The methods and definitions used in the present meta-analysis were broadly similar to those used in the previous meta-analysis.1
Identification of trials
Details of each trial included in the analysis are available on bmj.com. The aim was to identify all trials, published or otherwise, that were available by September 1997 and that compared an antiplatelet regimen with a control or one antiplatelet regimen with another among patients considered to be at high annual risk (for example, over 3% a year) of vascular events because of evidence of pre-existing disease (previous occlusive event or predisposing condition). We included only those trials that were believed to have used a randomisation method that precluded prior knowledge of the next treatment to be allocated (thus, alternation or odd or even dates would not suffice) and were “unconfounded”—that is, contained two randomised groups that differed only with respect to the antiplatelet comparison of interest. Trials of oral antiplatelet regimens were eligible only if they had assessed more than one day of treatment, but we included trials of parenteral antiplatelet regimens of any duration. An antiplatelet drug was defined as one whose primary effect on the vascular system is to inhibit platelet adhesion, platelet aggregation, or both.1
We identified relevant trials by searching several electronic databases (Medline, Embase, Derwent, Scisearch, and Biosis; search strategy available on request); searching the trials registers of the Cochrane Stroke and Peripheral Vascular Disease Groups; manual searching of journals, abstracts, and proceedings of meetings; scrutinising the reference lists of trials and review articles; and inquiry among many colleagues, including representatives of pharmaceutical companies.
Definition of outcomes
The primary measure of outcome was a “serious vascular event” (that is, non-fatal myocardial infarction, non-fatal stroke, or death from a vascular cause and including any death from an unknown cause because most deaths in high risk patients are likely to be due to vascular causes). In order to allow the number of serious vascular events to be derived by adding the numbers of non-fatal myocardial infarctions, non-fatal strokes, and vascular deaths, we considered an event non-fatal only if the patient survived to the end of the scheduled follow up period (or died of a definitely non-vascular cause). Each contributing trialist's definition of a particular outcome (such as myocardial infarction) was used for counting vascular events, and we included all events classified by the trialist as probable or definite.
Deaths were divided into those with a vascular cause (defined as cardiac, cerebrovascular, venous thromboembolic, haemorrhagic, other vascular, or unknown cause) and those that were considered definitely non-vascular. Strokes were subdivided into intracranial haemorrhages (including intracerebral, subdural, subarachnoid, and extradural haemorrhages) and strokes of ischaemic or unknown aetiology; transient ischaemic attacks were not to be included. Major extracranial bleeds were those occurring outside the cranial cavity that were considered by the trialist to be serious (which, in general, meant that the patient required admission to hospital or blood transfusion). If during the trial a patient experienced more than one type of non-fatal outcome—for example, a myocardial infarction followed by a stroke—both events were recorded, but such patients contributed only once to the composite outcome of serious vascular event. If during the trial a patient experienced more than one non-fatal event of the same type (for example, two myocardial infarctions) or more than one pathological type of stroke (for example, a haemorrhagic stroke and an ischaemic stroke), only the first was to be recorded.
We asked the coordinators of all potentially eligible trials for details about method of randomisation, blinding of treatment allocation, scheduled duration of treatment, and, if different, scheduled duration of follow up. Investigators for trials that had randomised at least 200 patients were asked to contribute, for each patient originally randomised, data on baseline characteristics (age, sex, blood pressure, and medical history) and dates of randomisation, follow up, and any vascular events that had occurred. In addition, we asked them for a tabular summary of the numbers of patients originally allocated to each treatment group (that is, without any post-randomisation exclusions) and the numbers of patients experiencing particular outcomes during the scheduled follow up period. These outcomes were non-fatal myocardial infarction, non-fatal stroke (haemorrhagic or other), non-fatal or fatal pulmonary embolism, death from a vascular or unknown cause, death from a definitely non-vascular cause, and major extracranial bleeding. Investigators responsible for trials that had randomised fewer than 200 patients were asked only for the tabular summary of the numbers of patients and outcomes (although a few such studies did contribute individual patient data).
In trials assessing a month or more of treatment, we intended that analyses would be of events occurring during the scheduled treatment period, but in two trials follow up data were available only for a period in excess of the scheduled treatment period (see bmj.com for details).8,9
In trials with shorter courses of treatment, we analysed events during a period as close as possible to one month after randomisation. We checked data both for internal consistency and for consistency with relevant published reports and referred queries back to trial coordinators. Especially when data on individual patients were provided, the calculated numbers of vascular events may differ slightly from those reported in trial publications. Occasionally, when trial data had been discarded by investigators or were otherwise not available, the numbers of vascular events could be determined only from published reports.
Proportional and absolute effects of treatment
We stratified analyses by trial to avoid direct comparisons between individuals in different studies. We calculated the observed minus the expected number of events, and its variance, from standard 2×2 tables of outcome by treatment. These were then summed over trials to give the grand total for observed minus expected events (O−E) and its variance (V). We then based significance tests on comparison of z=(O−E)/√V with the standard normal distribution; P denotes the two sided significance level and P>0.05 is non-significant by convention. The typical odds ratio for these trials was calculated by the one step method10
from b=(O−E)/V, either as exp(b) or, for rare events, as (2+b)/(2−b). For odds ratios between 0.5 and 2 these two methods give almost identical answers.
Some trials used a deliberately unequal randomisation ratio and so had a substantial imbalance in the numbers of patients in treatment and control groups. We multiplied the control group in such trials by an appropriate integer1
when calculating “adjusted” control totals (although not when making other calculations). When comparing the percentages affected in the treatment and in the adjusted control groups, we calculated the standard error (SE) of the difference (D) between these percentages as D/z.
Effects in specific categories of trials
We compared different trials or groups of trials using standard χ2
tests for heterogeneity or, where appropriate, tests for trend between the observed effects on vascular events (with appropriate allowance made for multiple comparisons). But, even where there is significant heterogeneity, groups of patients in whom treatment is particularly advantageous or relatively ineffective can be difficult to identify reliably. Especially when small numbers of patients in a particular category have been studied, it is important that “lack of evidence of benefit” when that category is considered on its own is not misinterpreted as “evidence of lack of benefit.”11
As antiplatelet therapy reduces vascular events in a wide range of patients at high risk of occlusive vascular disease, the relevant question in any particular category is whether there is convincing evidence that there is no material benefit from treatment.12,13
Description of trials
We identified 448 apparently randomised trials comparing an antiplatelet regimen with a control or one antiplatelet regimen with another among high risk patients. After review and, in cases of doubt, consultation with trial coordinators, 166 trials were excluded: 52 were not properly randomised, 24 were confounded, three had large numbers lost to follow up, 13 were abandoned before any outcome data were collected, 20 had a crossover design, and 54 had not systematically recorded any of the relevant outcome events. In addition, since the focus of the present analyses was on patients at high risk of occlusive arterial disease, we excluded trials among patients with dementia or occluded retinal veins (even if they had been included in the 1994 meta-analysis1
). Insufficient information was available from 19 eligible trials among 3427 patients.
Details of the remaining 197 randomised trials that compared antiplatelet therapy versus control (195 with data on vascular events) and the 90 that compared different antiplatelet regimens (89 with data on vascular events) are available on bmj.com. Information on individual patients was available for trials that collectively included 59% of the vascular events, and in these trials fewer than 2% of patients were lost to follow up. (Further details of excluded trials and missing data are available on request.)