Health economic evaluation

Health economic evaluation compares alternative courses of action by their costs and consequences. Its results depend on the costs counted, the outcomes measured, the perspective and time horizon, and the assumptions built into the analysis.

Purpose and related disciplines

Health economics studies the efficient and effective allocation of health resources, usually focusing on interventions' costs and consequences. ^{1, 2, 3} An economic evaluation compares alternative courses of action in terms of both costs and consequences. ^{4, 2, 5} The usual premise is a fixed healthcare budget that should maximise the health of a nation's population or an insurance plan's members. ^{6, 1} Opportunity cost is then health forgone: spending on expensive care with questionable clinical benefit leaves less for other patients, who may consequently suffer or die. ⁶ England's National Health Service (NHS) can technically provide more care than it can afford, making choices unavoidable. ⁴

Pharmacoeconomics is the part of health economics concerned solely with pharmaceutical products. ² Health technology assessment (HTA) is broader, assessing drugs, devices, procedures, diagnostics and healthcare-delivery systems, processes and programmes to inform provision and policy decisions. ² Economic evaluations are important components of HTA programmes internationally. ⁵

Medical economics and health economics are synonymous, with health economics the standard term. ⁷ The American Medical Association established its Bureau of Medical Economics in 1931, although the discipline's beginning is usually attributed to Kenneth Arrow's 1963 comparison of healthcare economics with other goods and services. ⁷ Michael Grossman's health-production model treats individuals as producers and consumers of health, a form of capital that deteriorates without investment. ⁷ Health is both a consumption good giving direct satisfaction, such as better quality of life, and an investment good yielding productivity, fewer sick days and higher wages; investing takes time and resources away from other goals. ⁷ This theoretical approach has influenced practical economic analyses. ⁷

Health economics and outcomes research

Health economics and outcomes research (HEOR) concerns what matters to payers and patients in real practice, what differentiates an intervention and what justifies its cost. ^{3, 8} Its outcomes can be subjective, such as how a patient feels, or measured, such as blood-pressure change. ³ Outcomes research considers three types of outcome. ^{3, 2}

• Clinical outcomes: effectiveness, treatment outcomes and lives saved.
• Economic outcomes: direct and indirect costs, spent or saved.
• Humanistic outcomes: health-related quality of life, treatment preferences, patient-reported or preference-based outcomes, and utilities such as quality-adjusted life-years.

^{3, 2}

Outcomes research includes comparative effectiveness research and patient-reported outcomes, the latter evaluating how patients feel and function, often with standardised questionnaires. ⁹ Effectiveness means performance in an uncontrolled real-world situation, while healthcare resource utilisation counts services used, such as visits, hospitalisations, laboratory tests and drugs. ¹⁰

Evidence sources

Sources of HEOR evidence include the following. ¹¹

• Real-world data, including claims-database analyses, chart reviews, surveys and observational studies.
• Clinical outcome assessments.
• Patient-reported outcomes.
• Systematic reviews.
• Meta-analyses.
• Clinical studies, including phase 3, post-marketing, phase 4 and open-label studies.
• Economic analyses, including models.

¹¹

Registries are another source, and several study types are generally needed across a product's life cycle. ^{8, 12} Real-world evidence receives emphasis because randomised trials may not represent real-life outcomes. ¹²

Reasons for generating evidence

Academic institutions, pharmaceutical companies, health plans, healthcare professionals and others use HEOR evidence. ⁸ Reasons for generating it include the following. ¹³

• Identify unmet needs.
• Address evidence gaps.
• Supplement randomised trial data with real-world data.
• Promote patient-centred research and provide evidence about what actually happens to patients.
• Help develop and evaluate cost-containment strategies.
• Respond to changes in market environments.
• Adapt to regions and affiliates, healthcare decision makers, populations and local treatment patterns.
• Meet HTA submissions' additional economic-information requirements compared with regulatory submissions.

^{13, 8}

Clinical trials cannot answer many real-world questions, and HEOR evidence supports clinical guidelines, stakeholder decisions and demonstrations of product value. ¹⁴ Its uses include the following. ¹⁵

• Evaluate real-world outcomes outside controlled trials.
• Allow treatment comparisons unavailable directly from clinical trial data.
• Inform selection of appropriate, cost-effective therapy.
• Support evidence synthesis for direct or indirect comparisons of efficacy and safety.
• Inform clinical guideline development.
• Evaluate product cost, budget impact and cost-effectiveness.
• Illustrate product value.
• Guide formulary coverage and reimbursement decisions.

¹⁵

The product life cycle

In the United States in 2018, pharmaceutical companies were the most common users of HEOR evidence. ⁸ HEOR applications span the product life cycle. ¹⁶

Phase	Applications
Preclinical	Exploratory research, gap analysis, very early modelling, market assessment and key opinion leader research
Phases 1 and 2	Market assessment, early modelling, piggyback studies, gap analysis, burden-of-illness studies and patient-reported outcome development, testing and validation
Phases 3 and 3b	Payer assessment, registries, pricing and reimbursement, model development and validation, comparative effectiveness research, Academy of Managed Care Pharmacy dossiers, piggyback studies, value messages and global value dossiers
After launch	Phase 4 studies, comparative effectiveness research, retrospective studies, Academy of Managed Care Pharmacy dossiers, model refinement, database analyses, global value dossiers, piggyback studies, prospective observational studies, chart reviews, HTA, safety surveillance and registries
Loss of exclusivity	Safety surveillance, comparative effectiveness research, HTA, real-world studies and global value dossiers

¹⁶

Pharmacy and therapeutics committees in hospitals or insurance companies weigh clinical and economic outcomes when choosing drugs; members often include clinicians, a lawyer, quality-assurance staff and a lay member. ¹⁶ A hypothetical comparison illustrates the information considered. ^{17, 18}

Drug	Dose	Cost per tablet	Efficacy	Side effects
A	One tablet daily	$1.25	77%–80%	Nausea, vomiting and irreversible hepatotoxicity (5%)
B	One tablet daily	$1.50	78%–80%	Nausea and headache

^{17, 18}

Cost, price and value

Cost is the money used to produce something. ² For medicines it includes research and development for marketed and unsuccessful products, manufacturing, distribution through intermediaries and marketing; these costs are partly fixed for a product and influence price. ² Price is what is paid, often differing from cost and reflecting value to the buyer. ² A product can have several prices: in the UK in 2013, ex-manufacturer price meant purchase from the manufacturer, ex-wholesale price meant pharmacies' purchase from wholesalers, and public or list price meant cost to the public purse. ² Novelty, development costs, competition and worldwide alternatives help determine price. ²

Value is whether something is worth its price, a judgement that differs among payers, prescribers and patients. ² Greater perceived value can support a higher price without changing costs, and the payer is often neither the care provider nor its recipient, so stakeholders' differing valuations need consideration. ²

Cost categories

Economic analyses distinguish these cost types. ²

• Direct medical costs: treatment-related costs, such as medicine, physician visits and hospital stays.
• Direct non-medical costs: costs of using treatment outside the treatment itself, such as transport and childcare.
• Indirect costs: time lost to illness, including lost productivity.
• Intangible costs: effects on health and quality of life that are difficult to measure, including poor health or lost social activity.
• Opportunity costs: the alternative forgone to pursue the chosen intervention.
• Incremental costs: additional total costs compared with an alternative.
• Marginal costs: additional or reduced costs from slight treatment changes.

²

Examples make the first four categories more concrete. ^{19, 20}

Category	Examples
Direct medical	Medications, monitoring and administration; patient counselling and consultation; diagnostic tests; hospitalisations; clinic and emergency-department visits; home medical visits; ambulance and nursing services
Direct non-medical	Travel by bus, car or taxi; condition-related assistance such as meal delivery or homemaking; out-of-town hotel stays for patients or families; childcare
Indirect	Lost productivity of patients and unpaid carers, and productivity lost through premature mortality
Intangible	Pain and suffering, fatigue and anxiety

¹⁹

Types of analysis

A benefit is a favourable treatment outcome. ² Intermediate endpoints are markers of benefit, such as the blood-pressure reduction on antihypertensive therapy; final outcomes are the end results of treatment, measured in clinical, economic and humanistic terms. ²

Five forms of economic evaluation differ in how costs and benefits are expressed. ²¹

1. Cost-effectiveness analysis. The most common form compares treatments for the same condition, measuring costs in money and benefits in final outcomes or natural units such as life-years gained or symptom-free days. Costs and effectiveness form a ratio; a lower cost per unit of effectiveness is preferred where resources are limited.
2. Cost–benefit analysis. Costs and benefits are both monetary, allowing explicit comparisons; an intervention whose benefits exceed costs is considered worthwhile.
3. Cost-minimisation analysis. Equivalent health outcomes are assumed, and the comparison identifies the least costly way to achieve them.
4. Cost-utility analysis. This form of cost-effectiveness analysis uses utilities, numerical values for health improvements. Quality-adjusted life-years are the most common unit; healthy-years equivalents and disability-adjusted life-years are also used.
5. Cost-consequence analysis. Costs and consequences are computed and tabulated separately rather than combined in one ratio, allowing decision makers to select the components relevant to them.

^{21, 9}

Budget impact analysis additionally estimates an intervention's overall cost to an organisation or health plan, typically the net cumulative cost for a specified number of patients in a defined population. ^{22, 9} Practical examples illustrate these analyses. ²²

Analysis	Example
Cost-consequence	Compare drugs preventing transplant rejection using drug, rejection-treatment, graft-survival and dialysis costs and the probability of consequences, leaving conclusions to the reader
Cost-minimisation	Compare hospital costs of inpatient and outpatient care after prostaglandin E2 gel to stimulate labour, assuming equivalent outcomes
Cost-effectiveness	Compare three ulcer treatments by annual costs and healing rates, potentially using gastrointestinal symptom-free days to make the results meaningful to clinicians
Cost-utility	Compare oncology drugs by cost, months or years gained and effects on quality of life, incorporating patient outcomes and preferences
Cost–benefit	Compare a vaccination programme's monetary costs and benefits with not implementing it
Budget impact	Compare the cumulative direct costs of treating a defined number of multiple myeloma patients with one drug or another

²²

In 2013, NICE technology appraisals and diagnostics guidance used cost-effectiveness analysis, usually measuring effectiveness in quality-adjusted life-years with instruments such as EQ-5D and costs as NHS and social-services expenditure. ⁴ Its medical technologies programme used cost-consequence models because it asked whether greater benefit could be achieved at the same or lower cost, or the same benefit at lower cost. ⁴

Quality-adjusted life-years and utilities

Quality-adjusted life-years (QALYs), typically used in cost-utility analysis, combine gains or losses in length and quality of life, with one year in perfect health worth one QALY. ^{23, 9} This common measure permits comparisons across studies, interventions and diseases for resource allocation, but translates poorly to day-to-day individual decisions. ^{23, 10}

Utilities measure patients' or other stakeholders' preferences for outcomes, typically from 0 for death to 1 for perfect health, although values below zero can represent states worse than death. ^{24, 25} For example, patients facing bilateral lower-limb amputation might value their state at 0.2, while patients with moderate seasonal allergies might assign 0.9. ²⁴ A treating physician's perspective can also be used where relevant. ²⁴ Utilities quantify qualitative preferences while incorporating positive and negative effects on health and quality of life. ²⁴ QALYs weight quality by time: five years at utility 0.8 yield 4.0 QALYs, compared with 3.6 QALYs for six years at 0.6. ²⁵

A 2017 example of value thresholds treated drugs below $150,000 per QALY as typically good value, while allowing public-health benefit, disease severity and societal valuation to influence the decision. ¹⁰

Incremental ratios and willingness to pay

The incremental cost-effectiveness ratio (ICER), used in cost-effectiveness and cost-utility analyses, compares additional costs with additional benefits when switching treatments. ^{21, 26} It is calculated as (cost of A − cost of B) divided by (benefit of A − benefit of B). ^{21, 26} If it lies within the payer's or provider's acceptable range, there is little reason to reject treatment on cost-effectiveness grounds. ²¹ Positive ratios require other factors to establish which treatment is preferable in the circumstances. ²⁶

A treatment that is less expensive and more effective dominates its comparator, so willingness-to-pay cut-offs do not apply. ^{26, 25} Willingness-to-pay thresholds apply to discrete positive incremental cost-effectiveness or cost-utility ratios and differ among societies. ²⁵

Perspective, discounting and sensitivity analysis

Perspective identifies in advance which costs, or whose costs, an analysis measures, according to the study's purpose. ^{27, 22} A payer perspective considers direct medical costs accruing to an insurer or national healthcare system over the short or long term; a hospital perspective typically considers direct treatment costs and preventable readmissions. ²⁷ A societal perspective adds direct and indirect costs borne by patients and society, such as copayments and lost productivity, and is the most comprehensive approach. ²³ It is most applicable to nationalised healthcare and was not widely used in US studies in 2013. ²³

Discounting adjusts future costs and benefits, particularly beyond one year, to present economic value and is common in multiyear analyses. ²² In 2013, future costs were typically discounted by 3%–5% annually; discounting differs from the upward inflation adjustment also common in retrospective studies lasting more than a year. ²²

Sensitivity analysis examines uncertainty by changing selected assumptions and observing their effect on overall results, a widely accepted way to assess a result's strength. ²² For example, an analysis assuming 3% mortality can be repeated at 1% and 5%, and assumptions about drug costs, hospital-stay length or treatment duration can also be varied. ²²

Models

Economic research depends heavily on modelling, sometimes requiring dedicated computer software. ²⁸ A decision tree is a common first step, systematically comparing decisions and subsequent patient outcomes. ²⁸ One pressure-ulcer example compares two treatments, with epithelialised, proliferative or necrotic states immediately after treatment and later closure, granulation or chronic wounds; a more detailed tree would give final-outcome probabilities on its branches. ²⁸ Decision trees suit limited treatment choices, treatment duration and timeframes, with manageable numbers of states. ²⁸

Markov models are more appropriate for complex situations such as multiple diabetes treatments over time, ongoing risk, multiple outcomes, uncertain event timing or recurrent events such as myocardial infarction. ²⁹ They use a theoretical cohort, often from trial data, with each patient always in one of finitely many health states and events represented by transitions; the model usually ends at death or another predefined endpoint. ²⁹ A simple example permits transitions from asymptomatic disease to progression to death, and back and forth between asymptomatic and progressive disease. ²⁹

A budget impact model compares introducing an intervention with a scenario without it, frequently over five years. ²⁵ Population size and disease profiles can change, for example through lower infection incidence after vaccination or lower chronic-disease prevalence after more effective treatment. ²⁵ Cohort- or patient-level models can account for entry and exit, changes in case mix and disease severity, and resulting management costs. ²⁵

Key budget impact model elements include the following. ²⁵

• Restricted access to therapies.
• All therapy costs and projected changes.
• Time horizon.
• Health-plan characteristics, including eligible population size and current and projected case mix.
• Model inputs.
• Model outputs.
• Ratification or validation, including decision makers' assessment of face validity and verification of calculations.
• Projected uptake of the new intervention and changes in existing therapies' use.
• Sensitivity analyses varying key assumptions such as discount rates in alternative scenarios selected from the budget holder's perspective.
• Type of sensitivity analysis, including deterministic or probabilistic.

²⁵

A hypothetical five-year budget model for a treatment called spastex produced lower per-patient and per-member annual costs than its reference scenario. ²⁵ Its tornado plot varied cost, adherence and eructation-rate assumptions by 5%, with treatment cost showing the largest displayed influence. ²⁵

Models should present familiar, relevant outcomes and let budget holders vary assumptions, often through spreadsheet dropdown menus, to explore their effects on outputs. ²⁵ They should be tested with alternative software platforms or versions because users will use different applications. ²⁵

Data inputs and registry examples

Accuracy depends on data quality and applicability, with clinical outcomes, associated costs and patient preferences the primary inputs. ²⁴ Clinical outcomes and costs often come from trials of the treatment being assessed, and economic evaluations may run alongside those trials; otherwise inputs can come from published research, existing databases or expert opinion when evidence is limited. ²⁴ Determining whether cost data generalise to the population being assessed is a major difficulty. ²⁴ Other investigators' trials, medical records, reimbursement claims, standard diagnosis-related-group reimbursement information and hospital cost databases can supply inputs. ²⁴ In 2013, the Healthcare Cost and Utilization Project's free HCUPnet query system supplied diagnosis- and procedure-specific US inpatient and emergency-department data and associated charges and costs. ²⁴

Retrospective database analysis was described in 2013 as the most common format for HEOR results: payer and other databases track routine-care utilisation in large populations over time at relatively low cost. ²² Data can be hard to obtain and require numerous assumptions. ³⁰

Registries often supply cost-utility, cost-effectiveness and other pharmacoeconomic analyses, as well as disease-burden and healthcare-utilisation information. ^{31, 32} In an adult hypopituitarism registry analysis, treatment became linearly more cost-effective as women's quality-of-life impairment increased; in men there was a large change between mild and moderate impairment and almost none between moderate and severe impairment. ^{32, 33, 34}

A paediatric growth registry distinguished expected responders, high responders and children not responding as expected. ³⁴ High responders could potentially receive lower doses at lower cost without losing response, while poor responders first needed checks on whether growth hormone was being injected before considering whether treatment made sense for genuinely non-responding patients. ³⁵ Modelling in German patients showed that individualising treatment this way could save 7% of the total growth-hormone budget. ³⁶