A hospital considers replacing an established diagnostic test with a promising new one. The manufacturer reports excellent accuracy, but the hospital also needs to know which patients will benefit, how many false alarms the test produces, whether clinicians can act on its results and what the whole service will cost. A single number cannot answer those questions. Health Technology Assessment brings them into one accountable decision.
The gold standard of health technology assessment is a transparent comparison of clinical benefit, safety, economics, equity, organisation and ethical consequences for a defined population. This guide uses eighteen practical controls, two fictional numerical cases, FAQs and the linked eduKate health and evidence library to explain how responsible appraisal works.
This subject joins The Gold Standard. It is general education and does not constitute medical advice, formal HTA approval, a purchasing recommendation or regulatory certification.
Quick information: five questions every HTA must answer
- What decision and patient population are under review?
- Which relevant existing treatment, test or service is the comparator?
- What are the demonstrated benefits, harms and evidence gaps?
- What resources, equity effects and delivery constraints are involved?
- Which organisation has the authority to appraise, approve and revisit the decision?
Chapter index
- 1. Begin with a decision that a health system must make
- 2. Distinguish HTA from regulatory marketing authorisation
- 3. Scope the population, intervention and comparator
- 4. Choose outcomes that matter to patients
- 5. Build the clinical evidence comparison transparently
- 6. Separate efficacy in trials from effectiveness in practice
- 7. Understand harm, adverse events and safety uncertainty
- 8. Assess diagnostic accuracy as part of a pathway
- 9. Check medical devices, operators and learning curves
- 10. Treat software and clinical AI as changing technologies
- 11. Define perspective and resources before economic modelling
- 12. Distinguish cost-effectiveness from budget impact
- 13. Keep QALYs and thresholds in their proper setting
- 14. Examine distributional effects and access
- 15. Include the organisation that must deliver the service
- 16. Maintain transparency, governance and conflict-of-interest controls
- 17. Understand WHO, EU and national HTA roles
- 18. Plan reassessment when evidence and technology change
1. Begin with a decision that a health system must make
Health technology assessment, or HTA, systematically evaluates the effects and implications of health technologies and interventions to inform decisions. A health technology may be a medicine, device, diagnostic, procedure, digital service or organisational system. The question is rarely whether the product is exciting; it is whether it serves a defined population better than feasible alternatives.
The World Health Organization describes HTA as multidisciplinary, encompassing economic, social, organisational and ethical issues as well as clinical evidence. The gold standard defines the decision, comparator, expected benefits and harms, implementation demands and evidence gaps before seeking a favourable conclusion.
2. Distinguish HTA from regulatory marketing authorisation
Product regulators and HTA bodies can ask overlapping but different questions. Regulatory processes may focus on evidence needed to permit marketing and safe use of a product within their mandates. HTA asks how a technology compares with available options and what using it would mean for health systems, patients and resources.
A device’s clearance or a medicine’s authorisation does not automatically establish better value than standard care or guarantee public reimbursement. Likewise, an economic evaluation cannot replace safety and regulatory obligations. Specify which authority is responsible for each decision and do not use one institution’s approval as a substitute for another’s assessment.
3. Scope the population, intervention and comparator
The population must reflect who could actually receive the technology. Define the condition, disease severity, age, previous treatments and relevant setting. Specify the intervention accurately, including dose, device configuration, clinician training or software version. Select comparators representing credible current practice rather than an artificially weak alternative.
A new test compared only with no testing may seem impressive while offering little advantage over an established diagnostic strategy. Record the decision problem using a consistent framework such as population, intervention, comparator and outcomes, and identify which choices vary across clinical settings.
4. Choose outcomes that matter to patients
A biomarker change, faster workflow or improved image resolution can be useful but does not automatically mean a person lives longer or better. Relevant outcomes may include survival, symptoms, quality of life, adverse effects, diagnostic certainty, daily functioning and caregiver burden.
Link intermediate endpoints to patient-important consequences with evidence. State the timeframe and whether the benefit was measured directly or estimated through a model. If a technology saves staff time but provides no measured patient outcome, report its operational value without inventing clinical benefit.
5. Build the clinical evidence comparison transparently
Randomised trials, pragmatic studies, registries and observational analyses can contribute different evidence. Systematic review aims to find and critically compare relevant studies rather than selecting the most positive result. Risk of bias, missing outcomes, selective reporting and differences in follow-up can affect conclusions.
Preserve search criteria, inclusion rules, study features and reasons for exclusion. Different populations or comparators may require separate interpretation. A statistically significant difference is not automatically clinically important; absolute changes and uncertainty should be shown beside relative measures.
6. Separate efficacy in trials from effectiveness in practice
A technology may work under highly supervised trial conditions yet perform differently in routine care. Patients may vary in adherence, comorbidity, access and follow-up. Operators may differ in experience, and real-life systems may have less time or fewer staff than a research setting.
Assess how trial evidence transfers to the intended service. Identify additional training, monitoring or support needed to preserve the benefit. Do not dismiss well-conducted trials, but do not assume that their idealised delivery is the ordinary operating model for every clinic.
7. Understand harm, adverse events and safety uncertainty
Benefits need to be weighed against possible harms. The risk of adverse events depends on the intervention, population, duration and the ability to recognise and respond to problems. Studies with limited sample size or short follow-up may fail to detect rare or delayed harms.
Report adverse events by appropriate denominators, definitions and severity. For devices and digital products, consider misuse, malfunction, cybersecurity, data errors and human factors where relevant. A favourable average outcome can still coexist with risks concentrated in vulnerable groups. Safety claims require context and ongoing evidence.
8. Assess diagnostic accuracy as part of a pathway
A diagnostic test can have strong sensitivity and specificity yet produce substantial false positives or miss important cases, depending on disease prevalence and the threshold used. Accuracy is not equivalent to clinical utility. What matters is how test results change decisions, treatment, harms and outcomes.
Specify intended use, reference standard, population and the consequences of each result. A test that finds a condition earlier may create extra follow-up or overdiagnosis. Health technology assessment must examine the full pathway, not solely the accuracy statistic provided by a manufacturer.
9. Check medical devices, operators and learning curves
The performance of a medical device may depend on the skill of its operators, calibration, consumables, maintenance, service interruptions and training. Device generations can change rapidly, making evaluation more complicated than assessing a fixed pharmaceutical product.
Record the technology version and real implementation conditions. Compare learning curves and training requirements, and identify when early results may be affected by inexperience. Avoid evaluating a highly skilled demonstration as though the same outcomes were established throughout a health network.
10. Treat software and clinical AI as changing technologies
Digital health services and AI systems can change through updates, data drift and integration with different record systems. A retrospective accuracy score is insufficient to show clinical benefit or safety in the intended workflow. Models may behave differently when populations, codes, language or staffing patterns change.
Document the intended use, version, validation population, false alerts, human oversight, data governance and monitoring plan. If a decision-support algorithm is updated, determine what evidence needs reassessment. Automation should not obscure who is accountable for patient-relevant decisions.
11. Define perspective and resources before economic modelling
Economic comparisons require a viewpoint: hospital, payer, health system or society. Those perspectives may count different resources, such as staff time, equipment, travel, caregiver time and productivity. Prices and actual opportunity costs may differ, and both can change over time.
Record resource quantities separately from their valuations, including price year and currency. Capital costs, consumables and future maintenance belong in an explicit time horizon. An impressive unit-price discount may be outweighed by expensive implementation, training or repeated testing. The economic question is wider than purchase price.
12. Distinguish cost-effectiveness from budget impact
An intervention may offer a favourable expected relationship between costs and health benefits over a long horizon while requiring more near-term funding than a system can provide. Cost-effectiveness, affordability and available service capacity are separate considerations.
Estimate eligible populations, uptake, unit costs, timing and possible offsets. Avoid multiplying a best-case per-person benefit across everyone without modelling eligibility and access. Local reimbursement decisions also include policy, clinical and ethical considerations that cannot be reduced to a single ratio.
13. Keep QALYs and thresholds in their proper setting
Quality-adjusted life years combine life duration and health-related utility under specified valuation methods, helping compare otherwise different health effects. Such measures carry assumptions about health states, instruments, discounting and whose values are represented. They do not express the total moral value of a person.
NICE’s technology-appraisal manual was last updated on 31 March 2026 and describes methods used in its UK setting. Its thresholds and procedural rules should not be imported automatically into Singapore or another jurisdiction. State the actual local decision framework and the uncertainty surrounding modelled outcomes.
14. Examine distributional effects and access
A technology may improve outcomes on average while being less accessible to rural patients, lower-income groups or those needing language and disability accommodations. Subgroup evidence may be limited and cannot always support precise separate estimates. Nonetheless, unequal distribution deserves explicit examination.
Describe who is expected to benefit, who bears costs, and what arrangements would support equitable delivery. Patient and caregiver experience may reveal constraints not visible in an efficacy chart. An assessment should avoid claiming fairness solely because every eligible person theoretically meets identical written rules.
15. Include the organisation that must deliver the service
A technology may require specialist staff, laboratory capacity, space, equipment integration, digital records, referral systems and aftercare. If those are unavailable, estimated clinical benefits may not be achievable. Opportunity cost includes staff diverted from other needed services.
Map the actual care pathway from referral to follow-up. Test throughput, training, waiting times and failure modes. An innovative scan that produces results faster than clinicians can review them may shift the bottleneck rather than improve the patient’s overall experience.
16. Maintain transparency, governance and conflict-of-interest controls
Technology assessment can involve manufacturers, patients, clinicians, payers, researchers and public agencies with legitimate but differing interests. Financial relationships, preferred outcomes and selective evidence can weaken trust unless disclosure and independent review are built into the process.
Publish methods, evidence selection, key assumptions and reasons for recommendations within lawful confidentiality boundaries. Give affected stakeholders meaningful participation without allowing any one group to determine scientific conclusions. The WHO’s institutionalisation work emphasises governance arrangements, transparent processes and monitoring.
17. Understand WHO, EU and national HTA roles
WHO provides guidance on building HTA capability and making transparent decisions. Within the European Union, Regulation (EU) 2021/2282 has applied since 12 January 2025, introducing joint clinical assessment for certain technologies on a staged basis. From that date, certain new cancer medicines and advanced therapy medicines entered the initial scope.
An EU joint clinical assessment provides a shared scientific analysis of relative clinical effects but does not itself set each Member State’s medicine price, reimbursement or national use decision. The European Commission reported further developments for selected medical devices during 2026. NICE operates a distinct UK appraisal framework. These arrangements cannot be conflated into one worldwide approval system.
18. Plan reassessment when evidence and technology change
An HTA report is an assessment at a time, using an identified evidence base. Further trials, real-world outcomes, price changes, software versions or new comparators can alter the decision. A strong process therefore includes conditions for review and evidence collection after adoption.
Specify what new information would materially change the recommendation and who must monitor it. The transferable skill is to link benefits, harms, costs, equity and deliverability without mistaking a model’s output for certainty. A decision can be responsible today while remaining explicitly revisable tomorrow.
Worked fictional case A: diagnostic sensitivity is not predictive value
In an imaginary population of 1,000 tested people, 100 truly have a defined condition and 900 do not. Assume a test has 90% sensitivity and 95% specificity in this fictional setting. Of the 100 affected people, 90 test positive and 10 test negative. Of the 900 unaffected people, 855 test negative and 45 test positive. There are therefore 90 true positives and 45 false positives among the 135 positive results.
The positive predictive value is 90 ÷ 135 = approximately 66.7%, not 90%. The negative predictive value is 855 ÷ (855 + 10) = approximately 98.8%. These invented results show why sensitivity, specificity and prevalence must be interpreted together. Whether the test improves health still depends on the downstream care pathway and consequences of false and missed results.
Worked fictional case B: cost-effectiveness and first-year affordability differ
Consider two imaginary alternatives under a single assumed payer perspective. Existing care costs 5,000 currency units per eligible person and produces 1.00 expected QALY over a declared horizon. A new intervention costs 9,000 units and produces 1.25 expected QALYs. The added cost is 4,000 and the added health gain 0.25 QALY, giving a simplified incremental cost-effectiveness ratio of 4,000 ÷ 0.25 = 16,000 currency units per additional QALY.
Suppose the first year has 600 eligible people and only half are expected to receive the new intervention. Under the simplified added net cost of 4,000 per treated person and ignoring additional start-up costs, first-year budget impact for 300 users is 1,200,000 currency units. A favourable ICER under a particular decision framework does not automatically supply that budget, or prove clinical feasibility. Every value here is fictional, and no real reimbursement threshold is implied.
Health technology assessment audit
- Define intervention, eligible patients and a credible comparator.
- Check trial quality, effect size and patient-important outcomes.
- Separate safety, diagnostic accuracy and practical clinical utility.
- State technology version, user training and organisational dependencies.
- Reconstruct costs, perspectives, outcome definitions and uncertainty.
- Distinguish long-term cost-effectiveness from near-term budget impact.
- Assess fair access, patient and caregiver participation and ethical impacts.
- Identify jurisdiction-specific decisions, conflicts and reassessment triggers.
Frequently asked questions
What is health technology assessment?
A systematic multidisciplinary evaluation of a health technology’s effects and implications to inform health-system decisions.
Does HTA only assess medicines?
No. It can examine devices, diagnostic tests, procedures, digital systems and other interventions.
Is an authorised product automatically cost-effective?
No. Regulatory authorisation and comparative health-system value are different questions.
What is the comparator in HTA?
The feasible current or alternative care pathway against which a technology is evaluated.
Why do patients matter to HTA?
Patients can identify outcomes, burdens and access problems that standard clinical or financial measures overlook.
Is sensitivity the chance a positive test result is correct?
No. Sensitivity is the probability of a positive result among people with the target condition; positive predictive value uses a different denominator.
Can a diagnostic test improve accuracy but not health outcomes?
Yes. A more accurate result needs a beneficial, feasible pathway to influence patient care.
What is the difference between cost-effectiveness and budget impact?
Cost-effectiveness compares value relative to alternatives over a horizon; budget impact estimates financial consequences for a payer over specified periods.
Is a QALY the value of a human life?
No. It is a defined health-outcome measure used in some economic evaluations and does not capture every ethical concern.
Do EU joint clinical assessments determine national reimbursement?
No. EU joint assessments support national HTA, while Member States retain national use, pricing and reimbursement responsibilities.
Can an AI medical tool be assessed once and used forever?
Not necessarily. Updates, drift and new settings may require further validation and review.
What makes HTA transparent?
Clear questions, evidence selection, assumptions, uncertainty, stakeholders, conflicts and reasons for conclusions.
Connect the health, economics and evidence library
Health economics studies resource trade-offs; epidemiology and biostatistics examine outcomes and uncertainty; experimental design helps identify causal effects; health informatics deals with meaningful records; and decision theory clarifies risk. HTA combines their contributions while keeping the final policy decision with the relevant authority.
- The Gold Standard
- The Gold Standard of Health
- The Gold Standard of Health Economics
- The Gold Standard of Health Informatics
- The Gold Standard of Epidemiology
- The Gold Standard of Biostatistics
- The Gold Standard of Experimental Design
- The Gold Standard of Decision Theory
- The Gold Standard of Statistics
- The Gold Standard of Information Security
- How X Works hub
Current international frameworks and authoritative sources
- WHO: Health Technology Assessment, Global Overview
- WHO: Health Technology Assessment, Assessment Scope
- WHO: Institutionalising Health Technology Assessment
- NICE: Technology Appraisal Manual (31 March 2026 Update)
- European Commission: EU Health Technology Assessment Overview
- European Commission: Joint Clinical Assessments for Medicinal Products
- European Commission: Joint Clinical Assessments for Medical Devices
The WHO, NICE and European Commission materials were reviewed on 24 October 2026. NICE’s manual was updated on 31 March 2026 for its UK process. EU joint clinical assessment operates under a separate framework applying from 12 January 2025; national decisions about coverage, pricing and reimbursement remain distinct. These are not worldwide or Singapore-specific approval rules. Every test result, cost and QALY in the worked examples is invented solely for education.
Return to chapter index · The Gold Standard hub · How X Works hub