protocol as implemented had as its primary analyses a comparison of olmesartan initiators to initiators of other ARBs and of ACE inhibitors among persons who had not previously used either other ARBs or ACE inhibitors. Secondary analyses included a subgroup of persons with prior use of hypoglycemic agents, a comparison with ARBs initiators among persons with prior ACE inhibitor use, comparisons of olmesartan with specific ARBs (they are not reported here because they are not different from the overall comparison) and sensitivity analyses, these being intent-to-treat analyses in the primary cohort and a more stringent definition of continuous use in the as-treated analyses. (The sensitivity analyses are also not reported here, as they gave essentially the same results as those reported in the succeeding sections.) After seeing the study results, which included a nonsignificant elevation in risk of about twofold among olmesartan users in the primary cohort among persons with prior use of hypoglycemic agents, the FDA requested additional subgroup analyses in persons with prior hypoglycemic use and prior use of ACE inhibitors, persons whom the primary cohort had omitted.
349
Source population A large US commercial health insurance company provided insurance claims and enrollment data for this work. The insurance included drug benefits along with coverage for services. Although people of all ages were represented, the proportion of subjects aged 65 years and older (ages at which most Americans have health insurance through Medicare) was smaller than in the USA overall. For 2009, data relating to approximately 13.3 million individuals with both medical and pharmacy benefit coverage were available. Persons in the study were those who at age 20 years or older received a first dispensing for olmesartan, another ARB or an ACE inhibitor, the “study drugs,” between 2002 and 2009. For inclusion, an individual’s first dispensing of a study drug had to come after a period of at least 183 days during which the subject was eligible for insurance benefits. The individual had to have had no prior dispensing of any ARB. Individuals remained under observation until they terminated treatment with the initiating product, left insurance coverage, entered long-term care, died or reached the close of observation at the end of 2009. Termination of treatment was defined as the first occurrence of a gap of 30 days beyond the treatment that would be ascribable to a dispensing (calculated as dispensing date plus days of medication supplied), with no new dispensing. Additionally, patients who initiated an ARB while a member of the olmesartan or ACE inhibitor cohorts were censored at the first dispensing of an ARB, and patients who initiated olmesartan while a member of the other ARB or ACE inhibitor cohorts were censored at the first dispensing of olmesartan. Persons who initiated treatment with two different study drugs on the same day were not included. The end of insurance coverage was taken to have occurred 32 days following an administrative record of discontinuation of eligibility without an indication of reinstatement. If an individual terminated enrollment with the insurer and then enrolled again at a later date, each enrollment period was considered separately, so that a person could potentially leave and reenter the cohort through discontinuation and reenrollment. (Because the outcome under study was death, no person could suffer an event in two different enrolment periods.) All the cohort entry criteria had to be met in each enrollment period. Covariates
† Personal communication to Allen Feldman, Daiichi Sankyo, Incorporated, 3 June 2011.
Copyright © 2013 John Wiley & Sons, Ltd.
Definitions of baseline characteristics were keyed to the date of first treatment with a study drug. In addition Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
350
a. m. walker et al.
to age and sex, we noted health-care utilization and drugs dispensed in the 183 days prior to treatment initiation. Drugs were considered both in narrow therapeutic classes and as summary counts of the numbers of distinct drugs dispensed. Diagnoses associated with insurance claims for physician and hospital care were identified in the 30 days before treatment initiation and during the preceding 153 days.
Comparison groups Comparison groups were built without knowledge of outcome status, as the cohorts needed to be constructed in order to know which records to submit to the US National Death Index (NDI) for linkage. Within each of the eight calendar years of cohort accrual from 2002 through 2009 and within strata defined by the prior use of ACE inhibitors or not, we formed cohorts of initiators of the study drugs, with covariate balance achieved through matching initiators of other ARBs or ACE inhibitors to initiators of olmesartan, using a propensity score particular to that calendar year and prior-use stratum. In strata without prior use of ACE inhibitors, initiators of olmesartan could be compared both with initiators of other ARBs and of ACE inhibitors. In strata with prior use of ACE inhibitors, only the contrast of olmesartan initiators versus initiators of other ARBs was available. The propensity score models were initially derived by backward elimination in a logistic regression that modeled the choice of olmesartan versus either other ARBs or ACE inhibitors as a function of all the covariates. All retained predictors in the model had a p-value of 0.1 or less. After propensity score models had been estimated for each comparison (olmesartan versus other ARBs and olmesartan versus ACE inhibitors) in each stratum, we used the union of all retained predictor variables from all comparisons to fit a final common logistic model that had the same covariates for every stratum. Olmesartan patients were matched individually to two comparators from other ARB and ACE inhibitor initiators in the stratum with no prior ACE inhibitor use in the same calendar year, except in 2009 where each olmesartan patient was matched to one ARB patient, as there were not enough other ARB initiators for a 2:1 match. In the strata with prior ACE inhibitor use, olmesartan initiators were matched to two ARB initiators. Matching used a nearestneighbor algorithm with a caliper of 0.1 on the logit propensity scale, a value that represented approximately 0.2 times the average standard deviation of the year-specific logit propensity scores.3,4 Copyright © 2013 John Wiley & Sons, Ltd.
Outcomes The outcomes were sudden cardiac death (SCD) and death from any cause. Another planned outcome of in-hospital deaths from myocardial infarction yielded fewer than two additional deaths in any cohort, did not change any qualitative finding and is not reported here. In order to ascertain deaths, a business affiliate of the insurer (Optum) submitted personal identifying information of all 178 003 cohort members who terminated participation in the insurance plan from 2002 to 2010 to the NDI for linkage to the NDI records. If different enrollment periods for an individual were associated with different data (e.g., different family names or dates of birth), records corresponding to each data variant were submitted to the NDI. Potential matches returned by the NDI were accepted according to the NDI’s suggested guidelines. SCD was classified according to an algorithm proposed by Chung et al.5 This involved an NDI match to one of the submitted records with an underlying cause of death from Table S1. The match had to occur on or before termination of enrollment, with the additional requirement that there be no evidence of a terminal hospital stay. For 70 subjects for whom there were insurance claims recorded more than 30 days after an NDIdetermined death, or for whom the NDI date of death differed by more than seven days from an apparent death recorded in the insurance claims, two adjudicators examined the chronological insurance claims profile and the NDI record without knowledge of the individual’s exposure status. When the later claims reflected billings that could recur automatically (such as durable equipment rentals) or medical–legal services, the deaths were retained (32 occurrences). Individuals whose claims showed an uninterrupted sequence of billings spanning the putative date of death from the NDI were classed as not having died during study follow-up (38 occurrences). Thirty-three of the latter cases were persons for whom there was no Social Security number in the submitted identifying data from the insurer, so that the rejected, apparently false-positive NDI matches had been based only on similarity of names and dates of birth. There were in addition 10 persons for whom Social Security number was unavailable in the insurance files and who had no match in the NDI, but for whom there was an International Classification of Diseases, 9th Revision (ICD-9) code indicating death in the insurance claims, with no insurance claims more than seven Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
cardiac mortality in users of antihypertensive agents
days after. These events were retained in the all-cause mortality analyses but did not qualify as SCD cases. Analysis For each of the study endpoints, we undertook a common proportional hazards analysis over strata defined by calendar year and prior use of ACE inhibitors. In persons without prior use of ACE inhibitors, initiators of olmesartan were compared with initiators of other ARBs and with initiators of ACE inhibitors in separate analyses. In cohort members with prior use of ACE inhibitors, it was only possible to compare initiators of olmesartan and other ARBs. There was a prespecified subgroup analysis in persons with prior use of hypoglycemic agents and prior use of ACE inhibitors. The original protocol specified that the primary analysis would be in persons who had not used ACE inhibitors previously and that the combined results (including those with and without prior ACE inhibitor use) would be a secondary comparison. After it had reviewed (i) the overall combined findings and (ii) the subgroup analysis in persons who had not previously used ACE inhibitors and who had used hypoglycemic agents, the FDA requested that the analysis be expanded so that the subgroup of persons with prior hypoglycemic use would additionally include prior users of ACE inhibitors, permitting a combined analysis. To check for differential censoring by treatment group in the compared cohorts, we conducted proportional hazards analyses with censoring for any reason taken as the dependent variable and study drug group (olmesartan, other ARBs, and ACE inhibitors) as predictors. The hazard ratios for termination of follow-up, comparing olmesartan with comparators, were in both cases equal to 0.99. For the analyses in prior users of hypoglycemic agents, we checked the overall covariate distributions between olmesartan and each of the comparators. We added to the outcome model new terms that summarized covariates that were out of balance in this subgroup. In a combined as-treated analysis stratified by prior ACE inhibitor use, a subject could be an initiator of an ACE inhibitor in the primary cohort in one year, and then in a subsequent year start olmesartan or another ARB, qualifying as an olmesartan (or a matching other-ARB) initiator in the later year. We accounted for the second cohort membership by generating a new follow-up record for the second exposure and invoking a robust variance estimator. Copyright © 2013 John Wiley & Sons, Ltd.
351
Power calculations indicated that sizes for the olmesartan cohort above 25 000, with two comparators for each olmesartan initiator, would have a greater than 90% power to detect RRs of 2 or greater for SCD. RESULTS Characteristics of the source population and cohorts From 2002 through 2009, there were 42 410 olmesartan initiators who had not previously used ACE inhibitors and 21 930 who had. The corresponding counts for initiators of other ARBs (cohort candidates) with no prior ACE inhibitor use and initiators of ARBs with prior ACE inhibitor use were 130 225 and 68 439, respectively. Over the same time, there were 392 543 initiators of ACE inhibitors. The propensity models retained 205 predictor variables. Matching on propensity score yielded cohorts with close balance on demographics, diagnoses, and drug classes dispensed, as well as on a variety of health-care utilization measures. For no baseline characteristic was the prevalence difference between cohorts (summed over all the years together) more than 0.5%. The final matched cohort sizes for the comparison between olmesartan and other ARBs were 38 750 and 72 326, respectively, for patients with no prior use of ACE inhibitors. (Recall that in 2009, only a 1:1 match was possible, so the overall ratio fell short of 2:1.) These cohorts had average observation times of 8.3 and 8.7 months, respectively. The corresponding matched cohort sizes for the olmesartan (with no prior use of ACE inhibitors) versus ACE inhibitor comparisons were 41 801 and 83 602 (8.4 months average observation time in each). Among those with prior use of ACE inhibitors, there were 18 373 initiators of olmesartan (9.3 months average observation time) and 36 746 matched initiators of other ARBs (9.5 months), making for a total of 57 123 olmesartan initiators matched to 109 072 initiators of ARBs including groups with and without prior use of ACE inhibitors. All cohorts in this commercially insured population were predominantly middle-aged, approximately 52% being in the range 45–64 years. All cohorts were either 53 or 54% men. Table 1a through 1d shows demographic characteristics and summary measures of health-care utilization for cohort members without and with prior use of ACE inhibitors, and for the subsets of each of those groups who received hypoglycemic agents before cohort entry. The users of hypoglycemic agents were about 7% of the olmesartan users and Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
352
a. m. walker et al.
Table 1a. Age, sex, and health-care utilization in the period 183 days prior to cohort entry: persons without prior ACEI use All patients Drug N
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Olm
ARB
ACEI
Olm
ARB
Olm
ACEI
42 410
130 225
392 543
38 750
72 326
41 801
83 602
%
%
%
%
%
%
%
8.2 22.0 35.5 27.0 5.6 1.6
6.8 19.9 33.6 29.6 7.3 2.7
8.1 20.6 33.8 29.2 6.4 1.9
7.6 21.5 35.1 28.0 6.0 1.8
7.6 21.6 35.1 28.0 6.0 1.7
8.2 22.0 35.4 27.1 5.7 1.6
8.0 21.7 35.5 27.4 5.7 1.7
53.9 46.1 Mean 7.7 3.7 3.9 0.06 2956 510 1734 712
51.5 48.5 Mean 7.9 3.8 3.9 0.08 3560 551 1838 1172
58.2 41.8 Mean 7.5 3.7 3.3 0.14 4955 531 1809 2614
53.1 46.9 Mean 7.6 3.6 3.8 0.06 2982 502 1734 747
53.3 46.7 Mean 7.6 3.6 3.8 0.06 3126 513 1778 836
54.1 45.9 Mean 7.6 3.6 3.8 0.06 2940 501 1718 721
54.2 45.8 Mean 7.7 3.6 3.8 0.06 3172 526 1838 808
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
Table 1b. Age, sex and health-care utilization 183 days prior to cohort entry: persons with prior ACEI use All Drug N
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Olm
ARB
Olm
ARB
21 930
68 349
18 373
36 746
%
%
%
%
3.8 16.2 34.6 36.3 7.6 1.5
3.6 15.0 33.1 37.7 8.6 2.0
3.9 15.7 33.9 37.0 7.9 1.7
3.8 15.9 33.4 36.9 8.2 1.8
55.2 44.8 Mean 9.1 6.3 4.7 0.08 4321 936 2113 1272
53.2 46.8 Mean 9.8 6.8 5.0 0.14 5753 1035 2451 2268
54.1 45.9 Mean 9.1 6.3 4.6 0.09 4474 941 2140 1393
54.0 46.1 Mean 9.1 6.3 4.6 0.09 4593 953 2181 1459
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
(before matching) over 10% of the users of other ARBs and ACE inhibitors. All the tables give the distributions of characteristics in the general population Copyright © 2013 John Wiley & Sons, Ltd.
(before matching) and in the matched populations that form the basis of the analysis. Persons who filled prescriptions for hypoglycemic agents before they Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
353
cardiac mortality in users of antihypertensive agents
Table 1c. Age, sex, and health-care utilization in the period 183 days prior to cohort entry: persons without prior ACEI use, subset with prior use of hypoglycemic agents All patients
Matched
Drug
Olm
ARB
ACEI
Olm
ARB
Olm
ACEI
N
2991
13 765
56 421
2892
5536
2982
6403
%
%
%
%
%
%
%
5.6 16.5 33.5 33.4 8.9 2.2
4.8 15.2 31.3 36.4 9.5 2.8
7.2 18.4 32.7 32.4 7.7 1.8
5.3 16.0 33.3 34.0 9.1 2.3
5.9 17.4 33.0 34.4 7.4 2.0
5.6 16.5 33.6 33.3 8.9 2.3
6.5 19.5 34.6 30.3 7.4 1.8
51.4 48.7 Mean 9.5 6.6 4.7 0.09 4489 1045 2178 1267
52.2 47.8 Mean 9.7 6.8 4.8 0.13 5436 1138 2334 1963
57.8 42.2 Mean 8.9 6.3 4.2 0.15 5757 1037 2116 2604
51.3 48.7 Mean 9.5 6.5 4.7 0.10 4527 1049 2193 1285
52.5 47.5 Mean 9.0 6.4 4.4 0.09 4544 1026 2208 1311
51.4 48.6 Mean 9.5 6.6 4.7 0.09 4484 1037 2177 1271
53.3 46.7 Mean 8.9 6.1 4.4 0.08 4168 988 2175 1005
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensin-converting enzyme inhibitor.
Table 1d. Age, sex and health-care utilization 183 days prior to cohort entry: persons with prior ACEI use, subset with prior use of hypoglycemic agents All
Age (years) 20–34 35–44 45–54 55–64 65–74 75+ Sex Male Female Utilization Diagnosis codes Distinct drugs Physician visits Inpatient stays Total cost ($) Pharmacy cost ($) Outpatient cost ($) Inpatient cost ($)
Matched
Drugs
Olm
ARB
Olm
ARB
N
4398
17 566
4037
8211
%
%
%
%
2.2 12.9 32.6 41.0 9.7 1.6
2.6 11.9 30.7 42.5 10.4 2.0
2.3 12.7 32.2 41.2 10.0 1.7
2.7 12.8 31.3 41.1 10.2 1.8
55.7 44.3 Mean 10.6 9.1 5.4 0.12 6038 1622 2473 1942
55.1 45.0 Mean 11.1 9.3 5.6 0.19 7498 1632 2888 2978
55.5 44.5 Mean 10.6 9.1 5.3 0.13 6129 1620 2477 2032
56.2 43.8 Mean 10.2 8.7 5.1 0.12 5828 1564 2471 1794
Olm, olmesartan; ARB, angiotensin-receptor blocker; ACEI, angiotensinconverting enzyme inhibitor.
entered the cohort were older, carried more diagnoses and were heavier consumers of health care. Table S2a through S2d shows the 15 diagnoses most frequently Copyright © 2013 John Wiley & Sons, Ltd.
seen in each of the groups during the period from 183 through 31 days prior to cohort entry. For all groups, there was substantial balance in demographics, health-care utilization, and diagnoses after matching. As detailed below, a number of characteristics were nonetheless sufficiently out of balance in the subsets with prior hypoglycemic agent use to warrant further covariate control.
Mortality in comparison with initiators of other angiotensin-receptor blockers and angiotensin-converting enzyme inhibitors The analyses for SCD in olmesartan versus other ARBs and olmesartan versus ACE inhibitors are given in Table 2. Note that the comparison under the heading “All initiators” in the table includes persons with prior ACE inhibitor use and is necessarily restricted to ARBs. For both comparisons, the mortality from SCD was lower in the olmesartan group than in the comparator, and for both, the CIs included an RR of 1.0. All-cause mortality is shown in Table 3. The rates were lower in the olmesartan initiators than in the comparators, with upper confidence bounds for the RR at or just below 1.0. Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
354
a. m. walker et al.
Table 2. Mortality from sudden cardiac death Mortality Deaths
Person-years
Rate
95%CI
Ratio
95%CI
Olmesartan Other ARBs
14 32
Among persons without prior use of ACEIs 26 774.0 0.5 52 189.2 0.6
0.3–0.9 0.4–0.9
0.9
0.5–1.6
Olmesartan ACEIs
13 42
29 116.4 58 240.1
0.2–0.8 0.5–1.0
0.6
0.3–1.1
Olmesartan Other ARBs
24 60
0.8
0.5–1.3
0.4 0.7
All initiators (with or without prior use of ACEIs) 41 080.5 0.6 0.4–0.9 81 414.9 0.7 0.6–0.9
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and (for “All initiators”) by prior use of ACEIs. ARBs, angiotensin-receptor blockers; ACEIs, angiotensin-converting enzyme inhibitors; CI, confidence interval.
Table 3. All-cause mortality Mortality Deaths
Person-years
Rate
95%CI
Ratio
95%CI
Olmesartan Other ARBs
49 133
Among persons without prior use of ACEIs 26 774.0 1.8 52 189.2 2.5
1.4–2.4 2.1–3.0
0.7
0.5–1.0
Olmesartan ACEIs
51 150
29 116.4 58 240.1
1.3–2.3 2.2–3.0
0.7
0.5–0.9
Olmesartan Other ARBs
101 253
0.8
0.6–1.0
1.8 2.6
All initiators (with or without prior use of ACEIs) 41 080.5 2.5 2.0–3.0 81 414.9 3.1 2.7–3.5
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and by prior ACEI use. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
Users of hypoglycemic agents
DISCUSSION
The rate of SCD was slightly lower among olmesartan initiators who had used hypoglycemic agents previously than it was for similar ARB initiators overall. The overall RR, compared with other ARB users both with and without prior use of ACE inhibitors, was 0.8 with an upper confidence bound of 2.2 (Table 4). In the persons with prior use of hypoglycemic agents who moreover had not previously used ACE inhibitors, the SCD rate in olmesartan users exceeded that in comparators by twofold. This estimate is based on three deaths only in each group and has confidence bounds that run from 0.4 to greater than 10-fold. Although the point estimates in strata with and without prior use of ACE inhibitors were very different, the very small number of events meant that the differences were compatible with chance ( p = 0.18). Table 5 presents the all-cause mortality for the same comparisons. Mortality rates were essentially the same in each instance.
The size of the source data and the opportunity for extensive confounder control mean that these results provide reassurance about the relative safety of olmesartan as it has been used in the USA since market introduction in 2002. Because it reflects actual use, the population studied here differs from the diabetic olmesartan users in the ROADMAP and ORIENT studies. The overall finding of lower cardiac mortality in olmesartan users versus comparators effectively rules out an important mortality-increasing effect of olmesartan in general. Prior users of hypoglycemic agents proved to be a small subgroup, with large variations in point estimates of RR, but no overall indication of excess risk. Nonetheless, for statistical reasons, the study provides only modest assurance about the relative safety of olmesartan in diabetic populations. The signals that gave rise to this study arose in two trials of persons with type II diabetes, and so the
Copyright © 2013 John Wiley & Sons, Ltd.
Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
355
cardiac mortality in users of antihypertensive agents Table 4. Mortality from sudden cardiac death in prior users of hypoglycemic agents Mortality Deaths
Person-years
Rate
95%CI
Olmesartan Other ARBs
3 3
Among persons without prior use of ACEIs 1 881.6 1.6 0.3–4.7 3770.0 0.8 0.2–2.3
Olmesartan ACEIs
3 3
1951.3 4207.3
Olmesartan Other ARBs
6 15
1.5 0.7
Adjusted rate ratio
95%CI
2.1
0.4–11.7
2.0
0.4–10.7
0.8
0.3–2.2
0.3–4.5 0.1–2.1
All initiators (with or without prior use of ACEIs) 4811.8 1.2 0.5–2.7 10 101.7 1.5 0.8–2.4
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and with adjustment for covariates imbalanced because of restriction to a subset of the balanced cohort. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
Table 5. All-cause mortality in prior users of hypoglycemic agents Mortality Deaths
Person-years
Rate
95%CI
Olmesartan Other ARBs
8 16
Among persons without prior use of ACEIs 1881.6 4.3 1.8–8.4 3770.0 4.2 2.4–6.9
Olmesartan ACEIs
8 16
1951.3 4207.3
Olmesartan Other ARBs
25 53
4.1 3.8
1.8–8.1 2.2–6.2
All initiators (with or without prior use of ACEIs) 4811.8 5.2 3.4–7.7 10 101.7 5.2 3.9–6.9
Adjusted rate ratio
95%CI
1.0
0.4–2.4
0.9
0.4–2.1
1.0
0.6–1.6
Rate is per 1000 person-years; the ratios compare each olmesartan cohort with its respective comparator, as derived from a proportional hazards analysis stratified by calendar year and with adjustment for covariates imbalanced because of restriction to a subset of the balanced cohort. ARBs, angiotensin-receptor blockers; ACEI, angiotensin-converting enzyme inhibitor; CI, confidence interval.
subgroup with prior hypoglycemic use deserves special scrutiny. ROADMAP excluded persons who had used ACE inhibitors (or other ARBs) in the 6 months before study entry.1 In ORIENT, by contrast, approximately 73% of the patients had used ACE inhibitors prior to study entry, and participants were required as part of the study protocol to continue treatment.2 In the present study, the primary cohort without prior ACE inhibitor use corresponds therefore to the ROADMAP population. The combined results respond more nearly to the two external signals that prompted the study. Both ROADMAP and ORIENT were trials of olmesartan as a preventive therapy in a setting in which placebo was an appropriate comparator. Any risk excess associated with treatment could be seen as an absolute risk of therapy. In testing whether the concerns raised by ROADMAP and ORIENT might affect people treated for hypertension, we and the regulatory agency that requested this study moved into a related question of comparative safety. Nontreatment Copyright © 2013 John Wiley & Sons, Ltd.
of people with hypertension is not a therapeutic option, and so the question became: does olmesartan differ from other treatments, in particular other ARBs or ACE inhibitors, with respect to the risk of cardiac death? This study cannot answer the question of absolute risk in persons with hypertension that might be approached in a hypothetical (as well as unethical and clinically irrelevant) comparison with no treatment at all. In addition to shedding light on therapeutic alternatives, an observational study with active comparators rather than nontreatment as a reference group runs a smaller risk of bias. By comparison with expected medical differences between treated and untreated people, there is an anticipated rough similarity in characteristics between people receiving alternative therapies. A starting covariate balance between compared therapies substantially reduces the scope for confounding by indication, the broad category of biases introduced by the reasons for treatment, which often involve patient characteristics that are difficult to measure or to quantify.6 Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
356
a. m. walker et al.
Fully known covariates and realistic parameterization are assumed in statistical adjustment and matching. The effect of deviations from this ideal state is small when the poorly measured (or unknown) covariates have similar distributions in the compared groups. More than 90% of the olmesartan initiators in the database could be matched to initiators of other ARBs and of ACE inhibitors with very similar patterns of diagnoses associated with insurance claims, drugs dispensed, health-care utilization and demographic characteristics. The average follow-up times of 8 to 9 months from treatment initiation were adequate to evaluate effects of the study drugs over a corresponding term, but were too short to permit a duration–response analysis and they do not rule out effects that would occur with longer follow-up. (Both ROADMAP and ORIENT had average durations of follow-up of about 3 years.) Nor was the study size adequate to permit a dose–response analysis. Because the cohorts had to be formed in order to know which records to submit for NDI matching, it was not possible to screen for potential instruments among the 205 retained predictors. Given the small numbers of events, there is no post hoc technique that could accurately designate instruments. The possibility remains therefore that the analysis may have amplified the impact of unmeasured confounders through what has been called “Z-bias” or “bias amplification.”7,8 The restriction of the initial analysis to persons without prior use of ACE inhibitors had the effect of focusing on a population of treatment initiators who may have been substantially healthier than those with prior ACE inhibitor use. Accordingly, the analyses most representative of the experience of olmesartan initiators generally may be the combined analyses that include both persons who had and who had not used ACE inhibitors before cohort entry. CONFLICT OF INTEREST This work was funded by research contracts between Daiichi Sankyo, Incorporated, and World Health Information Science Consultants, LLC (WHISCON), and between WHISCON and Optum. Crawford Parker and Allen Feldman are employees of Daiichi Sankyo, Incorporated. KEY POINTS • Persons prescribed olmesartan, other ARBs and ACE inhibitors for the first time had similar rates of sudden cardiac death and all-cause mortality during an average of eight months of follow-up. • The presence or absence of diabetes mellitus and of prior treatment with ACE inhibitors did not affect the relative risks. Copyright © 2013 John Wiley & Sons, Ltd.
ETHICS STATEMENT The original protocol was completed on 15 September 2011, and the revision was finalized on 6 March 2012. The New England Institutional Review Board considered and approved the protocol and its revision on 7 November 2011 and 9 April 2012. The FDA requested supplementary analyses in September 2012, and these are reported here, along with additional results requested by reviewers. REFERENCES 1. Haller H, Ito S, Izzo JL, et al. Olmesartan for the delay or prevention of microalbuminuria in type 2 diabetes. N Engl J Med 2011; 364: 907–917. 2. Imai E, Chan JCN, Ito S, et al. Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study. Diabetologia 2011; 54: 2978–2986. 3. Rassen JA, Doherty M, Huang W, Schneeweiss S. Pharmacoepidemiology Toolbox. Boston, MA. http://www.hdpharmacoepi.org. 4. Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 2011; 10: 150–151. 5. Chung CP, Murray KT, Stein M, Hall K, Ray WA. A computer case definition for sudden cardiac death. Pharmacoepidemiol Drug Saf 2010; 19: 563–572. 6. Walker AM. Confounding by indication. (Editorial) Epidemiology 1996; 7: 335–336. 7. Brookhart MA, Stürmer T, Glynn RJ, Rassen J, Schneeweiss S. Confounding control in healthcare database research: challenges and potential approaches. Med Care 2010; 48(6 Suppl): S114–S120. 8. Pearl J. Invited commentary: understanding bias amplification. Am J Epidemiol 2011; 174(11): 1223–1227.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article: Table S1. ICD-10 codes for underlying cause of death that define sudden cardiac death in non-hospitalized patients. Adapted from Chung et al. (2010) Table S2a. 15 most common diagnoses in the period 31–-183 days prior to cohort entry: persons without prior ACEI use Table S2b. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons with prior ACEI use Table S2c. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons without prior ACEI use, subset with prior hypoglycemic agent use Table S2d. 15 most common diagnoses in the period 31-–183 days prior to cohort entry: persons with prior ACEI use, subset with prior use of hypoglycemic agents
Pharmacoepidemiology and Drug Safety, 2014; 23: 348–356 DOI: 10.1002/pds
