arznei-telegramm 2003; 34: 73-4 | ||||||||||
STOP-NIDDM STUDY WITH ACARBOSE ... slipshod work, manipulation, deception One year after first publication of the STOPP-NIDDM study* with acarbose (GLUCOBAY) in the Lancet (1) (a-t 2002; 33: 72-3), an evaluation concerning cardiovascular diseases (a-t 2003; 34: 38), used in advance by the study sponsor Bayer for promotional purposes, has now appeared in the Journal of the American Medical Association (JAMA)(2). By lowering postprandial glycaemia, acarbose is said to reduce the risk of hypertension and cardiovascular disease (2). 714 patients with impaired glucose tolerance took acarbose over a period of 3.3 years on average, 715 received a placebo. The primary endpoint was a surrogate criterion: diabetes rate diagnosed by oral glucose tolerance test. 61 patients were excluded from the analysis, allegedly because their data was incomplete or because their glucose tolerance was not impaired. Subsequently 682 were evaluated in the acarbose group and 686 in the placebo group ("modified population"). 211 (31%) acarbose users discontinued treatment, mostly because of adverse events. In the placebo group 130 (19%) discontinued treatment (1,2). "Blind" study conditions, which had been aimed for, were not met probably because of the side effects. 79% of patients in the verum group and 69% of doctors allocated the acarbose treatment correctly (2). The primary endpoint decreased under acarbose from 42% to 32%. However, after discontinuation the effect did not last: In a three-month placebo run-out phase the number of diagnosed diabetics in the original drug group exceeded their number in the placebo group (15.4% vs 10.6%) (1). Six patients died under acarbose treatment, three under placebo (1), of these three (acarbose) and one (placebo) from non-cardiovascular causes. According to the prepublished study protocol, cardiovascular events were to be recorded prospectively. However, defined were only myocardial infarction, cerebrovascular accident and congestive heart failure (3), and even those only inadequately. During or after conclusion of the study "angina", "revascularization procedures" and "peripheral vascular disease" were added (2). No publication clearly indicates which cardiovascular events were actually predefined secondary endpoints of the study. If at all they constitute one of many secondary objectives (3). In the "modified group" cardiovascular events occurred in 15 patients under acarbose and in 32 taking placebo (2). The difference of p = 0.03 (2) did not achieve the statistical significance of <0.01 required for multiple testing. At the presentation of the STOP-NIDDM data at the European Diabetes Congress in Budapest in 2002 the p-value for this difference was still 0.05 (4) and therefore did not even reach the 5% significance level. Data from the safety analysis of the study can be viewed on a Bayer internet site, including all 1,429 patients (5). Accordingly 33 patients had a cardiovascular event during and up to seven days after taking acarbose and 39 patients after taking placebo. These figures also give no complete picture of an intention-to-treat analysis. In addition it is not possible to properly reconstruct, which particular events are involved, especially in the case of the acarbose group. As opposed to the data from the "modified population", on that internet page a myocardial infarction is recorded in two instead of one and angina in six instead of five patients of the acarbose group. The evaluation presented in JAMA is therefore based on a retrospective patient selection and thus invalid. Moreover, the STOP-NIDDM publications contain numerous other "modifications", omissions, inconsistencies and inaccuracies. In the "modified study group" new cases of hypertension are reported to occur significantly less often under acarbose with 11% versus 17%. However, it is not clear how many patients already had hypertension at the start of the study: according to the Lancet publication the figure was 46% (1), according to JAMA 51% (2). The result can therefore not be interpreted The prepublished study protocol and the final publications contradict each other in the number of randomised patients and the randomisation period (1-3). Whereas the Lancet article (1) indicated that eight participants in the drug group and nine in the placebo group did not fulfil the inclusion criteria, precisely the reverse was the case in JAMA. Lacking diligence? Obviously not, because at the same time the number of subjects for whom no assessable data was available changed accordingly - therefore "creative accounting" becomes more likely? There is also contradictory information relating to the planned duration of the follow-up period (1,2,5,6). The average weight reduction in the drug group - in itself not surprising, given the high rate of acarbose-related gastrointestinal symptoms - was first quoted as 0.5 kg (1), later as 1.15 kg (2). The dietary advice for weight reduction chosen for STOP-NIDDM was by contrast obviously ineffective: Contrary to previous study experiences where body weight was reduced by several kilograms after this intervention, it increased slightly in the placebo group despite the advice (2). Antihypertensive treatment during STOP-NIDDM, using calcium antagonists as the treatment of choice, combined if necessary with alpha-blockers and ACE inhibitors and avoiding diuretics and beta-blockers, was also not optimal and did not correspond to standard practice. Calcium antagonists, alpha-blockers and ACE inhibitors are inferior to the antihypertensive treatment with diuretics and beta-blockers with respect to cardiovascular sequelae (7-9). As in the Lancet publication, the role of the sponsor in the planning, implementation and evaluation of the study was also concealed in JAMA. In both publications the authors emphasise that Bayer, as the sponsor, had not played a (direct) (2) role in preparing the study protocol, data analysis and data interpretation (1,2). This is incorrect. The study protocol was drafted by the STOP-NIDDM Trial Research Group, which included two Bayer staff members (3). These staff members appear on the original list of group members (3), however in both final publications their names no longer appear (1,2). Bayer representatives were five of the eleven members of the Steering Committee, which among other things receives all the important study data (3). In the Lancet the STOP-NIDDM authors do not report any conflict of interest (1). In JAMA three authors state that they have received fees and research funds from Bayer (2). Despite regular presentations at Bayer conferences HANEFELD does not reveal his financial associations with Bayer here either (2) (cf. a-t 2002; 33: 72-3).
|
(R = randomised study, M = meta-analysis) | ||
R | 1 | CHIASSON, J.L. et al. : Lancet 2002; 359: 2072-7 |
R | 2 | CHIASSON, J.L. et al. : Jama 2003; 290: 486-94 |
3 | CHIASSON, J.L. et al. : Diabetes Care 1998; 21: 1720-5 | |
4 | CHIASSON, J.L. et al. : Diabetologia 2002; 45 (Suppl.2): A104 | |
5 | http://www.stop-niddm.com; to be found under -> sitemap -> slides -> complete slideshow -> slide 48 -> "Serious Treatment-Emergent Adverse Events", Access: 6. August 2003 | |
6 | "STOP-NIDDM", supplement Dtsch. Med. Wschr. 1997; 122 (Issue 38): 1-4 | |
M | 7 | PAHOR, M. et al.: Lancet 2000; 356: 1949-54 |
R | 8 | ALLHAT Officers and Coordinators: JAMA 2002; 288: 2981-97 |
R | 9 | ALLHAT Officers and Coordinators: JAMA 2000; 283: 1967-75 |
* |
|
© arznei-telegramm 08/03 |