Statistically significant differences versus convincing evidence of real treatment effects: an analysis of the false positive risk for single-centre trials in anaesthesia

BACKGROUND

The American Statistical Association has highlighted problems with null hypothesis significance testing and outlined alternative approaches that may 'supplement or even replace P-values'. One alternative is to report the false positive risk (FPR), which quantifies the chance the null hypothesis is true when the result is statistically significant.

METHODS

We reviewed single-centre, randomised trials in 10 anaesthesia journals over 6 yr where differences in a primary binary outcome were statistically significant. We calculated a Bayes factor by two methods (Gunel, Kass). From the Bayes factor we calculated the FPR for different prior beliefs for a real treatment effect. Prior beliefs were quantified by assigning pretest probabilities to the null and alternative hypotheses.

RESULTS

For equal pretest probabilities of 0.5, the median (inter-quartile range [IQR]) FPR was 6% (1-22%) by the Gunel method and 6% (1-19%) by the Kass method. One in five trials had an FPR ≥20%. For trials reporting P-values 0.01-0.05, the median (IQR) FPR was 25% (16-30%) by the Gunel method and 20% (16-25%) by the Kass method. More than 90% of trials reporting P-values 0.01-0.05 required a pretest probability >0.5 to achieve an FPR of 5%. The median (IQR) difference in the FPR calculated by the two methods was 0% (0-2%).

CONCLUSIONS

Our findings suggest that a substantial proportion of single-centre trials in anaesthesia reporting statistically significant differences provide limited evidence of real treatment effects, or, alternatively, required an implausibly high prior belief in a real treatment effect.

CLINICAL TRIAL REGISTRATION

PROSPERO (CRD42023350783).

Interpreting frequentist hypothesis tests: insights from Bayesian inference

Randomized controlled trials are one of the best ways of quantifying the effectiveness of medical interventions. Therefore, when the authors of a randomized superiority trial report that differences in the primary outcome between the intervention group and the control group are "significant" (i.e., P ≤ 0.05), we might assume that the intervention has an effect on the outcome. Similarly, when differences between the groups are "not significant," we might assume that the intervention does not have an effect on the outcome. Nevertheless, both assumptions are frequently incorrect.In this article, we explore the relationship that exists between real treatment effects and declarations of statistical significance based on P values and confidence intervals. We explain why, in some circumstances, the chance an intervention is ineffective when P ≤ 0.05 exceeds 25% and the chance an intervention is effective when P > 0.05 exceeds 50%.Over the last decade, there has been increasing interest in Bayesian methods as an alternative to frequentist hypothesis testing. We provide a robust but nontechnical introduction to Bayesian inference and explain why a Bayesian posterior distribution overcomes many of the problems associated with frequentist hypothesis testing.Notwithstanding the current interest in Bayesian methods, frequentist hypothesis testing remains the default method for statistical inference in medical research. Therefore, we propose an interim solution to the "significance problem" based on simplified Bayesian metrics (e.g., Bayes factor, false positive risk) that can be reported along with traditional P values and confidence intervals. We calculate these metrics for four well-known multicentre trials. We provide links to online calculators so readers can easily estimate these metrics for published trials. In this way, we hope decisions on incorporating the results of randomized trials into clinical practice can be enhanced, minimizing the chance that useful treatments are discarded or that ineffective treatments are adopted.

Reverse-Bayes methods for evidence assessment and research synthesis

It is now widely accepted that the standard inferential toolkit used by the scientific research community—null‐hypothesis significance testing (NHST)—is not fit for purpose. Yet despite the threat posed to the scientific enterprise, there is no agreement concerning alternative approaches for evidence assessment. This lack of consensus reflects long‐standing issues concerning Bayesian methods, the principal alternative to NHST. We report on recent work that builds on an approach to inference put forward over 70 years ago to address the well‐known “Problem of Priors” in Bayesian analysis, by reversing the conventional prior‐likelihood‐posterior (“forward”) use of Bayes' theorem. Such Reverse‐Bayes analysis allows priors to be deduced from the likelihood by requiring that the posterior achieve a specified level of credibility. We summarise the technical underpinning of this approach, and show how it opens up new approaches to common inferential challenges, such as assessing the credibility of scientific findings, setting them in appropriate context, estimating the probability of successful replications, and extracting more insight from NHST while reducing the risk of misinterpretation. We argue that Reverse‐Bayes methods have a key role to play in making Bayesian methods more accessible and attractive for evidence assessment and research synthesis. As a running example we consider a recently published meta‐analysis from several randomised controlled trials (RCTs) investigating the association between corticosteroids and mortality in hospitalised patients with COVID‐19.

The reproducibility of research and the misinterpretation of p-values

We wish to answer this question: If you observe a 'significant' p-value after doing a single unbiased experiment, what is the probability that your result is a false positive? The weak evidence provided by p-values between 0.01 and 0.05 is explored by exact calculations of false positive risks. When you observe p = 0.05, the odds in favour of there being a real effect (given by the likelihood ratio) are about 3 : 1. This is far weaker evidence than the odds of 19 to 1 that might, wrongly, be inferred from the p-value. And if you want to limit the false positive risk to 5%, you would have to assume that you were 87% sure that there was a real effect before the experiment was done. If you observe p = 0.001 in a well-powered experiment, it gives a likelihood ratio of almost 100 : 1 odds on there being a real effect. That would usually be regarded as conclusive. But the false positive risk would still be 8% if the prior probability of a real effect were only 0.1. And, in this case, if you wanted to achieve a false positive risk of 5% you would need to observe p = 0.00045. It is recommended that the terms 'significant' and 'non-significant' should never be used. Rather, p-values should be supplemented by specifying the prior probability that would be needed to produce a specified (e.g. 5%) false positive risk. It may also be helpful to specify the minimum false positive risk associated with the observed p-value. Despite decades of warnings, many areas of science still insist on labelling a result of p < 0.05 as 'statistically significant'. This practice must contribute to the lack of reproducibility in some areas of science. This is before you get to the many other well-known problems, like multiple comparisons, lack of randomization and p-hacking. Precise inductive inference is impossible and replication is the only way to be sure. Science is endangered by statistical misunderstanding, and by senior people who impose perverse incentives on scientists.