Clinical Trial

Clinical Trial Evidence for Health Technology Assessment and Economic Evaluation

What makes a study a clinical trial is that the study protocol decides in advance which intervention each participant receives and then follows what happens to them. Randomisation, control groups and blinding are design features that a trial may or may not have, and the strength of its evidence depends on them. In health economics, randomised trials are the preferred source of relative treatment effects in NICE appraisals and often the vehicle for collecting costs and quality of life data, and the NICE manual (PMG36) sets out what an appraisal takes from them.

What qualifies as a clinical trial

The defining feature is prospective assignment of an intervention under a research protocol. The WHO definition used for trial registration covers any research study that assigns human participants or groups of humans, ahead of time, to one or more health-related interventions in order to evaluate the effects on health outcomes. Its examples of interventions run from drugs and devices to behavioural treatments and process-of-care changes, and it covers Phase I to Phase IV.

The International Committee of Medical Journal Editors (ICMJE) uses a similar definition and adds that a trial may run with or without concurrent comparison or control groups. The ICH E8(R1) guideline on clinical studies draws the same line. When the study procedures control which drug a participant receives, the study is interventional; when exposure is only observed, it is an observational study. A single-arm study in which everyone receives the new treatment is therefore a clinical trial, but its comparison with any other treatment rests on external data. The randomised controlled trial is the subset in which allocation is decided by chance.

Phases and designs

Drug trials are often described by phase. The United States regulation on investigational new drugs (21 CFR 312.21) describes Phase 1 as the first introduction of the drug into humans, generally with 20 to 80 participants; Phase 2 as controlled studies of effectiveness and common short-term risks, usually no more than several hundred participants; and Phase 3 as expanded trials, usually several hundred to several thousand participants, that gather the information needed for an overall benefit-risk judgement. WHO describes Phase IV studies as taking place after approval, in a wide population over a longer period. ICH E8(R1) adds that the phase concept is a description, not a requirement, and that phases may overlap or be combined.

The phase says little about design. ICH E8(R1) notes that randomised designs range from simple parallel groups to adaptive designs, which allow prospectively planned modifications, and platform studies, in which several drugs enter and leave under prespecified decision rules. Crossover trials, adaptive trial designs and non-inferiority trials have their own pages. For economic evaluation the more useful distinction is between explanatory trials, which test efficacy under tightly controlled conditions, and pragmatic trials, which test effectiveness in routine practice against current care.

Pinning down the question with the estimand

A trial answers one precise question, and the ICH E9(R1) addendum on estimands sets out how to state it. An estimand is described by its attributes, which are the treatment and its comparator, the population targeted by the clinical question, the variable or endpoint measured for each patient, and a population-level summary used to compare treatments. Events after treatment starts that affect the interpretation or existence of the measurements, such as discontinuing treatment, using an additional medication or switching between treatments, are called intercurrent events, and the estimand says how each is handled.

Under the treatment policy strategy, the outcome is used whether or not the intercurrent event occurs. The addendum links this to the effect of a treatment policy described under the intention-to-treat principle. The choice matters for economic evaluation. A cost-effectiveness model asks what happens when a health system adopts a treatment, including the people who stop it, so a treatment policy estimand often fits the decision better than a hypothetical one in which everyone stays on treatment, provided the treatments people move on to in the trial match routine care.

What health technology assessment takes from a trial

NICE states in its technology appraisal manual (PMG36, section 3.3.2) that for relative treatment effects there is a strong preference for high-quality randomised controlled trials. The manual explains that randomisation makes baseline differences between groups, including unmeasured ones, a matter of chance, so a trial should in principle give a minimally biased estimate of relative benefits and risks (section 3.3.5). It judges relevance by internal validity (blinding where appropriate, randomisation method, allocation concealment, completeness of follow-up, size and power, outcomes and intention-to-treat analysis) and by external validity, whether results apply to wider patient groups and routine practice (section 3.3.6).

The manual also lists limits (section 3.3.8). Trials may not quantify the effect over the course of the condition, may be unethical or infeasible, or may not generalise because of the comparator, population or treatment pathway. Single-arm trials compared with an external control are classed with non-randomised evidence, which the manual describes as tending to be at high risk of bias because the factors that drive treatment assignment may predict outcomes (section 3.3.9). When trials are combined in a network meta-analysis, randomisation must be preserved; comparing single arms from different trials is not acceptable (section 3.4.20).

Trials rarely supply every model input. The manual notes that trial data may not estimate the risk of some outcomes in the population of interest, or over a long enough period, for the economic analysis (section 4.6.15), and suggests applying relative effects from randomised trials to baseline risks for the population of interest, which observational studies can inform (section 4.6.16).

Economic evaluation alongside a clinical trial

Collecting resource use and quality of life data within a trial gives patient-level costs and outcomes linked to randomised groups. Petrou and Gray, in their BMJ guide, note that funders such as the NIHR Health Technology Assessment Programme routinely ask for cost-effectiveness to be built into trial design. The ISPOR task force report on cost-effectiveness analysis alongside clinical trials recommends designing trials to evaluate effectiveness rather than efficacy where possible, obtaining resource use and health state utilities directly from participants, running an incremental analysis on an intention-to-treat basis and characterising uncertainty.

Several features of trial data shape the analysis. Petrou and Gray point out that power calculations are almost always based on the primary clinical outcome, and that the joint distribution of cost and effect differences can show clear cost-effectiveness when neither difference is individually significant. The arithmetic mean cost is the quantity of interest even though costs are often right skewed, and multiple imputation is increasingly favoured over complete-case analysis for missing data. Explanatory trials can limit generalisability through strict inclusion criteria and protocols, and protocol-driven costs should be removed from the cost-effectiveness calculation.

The deeper limit is scope. Sculpher, Claxton, Drummond and McCabe argue that a decision needs costs and effects over an appropriate time horizon, for all relevant alternatives and patient groups, using all the available evidence, so that in most circumstances a single trial is an inadequate and partial basis for decision making. On their account a trial is one source of evidence within a wider synthesis and decision model.

Worked example with a within-trial result and two extrapolations

The numbers are illustrative. A 12-month trial compares a one-off intervention with usual care. Within the trial the intervention adds GBP 1,500 in mean cost per patient and 0.05 QALYs, so the within-trial ICER is 1,500 / 0.05 = GBP 30,000 per QALY. At an illustrative threshold of GBP 20,000 per QALY the net monetary benefit is 20,000 × 0.05 − 1,500 = −GBP 500, and the intervention does not look worthwhile.

PMG36 (section 4.2.24) asks for scenarios when effects are estimated beyond the clinical studies, including one in which the technology gives no further benefit, alongside more optimistic assumptions. Three scenarios cover four years after the trial, with no further costs and discounting at the reference-case rate of 3.5% a year (section 4.5.1), with discounted incremental QALYs given by the formula below.

$$\Delta E = \Delta q_{1} + \sum_{t=1}^{4} \frac{\Delta q_{t+1}}{(1.035)^{t}}$$

where $\Delta E$ is the discounted incremental QALYs over five years, $\Delta q_{1} = 0.05$ is the gain observed in the trial year, $\Delta q_{t+1}$ is the assumed gain in year $t+1$ and $t$ counts years after the trial. The discount factors for years 2 to 5 are 0.9662, 0.9335, 0.9019 and 0.8714, which sum to 3.6731.

In scenario B the extra QALYs are 0.05 × 3.6731 = 0.1837, giving a total of 0.2337, and the net monetary benefit is 20,000 × 0.23365 − 1,500 = GBP 3,173. In scenario C the gains of 0.04, 0.03, 0.02 and 0.01 discount to 0.03865 + 0.02801 + 0.01804 + 0.00871 = 0.09341, giving 0.1434 in total and a net monetary benefit of GBP 2,868 − GBP 1,500 = GBP 1,368. For scenario B the ratio is calculated as follows.

$$\text{ICER}_{B} = \frac{\Delta C}{\Delta E} = \frac{1500}{0.23365} = 6420$$

where $\Delta C$ is the incremental cost in GBP and $\Delta E$ the discounted incremental QALYs.

The decision turns on what happens after follow-up, which the trial cannot observe. In both scenarios where benefit persists, the intervention is cost-effective at the illustrative threshold, while the within-trial result says it is not. PMG36 (section 4.6.20) asks that extrapolation assumptions be judged for clinical and biological plausibility and against external data, so the case rests on longer-term evidence and the extrapolation method as much as on the trial.

Registration and reporting

WHO regards registration of every interventional trial as a scientific, ethical and moral responsibility, partly because publication bias and selective reporting undermine informed decisions, and cites the Declaration of Helsinki requirement to register every trial in a public database before the first participant is recruited. The ICMJE recommends that journals require registration at or before the time of first patient consent as a condition for considering a trial for publication.

For reports of randomised trials, the CONSORT 2025 statement provides a 30-item checklist and a participant flow diagram, and supersedes CONSORT 2010. Its new open science section groups trial registration, access to the protocol and statistical analysis plan, sharing of de-identified participant data, and funding and conflicts of interest. For an economic analyst, these items show whether outcomes were prespecified and whether patient-level data may be available.

Common misreadings

Most errors come from reading the label "clinical trial" as a guarantee of a particular design or of decision relevance. Each can be checked in the trial report.

• Trial taken to mean randomised. A single-arm or non-randomised trial is still a trial, and NICE treats single-arm comparisons with external controls as non-randomised evidence.
• Within-trial ICER read as a lifetime result. A truncated horizon can reverse the decision, as the example shows.
• Surrogate outcomes taken as patient benefit. PMG36 (section 4.6.6) reserves level 1 evidence for a surrogate endpoint for randomised trials showing a commensurate effect on the final outcome.

Sources

• World Health Organization. Clinical trials: questions and answers. 16 January 2020. https://www.who.int/news-room/questions-and-answers/item/clinical-trials
• World Health Organization. Clinical trials (health topic). Accessed 9 October 2026. https://www.who.int/health-topics/clinical-trials
• International Committee of Medical Journal Editors. Recommendations: clinical trial registration. Accessed 9 October 2026. https://www.icmje.org/recommendations/browse/publishing-and-editorial-issues/clinical-trial-registration.html
• International Council for Harmonisation. ICH E8(R1) General Considerations for Clinical Studies. Step 4, 6 October 2021. https://database.ich.org/sites/default/files/E8-R1_Guideline_Step4_2021_1006.pdf
• International Council for Harmonisation. ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials. Step 4, 20 November 2019. https://database.ich.org/sites/default/files/E9-R1_Step4_Guideline_2019_1203.pdf
• United States Code of Federal Regulations. 21 CFR 312.21 Phases of an investigation. eCFR, accessed 9 October 2026. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-D/part-312/subpart-B/section-312.21
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Chapters 3 and 4. https://www.nice.org.uk/process/pmg36/chapter/evidence-2
• Petrou S, Gray A. Economic evaluation alongside randomised controlled trials: design, conduct, analysis, and reporting. BMJ 2011;342:d1548. doi:10.1136/bmj.d1548. https://pmc.ncbi.nlm.nih.gov/articles/PMC3230107/
• Ramsey SD, Willke RJ, Glick H, et al. Cost-effectiveness analysis alongside clinical trials II: an ISPOR Good Research Practices Task Force report. Value in Health 2015;18(2):161-172. doi:10.1016/j.jval.2015.02.001.
• Sculpher MJ, Claxton K, Drummond M, McCabe C. Whither trial-based economic evaluation for health care decision making? Health Economics 2006;15(7):677-687. doi:10.1002/hec.1093.
• Hopewell S, Chan AW, Collins GS, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ 2025;389:e081123. doi:10.1136/bmj-2024-081123. https://pmc.ncbi.nlm.nih.gov/articles/PMC11995449/