Sample size re-estimation (SSR) refers to the recalculation of the sample size using the existing trial data as original planned to ensure that the final statistical test achieved the pre-defined goals. SSR can enhance research efficiency, save trial costs, and accelerate the research process. Depending on whether the group assignment of the patients is known, SSR is divided into blinded sample size re-estimation and unblinded sample size re-estimation. Blinded sample size re-estimation can estimate the variance of the primary evaluation index through the EM algorithm or single sample variance re-estimation method, and then calculate the sample size. Unblinded sample size re-estimation can calculate the sample size by estimating the overall variance or therapeutic effect difference, but it needs to control the family wise type I error (FWER) rate. Cui-Hung-Wang method, conditional rejection probability method, P-value combination method, conditional error function, and promising zone are common methods used to control FWER. Currently, there are application examples of SSR methods. With the maturation of related theories and the popularization of methods, it is expected to be widely applied in clinical trials, especially in traditional Chinese medicine clinical trials in the future.
Clinical prediction models refer to models that can predict the probability of the occurrence of a certain clinical outcome event of the research objects, and they have important value in fields such as disease risk stratification, prognosis prediction, and precision medical decision - making. To further standardize this methodology, in 2024, an international multidisciplinary expert group composed of institutions from Switzerland, the Netherlands, the United Kingdom, and others, based on the TRIPOD statement and the PROBAST assessment tool, jointly released the "Step - by - step guide for developing clinical prediction models". This guide systematically constructs 13 steps: defining the objective, creating a team, conducting a literature review, developing a protocol, choosing to develop a new model or update an existing model, defining the outcome measure, identifying candidate predictors, collecting and checking data, determining the sample size, handling missing data, fitting the prediction model, evaluating the performance of the prediction model, determining the final model, performing decision curve analysis, evaluating the predictive ability of individual predictors, writing a report and publishing the results. This paper deeply analyzes the steps of this guide, aiming to provide a reference for clinical researchers.
Response-adaptive randomization (RAR) dynamically adjusts the probability of assigning patients to different groups, optimizing treatment efficacy and participant welfare. It is particularly suitable for clinical studies involving multiple interventions or dose-finding and seamless phase II/III trials. This paper systematically introduces the concept, principles, and types of RAR, as well as its application in clinical trials (including traditional Chinese medicine research). It also provides R implementation code, offering researchers practical tools aimed at promoting the adoption of RAR in clinical practice.
The covariate-adjusted response-adaptive randomisation (CARA) design combines the advantages of response-adaptive randomisation and covariate-adaptive randomisation, and improves the efficiency and reliability of clinical trials by combining analytical results and covariates and dynamically adjusting the allocation of subsequent patients. This paper describes in detail several methods of CARA design and their example applications of various methods, including the dominant confidence method, the urn model, the generalized linear model, and the Atkinson model, and provides the corresponding R codes in anticipation of a wider application of the provided R codes in clinical trials.