گزیده خلاصه مقالات اپیدمیولوژی و آمار-شماره 14

Routine data are not 'dirty' data: how to approach using routine data for research

1. BMJ Health Care Inform. 2026 May 29;33(1):e102018. doi: 10.1136/bmjhci-2025-102018. The rise of electronic patient records has led to an abundance of routine healthcare data, and the development of NHS Secure Data Environments (SDEs) is transforming their potential for research. However, routine data differ fundamentally from research data; they are generated to support care processes rather than answer predefined research questions. This creates challenges when they are analysed without understanding the context in which they were produced.Using real examples from practice, this paper illustrates the variety of ways routine data can be shaped and how this leads to misinterpretation if treated as merely 'dirty' research data. These examples underpin practical strategies for researchers, including analysis at or by source and working closely with local data providers to understand how a dataset was created.We argue that SDEs must foster strong links between data users and creators to preserve data provenance and support accurate interpretation. By approaching routine data as meaningful records of healthcare processes-rather than flawed research datasets-researchers can fully realise their potential to drive healthcare improvement.


Practical statistics for bioimage analysis - a guide to experimental design and data interpretation

Bioimage analysis is a powerful tool for investigating complex biological processes, but its robustness depends on technical precision and rigorous experimental design. In particular, the use of appropriate controls and experimental repetition is critical for drawing meaningful conclusions. However, there are times when both are inadequately applied or overlooked in favour of 'statistical significance', often derived from misused or misinterpreted statistical tests. In this Perspective, we reanalyse publicly available image datasets to highlight the crucial role of robust experimental design in interpreting results. Our findings underscore the importance of focusing on effect sizes and biological relevance over arbitrary statistical thresholds. We also discuss the diminishing returns of increased data collection once statistical stability has been achieved. By refining control usage and emphasising effect sizes, this Perspective offers guidance to enhance the reproducibility and robustness of research findings. We provide open access code to allow researchers to engage with the dataset, promoting better practices in experimental design and data interpretation.


Understanding regression models in clinical research

Medical research often employs regression models. However, many clinicians are not confident in their ability to interpret these studies, which can lead to misinterpretation and negatively affect clinical practice. This article aims to provide a basic framework for understanding regression models in clinical research. In this article, we explain what regression models are, the purposes they serve, and some of the most common regression models, with examples illustrating how to interpret them in plain language. We discuss common errors and shortcomings in regression models.


The history of statistics in statistically valid regulation

This paper uses the introduction at the US Food and Drug Administration of dose-response extrapolation for ascertaining toxicity between 1950 and 1980 to analyse the negotiations between statistical knowledge and regulation. Those statistical methods enabled experimental results to be translated into a 'safe dose' of a substance for human consumption, but different methods continued to give different estimates of effects, and there was little basis for determining which methods were most accurate. I argue that statisticians were not proposing their discipline so much as a tool for mechanical decision making than as a set of methods for establishing a regulatory procedure that made assumptions and judgements visible. Consequently, their use did not bring debates about low-dose toxicity to a close even as they promoted a regulatory ethos and enabled regulators to act, even in cases of inherent uncertainty and inescapable variability.


The win ratio in contemporary clinical trials: growing adoption and interpretation challenges-a meta-epidemiological study

Objectives: The win ratio (WR) has been used with increasing frequency in clinical trials with hierarchical composite end points. This meta-epidemiological study aims to systematically evaluate the contemporary use of the WR in clinical trials. Specifically, we sought to (1) describe temporal trends in WR application, (2) quantify its use across therapeutic areas, and (3) examine component directional contributions and conclusion consistency between analyses summarized by the WR and the original analyses. Methods: We conducted a systematic review (PRISMA 2020) of PubMed, Embase, and Web of Science (search date: 26 March 2025). Eligible studies were published clinical trials or reanalyses that summarized treatment effects using the WR based on individual data. Studies were classified by application type (prespecified vs reanalysis) and mapped to International Classification of Diseases (ICD-10) categories. After inclusion, we further screened for studies with comparable data, then performed two comparisons: first, the component directional contribution between end point component and the overall WR result; and second, the conclusion-level consistency between each WR-reported analyses and the conclusions of the original trials. Results: A total of 90 studies were included: 35 prespecified end point analyses and 55 reanalyses. Circulatory system trials accounted for about 80% of the applications. We included 57 studies for comparison, and discrepancies between component-level results and overall conclusions were common. In prespecified end point analyses, 36 of 139 components (25.9%) had directional discordance or neutrality with the overall WR point estimate, compared with 17.0% in reanalyses, although the difference was not statistically significant (generalized linear mixed-effects model with study as a random effect, P = .114). Among the 35 studies included for conclusion-level comparison, 22 reached WR-reported conclusions that were consistent with those of the original trials. Nearly all observed discrepancies in conclusions were associated with changes in the selection or prioritization of end point components. Conclusion: The use of the WR in clinical research has steadily increased and is expanding beyond cardiovascular trials. Hierarchical composite end points constructed using the WR may differ from conventional composites in component selection, and directional discordance among components is not uncommon. In WR-reported reanalyses, changes to end point components may influence clinical interpretation. Careful attention to the appropriate use of the analytical framework, end point definition, and result interpretation is essential to ensure the interpretability and validity of WR-reported analyses.


The meaning of "p-value": Pursuing clinical and statistical significance. A narrative review

The evolution of statistical thinking within the scientific community has developed over nearly three centuries. The concept of "probability," commonly represented by the p-value, has been applied in medical research for more than a century. Its integration into hypothesis testing has played an important role in advancing medical science. Nevertheless, the p-value has also become one of the most misunderstood and misapplied statistical measures. The common practice of rigidly categorizing the probability value as being above or below the 0.05 threshold has often taken on a dogmatic quality, at times resembling a doctrine rather than a purely scientific criterion. In reporting experimental results and conducting hypothesis testing, references to "statistical significance" are frequently equated with "truth", while findings deemed "not statistically significant" are mistakenly interpreted as "false". This is a conflation and it is incorrect. The objective of this article is to provide a concise review of the history of statistics in medicine, clarify the development of the "probability value" concept, discuss its role in hypothesis testing, and consider its interpretation alongside clinical significance.


The Placebo Effect in Rare Disease Clinical Trials: Measurement, Impact, and Statistical Approaches for Patient-as-Own-Control Designs

7. Clin Transl Sci. 2026 May;19(5):e70550. doi: 10.1111/cts.70550. A frequently cited concern regarding patient-as-own-control trial designs in rare disease is the potential for placebo and related effects to inflate apparent treatment efficacy. Whether this concern is disqualifying or manageable has not been systematically evaluated. We reviewed meta-analyses quantifying placebo effect magnitude by endpoint type, reporter modality, and trial duration and evaluated statistical methods available for post-trial placebo adjustment in own-control designs. Placebo effects depend heavily on endpoint type. For objective endpoints (enzyme activity, serum biomarkers, imaging volumetrics)-which constitute the majority of primary endpoints in approved rare disease therapies-placebo effects are consistently small and in most meta-analyses statistically indistinguishable from zero (standardized mean difference [SMD] < 0.10). For subjective endpoints (patient-reported pain, caregiver-rated function), effects are larger (SMD 0.20-0.50) but well-characterized and correctable. Placebo effects peak early and decay over weeks, providing a temporal signature distinguishable from sustained pharmacological effects. Multiple complementary analytical methods-including temporal trajectory modeling, objective-subjective concordance analysis, Bayesian informative priors, extended run-in observation designs, and blinded outcome assessment-are available to quantify and manage placebo contributions. Importantly, the randomized controlled trial's structural advantage in canceling placebo is itself degraded in small samples, where asymmetric allocation of placebo responders can distort between-arm comparisons. The placebo objection to own-control designs is manageable rather than disqualifying. For objective endpoints, correction is minimal. For subjective endpoints, a rich analytical toolkit supports credible decomposition of drug and placebo components. These findings support the broader adoption of own-control designs in rare disease clinical trials.


Goodbye P<0.05. P-value is simply one item among many to gauge scientific evidence

For a century P-value is routinely used in almost every research paper with a threshold of 0.05 to reject the null hypothesis. The aim of this short review was to discuss the validity of this arbitrary (yet sacred) threshold. The history of P-value shows that very quickly, practitioners had found a simple method that appealed to them, while statisticians saw no great need to curb this enthusiasm, which seemed consensual. However, heavy reliance on P-values is of concern because of potential misuse and misinterpretation. The main pitfalls of P<0.05 are the dichotomized approach with a black-or-white judgement, possible false positive results, lack of information about magnitude of the effect, clinical relevance, and use out of context. These pitfalls explain why several statisticians and researcher recommend abandoning, not the P-value itself but the threshold of 0.05 and the term "statistical significance". We are faced with a paradigm shift by demoting P-value from its threshold-screening role and using alternative tools such as Bayesian methods, effect size with confidence intervals, more stringent thresholds, pragmatic trials, and the minimal clinically important difference. This will require statisticians, researchers, publishers, and health care decision makers to radically change the way they interpret scientific data by abandoning century-old dichotomous analysis-a true revolution to come.


Supply and Demand in the Mathematics of Rare Disease Drug Development: Why Choosing the Right Model Is Crucial

Clinical trials for rare diseases face a fundamental mathematical challenge that conventional randomized controlled trial (RCT) designs cannot overcome. With approximately 95% of the estimated 10,000-16,000 rare diseases lacking approved therapies, and drug development programs failing at rates exceeding 75% in non-oncology indications, the field confronts a stark reality: Traditional trial designs demand patient numbers that simply do not exist. This perspective article examines the critical mismatch between the statistical requirements of different trial designs (the "demand") and the actual patient populations available for study (the "supply"). We demonstrate mathematically that alternative trial designs-particularly patient-as-own-control and natural history comparator models-can reduce required sample sizes by 5- to 20-fold while maintaining statistical rigor. We further point out that a substantial proportion of rare disease trial failures stem not from therapeutic inefficacy but from recruitment and retention challenges inherent to underpowered RCT designs-challenges that are directly addressable through appropriately matched trial design. Given that most rare disease development programs receive only one opportunity to demonstrate efficacy, the continued application of inappropriate statistical models represents both a scientific failure and an ethical and economic challenge to the rare disease community. We propose that regulatory agencies formalize acceptance of alternative trial designs for rare diseases, supported by explicit mathematical frameworks that transparently account for genetic heterogeneity, pediatric populations, and the statistical efficiency gains achieved through within-subject correlation.


Non-linear regression modelling for medical professionals; making curved paths straightforward

Regression models provide information on complex relationships between patient factors, investigations, diagnoses, treatments, and outcomes. These inferences underpin evidence-based medicine. However, by default regression models assume straight-line relationships, and the common approach of splitting continuous variables into groups has several disadvantages. We discuss pitfalls with these common approaches, and provide an interactive regression model playground, which acts as a point-and-click showcase of these concepts. More flexible regression modelling techniques are available, which allow non-linear relationships between predictors and outcome to be captured. However, they have been shown to be underused in medical research. We feel a major contributor to this is that more flexible non-linear models are typically explained for a statistical audience, creating a barrier for medical professionals. In this article, we introduce non-linear regression for medical researchers. Specifically, we focus on restricted cubic splines (RCS), which allow curved relationships to be fit, within the context of regression models. This has the benefit that the overall structure of the regression model and its outputs, which are familiar to medical researchers, stays the same, with the simple addition of non-linear modelling of specific variables. We implement RCS in a case study, with accompanying example R scripts (available on GitHub). We also launch an R package ("rmsMD") which aims to make this technique approachable to medical researchers, as well as creating publication-ready tables and plots. Overall, this article equips medical researchers with an intuitive understanding of non-linear modelling, which can then be applied with the easy-to-use tools provided.


Statistical Inference in Database Exploitation: Sample Size and Confounding Factors

Database (DB) exploitation is an essential tool in medical and epidemiological research, enabling the extraction of hidden insights from large volumes of information through statistical analysis. This work provides a methodological reflection on two critical aspects of such studies, namely sample size and confounder detection via multivariate analysis. When working with databases, researchers must perform statistical inference based exclusively on the available data and therefore often ask whether the fixed sample size is sufficient to detect relationships of interest. In most cases, no threshold value separates valid sizes from invalid ones, as statistical power increases gradually and depends on multiple parameters, with no fixed cutoff points. Even when the results are inconclusive (high p-values), possibly because of limited sample size, their publication is essential to feed future meta-analyses that may provide more solid conclusions. DB analysis requires discriminating between general and circumstantial associations. Although the relationships revealed by univariate analysis maintain their descriptive value, the accuracy of their interpretation can be increased by identifying possible confounding factors.