Clinical Outcome Assessment

Clinical Outcome Assessment Types, Fitness for Purpose and the Route to QALYs

A clinical outcome assessment (COA) is the instrument behind most trial endpoints about symptoms and daily functioning, whether a questionnaire, a clinician rating, a caregiver diary or a timed task. The four kinds differ in whose judgement shapes the score. For health economists, COA results carry the trial evidence on how a treatment changes symptoms and functioning, yet they rarely arrive in the form a cost-utility model needs.

Four types of COA, sorted by whose judgement shapes the score

The ISPOR task force report by Walton and colleagues classifies COAs by whether judgement can influence the measurement, and if so, whose, which gives four categories. The US Food and Drug Administration (FDA) uses the same four in its 2025 guidance on fit-for-purpose COAs, adding that the choice depends on the concept to be measured, the best source of that measurement and the context of use.

A report by someone answering as if they were the patient is a proxy-reported outcome, which the FDA discourages; the 2009 PRO guidance states that a proxy report is not a PRO. Where self-report is impossible, the FDA recommends an ObsRO limited to what can be observed. An observer cannot validly report an infant's pain intensity but can report behaviour thought to be caused by pain.

Where the COA family stops at biomarkers and survival

Walton and colleagues draw the outer boundary through motivation and judgement. A COA is any assessment that may be influenced by human choices, judgement or motivation, whereas a biomarker is subject to little, if any, motivational or judgemental influence. A walk test therefore stays a COA although no one rates it, because the patient's effort shapes the score; a serum cholesterol level does not. Biomarker evidence, including surrogate endpoints, raises the separate question of whether the marker predicts patient outcomes.

Survival sits at the edge. The FDA COA web page defines a COA as a measure that describes or reflects how a patient feels, functions or survives. The 2025 guidance defines it as a measure of how a patient feels or functions, and states that the guidance covers feeling and functioning, not survival.

From meaningful aspect of health to context of use

The FDA guidance builds its approach from three linked ideas. Clinical benefit is a positive effect on how an individual feels, functions or survives, and to provide it a treatment should affect a meaningful aspect of health, meaning some aspect of feeling or functioning in daily life that matters to patients. The concept of interest is what the COA specifically measures, which may equal that aspect, as with nausea intensity, or reflect only part of it. The guidance's example is the 6-minute walk test, which measures functional walking capacity, one of several aspects of lower limb function.

The context of use states how scores will serve as the basis of an endpoint. It covers the target population, the role of the COA in the trial (for example, the mean score at 12 weeks), and how, where and by whom it is collected.

Showing that a COA is fit for purpose

The FDA treats a COA as fit for purpose when the level of validation is sufficient to support its context of use. A sponsor should state the concept of interest and context of use clearly and give an evidence-based rationale for reading scores as measures of that concept. Evidence for each part of the rationale is weighed against the uncertainty about that part, and may come from literature reviews, natural history studies, qualitative research with patients or caregivers, and quantitative studies.

The guidance lists eight components such a rationale may need, among them whether the COA captures all the important parts of the concept (previously called content validity), whether respondents understand the items as intended, whether scoring is appropriate, whether scores are overly influenced by measurement error, and whether scores correspond to the meaningful aspect of health. The guidance follows the measurement field in treating validity as one concept supported by different sources of evidence. Properties such as construct validity, test-retest reliability, measurement error and responsiveness supply evidence for these components.

Formal qualification is optional. The FDA describes qualification as a conclusion that a COA is well defined and reliable within its context of use, and states that sponsors are not required to use a qualified COA to support a trial endpoint.

From score to endpoint, and what happens when patients die

The 2025 guidance separates three things. The COA is the instrument with its instructions and scoring rules; a score is the value it produces; and an endpoint is a precisely defined variable that is statistically analysed to answer a research question, such as the average daily symptom score over the 7 days before the week 12 visit.

The FDA's 2023 draft guidance on COA-based endpoints notes that analysing the continuous or ordinal score usually maximises statistical power, while dichotomised responder endpoints are well established and can be reasonable when the question is the chance of reaching a clearly defined and important health state, such as complete symptom resolution, with a single prespecified threshold justified in the endpoint rationale. It also distinguishes missing data from intercurrent events such as death, which happen after randomisation and affect whether the endpoint can be observed or how it is interpreted. The ICH E9(R1) addendum on estimands sets out strategies for these events. A composite strategy can give people who die a value reflecting their lack of ability to function, a while-on-treatment strategy uses values up to the event, and a hypothetical strategy envisages a scenario in which the event does not occur. A treatment policy strategy cannot be used for a variable that cannot be measured because of death.

Reading change through meaningful score differences

The FDA's 2023 draft introduces the meaningful score difference, the size of difference between two COA scores that patients would regard as meaningful, and describes anchor-based methods for estimating it. An anchor is an external variable, not derived from the COA, whose meaningful differences are already known, such as a patient's rating of change from much better to much worse. The draft states that distribution-based methods do not directly consider the patient voice and cannot be the sole basis for a threshold, and it recommends several anchors. The minimal clinically important difference page covers anchor-based estimation in more detail.

How COA evidence reaches QALYs in a cost-utility analysis

Most COA scores cannot enter a cost-utility analysis directly. The NICE manual (PMG36, section 4.3.2) explains that deriving a utility has two elements, measuring health-related quality of life in people in a health state, and valuing it by preferences relative to other states, usually full health and death. A disease-specific symptom scale supplies only the first element. A fall in a fatigue score has no weight on the scale that QALYs need until a preference-based valuation is attached.

The NICE reference case sets the route (PMG36, sections 4.3.3 to 4.3.10). Health-related quality of life should be measured directly by patients, and when that is not possible by the person who acts as their carer rather than by healthcare professionals. The EQ-5D is the preferred measure in adults, with preference values applied to patients' responses. If trials lack EQ-5D data, values can come from the literature or be estimated by mapping other measures to EQ-5D with a mapping function, which NICE treats as a departure from the reference case and expects to be tested in sensitivity analysis. To argue that the EQ-5D is inappropriate for a condition, a company should give qualitative evidence that it lacks content validity, supported by evidence of poor construct validity and responsiveness, which is the same evidence a COA developer would assemble.

COA results also shape model structure. Responder status or severity bands defined by a COA often become model health states, so the threshold between states drives the QALY gain credited to a treatment.

Worked example of turning a responder endpoint into QALYs

All numbers are illustrative. A trial of a new treatment for fatigue in a chronic condition uses a validated PRO and a pre-specified responder definition. At 12 weeks, 60% of patients on the new treatment respond against 40% on standard care. EQ-5D collected in the trial gives a utility of 0.78 for responders and 0.66 for non-responders. The model assumes response status is reached at once and held for one year, and the new treatment costs GBP 600 more per patient over that year.

1. Expected utility in each arm. For the new treatment, 0.60 × 0.78 = 0.468 and 0.40 × 0.66 = 0.264, a total of 0.732. Under standard care, 0.40 × 0.78 = 0.312 and 0.60 × 0.66 = 0.396, a total of 0.708.

2. QALY gain per patient. Over one year the gap between 0.732 and 0.708 is 0.024 QALYs, which the general form gives directly as follows.

$$\Delta Q = (p_T - p_C)(u_R - u_N)\, t = (0.60 - 0.40)(0.78 - 0.66) \times 1 = 0.20 \times 0.12 = 0.024$$

where $\Delta Q$ is the QALY gain per patient, $p_T$ and $p_C$ are the responder proportions with the new treatment and standard care, $u_R$ and $u_N$ are the utilities of responders and non-responders, and $t$ is the time in years for which response is held.

3. Cost per QALY. GBP 600 / 0.024 = GBP 25,000 per QALY gained.

4. Change the utility source. If EQ-5D had not been collected and a mapping function gave utilities of 0.76 and 0.68, the gap would be 0.08. The gain falls to 0.20 × 0.08 = 0.016 QALYs and the ratio rises to GBP 600 / 0.016 = GBP 37,500 per QALY.

5. Interpretation. A 20 percentage point difference in responders, one extra responder for every 5 patients treated, buys only 0.016 to 0.024 QALYs a year, because the COA difference reaches the QALY only through the utility gap between states. The decision can turn on whether utilities were measured or mapped, whether response lasts, and whether the responder threshold reflects change patients regard as meaningful.

Boundaries with nearby concepts

A COA is one kind of outcome measure; outcome measures also include survival, hospital use and costs. The clinical outcome is what is being measured, the COA is the tool, and the endpoint is the analysed variable. The EQ-5D is completed by patients and so is patient-reported, but its preference weights make it a utility instrument, which most disease-specific COAs are not.

Where COA-based claims usually go wrong

Weak COA claims often rest on a familiar instrument rather than on evidence for the stated use. The errors below recur in trial reports and in the economic submissions built on them.

• A familiar instrument is validated in general. Fitness for purpose holds only for a stated concept and context of use, so a measure developed in adults with mild disease needs fresh evidence for children or severe disease.
• A proxy answer counts as a PRO. A carer answering as if they were the patient produces a proxy report, which the FDA does not treat as a PRO.
• A score change equals a utility change. A COA difference becomes a QALY gain only through a preference-based valuation, and mapped values add uncertainty that belongs in sensitivity analysis.
• Capacity equals daily life. A PerfO such as a walk test measures capacity on a set task, which may cover only one part of the aspect of health that matters.
• Death is a missing data problem. Death is an intercurrent event, and the estimand must say how it enters the endpoint before the analysis is chosen.

Sources

• Food and Drug Administration. Guidance for Industry. Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims. Silver Spring, MD: FDA; December 2009. https://www.fda.gov/media/77832/download
• Food and Drug Administration. Patient-Focused Drug Development: Incorporating Clinical Outcome Assessments Into Endpoints for Regulatory Decision-Making. Draft guidance (PFDD Guidance 4). Silver Spring, MD: FDA; April 2023. https://www.fda.gov/media/166830/download
• Food and Drug Administration. Patient-Focused Drug Development: Selecting, Developing, or Modifying Fit-for-Purpose Clinical Outcome Assessments. Guidance (PFDD Guidance 3). Silver Spring, MD: FDA; October 2025. https://www.fda.gov/media/159500/download
• Food and Drug Administration. Clinical Outcome Assessment (COA): Frequently Asked Questions. Content current as of 2 December 2020, accessed 9 October 2026. https://www.fda.gov/about-fda/clinical-outcome-assessment-coa-frequently-asked-questions
• International Council for Harmonisation. ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials to the Guideline on Statistical Principles for Clinical Trials. Step 4, 20 November 2019. Section A.3.2. https://database.ich.org/sites/default/files/E9-R1_Step4_Guideline_2019_1203.pdf
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 31 March 2026. Section 4.3. https://www.nice.org.uk/process/pmg36
• Walton MK, Powers JH, Hobart J, Patrick D, Marquis P, Vamvakas S, Isaac M, Molsen E, Cano S, Burke LB. Clinical outcome assessments: conceptual foundation. Report of the ISPOR Clinical Outcomes Assessment Emerging Good Practices for Outcomes Research Task Force. Value in Health. 2015;18(6):741-752. https://doi.org/10.1016/j.jval.2015.08.006