1405
LETTERS to the EDITOR
WHO
clofibrate/cholesterol trial: clarifications
SIR,-Reports of the World Health Organization primary trial of ischaemic heart disease with clofibratel-3 may need clarification if the data are to be included accurately in overviews of clinical trials of cholesterol lowering. Confusion seems to have arisen over the terms "in trial" and "out of trial" used in the published reports, particularly in relation to analysis by intention to treat.4--7 Nor does the unusual completeness of the mortality follow-up seem to have been appreciated. The debate on cholesterol lowering has flared up again, this time in the BM],4--7 and Lancet readers may ask what difference the following re-presentation of data for the WHO clofibrate trial makes to the conclusions of overviews or meta-analyses. In our opinion, the explaining away of the excess mortality reported in the WHO trial, which remains the largest long-term, double-blind study of reducing hypercholesterolaemia, needs more serious scrutiny. We hope that these clarifications will make it even more evident that the adverse features of this trial can no longer be rationalised. In the WHO trial, when a patient had a myocardial infarction his physician was free to treat him with or without clofibrate and he was withdrawn. Withdrawals from the trial for other medical reasons also occurred, and some men "lapsed" (ie, opted out, failed to attend, or moved away).l Closure of the trial as a whole took place from July, 1975, treatment being stopped and men still in the trial withdrawn at the next visit after that date. Events before lapse, withdrawal, or closure were classified as "in trial"-ie, patient still receiving "treatment" (clofibrate or placebo). Events after closure, lapse, or withdrawal were "out of trial"-ie, patients no longer receiving treatment. Deaths after lapse or withdrawal for reasons other than closure of the trial as a whole were "out of trial". In the 1978 paper,’ incidence rates (IHD) were calculated on the basis of man-years of exposure up to the time of leaving the trial, whether because of withdrawal, lapse, or closure. The basis was not, therefore, "intention-to-treat" because events between leaving the trial and its closure for men who left it early were excluded. Mortality from ischaemic heart disease (IHD) and other causes up to one year after leaving the trial was also reported, and this too was not fully on the basis of intention to treat.
prevention
TABLE I-DEATHS BEFORE AND AFTER CLOSURE OF TRIAL BY MAIN CAUSE AND TREATMENT GROUP
In the follow-up papers, on the other hand, mortality data were obtained for all men up to the end of 19782and then 1982,33 irrespective of whether they had been withdrawn, had lapsed, or had stayed in the trial until it closed. The numbers and rates in these papers are on the basis of intention to treat, and all men originally randomised or selected for the treatment groups are included in the population at risk. In some tables, however, rates were calculated according to when the death occurred, either in trial or in various intervals after leaving the trial. In table ni of the 1984 paper3 deaths were
classified
as
"In the
treatment
phase" (in trial)
or
"After
treatment up to Dec 31,1982" (out of trial)—ie, not on the basis of intention to treat which would have the inclusion of all
implied
deaths before and after a certain time or before and after trial closure. For those seeking intention-to-treat data we present here the 1982 follow-up mortality information before and after closure of the trial. This date for a man who had left early was deemed to be the anniversary of his admission to the trial, which fell between July, 1975, and July, 1976. Men who died after leaving the trial (ie, ceasing treatment) but before their closure date, thus defined, have therefore been grouped together with those who died in trial by the old definition as dying before closure of the trial. Men who died after their closure date were classified as dying after closure of the trial. Tablei shows a total of 542 deaths from all causes, in all three groups, before trial closure, an increase of 254 over the 288 previously reported in-trial deaths. Of these 254, 108 were in the clofibrate treated group (group I), and 94 and 52 in the high and low cholesterol controls (groups II and III), respectively. Table 11 gives information on deaths in groups I and II in a manner analogous to table ill in the 1984 papers except that the deaths are classified as before or after trial closure instead of in treatment or after treatment. The ratios (1/11) of deaths from all causes, and from all causes except IHD, before closure are lower than those in-trial reported earlier (shown in parentheses), but they remain significantly different from unity. For cancer, the probability level is p < 0 10. The increase in the ratio for IHD is not significant. (Incidentally the ratio for deaths from accidents and violence is close to unity by either method of analysis.) Though analysis by intention to treat can remove important biases the fact that it ignores the point at which treatment stopped in this study causes us to question whether it should always be the only method used. TABLE II-NUMBER OF DEATHS AND RATIOS IN GROUPSI AND II FOR IHD, CANCER, AND ALL CAUSES EXCEPT IHD
Group 1=clofibrate (5331 men), group II=high 111= low cholesterol group (5118)
cholesterol control
(5296),
group
Ratios from previous report’ in parentheses. Age-standardised rates not shown because previous report indicated that ratios of deaths and ofstandardised rates were very similar 0-01. Significantly different from 1 00 at *p < 0 05 or tp
LETTERS to the EDITOR
WHO
clofibrate/cholesterol trial: clarifications
SIR,-Reports of the World Health Organization primary trial of ischaemic heart disease with clofibratel-3 may need clarification if the data are to be included accurately in overviews of clinical trials of cholesterol lowering. Confusion seems to have arisen over the terms "in trial" and "out of trial" used in the published reports, particularly in relation to analysis by intention to treat.4--7 Nor does the unusual completeness of the mortality follow-up seem to have been appreciated. The debate on cholesterol lowering has flared up again, this time in the BM],4--7 and Lancet readers may ask what difference the following re-presentation of data for the WHO clofibrate trial makes to the conclusions of overviews or meta-analyses. In our opinion, the explaining away of the excess mortality reported in the WHO trial, which remains the largest long-term, double-blind study of reducing hypercholesterolaemia, needs more serious scrutiny. We hope that these clarifications will make it even more evident that the adverse features of this trial can no longer be rationalised. In the WHO trial, when a patient had a myocardial infarction his physician was free to treat him with or without clofibrate and he was withdrawn. Withdrawals from the trial for other medical reasons also occurred, and some men "lapsed" (ie, opted out, failed to attend, or moved away).l Closure of the trial as a whole took place from July, 1975, treatment being stopped and men still in the trial withdrawn at the next visit after that date. Events before lapse, withdrawal, or closure were classified as "in trial"-ie, patient still receiving "treatment" (clofibrate or placebo). Events after closure, lapse, or withdrawal were "out of trial"-ie, patients no longer receiving treatment. Deaths after lapse or withdrawal for reasons other than closure of the trial as a whole were "out of trial". In the 1978 paper,’ incidence rates (IHD) were calculated on the basis of man-years of exposure up to the time of leaving the trial, whether because of withdrawal, lapse, or closure. The basis was not, therefore, "intention-to-treat" because events between leaving the trial and its closure for men who left it early were excluded. Mortality from ischaemic heart disease (IHD) and other causes up to one year after leaving the trial was also reported, and this too was not fully on the basis of intention to treat.
prevention
TABLE I-DEATHS BEFORE AND AFTER CLOSURE OF TRIAL BY MAIN CAUSE AND TREATMENT GROUP
In the follow-up papers, on the other hand, mortality data were obtained for all men up to the end of 19782and then 1982,33 irrespective of whether they had been withdrawn, had lapsed, or had stayed in the trial until it closed. The numbers and rates in these papers are on the basis of intention to treat, and all men originally randomised or selected for the treatment groups are included in the population at risk. In some tables, however, rates were calculated according to when the death occurred, either in trial or in various intervals after leaving the trial. In table ni of the 1984 paper3 deaths were
classified
as
"In the
treatment
phase" (in trial)
or
"After
treatment up to Dec 31,1982" (out of trial)—ie, not on the basis of intention to treat which would have the inclusion of all
implied
deaths before and after a certain time or before and after trial closure. For those seeking intention-to-treat data we present here the 1982 follow-up mortality information before and after closure of the trial. This date for a man who had left early was deemed to be the anniversary of his admission to the trial, which fell between July, 1975, and July, 1976. Men who died after leaving the trial (ie, ceasing treatment) but before their closure date, thus defined, have therefore been grouped together with those who died in trial by the old definition as dying before closure of the trial. Men who died after their closure date were classified as dying after closure of the trial. Tablei shows a total of 542 deaths from all causes, in all three groups, before trial closure, an increase of 254 over the 288 previously reported in-trial deaths. Of these 254, 108 were in the clofibrate treated group (group I), and 94 and 52 in the high and low cholesterol controls (groups II and III), respectively. Table 11 gives information on deaths in groups I and II in a manner analogous to table ill in the 1984 papers except that the deaths are classified as before or after trial closure instead of in treatment or after treatment. The ratios (1/11) of deaths from all causes, and from all causes except IHD, before closure are lower than those in-trial reported earlier (shown in parentheses), but they remain significantly different from unity. For cancer, the probability level is p < 0 10. The increase in the ratio for IHD is not significant. (Incidentally the ratio for deaths from accidents and violence is close to unity by either method of analysis.) Though analysis by intention to treat can remove important biases the fact that it ignores the point at which treatment stopped in this study causes us to question whether it should always be the only method used. TABLE II-NUMBER OF DEATHS AND RATIOS IN GROUPSI AND II FOR IHD, CANCER, AND ALL CAUSES EXCEPT IHD
Group 1=clofibrate (5331 men), group II=high 111= low cholesterol group (5118)
cholesterol control
(5296),
group
Ratios from previous report’ in parentheses. Age-standardised rates not shown because previous report indicated that ratios of deaths and ofstandardised rates were very similar 0-01. Significantly different from 1 00 at *p < 0 05 or tp
