NICE Reference Case

What it is

— The NICE Reference Case is the fixed methodological specification that the National Institute for Health and Care Excellence (England) requires for the base-case economic evaluation in its technology appraisals and highly specialised technologies (HST) evaluations. It is maintained by NICE and lives as Chapter 4 ("Economic evaluation") of NICE health technology evaluations: the manual (process guide PMG36, published 31 January 2022, superseding the 2013 Guide to the methods of technology appraisal, PMG9). The reference case is not a free-standing checklist or a reporting tool; it is a normative methods standard that pins down the analytic choices most likely to drive a cost-effectiveness result so that the incremental cost-effectiveness ratio (ICER) submitted for one technology is comparable with the ICER for every other. Its core elements: the perspective on outcomes is all direct health effects for patients and, where relevant, carers; the perspective on costs is the NHS and Personal Social Services (PSS) (productivity and broader societal costs are excluded from the base case); the time horizon is long enough to capture all material differences in cost and outcome (usually lifetime when survival is affected); health gain is measured in quality-adjusted life years (QALYs) with the EQ-5D as the preferred utility instrument valued by a UK general-population tariff; costs and effects are discounted at the same annual rate, currently 3.5%; comparators are those defined in the NICE scope; and uncertainty must be characterised through probabilistic sensitivity analysis (PSA). NICE permits departures from the reference case, but only when each departure is justified and the reference-case base case is reported alongside it.

When to use

— Use the reference case whenever you are building or appraising a cost-utility / cost-effectiveness model intended for a NICE HTA decision in England (technology appraisal or HST). It governs the base case: the manufacturer/company submission, the Evidence Review Group / External Assessment Group critique, and the appraisal committee all read the model against it. Decision rules for which framework applies: (1) PMG36 (2022) is the current manual and applies to evaluations started under the integrated process; PMG9 (2013) governed earlier appraisals and is the correct reference for historical or in-flight submissions begun under it — match the manual to the appraisal's timeline, do not retrofit. (2) For reporting an economic evaluation in a journal or as a submission appendix, the reference case is not the reporting instrument — pair it with CHEERS 2022 (the reporting checklist for health economic evaluations). (3) The reference case is NICE-specific: for other agencies use their own reference case (e.g., ICER value framework in the US, CADTH in Canada, PBAC in Australia, the EU Joint Clinical Assessment for relative effectiveness) — methods such as the discount rate and perspective differ by jurisdiction and are not transferable.

What it requires

— The reference case enforces a small set of high-leverage methodological commitments, and for real-world evidence (RWE) feeding the model each one creates concrete obligations: (a) NHS/PSS cost perspective — unit costs and resource use must be costed from an NHS payer view; RWE cost inputs drawn from non-UK or societal datasets must be re-based to NHS/PSS, and the cost concept (all-cause vs disease-attributable, per-patient-per-month vs per-episode) must be stated explicitly. (b) Lifetime horizon — short-horizon RWE almost always requires extrapolation of survival and treatment effect beyond observed follow-up, with the extrapolation method and its uncertainty pre-specified. (c) QALYs via EQ-5D — utilities must map to a UK EQ-5D tariff; mapping algorithms from disease-specific instruments must be justified. (d) Comparators from the scope — the effectiveness contrast must be against the relevant comparator(s), which is exactly where comparative RWE design discipline matters: a defensible effectiveness estimate demands an active-comparator, new-user structure, time-zero alignment, confounding control, and a clearly stated estimand with handling of intercurrent events. (e) Single-arm / external-control evidence (common in HST and oncology) must address comparability of the external control to the trial arm and the attendant selection and confounding biases. (f) Discounting at 3.5% for both costs and effects. (g) PSA to propagate parameter uncertainty into the ICER, plus scenario/deterministic analysis for structural and judgemental assumptions; bias in RWE inputs (measurement error in phenotypes, residual confounding) should be probed with quantitative bias analysis and negative controls rather than assumed away.

When NOT to use — limitations and common misapplications

— (1) It is a methods standard, not a reporting checklist. Completing the reference case does not satisfy reporting expectations; CHEERS 2022 does that, and a clean reference-case base case can still be opaquely reported. (2) A reference-case-compliant model is not automatically credible — compliance fixes comparability of methods, not the validity of the inputs; a model with perfect perspective, horizon, and discounting but a confounded RWE effectiveness estimate produces a precise, comparable, wrong ICER. (3) Presenting departures without the base case. Analysts sometimes submit only their preferred (non-reference-case) analysis; NICE requires the reference-case base case and the justified departure side by side. (4) Wrong manual for the timeline — applying PMG36 to an appraisal conducted under PMG9 (or vice versa) misstates the applicable rules (e.g., the 2022 severity modifier replacing the earlier end-of-life modifier). (5) Importing foreign-perspective or societal costs unadjusted — US claims-derived or societal-perspective costs cannot be dropped into an NHS/PSS base case without re-basing. (6) Treating PSA as optional or as a cosmetic add-on rather than the required characterisation of decision uncertainty. (7) Conflating the reference case with the Budget Impact Test — the BIT (the £20m-per-year commercial-arrangement trigger) is a separate NICE process about affordability, not the cost-utility reference case; budget-impact analysis is governed by its own conventions, not by the reference case.

How it maps to this catalog

— The reference case sets the requirements; these catalog concepts implement the RWE inputs that feed a reference-case model: comparative effectiveness from observational data is built with active-comparator-new-user and formalised as a target-trial-emulation, balanced with propensity-score-methods-psm-iptw and, when confounders are unmeasured, high-dimensional-propensity-score-hdps-rwe; the causal target is specified with estimands-ate-att-intercurrent-events-rwe. Where the evidence is single-arm, see single-arm-external-control, rare-disease-external-controls-rwe, and external-adjustment-validation-substudy-bias-correction-rwe. Endpoints and exposures from claims/EHR are defined with diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe and validated via claims-outcome-algorithm-ppv-sensitivity-rwe; follow-up integrity is governed by attrition-and-loss-to-follow-up-rwe. Cost inputs for the NHS/PSS perspective draw on healthcare-costs-pppm-pppy-pmpm, all-cause-vs-attributable-costs-rwe, and cost-outlier-handling-rwe; the 3.5% discounting requirement is implemented in discounting-costs-effects-rwe; whether the RWE population maps to the NHS decision population is assessed with generalizability-transportability-external-validity-rwe. Residual bias in RWE inputs is stress-tested with quantitative-bias-analysis-toolkit-rwe, negative-control-outcomes-rwe, and empirical-calibration-negative-controls-rwe. Applied note (claims/EHR/registry RWE): when a UK or non-UK claims/EHR source supplies the comparative effectiveness or cost input to a NICE submission, the catalog requirement is to (i) state the estimand and emulate the relevant trial, (ii) re-base costs to NHS/PSS and declare the cost concept, (iii) extrapolate to a lifetime horizon with pre-specified survival models, and (iv) carry the resulting parameter and structural uncertainty into the PSA — and to report all of this against CHEERS 2022, not merely assert reference-case compliance.