Justification for no intervention controls in the cervical cancer screening trials in India

The crucial point of disagreement is whether the placebo trials were really necessary to answer the policy relevant question: Could equivalence trials answer give policy makers the answers that they need? The Indian researchers, their funders, and WHO have clearly endorsed the randomized controlled trials with no screening in the control group. The critics disagree. What are the arguments?

Those who defend the trials claim that a local standard of care is necessary because that is to only way to obtain scientifically valid results. The critics, including Ruth Macklin, argue that all arms in a trial should get some form of cervical cancer screening. Again, it is Macklin who provides the argument. It is worth quoting at length from professor Macklin’s paper where she spells out this argument in detail:

It is simply not true that “only a ‘no care’ control arm can give definitive results.” Although the randomised controlled trial is the “gold standard” in clinical research methodology, this does not mean that the control arm must be a placebo. In settings in which the standard diagnostic method is a proven intervention and researchers want to test a new method, or even a less expensive method, it would be unethical to withhold the proven diagnostic method from the participants. The research design would then be a non-inferiority trial, which would test the experimental procedure against the proven intervention to see whether the former is as good (or almost as good) as the latter. That is a perfectly acceptable research design, although it would involve more research subjects and take longer than a placebo-controlled trial. The idea that it is ethically acceptable to design a study in resource-poor settings in which the participants do not have access to a proven diagnostic method outside the trial is flawed. If researchers in India wanted to study VIA to determine whether it is as good (or almost as good) as the Pap smear, they could do so in a tertiary care setting which has the equipment and trained personnel to allow for the routine use of the cytology-based screening method. Using the existing baseline data on the incidence of cervical cancer in India, the efficacy of the experimental method (VIA) could then be ascertained [My emphasis]

It is certainly true that a non-inferiority trial can be designed to establish that a new intervention is or is not equivalent or non-inferior to an existing intervention. The question is whether this design is appropriate in this case for the purpose of answering the question policy makers have. This depends on how we can interpret the results of such a trial. Let us proceed separately with the two possible scenarios, first that the trial establishes that the two interventions are equivalent, and second that they are not.

If the trial establishes that the two interventions are equivalent, we cannot from that fact alone conclude that the new intervention is generally effective due to a problem that Susan Ellenberg and Bob Temple  have identified as lack of assay-sensitivity. This is the ability of a test (clinical trial) to identify an effect reliably. If an equivalence trial demonstrates that two interventions have a similar effect, we can only conclude that the new intervention is actually effective if it is known that the control intervention is effective. In the case of cervical screening trials the only intervention that was identified as effective was systematic screening every three years for at least a decade. An equivalence trial would have to include this intervention as a control if we want the trial to establish equivalence to a known effective intervention. The active intervention arm, however, would likely only include a single screening visit, as this is the feasible policy option to be tested. But it is highly unlikely that any trial would establish the equivalence between this intervention and the worldwide best standard of care. A trial that aims to establish equivalence would therefore be a complete waste of resources. The only sensible aim and outcome is therefore a trial that establishes that the new intervention is inferior to the established intervention.

If the trial establishes that the two interventions are not equivalent, i.e. that the difference between the standard intervention and new intervention is greater than what has been identified as clinically significant, we can indeed conclude that the new intervention is inferior to the established intervention. But that does not mean that we automatically should reject the new intervention as unsuitable in a low income setting: even though it is inferior to the best current intervention. Policy makers in resource poor settings need to know how much better the new intervention is to what is available locally now. If the trial had included an arm with the local intervention, the trial data would allow us to say something about that. In the absence of such an arm in the equivalence trial, we have to use, as Macklin points out, “existing baseline data” as a comparator, and make the assumption that these data are the same as we would have found had we included a control arm in the population under study. This study would basically then have the status of a regular historical control study, and we have seen in the previous post that such data were not sufficient to establish the effectiveness of VIA. 

An equivalence trial will therefore simply not provide the knowledge that policy makers want. They do not want to know how VIA done under ideal conditions compare with PAP smear screening, i.e. done every three year with access to high quality labs. Nor do they want to know how VIA done under field conditions compares with PAP smears done under field conditions. PAP smears done under these conditions may be ineffective: establishing that PAP smear is equivalent to VIA does not establish the effectiveness of VIA. Macklin’s proposal will simply not provide us with any useful results. What we do want to know is how effective VIA is done in field conditions, with testing once or only a few times, not regularly every three years. The reason for this is directly linked to the two problems pointed out with regard to equivalence trials: the effect of VIA varies enormously with regard to how it is carried out, and the relevant comparison is not how VIA done under ideal conditions compares with PAP smears under ideal conditions, but how effective a feasible VIA screening program compares with the current situation of no screening. If that is what we want to know, then we have to have a no-intervention control.

An examination of the results of the three trials supports this analysis. The first trial was published in 2007. It demonstrated that a single VIA screening caused a 35% reduction in deaths from cervical cancer. This is a significant reduction, especially because of the simplicity of the intervention. Had they used any PAP smear intervention in the control arm, it is likely that the results would be uninterpretable. Use of a single PAP smear might have shown no difference between PAP smears and VIA, or that PAP smear screening was inferior.  Since very little is known about the effectiveness of a single PAP smear, the trial results provide no useful policy advice. The results of the second reported trial confirm this analysis. This trial had four arms, one with PAP smears, one with VIA, one with HPV testing and one control. It only found that HPV testing was effective in reducing cancer. Both PAP and VIA were not effective compared with the control group. Had the trial not included a control group we would have correctly concluded that HPV testing is more effective than PAP and VIA, in line with the argument presented by Macklin for an equivalence trial. But we would also have concluded that VIA is effective, because it is assumed that PAP is effective, and there was no difference between them. It turns out, nor surprisingly, that two screening method used in this trial, using PAP and VIA, is not effective at all, even though, under ideal conditions they might be.

In 2009, then, there are results of two trials, one showing that a single screening is effective, one showing that it is not. It is not clear why there is this difference between the two trials. But it does show that how VIA is implemented is crucial for its effect, including any follow up treatment on the basis of a positive result. To repeat the obvious, one simply cannot rely on data from outside a particular trial to infer what the results would be for the trial subjects had one used a particular intervention, or no intervention.

The results of the final and third trial was published in 2014. It was the publication of these results that prompted the last round of criticism in the Indian Journal of Medical Ethics. It demonstrated that screening four times with VIA reduced cancer mortality by 30%, showing that a relatively simple intervention, but still less than PAP smears every three years, could have a dramatic effect on cervical cancer mortality. This is a result that simply could not have been know without the results from this trial, and without a control group with no intervention.

Perhaps what is most damaging for the critics is that any alternative design would also expose a significant number of women to known inferior screening programs, without yielding any useful results. No research design avoids exposing women to known inferior interventions, but only a placebo control trial can provide knowledge useful for policy. Having a state of the art control group in the three trials discussed here would have exposed all the other women in the trial to what is known to be an inferior screening program. Having a truncated Pap smear screening program in the control group would expose all women to known inferior treatments. 

Therefore, even if we agree that these trials provided results that are essential for policy, it does not follow directly from this that the trials were justified. We also have to argue that we are not exposing the individuals in the trial to unjustified risk. Even those who advocate equivalence trials have to do this, since they also expose individuals to known inferior interventions. What alternatives are there in 1998? There is no national screening program. Only the wealthy and educated in India have access to cervical screening programs. There are no plans to introduce any national screening program, and there is agreement that the known, effective screening program, Pap-smears, is unimplementable in the foreseeable future. In fact, 20 years later there is still no national screening program in India. In 1998 there was agreement that more research is necessary to identify a suitable screening program. 

The challenge for the critics is that any research design will expose a large group of people to a known inferior screening program.  And if the design of these trials is not able to answer the policy question, then the critics are exposing all these women to an inferior intervention for no good reason whatsoever. The only way to avoid exposing women to known inferior interventions is to do no research at all. But that would mean that no women get access to something that may be of benefit to them. Doing the research will at least save some lives, even in the control group because of the increased monitoring of this group, and useful knowledge is generated that may be used to benefit future women. Finally, even with a no intervention control group, no woman in the control group is denied any treatment that they would be able to get outside of the clinical trial.

Let me conclude this examination of the ethics of the cervical cancer screening trials by making a few remarks. Most people intuitively object to having no intervention in a control group when there is an effective intervention available, even somewhere else. In an ideal world everyone should have access to useful interventions. However, it should also matter morally whether a given scientific design actually produces useful results, and whether alternative designs allow us to avoid the moral challenge of providing everyone with known effective interventions. The role of bioethicists should be to examine carefully the arguments provided by those who argue in favor of having a no intervention control arm. It is simply not acceptable to simple refer to the Helsinki Declaration, and think that this settles the question of the suitability of alternative scientific designs, in the way that writers in the Indian Journal of Medical Ethics and the Ethics Dumping project tend to do. Once it is clear what the scientific issues are, it is still necessary to weigh the harms and benefits of alternative courses of actions, and then reach a conclusion about how to proceed. But simply claiming that it was already known in 2000 that VIA was an effective screening method, or that equivalence trials can answer the relevant policy questions, do not change the empirical facts that are needed to make morally sound judgments.