Evaluating non-randomised intervention studies - NIHR Health ...

Health Technology Assessment 2003; Vol. 7: No. 27against smoking, treatment of raised bloodpressure and antiplatelet drugs, although itwas noted to vary between centres and over theyears.The primary outcome was death or major stroke.Overall there was a non-significant difference inoutcome (37.0% of surgery-group patients versus36.5% of control-group patients), although therewas a relationship between benefit of surgery anddegree of stenosis, surgery being of value when thedegree of stenosis was >80%.Baseline characteristics in the ECSTEight variables for baseline characteristicsassociated with prognosis were included in thedata set. Details of the values of these variables arenot given in the report as they will soon bepublished in other research publications producedby the trial collaborative group. Thesecharacteristics were collected during therandomisation process, and were thereforeavailable for every randomised patient.Adaptations made to the ECST for theresampling projectTo use the ECST dataset for this project, themulticentre nature of the trial was exploited toconstruct a series of smaller randomised and nonrandomisedstudies, in the same way as for theIST. Centres falling within geographical regionswere grouped together such that sufficientparticipants were accrued in each arm for theresampling to be undertaken. In addition, thesequential nature of the study recruitment wasused to split participants into ‘early recruits’ and‘later recruits’ to enable historically controlledstudies to be constructed. For all regions otherthan Region 4, early recruits were those recruitedup to and including 5 December 1987; laterecruits were those recruited after this date. A cutoffdate of 19 December 1988 was used for Region4. These dates were chosen to maximise thenumber of regions available for inclusion in thestudy.Sample sizes used in the resampling for the ECSTstudy were smaller than those for the IST. To beconsidered as a sub-study for the concurrentcohort design, we required that there were at least40 participants within each arm of each trial. Forthe historical cohort design we required that whenthe arms were divided into early and late recruitsthere were at least 40 early control recruits and 40late treatment recruits.Eight geographical regions were constructed onthis basis for the main analysis for both concurrentcohort and historical cohort designs. As centreswere smaller than for the IST, neighbouringcountries were grouped where necessary. Theseeight regions recruited 2833 (94%) of trialparticipants. For reasons of future publications,the regions are not identifiable in the tables wepresent.The ECST is one of two trials investigating theefficacy and safety of carotid endarterectomy.Both the ECST and the second trial, NASCET, 135published early results during the recruitmentperiod of the ECST, which led to the datamonitoring committee of the ECST changingthe inclusion criteria of the ECST in 1990.Cases recruited after this change were deletedfrom the data set where this change would beconfounded with the generation of groups(e.g. for the historically controlled studyanalyses).The distribution of the outcome of dead or majorstroke, and of the baseline characteristics in thesesub-studies are described in Table 41. Table 42describes the baseline characteristics for thecohorts split according to ‘early recruits’ (before)and ‘late recruits’ (after). Differences are evidentin both tables in the frequency of the outcome andfor some of the baseline characteristics, but aregreatest in the historical comparisons. In thesecentres the event rates are lower in the secondperiod in all eight centres, whereas thepercentages with high degrees of stenosis increasewith time in all but one centre. The patternobserved with degree of stenosis is likely to reflecta change in case-mix due to emerging evidence ofthe dangers of surgery for low-grade surgeryduring the trial’s recruitment period.173© Queen’s Printer and Controller of HMSO 2003. All rights reserved.