Selective reporting in clinical trial announcements occurs when the results presented give an incomplete or distorted view of the evidence. A release may emphasize a favorable subgroup, omit an unsuccessful endpoint, or report response rates only among patients who remained assessable. The numbers can be accurate while the impression they create is misleading. For investors, the task is to identify what was tested, which patients contributed to the result, and which findings remain undisclosed.
This is a focused part of reading clinical trial results and assessing their investment significance. Treat each announcement as a set of claims to reconcile against the trial plan. Separate confirmed findings from exploratory observations, and distinguish missing patient outcomes from information absent from the release. Neither an omission nor a disappointing result proves misconduct. Both can leave questions that prevent a confident assessment of the drug.
Recognize What the Announcement Leaves Out
Start with a disclosure check rather than the headline adjective. The CONSORT 2025 reporting checklist for randomized trials calls for patient counts, results for each primary and secondary outcome, effect estimates and their precision, and a distinction between planned and post hoc analyses. A corporate release is not a full trial report, so use those items as an assessment framework, not a claim that every release must reproduce a journal article. Mark an undisclosed result as unknown. Do not fill the gap with management’s description of the trial as encouraging.
| Announcement feature | What to check | Question to resolve |
|---|---|---|
| A secondary outcome leads the headline | The primary outcome and planned testing order | Did the trial achieve its stated main objective? |
| A favorable subgroup dominates the presentation | Its definition, timing and statistical role | Was this a planned claim or a finding selected after reviewing results? |
| Results refer to “evaluable patients” | Included and excluded patient counts, with reasons | Who disappeared from the denominator? |
| An endpoint is described without numerical results | The estimate, uncertainty and comparator result | What evidence supports the description? |
| Safety is described as manageable | Events and treatment discontinuations by group | What did patients experience, and how often? |
Reconstruct the plan before judging the result
Find the trial identifier and compare the announcement with earlier records. ClinicalTrials.gov retains a history of changes, allowing readers to inspect previous versions rather than relying only on the current entry; the National Library of Medicine’s description of study records and archived changes confirms that function. Check the outcome definition, assessment time, enrollment target and patient eligibility. Record when changes appeared. An amendment made before access to comparative results raises a different question from an unexplained change after those results became available. A registry timestamp alone may not establish when a decision was made or what the decision makers knew.
Build a small evidence file containing the announcement, registry history, protocol and statistical analysis plan where available. Add presentation slides and journal supplements when released. The purpose is reconciliation, not collecting documents for its own sake. Write down any mismatch in plain language: “The original primary assessment was at week 24; the release emphasizes week 12.” Then look for an explanation. The broader guide to researching biotech through trial registries, scientific papers and financial filings covers how these materials fit together.
Check Whether Favorable Numbers Answer the Planned Question
A favorable result does not automatically establish that a trial succeeded. Testing many outcomes creates more opportunities for a chance finding, so the planned procedure for controlling false positive conclusions matters. In a fixed testing sequence, failure at an earlier step can prevent later results from supporting a formal confirmatory claim under that procedure. Other designs allocate statistical error differently. Check the actual plan rather than assuming every trial follows the same sequence. The FDA guidance on multiple endpoints in clinical trials addresses endpoint families, testing order and multiplicity control. “Prespecified” identifies advance planning; it does not, by itself, establish that an analysis could support the claim being made.
Consider a hypothetical study with one primary endpoint, tested at a two-sided threshold of 0.05. Its result is p = 0.12. A secondary symptom score produces p = 0.02. If the plan requires primary endpoint success before formally testing that symptom score, the second number does not reverse the primary miss. Record the outcome accordingly: the primary objective was not met under the stated rule, and a favorable secondary observation merits further assessment. That is different from declaring the drug ineffective under every possible use. It is also different from calling the trial successful.
Keep an endpoint map beside the release. For each outcome, record its planned status, measurement time, analysis population and disclosed result. If the announcement emphasizes a different visit or a component of a broader measure, ask why that result deserves priority. Do not let the most attractive number choose the question retrospectively. Our guide to clinical endpoints and what they measure provides the background for assessing whether the highlighted outcome addresses a meaningful treatment benefit.
Assess Subgroup Claims Without Dismissing Useful Signals
Subgroup findings deserve attention, but their interpretation requires more than a favorable p-value. Advance definition, biological rationale, independent replication and results in the complementary subgroup all help assess whether a finding is persuasive. Interaction testing examines whether treatment effects differ between groups; an isolated interaction p-value cannot settle the clinical question, and a nonpersuasive test does not establish identical effects. The EMA guideline on subgroup analyses in confirmatory trials addresses these considerations. Distinguish a subgroup included in the formal testing strategy from one listed for exploratory analysis. Both can be planned, yet they serve different evidentiary purposes.
Inspect the definition, not just the label
Suppose an announcement highlights patients with “high biomarker expression.” Request the threshold, assay, number of patients in each treatment arm and the date that definition was chosen. Compare those details with the earlier plan. A label alone cannot tell you whether “high” meant the same thing before and after analysis. If the threshold changed, ask for results using the original definition and the reason for the revision. Check the proposed drug target and mechanism of action separately: a plausible scientific story should be evaluated alongside the clinical findings, rather than used to excuse missing details.
Compare treatment effects across groups
Consider a hypothetical study reporting p = 0.03 among biomarker-positive patients and p = 0.18 among biomarker-negative patients. Those numbers alone do not demonstrate that the drug works differently between the groups. Request the treatment effect and confidence interval in each, along with the interaction analysis where appropriate. The two estimates might be similar, with different precision. Alternatively, they might point in different directions. The guide to effect sizes, confidence intervals and p-values explains the statistical distinctions behind that comparison.
For an investment assessment, translate the subgroup claim into a testable next question. What evidence would confirm it? A useful proposal would fix eligibility and the biomarker threshold before a new study, identify its main outcome and define the analysis. Repeatedly presenting the original dataset with new cutoffs would not answer the same question. Review the proposed clinical trial design before treating a narrower development plan as established progress. Also revisit the commercial assumptions: if the investment case now rests on fewer eligible patients, the previous market estimate needs rebuilding rather than carrying forward unchanged.
Separate Missing Disclosure From Missing Patient Outcomes
An endpoint absent from a release and a patient outcome absent from the dataset are different problems. The first calls for fuller disclosure. The second requires an analysis that addresses unobserved outcomes. Missing data assessment should consider why observations are absent, how the analysis handles them, and whether alternative plausible assumptions change the interpretation. The expert recommendations in The Prevention and Treatment of Missing Data in Clinical Trials emphasize documenting reasons, defining the treatment effect of interest and examining sensitivity to assumptions. For investors, the useful question is not simply “How much is missing?” Ask what the missing observations could mean for the claimed benefit.
Reconcile every denominator
Consider a hypothetical single-arm study that enrolls 100 patients. The announcement reports 18 responses among 60 evaluable patients: a 30% response rate. The same 18 responses represent 18% of all enrolled patients. Neither calculation tells you the appropriate primary analysis without the outcome definition and patient accounting. Request an explanation for the other 40 patients. Were their assessments not yet due, were measurements unavailable, or did they leave before assessment? Do not automatically count every omitted patient as a nonresponder, or accept the smaller denominator without examination. The immediate task is to establish what each rate measures.
Create a patient reconciliation from enrollment through the reported assessment. Ask for counts assigned to each arm, treated, assessed and included in the analysis, followed by reasons for exclusions. Where a release uses “modified intention to treat,” request its exact definition rather than assuming the label resolves the issue. The CONSORT 2025 explanation of analysis populations and missing data reporting calls for clear analysis definitions, missing data methods and reporting that allows readers to examine the assumptions. Use those disclosures to check whether the headline result and the patient flow tell a consistent story.
Use hypothetical attrition to test your interpretation
Suppose a randomized trial assigns 120 patients to each arm. At the main assessment, outcomes are unavailable for 24 treated patients and six controls: 20% versus 5%. Now consider two possible explanations. In one, missed assessments followed unrelated scheduling problems. In another, treated patients left after worsening symptoms or intolerable side effects. Ask the company to show the documented reasons, timing and available observations before departure. Do not choose the explanation that best suits the investment thesis. Equally, do not infer that every departure reflects harm. The worked example identifies what needs investigation; it does not establish the direction or size of any bias.
Ask What Happened After Patients Stopped Treatment
Stopping treatment is not the same as leaving follow-up. A patient may discontinue medication while continuing study assessments. Treatment switching, rescue medication and death also require the trial question to be defined carefully. The estimand describes the treatment effect the study intends to estimate, including how such events are addressed. Missingness must be assessed relative to that question. The ICH E9(R1) guideline on estimands and sensitivity analysis distinguishes these issues and calls for sensitivity analyses that examine assumptions while targeting the same treatment effect.
Translate the terminology into a practical request. Does the headline estimate address outcomes after assignment regardless of later treatment changes, outcomes while patients remain on treatment, or a hypothetical situation without a particular event? Ask for the answer in the company’s own trial documentation. Then check which observations were collected and used. For a patient who stopped medication at week eight, were later symptoms still measured? If so, were those measurements included? If not, how was their absence handled? A long statistical label is no substitute for an understandable account of what happened to patients.
Request sensitivity results, not assurances
Ask which assumptions were changed in sensitivity analyses and how much the treatment estimate moved. Request the results themselves, including uncertainty, rather than accepting a statement that several analyses were supportive. A tipping point analysis can frame a useful question: how unfavorable would unobserved outcomes need to be before the interpretation changed? Treat that question as part of the evidence review, not an automatic pass or fail rule. Also check whether a purported sensitivity analysis retained the original treatment question. A result obtained after changing the population and outcome definition should be labeled clearly rather than presented as though it only tested a missing data assumption.
Keep pending assessments separate from lost observations in your notes. Request the data cutoff, follow-up distribution and the number of patients who had reached the assessment window. If two announcements use different cutoffs, reconcile the cohort and observation time before describing a change as improvement or deterioration. For an early response announcement, ask for the planned update that will address persistence of benefit. Set that update as an evidence checkpoint, not a promise that the result will strengthen.
Apply the Same Scrutiny to Safety Reporting
Safety needs its own patient accounting. The CONSORT Harms reporting recommendations call for clear definitions, assessment methods and denominators, including explanations of exclusions. Do not assume that the patients contributing to an efficacy analysis are identical to those included in safety reporting. Request adverse events and treatment discontinuations by arm, with the observation period stated. Keep serious events distinct from events described as severe, and ask what any attribution to treatment means in the report. A short description such as “well tolerated” leaves these questions unanswered.
In the hypothetical 100-patient study, suppose efficacy was highlighted among 60 evaluable patients while safety was reported among 98 patients who received treatment. That difference is a reconciliation task, not automatically an error. Ask whether the safety information accounts for patients omitted from efficacy assessment and whether treatment intolerance contributed to their exclusion. Read the relevant categories separately rather than treating one favorable total as reassurance about every risk. The guide to reading adverse events, discontinuations and benefit–risk balance covers the broader safety assessment.
Turn Unanswered Questions Into an Evidence Plan
Before revising a valuation, create a short record of what is known and what could change the assessment. This prevents an attractive exploratory result from acquiring the status of confirmed evidence through repetition. Keep company intentions separate from completed actions. “Plans to discuss a subgroup strategy” should remain a plan until there is a documented outcome. Likewise, expected presentation dates belong in a biotech catalyst schedule, with the missing disclosure each event is expected to address written beside it.
- Record the main result: The planned endpoint, population, effect estimate and uncertainty.
- Classify highlighted analyses: Confirmatory, exploratory, post hoc or not yet clear from available documents.
- Reconcile patient counts: Identify exclusions, unavailable outcomes and pending assessments.
- List unresolved assumptions: Include missing data handling and treatment changes.
- Define the next evidence checkpoint: State which document, analysis or new trial would resolve each question.
Adjust assumptions only as far as the evidence supports. A subgroup hypothesis can justify examining a future development path without justifying the same confidence as a successful planned analysis. Keep alternative scenarios visible rather than hiding unresolved questions inside a single optimistic forecast. The framework for estimating clinical and regulatory success probabilities provides a place to express that uncertainty. When the necessary disclosure is absent, write “not established from available results.” That is a usable research finding, and a better starting point for the next readout than guessing.