Contents
Academy · Research Data Analytics & AI · Biostatistics in clinical trials
Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan
Biostatistics shapes a clinical trial from its design and sample size to its analysis and reporting. This article explains the statistician’s role across the trial, then examines the step that most affects trust in the result: fixing the statistical analysis plan before the data are unblinded.
Biostatistics in a clinical trial is the planning, analysis and reporting of trial data so that the result answers the question the trial set out to ask, and can be trusted.
The statistician’s work runs through every phase of a trial: the design, the sample size, the analysis and the report. This article first outlines that role across the four phases of clinical development, then examines the statistical analysis plan (SAP), the document that fixes in advance how the data will be analysed. Its value depends almost entirely on timing: a plan fixed while everyone is still blind to the results protects the trial from analysis choices made to favour a result; a plan written afterwards protects nothing.
Biostatistics in clinical trials at a glance
- The statistician’s role
- Design, sample size, randomisation, interim monitoring, analysis and reporting1
- Analysis plan
- Required by ICH E9 and SPIRIT; a 55-item guideline sets out its content (Gamble et al., 2017)12
- Must be signed off
- Before database lock and unblinding; the primary analysis ideally before recruitment8
- Why it matters
- 62% of trials in one cohort changed, added or dropped a primary outcome3
- Interim looks
- Reviewed by an independent data monitoring committee, so the trial team stays blind10
- On this page
- 7 figures, 5 tables, a weak versus strong example and a sign-off checklist
What biostatistics does in a clinical trial
Biostatistics is the application of statistical methods to medicine and biology. In a clinical trial, the statistician is involved from the first protocol draft to the final report: choosing the design, working out how many participants are needed, deciding in advance how the data will be analysed, supporting interim monitoring, and reporting the results completely. ICH E9 sets out the statistical principles regulators expect at each of these stages1.
What the trial statistician does at each stage
| Stage | Statistical tasks | Why it matters |
|---|---|---|
| Design | Choose the design, randomisation and blinding; define the primary endpoint; calculate the sample size | Decides whether the trial can answer its question at all |
| Conduct | Monitor data quality; prepare closed reports for interim analyses by the data monitoring committee | Protects participants without unblinding the trial team |
| Analysis | Write and sign off the statistical analysis plan; analyse by intention to treat; handle missing data as planned | Prevents analysis choices from being driven by the results |
| Reporting | Report all pre-specified outcomes and any deviations, following CONSORT | Lets readers judge whether the result is trustworthy |
Principles from ICH E9 (Statistical Principles for Clinical Trials); the analysis and reporting stages are covered in detail below.
Even the first of these tasks is often done imperfectly. A review of trials in leading general medical journals found frequent gaps and errors in the sample size calculation that determines how many participants a trial recruits11.
Why the rest of this article focuses on the analysis plan
Design and sample size decide whether a trial can answer its question; the analysis plan decides whether readers can trust the answer it gives. Of all the statistician’s tasks, fixing that plan before unblinding is the one most often done late or incompletely, so the sections below examine it in detail.
What the evidence shows about planned and published analyses
Several cohort studies have compared what trials planned with what they published. The pattern is consistent: a large share of trials change, add or drop outcomes or analyses between plan and publication, and the changes lean towards statistically significant results.
Four studies of the gap between plan and publication
| Study | What was compared | Key finding |
|---|---|---|
| Chan et al. 2004, JAMA | Protocols of 102 Danish trials with their 122 publications | 62% changed, introduced or omitted a primary outcome; statistically significant outcomes were more likely to be fully reported (odds ratio 2.4 for efficacy, 4.7 for harm) |
| Mathieu et al. 2009, JAMA | Registry entries of 323 trials with their publications | Only 147 (45.5%) were adequately registered; of those, 31% showed discrepancies, and where the effect could be judged, 82.6% favoured significant results |
| Goldacre et al. 2019, Trials (COMPare) | 67 trials in NEJM, Lancet, JAMA, BMJ and Annals | Mean 76% of primary and 55% of secondary outcomes correctly reported; 5.4 undeclared outcomes added per trial; 87% needed a correction letter |
| Kahan et al. 2020, J Clin Epidemiol | 100 trials published in 2018 | Only 12 had a public pre-specified primary analysis (2 dated before recruitment); 11 of those 12 had unexplained discrepancies |
Figures as reported in each study’s abstract.
The changes are not random
When outcomes change, they tend to change towards significance. Chan and colleagues found significant efficacy outcomes had more than twice the odds of being fully reported, and 86% of surveyed trialists denied unreported outcomes existed despite clear evidence3. Mathieu and colleagues found that discrepancies favoured significant results in 19 of 23 assessable cases4. An analysis plan fixed before unblinding is what makes this kind of drift detectable.
Content pre-specified in a statistical analysis plan and the importance of timing
ICH E9 (International Council for Harmonisation, 1998) expects every planned analysis — the statistical test for the primary endpoint, the handling of missing data, the subgroups to be examined, and the distinction between planned and exploratory comparisons — to be fixed in writing before the data are unblinded. After unblinding, any change to the plan is, by definition, informed by knowledge of the result, which is precisely what pre-specification is intended to prevent. See the fuller overview of a trial statistician’s role for where the SAP sits relative to design and final reporting.
In practice, four aspects of the primary analysis decide whether a plan really closes off post-hoc choices. A 2017 consensus guideline sets out 55 items an SAP should address2; the four below are the ones most often left vague.
What the primary analysis must pin down, and how often protocols do
| Aspect | What to specify | Adequately specified in 99 published protocols |
|---|---|---|
| Analysis population | Who is analysed: intention-to-treat, modified ITT or per-protocol, with exact definitions | 8% |
| Analysis model | The statistical model and how it is implemented (for example, model type, correlation structure, estimation method) | 27% |
| Covariates | Which baseline variables are adjusted for, decided in advance, not by significance | 40% |
| Missing data | How missing outcomes are handled, in enough detail to reproduce (for example, imputation model and number of imputations) | 10% |
Aspects from the Pre-SPEC framework; percentages from Greenberg et al. (2018), who also found that no protocol adequately pre-specified all four.
Designing a plan that cannot be p-hacked
Pre-specification only works if it leaves nothing to choose after seeing the data. Kahan, Forbes and Cro’s Pre-SPEC framework sets out five points for achieving that, consistent with SPIRIT and ICH E98.
Pre-specified, but still open to p-hacking
“Stratification variables will be evaluated and retained in analyses where they are measured as significant or quasi-significant. Transformation of outcomes, including categorisation, may be undertaken to meet distributional assumptions…”
Missing data are not addressed, the mixed model is not specified closely enough to replicate, and two choices are left to judgement after seeing the data.
Fully pre-specified
Include everyone with at least one follow-up outcome; fit a linear mixed-effects model with an unstructured correlation matrix, estimated by restricted maximum likelihood; adjust for all stratification variables regardless of significance; do not transform the outcome; attach the planned statistical code.
Both versions are from Kahan, Forbes and Cro’s worked example: the left is a quoted protocol paragraph, the right summarises their rewrite.8
Interim monitoring while preserving the blind
A pre-planned interim analysis of accumulating trial data, usually conducted to check for an overwhelming safety signal or an already clear result, is reviewed by a Data Safety Monitoring Board (DSMB) using statistical output prepared specifically for that purpose, without unblinding the wider trial team or the sponsor. This structure ensures that knowledge of early data cannot influence how the remainder of the trial is conducted or analysed.
Who sees what while the trial is running
| Group | What they see | Why |
|---|---|---|
| Trial team and sponsor | Pooled, blinded data and open reports (recruitment, data quality) | Decisions about conduct must not be influenced by emerging results |
| Independent statistician | Unblinded data, to prepare closed reports | Someone must analyse by arm for the committee without informing the trial team |
| Data monitoring committee | Open and closed reports, including results by arm | Judges safety and whether pre-set stopping rules have been met |
| Trial statistician writing the SAP | Blinded data only, until sign-off and database lock | Keeps the plan independent of the results |
A typical arrangement; details vary by trial and should be fixed in the committee’s charter, as the DAMOCLES group recommends.
Interim analyses raise their own statistical problem: looking at the data repeatedly increases the chance of a false positive. Group sequential methods solve this by fixing the number of looks and a stopping boundary in advance9.
Sign-off checklist
Before the plan is signed off
- The primary analysis matches the protocol and registry entry, and was fixed before recruitment.
- One primary analysis is named; every other analysis is labelled sensitivity, supplementary or exploratory.
- Population, model, covariates and missing-data handling are each specified in enough detail to reproduce.
- Any data-dependent choice has a deterministic rule written down in advance.
- Interim analyses, stopping boundaries and the data monitoring committee’s role are described.
- The plan has been checked against the 55-item SAP content guideline.2
- The date of sign-off, and the blinding status of the statistician at that date, are recorded for the final report.
The point to take away
References
- International Council for Harmonisation. "ICH E9: Statistical Principles for Clinical Trials." read 8 October 2026. ICH, 1998 (FDA guidance for industry)
- Gamble C, Krishan A, Stocken D, et al. Guidelines for the content of statistical analysis plans in clinical trials. JAMA. 2017;318(23):2337-2343. read 13 November 2026. View via DOI
- Chan AW, Hróbjartsson A, Haahr MT, Gøtzsche PC, Altman DG. Empirical evidence for selective reporting of outcomes in randomized trials: comparison of protocols to published articles. JAMA. 2004;291(20):2457-2465. read 13 November 2026. View via DOI
- Mathieu S, Boutron I, Moher D, Altman DG, Ravaud P. Comparison of registered and published primary outcomes in randomized controlled trials. JAMA. 2009;302(9):977-984. read 13 November 2026. View via DOI
- Goldacre B, Drysdale H, Dale A, et al. COMPare: a prospective cohort study correcting and monitoring 58 misreported trials in real time. Trials. 2019;20(1):118. read 13 November 2026. View via DOI
- Greenberg L, Jairath V, Pearse R, Kahan BC. Pre-specification of statistical analysis approaches in published clinical trial protocols was inadequate. Journal of Clinical Epidemiology. 2018;101:53-60. read 13 November 2026. View via DOI
- Kahan BC, Ahmad T, Forbes G, Cro S. Public availability and adherence to prespecified statistical analysis approaches was low in published randomized trials. Journal of Clinical Epidemiology. 2020;128:29-34. read 13 November 2026. View via DOI
- Kahan BC, Forbes G, Cro S. How to design a pre-specified statistical analysis approach to limit p-hacking in clinical trials: the Pre-SPEC framework. BMC Medicine. 2020;18(1):253. read 13 November 2026. View via DOI
- O’Brien PC, Fleming TR. A multiple testing procedure for clinical trials. Biometrics. 1979;35(3):549-556. read 13 November 2026. View on PubMed
- DAMOCLES Study Group. A proposed charter for clinical trial data monitoring committees: helping them to do their job well. Lancet. 2005;365(9460):711-722. read 13 November 2026. View via DOI
- Charles P, Giraudeau B, Dechartres A, Baron G, Ravaud P. Reporting of sample size calculation in randomised controlled trials: review. BMJ. 2009;338:b1732. read 13 November 2026. View via DOI
Each source carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 8). Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan. Pubrica. https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/
Pubrica Academy. “Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan.” Pubrica, 8 Oct. 2026, pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/.
Pubrica Academy (2026) Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan. Pubrica. Available at: https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/ (Accessed: 15 October 2026).
Pubrica Academy. “Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan.” Pubrica. October 8, 2026. https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/
Pubrica Academy, ‘Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan’ (Pubrica, 8 October 2026) <https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/> accessed 15 October 2026.
Pubrica Academy. Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan. [Internet]. Pubrica; 2026 Oct 8 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/
Pubrica Academy, “Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan,” Pubrica, Oct. 8, 2026. [Online]. Available: https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/
@misc{on_biostatistics_and_clinical_trials_2026,
author = {{Pubrica Academy}},
title = {Biostatistics in Clinical Trials: Roles, Methods and the Statistical Analysis Plan},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/statistical/on-biostatistics-and-clinical-trials/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica assists in finalising a statistical analysis plan and ensuring that it is dated and version-controlled before unblinding, so that pre-specification can be demonstrated rather than merely asserted.