Let me begin with a disclaimer that I do not endorse chelation therapy, and do not view the results of the TACT trial as providing ‘actionable’ evidence for regulatory decision making or for guiding clinical practice. However, notwithstanding the flaws in trial design and conduct, I would like to make the following observations that address some of the key criticisms launched against the TACT trial. (The following has been adapted from earlier comments I made in response to a blog post on Respectful Insolence.)
1. Missing data, the elephant in the room
The amount of missing data, while vexing, is not unexpected given the onerous treatment protocol involving multiple infusions over a prolonged period of time. The missing data can challenge the interpretation of trial results, but the claim that it invalidates the results is not supported by careful examination of the supplementary information provided in the eAppendix. There do not appear to be major imbalances in either the key prognostic covariates or in the reasons for ‘missingness’ amongst the dropouts in the 2 randomized groups. Thus, careful scrutiny of the data doesn’t necessarily lead to the conclusion that ‘informative censoring’ did indeed occur in favor of chelation treatment. Moreover, the results of the sensitivity analyses, which I find quite reasonable, do not support the assertion that differential missing data bias the results in favor of the chelation treatment. I acknowledge that missing data are problematic in general as they challenge the interpretation of the overall results, but the onus is on those who claim that it invalidates the results to prove so! By the way, the problem of missing data (and sloppy trial conduct) is quite common even in clinical trials run by academic research organizations of the highest repute!
2. Potential investigator or patient unmasking
Impugning the integrity of the trial investigators, especially at CAM sites, requires a higher burden of proof than simply innuendo. These sites were audited and after implementation of remedial measures, they passed the scrutiny of NIH, FDA, and OHRP.
One would have expected a greater treatment effect as a result of potential trial misconduct at these sites, but the data do not suggest a differential treatment effect at CAM (HR 0.89) vs non-CAM sites (HR 0.72), interaction p = 0.28.
3. Subjective endpoints
It has been stated that the treatment effect was driven by the soft and somewhat subjective endpoints of CV hospitalizations and revascularization. For a variety of reasons, I am not a big fan of the type of composite endpoint used in TACT. However, the inclusion of these soft endpoints is driven primarily by trial feasibility considerations as these components are more prevalent. In fairness, the trial was not powered for ‘hard’ endpoints of mortality, MI or stroke. It is somewhat reassuring to note that the HR for the composite of cardiovascular death, nonfatal MI, or nonfatal stroke (stringent MACE), a prespecified secondary endpoint, was 0.84 (p=0.22), similar to the HR of 0.82 (p=0.035) for the primary quintuple composite endpoint (MACE plus). Finally, there doesn’t appear to be heterogeneity of treatment effect across the components of the composite endpoint, thus strengthening the observations.
It is important to understand that the inclusion of subjective endpoints like CV hospitalization and revascularization increases the potential for ascertainment error and misclassification that typically biases the results towards the null. So, if the objective of the trial is to prove superiority, as was the case in TACT, the bias would operate against it (conservative). Conversely, if the objective of the trial is to prove noninferiority, i.e., rule out unacceptable risk, the bias would operate towards it (anti-conservative).
4. Implausibility of treatment effect and Bayesian analysis
It has been argued that “there was no compelling clinical or preclinical evidence that chelation therapy has significant efficacy against atherosclerotic cardiovascular disease, and we know that chelation therapy can cause harm; so the risk was not minimal.” “A Bayesian analysis would not look kindly on the results due to the low prior probability of treatment effect.”
First, let me state there are several examples in medicine that have shone the spotlight on the slippery slope argument of plausibility (antioxidant vitamins, hormone replacement therapy, magnesium for MI, anti-reperfusion injury therapy come to mind). Thomas Huxley said it best, “the tragedy of science, the slaying of a beautiful (plausible) hypothesis by ugly facts!”
Second, the reason for the low prior probability of benefit assertion is not clear to me. The available evidence for treatment effects of chelation therapy in patients with coronary heart disease or peripheral arterial disease is mixed, and limited to case series and three small trials evaluating surrogate endpoints. No outcome data derived from properly designed clinical trials existed to support or refute treatment effect! Despite the lack of supportive evidence, the number of patients undergoing chelation therapy increased by >50% between 2002 and 2007. Thus, it was important to clarify the benefits and harms of chelation therapy in a controlled outcome trial. Accordingly, the argument that the trial was unethical, as has been suggested by some, doesn’t hold water!
Third, the principle of equipoise, an ethical prerequisite for patient randomization and enrollment, would require suspension of one’s belief in treatment effect as reflected in a pretrial null hypothesis probability of 50%. The p value of 0.03 translates into a minimum Bayes’ factor of 0.09, which reduces the null probability from 50% pretrial to about 9% post-trial. While this does not represent strong evidence against the null, it does reduce the level of skepticism surrounding chelation therapy.
Here is a formal Bayesian analysis of TACT that might be instructive:
Evidence (likelihood) = 0.82 (0.69, 0.99)
Prior (skeptical) = 1.0 (0.75, 1.33)
The mean is centered on null, and there is only 5% probability of an extreme treatment effect, i.e., >25% risk reduction or >33% risk increase. Many would argue that this is a reasonably skeptical prior distribution.
Posterior = 0.87 (0.75, 1.02)
The posterior probability of a >13% risk reduction is 50%, and the probability of a >10% risk reduction is 66%.
So if we start from a position of skepticism, integrating the results of the TACT trial reduces the degree of skepticism. This is exactly how Bayesian analysis helps upgrade knowledge and information. One would require a prior probability of 2-fold increase in CV risk with chelation (not supported by evidence to my knowledge) to completely nullify the results of the TACT trial, an arguably IMPLAUSIBLE conjecture!
5. Implications for clinical practice
Given the significant treatment by diabetes interaction, some have argued ‘tongue-in-cheek’ that “the recommendation from this trial should be that no non-diabetic with cardiovascular disease should be treated with chelation therapy because it definitely does not work for them.”
The presumption that the TACT trial provides actionable evidence for clinical practice is not endorsed by anyone including the TACT investigators. The best case interpretation of the data is that the evidence is inconclusive. Ideally, under such circumstances, the only worthwhile recommendation should be ‘additional studies are warranted to adjudicate the uncertainties’. That is how science and knowledge should progress.
6. Double standard for acceptance of convenient versus inconvenient evidence
I agree with the notion that there is clearly a double standard when it comes to accepting the results of trials evaluating so-called dubious quack cures such as chelation versus trials assessing promising de rigueur cures such as gene transfer or stem cell therapy.
Bottom line, in my opinion, the arguments that the TACT results are dubious or not valid are overstated. While the debate surrounding TACT is clearly warranted and welcome, I hope it generates more light than heat.