CLINAPTIS RESEARCH
Browse Learn
GUIDE · TRIAL RESULTS

Interpreting Clinical Trial Results

Assess what a trial establishes about benefit, risk and uncertainty before translating the result into an investment assumption.

By Bikram K. Singh, MBBS · Last updated 18 October 2026. BioCo examples are fictional teaching illustrations; their dates and figures are assumptions.

Write Down What the Trial Tested

Before reading the headline, write down the population, treatment, comparator, endpoint and follow-up. Add the phase, randomization and blinding.

Check the reported population, data cutoff and whether the analysis is interim or final against the protocol and analysis plan.

ICH E9: statistical principles

Read the Endpoint in Context

Selected oncology endpoints—not a universal ranking
EndpointWhat it measuresWhat to examine
Overall survival (OS)Time to death from any cause.Follow-up, subsequent therapies and crossover.
Progression-free survival (PFS)Time to defined progression or death.Assessment criteria, scan schedule and potential assessment bias.
Objective response rate (ORR)The proportion meeting specified tumor-response criteria.Response definition, duration, denominator and whether there is a control arm.

Tumor response and progression measures can provide evidence earlier than survival, but do not necessarily predict a survival benefit. Their regulatory usefulness depends on the disease, setting and evidence.

FDA: cancer trial endpoints, Tables 1–2 and Section III

Primary, secondary and exploratory describe the role of an endpoint in the analysis. They are different from whether an endpoint directly measures clinical benefit or serves as a surrogate. Accelerated Approval requires verification of benefit; the pathway should not be inferred from an endpoint name alone.

FDA: Project Confirm

Read the Size of the Effect

A relative effect needs a baseline and a time horizon. For a fixed-period binary outcome, report the event rate in each arm, the absolute difference and the relative comparison. Keep the numerator, denominator and direction of benefit visible.

FORMULA · FIXED-PERIOD BINARY OUTCOMEARR = pcontrol − ptreatedRR = ptreated / pcontrol   RRR = 1 − RRp: share of participants with the event by the end of the period (an adverse event here, so a benefit means ptreated < pcontrol). ARR: absolute risk reduction, in percentage points. RR: risk ratio. RRR: relative risk reduction. BioCo example below: 20% − 12% = 8 points; 12% / 20% = 0.60; 1 − 0.60 = 40%.

Do not treat a hazard ratio as a fixed-period risk ratio or convert it directly into months of survival gained. For time-to-event results, inspect the survival curves and reported time-based estimates alongside the relative effect. The clinical question remains how much benefit was observed, for whom and over what follow-up.

Interpret Uncertainty Without Overstating It

A p-value describes how incompatible the observed result—or a more extreme result—is with the specified null model, under its assumptions. It is not the probability that the treatment works, that the null is true or that approval will follow.

A 95% confidence interval comes from a procedure that would cover the true parameter in 95% of repeated applications under its assumptions. Read it with the effect estimate: an interval may remain compatible with benefits too small to matter clinically. Statistical significance alone does not settle clinical relevance.

ICH E9: statistical principles
DEFINITION · P-VALUE AND CONFIDENCE INTERVALp = P(data at least this extreme, given the null model and its assumptions)95% CI ≈ estimate ± 1.96 × SESE: standard error of the estimate. The 1.96 multiplier is the large-sample normal approximation; other methods use different multipliers or exact intervals. p is not the probability that the null is true given the data.

A Bayesian credible interval gives a posterior probability statement conditional on the data, prior and model. A posterior probability of exceeding a target depends on the prior, the data and the model. It is not an assumption-free probability of commercial success. FDA’s January 2026 draft guidance covers Bayesian methods in drug and biologic trials.

FDA: Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products (draft guidance, January 2026)

Check Which Analyses Can Support the Claim

With multiple endpoints, subgroups or interim looks, there are more opportunities for a favorable finding. Identify the prespecified testing strategy, the order of tests and any adjustments for multiplicity. A nominally small p-value for a secondary endpoint may not support a confirmatory claim if the planned testing sequence has failed.

FORMULA · CHANCE OF AT LEAST ONE FALSE POSITIVEP(at least one false positive) = 1 − (1 − α)mExample: α = 0.05, m = 5 → 1 − 0.955 = 22.6%m independent tests, each at significance level α. A Bonferroni threshold divides α by m: 0.05 / 5 = 0.01 per test. Correlated endpoints change the number, and a prespecified testing sequence can control the overall error rate without dividing α.

Keep exploratory findings visible as hypotheses for further work. A favorable result within one subgroup does not by itself establish that treatment effects differ between subgroups.

FDA: multiple endpoints

Also check the treatment effect being estimated and how discontinuation, switching treatment or rescue medication are handled. An intention-to-treat label does not resolve every missing-data problem. Look for the planned analysis and sensitivity analyses that test important assumptions.

ICH E9(R1): estimands and sensitivity analysis
DEFINITION · AN ESTIMAND HAS FIVE PARTS1 Population   2 Treatment   3 Variable (endpoint)4 Handling of intercurrent events   5 Population-level summaryIntercurrent events happen after treatment starts and affect how the measurement is interpreted or whether it exists, for example discontinuation, switching treatment or rescue medication. Source: ICH E9(R1).

Read Safety Alongside Benefit

Record adverse events by arm, including denominators, deaths, serious events, severity, treatment discontinuations and dose changes. Distinguish any event during treatment from an event attributed to the drug.

Serious and severe are different concepts. Severity describes intensity; seriousness uses criteria such as death, life-threatening events or hospitalization. A summary saying “no new safety signals” should lead you to the underlying tables, not replace them.

ICH E2A: safety definitions, Section II
DEFINITION · SEVERE IS NOT SERIOUSSevere: the intensity of an event (mild, moderate, severe).Serious: an outcome or action criterion, such as death, a life-threatening event or hospitalization.An event of modest intensity can still be serious if it leads to hospitalization, and a severe event can be non-serious. ICH E2A: seriousness, not severity, guides regulatory reporting.

As an analyst’s checklist, ask whether exposure and follow-up are comparable, whether missing safety data are explained, and what tolerability could mean for continued use. Small or immature datasets leave uncertainty about uncommon or late events.

Keep Advanced Designs in Proportion

Adaptive trials use prospectively planned changes based on accumulating data. Their credibility depends on the rules, safeguards and operating characteristics of the design. “Adaptive” does not mean uncontrolled flexibility after results appear.

FDA: adaptive designs for drugs and biologics, Section II

Bayesian borrowing uses assumptions about how relevant prior or external data are to the current trial. Review those assumptions and sensitivity to the prior. Historical controls are not automatically interchangeable with a concurrent randomized control arm.

FDA: Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products (draft guidance, January 2026)

BioCo: Read Beyond the Relative Benefit

Fictional arithmetic example. Assume a randomized study reports a prespecified adverse outcome by week 24 in 40 of 200 control participants and 24 of 200 treated participants. For this illustration, assume the outcome is known for everyone.

BioCo illustrative week-24 outcomes
MeasureCalculationResult
Control event risk40 ÷ 20020%
Treatment event risk24 ÷ 20012%
Absolute risk reduction20% − 12%8 percentage points
Risk ratio12% ÷ 20%0.60
Relative risk reduction1 − 0.6040%

“A 40% reduction” and “eight fewer events per 100 participants by week 24” describe the same illustrative comparison. Neither is a 40-percentage-point benefit, and this fixed-period risk ratio is not a hazard ratio.

What remains unanswered? The example does not supply an interval, test result, safety profile, endpoint definition or clinically meaningful target. It therefore does not establish statistical significance, an acceptable benefit–risk balance or an approvable product. Those are separate questions requiring evidence.

Connect the Finding to Explicit Assumptions

Write a short evidence summary before changing a model: what the study establishes; what it leaves uncertain; and which previous assumption now deserves revision. Consider the studied population, duration of benefit, safety, further evidence required and competitive context.

Keep clinical, regulatory and commercial judgments distinct. A positive trial result is not itself a forecast of uptake, price or equity value. Do not apply a standard probability uplift to a positive result. Revise each assumption only for the reason the evidence supports.

In the valuation model, identify where development risk is already represented before adding another adjustment. If a scenario includes stopping development, include the relevant cash flows and obligations. A Monte Carlo output is a model result, not observed clinical evidence.

See how assumptions enter the pre-commercial DCF →

A Checklist for the Next Readout

  • What exact study question and population does the result address?
  • Is the reported endpoint prespecified, and does the testing strategy support the claim?
  • What are the effect size, uncertainty and absolute outcome rates where applicable?
  • Are follow-up, data cutoff, analysis population and missing-data handling clear?
  • What does the safety evidence show, and what remains immature?
  • Are subgroup and cross-trial claims being given more weight than the design supports?
  • Which assumption changes, and what source supports that change?

Sources & Further Reading

Primary references for the methods discussed above. The BioCo example is a fictional teaching illustration, not a record of an actual trial.

Educational content, not investment advice. Terms