The architecture of clinical research

Author and Disclosure Information



In analyzing this trial, we could characterize it as a parallel controlled trial. Whether the allocation was done by randomization is not clear, but it was certainly an observational cohort study in that there were concurrent controls who were treated as similarly as possible except for the principal maneuver, which was the administration of citrus fruit. Already mentioned was the attention to averting susceptibility and performance bias. There was no evidence of compliance bias as the interventions were enforced, nor was there evidence of transfer bias because all subjects who were enrolled in the study completed the study because they were a captive group on a sailing ship. Finally, the likelihood of assembly bias seems small, as these sailors seemed to be representative of victims of scurvy in general, namely in being otherwise deprived of access to citrus fruits.

In terms of the statistical results of this study, subsequent analysis of the research showed that the impact of lemons and oranges was dramatic and showed a trend (P = .09) towards statistical significance. Notwithstanding the lack of a P < .05, Dr. Feinstein would likely say that this study satisfied the “intra-ocular test” in that the efficacy of the citrus fruit was so dramatic that it “hit you between the eyes.” He often argued that the widespread practice of prescribing penicillin for pneumococcal pneumonia was not based on the results of a convincing randomized controlled trial because the efficacy of penicillin in that setting was so dramatic that a randomized trial was not necessary (and potentially even unethical if the condition of “intra-ocular” efficacy was satisfied).

The final question to address in this lecture is whether randomized controlled trials, for all their rigor, always produce more reliable results than observational studies. This issue has been addressed by several authors.10–12 Sacks et al10 contended in 1983 that observational studies systematically overestimate the magnitude of association between exposure and outcome and therefore argued that randomized trials were more reliable than observational studies. Subsequent analyses tended to challenge this view.11,12 Specifically, Benson and Hartz11 compared the results of 136 reports regarding 19 different therapies that were studied between 1985 and 1998. In only 2 of the 19 analyses did the treatment effects in the observational studies fall outside the 95% confidence interval for the randomized controlled trial results. In this way, these authors argued that observational studies generally are concordant with the results of randomized trials. They stated that “our finding that observational studies and randomized controlled trials usually produce similar results differs from the conclusions of previous authors. The fundamental criticism of observational studies is that unrecognized confounding factors may distort the results. According to the conventional wisdom, this distortion is sufficiently common and unpredictable that observational studies are not liable and should not be funded. Our results suggested observational studies usually do provide valid information.”11

An additional analysis of this issue was performed by Concato et al,12 who identified 99 articles regarding 5 clinical topics. Again, the results from randomized trials were compared with those of observational cohort or case-controlled studies regarding the same intervention. The authors reported that “contrary to prevailing belief, the average results from well-designed observational studies did not systematically overestimate the magnitude of the associations between exposure and outcome as compared with the results of randomized, controlled trials on the same topic. Rather, the summary results of randomized, controlled trials and observational studies were remarkably similar.”12

On the basis of these studies, it appears that randomized control trials continue to serve as the gold standard in clinical research, but we must also recognize that circumstances often preclude the conduct of a randomized trial. As an example, consider a randomized trial of whether cigarette smoking is harmful, which, given the strong suspicion of harm, would be unethical in that patients cannot be randomized to smoke. Similarly, from the example before, a randomized trial of penicillin for pneumococcal pneumonia would be unethical because denying patients in the placebo group access to penicillin would exclude them from access to a drug that has “intra-ocular” efficacy. In circumstances like these, well-performed observational studies that are attentive to sources of bias can likely produce comparably reliable results to randomized trials.

In the end, of course, the interpretation of the study results requires the reader’s careful attention to potential sources of bias that can compromise study validity. The hope is that with Dr. Feinstein’s framework, you can be better equipped to think critically about study results that you review and to keenly ascertain whether there is any threat to internal or to external validity. Similarly, as you go on to design clinical trials yourselves, you can pay attention to these potential sources of bias that, if present, can compromise the reliability of the study conclusions internally or their applicability to patients outside of the study.

Next Article:

Basics of study design: Practical considerations

Related Articles