Abstract
Medical literature has historically focused on reporting p values and sharply contrasting results.1 However, notable steps have been taken to understand uncertainty and some researchers have suggested abandoning p values.2 Uncertainty is typically estimated by confidence intervals in frequentist statistics or by credible intervals, using the Bayesian approach. The benefits include the range of the possible results, as confidence intervals typically indicate whether results represent the true effect. Furthermore, the width of the interval indicates how likely it is that the effect estimate is correct. This provides additional value, instead of just giving dichotomised interpretation based on p-values, whether it is statistically significant or not.3 Table 1 presents an example, based on two hypothetical scenarios with identical absolute risks, to illustrate this point. This shows that reporting practices make a notable difference and that providing the absolute rates, relative rates and uncertainty estimates makes it easier for readers. However, I have noticed variations in reporting practices in most medical journals, including Acta Paediatrica. I searched the Web of Science database on 31 January 2023 for papers published in Acta Paediatrica in 2022 that reflected current result-reporting practices. This identified 157 full-length original studies. Other kinds of papers were excluded, such as brief reports, editorials and review. Five case series that had less than 30 participants were excluded, as they were unlikely to have enough power to conduct reasonable statistical testing. The papers were assessed using Covidence software (Veritas Health Innovation). The main outcome measure was using reporting practices to describe the differences or values of the main outcomes. I also extracted the use of confidence intervals and p values in the results sections. No ethical approval was required for this study. The 157 results were evaluated and 121 were included in the analysis. I found that 83% reported p values and 70% reported confidence intervals in the full-text results sections or tables. Both were reported in 58% of the papers and neither in 4%. p values were exclusively used in 26% and confidence intervals were exclusively used in 12%. The use of p values was more common than confidence intervals in the journal in 2022 and the exclusive use of p values was notably more frequent in original research papers. p value reporting is still more common than confidence interval reporting. As explained earlier, confidence intervals provide clear benefits over p values. But it should be noted that, although confidence intervals provide uncertainty estimates, the tendency for dichotomous interpretation of these still exists. It is simple to decide whether an intervention is effective or not by examining the overlap of one or zero, depending on the method used. However, this should not be the case, as the confidence intervals can be interpreted based on whether they include an effect that is compatible with either an increase or decrease in the outcome.4 For example, if an intervention has an observed risk ratio of 0.90 and the confidence interval is 0.50 to 1.30. It can be interpreted that the range of effect estimate includes both increase and decrease in the risks. However, in a similar case with risk ratio of 0.90, if the confidence intervals are 0.79 to 1.01, the effect sizes include a clear decrease, but unlikely increase. In this case, the finding should be interpreted as being more in favour of the intervention. If only p values had been used, this interpretation of the results would be nearly impossible. Previous studies have identified that authors are more likely to only present p values for so-called significant findings.5 A possible way to improve the reporting practice in medical journals has been the use of statistical editors, although it could be achieved by strict general editorial policies. Furthermore, the use of reporting checklists has been shown to improve reporting quality and unify reporting practices.6 However, more education for researchers and other parties, such as journals, editors, funders and universities, is the most desired goal when it comes to improving reporting practices in medical science. The aim of research should be to provide methodologically relevant papers, instead of focusing on statistical significance and impact.7 A clear limitation of this review is that it was only conducted by one author, which increases the likelihood of unintended mistakes in the extraction process. Furthermore, there are no uncertainty estimates in this report. This may seem ironic, but as this is just a descriptive report, these were left out. A further limitation was the lack of comparison to different journals or historical volumes of Acta Paediatrica, but the aim was to describe current reporting practices. Uncertainty is one of the key concepts in current medical reporting. Journals can decide and guide authors to embrace uncertainty more in reporting. Acta Paediatrica could achieve this by emphasising that authors should always provide uncertainty estimates in manuscripts. None. None.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have