 |
|
What is the background to this?
Clopidogrel can be used in combination with aspirin as prophylaxis against coronary events in patients who have received angioplasty. This nested cohort study, which was sponsored by the manufacturers of clopidogrel, looked at the rates of all-cause mortality among 13,663 patients in Quebec, Canada who underwent coronary stenting between January 2000 and September 2004. Until April 2007, the main health insurance provider in Quebec required specific approval before clopidogrel could be prescribed and dispensed. Physicians had to complete and fax a special authorisation form for approval.
The study compared mortality rates between those patients who obtained a supply of clopidogrel at the same time as other cardiovascular (CV) drugs, those where there was a delay in obtaining a supply of clopidogrel compared to obtaining their other CV drugs and those who did not obtain any supply of clopidogrel. They assumed that any delays in obtaining clopidogrel, or not obtaining supplies at all, were due to the approval requirements of the health insurance provider
What did this study claim?
Of the 13,663 patients, 1571 (11.5%) did not fill any clopidogrel prescriptions after a percutaneous coronary intervention despite doing so for other CV drugs, and 1174 (8.6%) patients filled their clopidogrel prescription with a delay of at least 1 day (median delay 5 days). After adjusting for certain co-variables and compared with those who obtained a prescription without delay, patients who did not fill any clopidogrel prescriptions had a 70% greater relative risk of dying from any cause (Hazard Ratio [HR] 1.70, 95% confidence intervals [95%CI] 1.34 to 2.15). Those in whom there was a delay in filling a prescription had a 34% greater relative risk of dying from any cause (HR 1.34, 95%CI 1.01-1.80) in the year following the coronary intervention. The authors attribute 50 deaths per year among Quebecois undergoing stenting to the policy of requiring specific approval for clopidogrel use.
So what?
This study has several flaws and limitations. First, and most obviously, is the assumption that delays in obtaining clopidogrel supplies, or not obtaining supplies at all, were due solely to the health insurance provider’s policy – this is asserted but the authors do not justify this statement or support it with evidence. The authors were also not able to adjust for contraindications to clopidogrel, so, for example, it might have been that some patients would not have received clopidogrel even if no special authorisation had been required. This is linked to the second limitation of the study – as the authors say, they cannot exclude the possibility that differences in other risk factors between the groups affected mortality. This is known as confounding. It is quite possible that those who did not receive clopidogrel or in whom its prescription was delayed were sicker (and so more likely to die) than those who received clopidogrel early.
However, a more subtle limitation is immortal time bias. This is discussed in an editorial accompanying this paper. Immortal time bias is not widely recognised but is very important when analysing cohort studies. You may wish to read another paper by the editorialist (Suisa S, Am J Respir Crit Care Med 2003; 168: 49-53), which explores the concept and shows how immortal time bias led authors of a cohort study to conclude that early introduction of inhaled corticosteroids reduced readmission rates after first hospitalisation for COPD, when in fact the data did not support this. In summary, immortal time is the time during follow-up when, by definition of the study group, a person cannot experience the event being examined. In this study, the effect of the bias arising from this phenomenon is that people in the “delayed clopidogrel” and “no clopidogrel” groups appeared to be at greater risk of death than those in the “no-delay” group.
It may be enough for readers to simply understand that immortal time bias can greatly exaggerate the benefits of an intervention in a cohort study, as with this study of clopidogrel. Those who wish to understand more can follow the link to the Suisa paper and/or read the notes about this below.
Action
There are two action points which arise from this publication. Firstly, the study does not support claims that administrative activity to limit access to clopidogrel was responsible for unnecessary deaths. Note, it does not categorically disprove that hypothesis, but it provides no useful information either way. A randomised controlled trial, which would avoid immortal time and confounding biases, or at least a better conducted observational study, would be required to answer that question. It might be the case that having a lower threshold for clopidogrel use (eg in relation to bleeding risk) might be harmful overall. That is plausible given some results from clinical trials (see the ACS floor of NPCi) – but could equally be untrue. However, lowering the threshold for clopidogrel use in the basis of this study would be unjustified and may be unwise.
Secondly the study highlights the need to be aware of the risk of immortal time bias when looking at observational studies. It is not necessary for front-line health professionals to understand the concept fully: it is enough to recognise that it may exist and should be considered. As the editorialist writes, immortal time bias may be present if:
-
It is an observational study (usually cohort studies)
-
The drug exposure changes over time
-
The data analysis does not properly account for the changes in drug exposure over time
-
A surprisingly tremendous drug benefit is reported
-
The result is hard to believe or too good to be true (it probably isn’t true)
Notes on Immortal Time Bias
Immortal time is a phenomenon found in cohort studies. The expression refers to a period during follow-up when, by definition of the study group, a person cannot experience the event being examined. There are two immortal time periods in this clopidogrel study.
First of all, the only patients included in the study were those who survived long enough to have at least one prescription for cardiovascular (CV) medication dispensed. The time between discharge and first dispensing is therefore immortal time for these people. In other words, everyone included in the study could not, by definition, have died in this time (ie they were immortal): people who died before they had chance to have any prescription dispensed (whether or not their first prescription would have included clopidogrel) were by definition not included in the study.
This introduces bias, because of the difference in mean time to first dispensing of a non-restricted medicine (6.9 days in the no-clopidogrel group versus 1.6 days in the delayed-clopidogrel group and 3.1 days in the no-delay group). Since the risk of fatal thrombotic complications is greatest soon after stenting, the “delayed clopidogrel” group would have included more patients who were at higher risk of dying than either of the other groups.
For example, consider two fictitious people who were discharged after coronary stenting: Sarah and Sue. They both had fatal thrombotic events on day 3 after discharge. Sarah, who collected a prescription for statins and beta-blockers on day 1 and who collected a prescription for clopidogrel on day 2, was included in the “delayed clopidogrel” group, and her death was recorded. Sue was intending to collect her prescription for clopidogrel and her other medicines together on day 3, but died before she could do so. She was not included in the “no-delay” group and her death was not recorded. Consider a third fictitious person, Karen, who survived long enough to collect her prescription for clopidogrel and her other medicines together on day 3. She was therefore included in the “no delay” group. Days 0-3 were “immortal time” for Karen since, if she had died in that time she would have been like Sue and not have been included in the analysis. That she survived long enough to collect her prescriptions suggests that she was at lower overall risk of dying than Sarah or Sue, owing to factors apart from whether or not she took clopidogrel. Thus the effect of this first immortal time bias is that people in the “delayed clopidogrel” group were at greater CV risk than those in the other groups.
The second period of immortal time is more complex, and relates to the time between collecting non-restricted CV medication and clopidogrel prescriptions. Consider four other fictitious people who were discharged after stenting: John, Martin, Charles and David. All survived long enough to collect their first prescription and so were included in the analysis. All four collected their prescriptions for non-restricted CV medicines on day 8 and all four had fatal thrombotic events on day 10. John collected his clopidogrel with the rest of his medication on day 8, and so was included in the “no-delay” group. Martin collected his clopidogrel prescription on day 9 and so was included in the “delayed clopidogrel” group. Charles was scheduled to pick up his clopidogrel prescription on day 11 but died before he was able to do so. David was not scheduled to receive clopidogrel at all. Both Charles and David were therefore included in the “no-clopidogrel” group. Days 0-8 were immortal time for all four men, since if they had died in that time they would not have been included in the analysis at all. But in addition, days 8-10 were immortal time for Martin with respect to his inclusion in the “delayed clopidogrel” group. If he had died in this time he would, like Charles, have been included in the “no-clopidogrel” group. Thus the effect of this second immortal time bias is that people in the “no-clopidogrel” group appear to be at greater risk of death than those in the “delayed-clopidogrel” and “no-delay” groups.
The authors compared the death rate in the “no-delay” group with the rate in the combined “delayed clopidgrel” and “no clopidogrel” groups. The adjusted hazard ratio was 1.55 (95% CI 1.27–1.89). This avoids the second immortal time bias. Does it therefore mean that anything other than receiving clopidogrel without delay is harmful (a 55% increased risk of death)? No it does not: the effects of the first immortal time bias may still be present and, perhaps more importantly, the study is still subject to confounding biases, as described above. It is quite possible that those who did not receive clopidogrel or in whom its prescription was delayed were sicker (and so more likely to die) than those who received clopidogrel early.
Immortal time bias is a complex, relatively recently recognised phenomenon affecting cohort studies. The correspondence to this paper may add further useful information when it becomes available, but we will be looking a lot more closely at cohort studies in the future to assess whether this new potential bias could be operating.
Feedback:
Please comment on this blog in the NPC discussion rooms, or using our feedback form.