28 The Clinical Study Report
When a trial ends, everything it produced has to be distilled into one document: the Clinical Study Report, or CSR, which accounts for what was planned, what was done, and what was found. A regulatory reviewer reads it with a critical eye, looking for any evidence of bias, error, or inadequate methodology, and has to decide from it whether the trial was conducted properly and whether the data are reliable enough to support the conclusions drawn from them. On that judgment rests the approval of a new therapy.
Regulatory reviewers are not the only readers. Inside the sponsor company the CSR sets development strategy: a positive report may trigger investment in further development, a negative one may end the program. The medical and scientific community may eventually see it, or publications based on it, so it has to meet standards of scientific rigor and transparency. Nor does it expire. Preserved as part of the permanent record, a CSR may be examined years or decades later in response to safety questions, litigation, or regulatory inquiries.
28.1 The ICH E3 Structure
CSR structure follows international guidelines, primarily ICH E3, which establishes a standardized format for presenting clinical study results. Table 28.1 shows the major sections, with indicative lengths. Note the proportions. The conclusions run one to two pages; the appendices behind them, which carry the protocol, the statistical analysis plan, the patient data listings, and other supporting materials, can run past a thousand.
| Section | Content | Typical Length | Key Elements |
|---|---|---|---|
| Title Page/Synopsis | Summary of entire study | 2-3 pages | Design, key results, conclusions |
| Introduction | Background and rationale | 2-5 pages | Disease context, drug mechanism, program fit |
| Study Objectives | What trial was designed to determine | 1-2 pages | Primary, secondary, exploratory objectives |
| Investigational Plan | Study design details | 10-30 pages | Randomization, blinding, dosing, procedures |
| Study Patients | Who enrolled | 10-20 pages | Disposition, demographics, baseline characteristics |
| Efficacy Results | Primary and secondary analyses | 30-100 pages | Endpoint analyses, subgroups, sensitivity |
| Safety Results | AEs, labs, vitals | 20-50 pages | AE tables, SAE narratives, lab shifts |
| Discussion | Interpretation of results | 5-10 pages | Clinical significance, limitations, context |
| Conclusions | Key findings | 1-2 pages | Summary statements |
| Appendices | Supporting materials | 100-1000+ pages | Protocol, SAP, TLFs, patient listings |
The opening sections orient the reader. A title page and synopsis condense the whole study into two or three pages for anyone who needs a quick overview. The introduction supplies context, and study objectives specify exactly what the trial was designed to determine, with efficacy questions usually primary and safety, pharmacokinetics, or other endpoints secondary. The investigational plan then describes the design in detail: the type of trial (randomized, controlled, blinded), the treatment groups, the dosing regimen, the duration of treatment and follow-up, and the inclusion and exclusion criteria that defined the population on paper.
The study patients section reports who actually enrolled: how many were screened, randomized, completed, and discontinued, with reasons. Demographic and baseline characteristics follow, and their job is to establish that the population the trial enrolled is the population it intended to enroll. Efficacy results then present the primary and secondary endpoint analyses together with the sensitivity and subgroup analyses that stress them; figures and tables display the data, and the text interprets it. Safety results present adverse events, serious adverse events, laboratory abnormalities, vital signs, and any other safety parameters collected, tabulated by treatment group, with notable events discussed in narrative form. The discussion interprets all of it for the investigational product, the disease, and the development program; the conclusions summarize. Everything else is appendix.
28.2 The Writing Process
A CSR takes several months and many hands. Medical writers sit at the center, synthesizing the locked database, the statistical analysis plan (SAP), and the tables, listings, and figures (TLFs) generated by the biostatistics team. Clinical operations contributes context on trial execution, site audits, and the operational challenges that may have affected the data. Medical and scientific reviewers keep the clinical narrative coherent, and regulatory affairs confirms that the document meets submission requirements and supports the intended marketing application.
The work runs in phases, and experienced teams start it early. A CSR “shell” or outline is often built months before the trial concludes, with the introduction, study objectives, and investigational plan drafted while the trial is still active. That front-loading is what keeps the post-lock period manageable: it allows early internal review of the study’s background and methodology, and leaves only the results and discussion for intensive drafting once the final data sets arrive. Results go in as statistical outputs become available, and the draft then goes through multiple rounds of review and quality control before finalization for regulatory submission. Increasingly, medical writers draft narrative sections with the help of large language model tools that generate first-draft text directly from the structured TLF and SAP inputs; medical judgment, discrepancy resolution, and final sign-off remain human-owned (Chapter 26 covers these tools and their current limits).
Not every clinical study warrants a full technical report. Under ICH E3 guidelines, abbreviated CSRs (using summarized data or with some sections deleted) may be acceptable for uncontrolled studies or other studies not designed to establish efficacy, for seriously flawed or aborted studies, or for controlled studies examining conditions clearly unrelated to the claim; a controlled safety study, however, is still reported in full (International Council for Harmonisation 1995). These criteria are keyed to a study’s design and role, not to the direction of its result: a negative but well-conducted controlled pivotal trial still requires a full CSR. Where an abbreviation applies, narrative sections such as detailed medical background or extensive investigator biographies may be summarized or omitted, which preserves the trial’s core results and safety data within the Trial Master File without producing a full publication-quality report.
28.3 Quality Standards
Every figure and data point in the report must match the underlying analysis datasets, because any discrepancy can erode reviewer confidence. Results sections have to align with stated objectives, terminology has to remain uniform across the document, and a missing regulatory element can trigger delays and supplemental inquiries. Transparency asks for something harder: an honest appraisal of the trial’s limitations and unexpected outcomes. Any attempt to minimize safety signals or obscure unfavorable findings will likely be discovered by inspectors, and it is the sponsor’s long-term credibility that pays (International Council for Harmonisation 1995).
The era of the CSR as a purely confidential document is ending. Driven by demands for scientific transparency and by ethical obligations to trial participants, regulators increasingly require public disclosure of clinical data. The EU Clinical Trials Regulation (CTR) 536/2014 requires sponsors to provide a summary of results for laypersons (European Medicines Agency 2022): plain language suitable for the general public, typically covering the trial’s rationale, who participated, what happened, and the side effects and benefits observed. Writing one well means simplifying complex medical terminology without losing scientific accuracy.
EMA Policy 0070 (European Medicines Agency 2014) goes further, requiring proactive publication of the clinical reports themselves once the regulatory decision is made, which hands the sponsor a redaction and anonymization problem: protect participant privacy while retaining data utility. Sponsors must document a re-identification risk assessment in an anonymization report before these documents are posted to public portals; EMA prefers quantitative methods (a qualitative approach has to be justified as a deviation) and expects the residual risk to sit at or below a pre-set target threshold, commonly 0.09.
What full CSR access is worth became concrete in the Tamiflu (oseltamivir) episode. For years, public claims that the drug prevented serious flu complications rested largely on journal articles and summaries. When independent reviewers obtained the clinical study reports themselves, a systematic re-analysis found that treatment shortened symptoms in adults by about 17 hours, from seven days to 6.3 days, and did not clearly support the larger public-health benefits that had justified government stockpiling (Jefferson et al. 2014). The CSR that regulators receive is increasingly the document that independent science will eventually read, so it must stand on its own, without unpublished context that only the sponsor holds.
28.4 Statistical Communication and Bayesian Reporting
A clinical study report has to make its numbers mean something to readers who range from trained statisticians to prescribing physicians.
Reporting Against the Pre-Specified Estimand
Since ICH E9(R1), the efficacy section of a CSR is expected to be organized around the trial’s pre-specified estimand, not around a bare treatment comparison. Each primary and key secondary estimand should be restated in full: the population, the treatment condition, the endpoint variable, the strategy for each intercurrent event, and the population-level summary measure. The report then confirms that the primary analysis estimates that estimand, and presents analyses built on alternative intercurrent-event strategies as labeled supplementary analyses, not as competing headline results: a different strategy defines a different estimand, whereas a sensitivity analysis keeps the estimand and tests the main estimator’s assumptions. E9(R1) is explicit that main, sensitivity, and supplementary results should be reported systematically, so a reviewer can judge how robust the primary conclusion is to the assumptions embedded in the chosen strategy. The estimands framework and the five intercurrent-event strategies are developed in Chapter 11 and Chapter 17; the CSR is where those design-stage choices are shown to have been honored (International Council for Harmonisation 2019).
From p-values to Probability Statements
The standard CSR statistical narrative reports a test statistic, a p-value, and a confidence interval. This format answers the question “Is the observed difference statistically significant?” It does not directly answer the question most relevant to regulators and clinicians: “How probable is it that this drug provides a clinically meaningful benefit?”
When a trial uses conventional frequentist methods and the p-value falls clearly below or above the threshold, the distinction between these two questions rarely matters for regulatory decisions. It matters in two cases. The first is a result near the threshold: a p-value of 0.04 in one trial and 0.06 in another carry different narrative weight despite containing similar evidence. The second is a trial using Bayesian methods, whose natural output is a probability statement, not a test statistic.
For Bayesian trials, the CSR should report the primary result in the same language the protocol and SAP specified: the posterior probability that the treatment is superior to control, the posterior probability of a benefit exceeding a clinically meaningful threshold, and the credible interval for the treatment effect estimate. The FDA review of REBYOTA (an approved fecal microbiota product) reported a 99.1% Bayesian posterior probability that the product was superior to placebo rather than a p-value, evaluating that probability against two pre-specified success thresholds of differing stringency: the result cleared the less stringent bar but not the more stringent one (US Food and Drug Administration, Center for Biologics Evaluation and Research 2022).
For frequentist trials, the effect size and its confidence interval are as important as the p-value and should appear in the synopsis, the summary of efficacy findings, and the conclusions. A result at p = 0.03 whose confidence interval reaches a hazard ratio of 0.99, nearly no clinical effect at one end, is meaningfully different from a result at p = 0.003 with an interval entirely below 0.85. Both are “significant.” They do not mean the same thing clinically, which is why the CSR narrative should discuss both the lower and the upper bound of the interval.
Communicating Uncertainty
Statistical reviewers work through the statistical appendix, but many readers of a CSR, clinicians in particular, may read little beyond the synopsis and the conclusions. A CSR that gives them only point estimates and binary pass/fail verdicts underserves the readers whose decisions it exists to inform. Those sections should communicate the effect size in clinically interpretable terms (what a hazard ratio of 0.78 means for a patient’s risk over a year), the range of plausible effects in the same terms, and the consistency of the primary result across pre-specified subgroups and secondary endpoints. For Bayesian analyses, they should also report the sensitivity of the conclusion to the prior specification, showing that the conclusion holds across alternative priors.
The aducanumab program (discussed in Chapter 27), whose two Phase III trials reached discordant results, shows why a report must address such discordance directly. When a result is unexpected, the report should say explicitly whether it is consistent with prior evidence, what alternative interpretations exist, and what additional information would be needed to resolve the uncertainty.
Reporting Adaptive Trial Results
The standard CSR template was designed for fixed-sample trials. An adaptive design has to document more than that template asks for: the adaptation rules pre-specified in the protocol and SAP, which adaptations were actually triggered, and against which dates and data cuts. It also reports the final sample size against the original plan, with the reason for any deviation. Unblinded interim analysis reports usually go in as appendices, with access controlled per the charter. The report also has to show how the alpha-spending or error-control method keeps the reported p-value valid despite the adaptations.
For platform trials, a CSR covering a single arm must make clear that its primary analysis used a shared control, and describe the concurrent controls (and the non-concurrent controls, if any) that contributed to the primary comparison, with the dates of contribution and the time-adjustment method applied.
28.5 Integration with Regulatory Submissions
In the Common Technical Document (CTD) format used for major regulatory submissions, CSRs appear in Module 5 (Clinical Study Reports); the CTD also includes summaries and overviews that synthesize information across multiple studies (International Council for Harmonisation 2016).
For pivotal studies, the CSR will be read closely: reviewers examine methodology, verify that the analysis followed the pre-specified plan, and assess whether the conclusions are supported by the data. A supportive study may draw a less intensive review, but its CSR must still give a reviewer enough to understand what was done and what was found. Either way the document outlives the review and remains the trial’s authoritative record long after the trial concludes.