Biostatistics for academic and pragmatic clinical trials

Yes. Biostatistical support pays off most when it comes in at the protocol stage, but arriving with the data already collected does not rule out a defensible analysis, as long as it is made explicit which decisions were specified in advance, which are set before analysing and which part should be interpreted as exploratory. This applies both to a trial that has not yet started, where the study is sized and the analysis plan written before recruitment, and to one that has closed, where the analysis plan is set before touching the data. What protects the validity of the primary analysis is pre-specifying the strategy before looking at the results, not having been involved from day one. If the results have already been explored, that must be documented and kept separate from the primary analysis that can be defended as pre-specified. When the trial arrives with the data collected, the work centres on a well-grounded plan, sound handling of missing data and the right sensitivity analyses.

The service covers the statistical cycle of the trial within the academic, pragmatic or implementation setting: sample size calculation, writing the statistical analysis plan (SAP), analysing the efficacy and safety endpoints, and preparing the results for publication or the end-of-study report. The client designs the intervention, runs the trial and collects the data; the service sizes, specifies, analyses and interprets. It does not include patient recruitment, clinical monitoring or the operational management of the trial, which belong to the research team or a CRO. What you receive is the statistical foundation of the study, from the sample size argument to the written methods section, plus the sensitivity analyses that help answer methodological objections from reviewers and committees.

The sample size is justified from the clinically relevant effect, the desired statistical power and the expected dropout rates, not as an isolated number but as a documented argument. This applies at the design stage, before recruitment, and it determines the statistical feasibility of the whole study. The calculation starts from the effect size the trial must be able to detect, set on clinical rather than statistical grounds, and is adjusted to the specific design: parallel, factorial or cluster-randomised, each of which calls for different considerations. Depending on the design, the significance level, the allocation between groups, the intracluster correlation or the longitudinal structure may also come into play. The full reasoning is documented so that a funding committee or a reviewer can see how the figure was reached. A sample size without its justification is the first decision a reviewer questions.

The statistical analysis plan (SAP) is the document that defines, before any analysis, a trial's analysis populations, endpoints, missing-data strategy and multiplicity adjustment. It is worth setting before seeing the data because pre-specification is one of the conditions that separate a confirmatory analysis from an exploratory one: choosing the method after the results are known opens the door to picking, consciously or not, the option that gives the desired result. The SAP is written following the ICH E9 structure as its methodological reference, including, where relevant, the intention-to-treat and per-protocol populations. In a trial that has already closed, setting the SAP before touching the data protects the validity of the primary analysis just the same.

No. The method is adapted to the actual design of the study and to the structure of the data, not to a default template. A cluster-randomised trial is not analysed like a parallel one, because ignoring the correlation within clusters underestimates the uncertainty; repeated measures are not treated like cross-sectional data, because observations from the same patient are not independent. For these cases the options include mixed-effects models for longitudinal data, intention-to-treat and per-protocol analyses, and multiple imputation when the missing-data structure requires it, among others depending on the endpoint and the design. Pragmatic and implementation trials, with the variability of real-world settings, are precisely where a generic template can produce inferences that are hard to defend, and where fitting the analysis to the specific design matters most.

Missing data are handled with a strategy defined in the SAP and backed by sensitivity analyses under different assumptions. This applies to most trials in practice, where dropout, missed visits or loss to follow-up have to be anticipated in the SAP. The strategy starts by characterising the plausible missing-data mechanism and choosing the method consistent with that assumption: multiple imputation is common, but it is not the only route. The key piece is the sensitivity analysis, which checks whether the conclusions hold under alternative assumptions about the missing data. If the result only holds under the most favourable assumption, that fragility needs to be known before peer review, not discovered during it. This is what makes it possible to tell a stable conclusion from one that depends too heavily on a favourable assumption.

Biostatistics for academic and pragmatic clinical trials covers the statistical cycle specific to the trial, including the analysis of the data once collected, and can include design decisions when the project comes in from the start. Study design and statistical protocols apply before data are collected, when the decision is how to set up the study: type of design, hypotheses, variables, measurement plan. They can be two phases of the same project or two different entry points: if you have not started the study yet, you mainly need design; if your trial is already under way or closed, you need the analysis plan and the analysis itself. They do not answer the same question: one helps you set up the study properly; the other helps you analyse it and defend it rigorously.

No. The service works with academic, pragmatic and implementation clinical trials, and prepares the statistical evidence for scientific publication, end-of-study reports and its defence in front of reviewers and funding or oversight committees. It does not cover preparing or managing a formal regulatory submission to agencies such as the EMA or the FDA, nor producing datasets and documentation to agency standards (CDISC, SDTM/ADaM), nor the tables, listings and figures (TLFs) required in a regulatory dossier. That is a different regulatory competence, with its own framework, and it lies outside ADI's scope. The reference to the ICH E9 structure in the analysis plan is good methodological practice in academic and pragmatic research, not a capability for submission to an agency. The line does not blur: academic and pragmatic trials, yes; a formal regulatory dossier, no.