Thinking Beyond Null Hypothesis and P Value
SunilKumar Raina · Amrita Journal of Medicine · 2025
Dear Sir, Went through with interest the article entitled “Statistics for Doctors: Hypothesis Testing and P value” published in Amrita Journal of Medicine with a focus on the role of statistics in research in medical sciences.[1] The authors need to be complimented for their efforts in laying emphases on the role of hypothesis testing and P value.[1] Although there is a wider agreement on the null hypothesis (often denoted as H0) and the statistical methods associated with, being fundamental to scientific research, including medical science, their application in medical science has limitations that tend to get ignored. The reason for agreement could primarily be because of an overwhelmingly strong affinity among medical researchers for Frequentist methods as the preferred methods of evaluating medical research, including but not limited to just hypothesis testing and P value. Medicine and public health deal with complex, multifaceted systems which are limited not just to the human body but also to the individuals as part of an extended social and physical environment interacting and influencing health outcomes, and therefore limiting the research conduct and evaluation into a simple “null” versus “alternative” hypothesis structure may not be appropriate. Although things are much more complex at the population level, even at the individual level, the diseases, treatments, and outcomes may not always have clear-cut, dichotomous outcomes. For example, in a clinical trial, a treatment might not just show a yes/no effect but could show a range of benefits or harms, which may get unnoticed in a “null” versus “alternative” hypothesis structure. The null hypothesis often focuses on “statistical significance,” typically relying on P value (e.g., <0.05), which essentially is arbitrary only, to determine whether an effect exists. But more importantly, statistical significance does not always translate to clinical significance. A statistically significant finding might not always have a meaningful impact on patient care or health outcomes. In identifying a cause for disease and an intervention, the null hypothesis relies largely on sample size, which again is derived from calculations arrived on from an observation of previous studies. If the sample size is small or the design is flawed, accepting the null hypothesis in these cases will endanger potentially beneficial treatments being incorrectly dismissed, leading to a slower pace of medical advancement. Significantly also, in medical research, there are often many variables to consider, such as treatment dosage, patient demographics, and comorbidities. Using the null hypothesis to test multiple hypotheses increases the risk of Type I errors (false positives), especially when corrections for multiple comparisons are not applied, leading to spurious findings. But that does not mean that there is no way around. Medical research is increasingly using other statistical methods that focus on more nuanced, probabilistic, or predictive models rather than simple null hypothesis testing. Bayesian statistics, for example, allows for updating probabilities as new evidence accumulates, providing a more flexible framework for decision-making. This approach is particularly useful in medicine, where patient outcomes can vary widely. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.