Scoping Review Maps Measure Alignment Errors in Hypertension Stroke Prevention Models

Decision-analytic and pharmacoeconomic models evaluating primary stroke prevention among adults with hypertension frequently risk major structural bias by treating disparate epidemiological statistics as interchangeable inputs, according to a scoping review published in Frontiers in Medicine. Researchers led by Jiajun Zhou, Hetong Zhang, Sha Li, Xuechen Chen, and Jie Jiang from Jinan University evaluated how epidemiological evidence is selected, mathematically converted, and assigned to state transitions within economic evaluations. State-transition and Markov models require cycle-specific and pathway-specific parameters that correspond precisely to clearly defined health states, cycle lengths, and transition directions. However, the study highlights that applied health-economic evaluations often focus disproportionately on tree structures, software syntax, and Markov state layouts while leaving the epidemiological compatibility between source study measures and target transition pathways without systematic methodological scrutiny.

The review investigated the disconnect that regularly arises before an empirical parameter is introduced into health-economic software. Clinical trials frequently report hazard ratios for first stroke, prospective cohort studies report incidence density per person-time, burden-of-disease projects report age-standardized mortality, clinical registries report short-term case fatality, and administrative claims databases report the proportion of hospitalized patients who died in a calendar year. Although each indicator describes stroke-related disease burden, each represents an entirely different node in the natural history of cardiovascular disease. When modelers treat these distinct epidemiological measures as numerically swappable probabilities, they risk conflating first events with post-event mortality, double counting disease-specific and background deaths, misinterpreting multi-year cumulative risks as single-cycle probabilities, or inappropriately applying instantaneous hazard ratios as unconditional risk ratios.

To investigate the extent and nature of these parameterization practices, the investigators conducted a scoping review guided by the Joanna Briggs Institute methodological guidance and reported in adherence to PRISMA-ScR standards. The team applied a Population-Concept-Context framework focusing on adults with hypertension eligible for primary stroke prevention, the identification and conversion of epidemiological transition measures, and decision-analytic modeling contexts. Three complementary search strands were executed across PubMed/MEDLINE, Web of Science Core Collection, and the Cochrane Database of Systematic Reviews through 7 August 2026. Across the electronic databases, searches identified 1,170 records, including 901 from PubMed, 254 from the Cochrane Database of Systematic Reviews, and 15 from the Web of Science Core Collection. After removing 284 duplicates, two independent reviewers screened 886 unique titles, assessed 164 abstracts, and evaluated 73 full-text articles, with unresolved discrepancies adjudicated by a third expert.

Ultimately, 30 database-derived studies and two additional sources retrieved through backward citation tracking met all inclusion criteria, yielding a final synthesis evidence set of 32 published sources. The authors classified included publications into four mutually exclusive evidence streams reflecting their primary methodological role rather than an ordinal hierarchy of evidence quality. Stream A contained two applied model-based reports, encompassing an economic evaluation and a projection model in primary stroke prevention. Stream B comprised 24 core methodological sources addressing transition estimation, rate-to-probability mathematics, relative effect conversions, competing risks, mortality decomposition, and model reporting standards. Stream C consisted of the primary clinical report from the China Stroke Primary Prevention Trial, which supplied empirical effect parameters for a didactic worked example. Stream D included five supporting context sources, such as cardiovascular risk-prediction tools and authoritative health-economic monographs.

Through thematic and methodological synthesis across the 32 included sources, the researchers identified four distinct domains of measure-to-pathway mismatch: effect measure mismatch, frequency scale mismatch, disease-process decomposition mismatch, and parameter migration or extrapolation mismatch. The team emphasized that the foundational criterion for selecting any model input parameter must be pathway validity rather than sheer numerical availability. Pathway validity requires verified consistency across four core dimensions: the target clinical population, the precise event definition, the observation time window, and the exact denominator and underlying risk set. Mathematical frequency conversions, relative effect approximations, and extrapolation methods remain defensible only when model developers explicitly establish that the underlying source measure aligns with the target transition node within the model disease structure.

In the domain of effect-measure mismatch, the authors examined how relative effects from clinical trials are transported to model transitions. A common practice in primary stroke prevention models involves applying a trial-derived hazard ratio for first stroke directly to a baseline annual event probability. The authors noted that while a hazard ratio can approximate a pathway-specific relative effect when proportional hazards hold, event rates are low, follow-up durations are comparable, and competing risks are minimal, hazard ratios describe instantaneous hazards across evolving risk sets rather than fixed-time cumulative risks. Directly multiplying a baseline cycle probability by a hazard ratio without converting through the hazard scale introduces systematic bias, particularly in older cohorts, extended model cycles, or states with multiple exits. Furthermore, applying an odds ratio from logistic regression directly as a risk ratio substantially overstates intervention benefits when clinical events are common.

Addressing frequency-scale mismatch, the review detailed errors that emerge when the time dimensions of source studies conflict with the operational cycle length of state-transition models. Health-economic frameworks regularly require monthly, quarterly, or annual transition probabilities, whereas published literature frequently reports multi-year cumulative risks, incidence rates per 1,000 person-years, or lifetime event proportions. Treating an incidence rate as a transition probability without conversion is mathematically invalid because rates and probabilities possess fundamentally different mathematical dimensions. Similarly, dividing a multi-year cumulative risk by follow-up time ignores non-linear hazard dynamics. The authors underscored that assuming a constant hazard over long analytical time horizons is clinically implausible for hypertension, where vascular risk changes substantially with advancing age, blood pressure levels, adherence, and comorbid diabetes. Where multiple mutually exclusive competing exits exist, such as ischemic stroke, hemorrhagic stroke, and non-stroke mortality, converting marginal probabilities independently risks yielding transition sums exceeding one.

The review also detailed disease-process decomposition mismatches, highlighting the critical epidemiological divergence between population stroke mortality and post-stroke case fatality. Population stroke mortality incorporates both stroke incidence across an entire community and survival following an event, whereas case fatality reflects the proportion of incident stroke patients who die within a specified post-event observation window. When an intervention primarily prevents first events, its clinical benefit belongs on the transition from health to incident stroke. If modelers take an unadjusted stroke mortality reduction from literature and simultaneously apply it to both first-stroke incidence and post-stroke case-fatality pathways, the intervention benefit is counted twice. The authors demonstrated that decomposing population mortality into separate incidence and fatality parameters requires exceptionally restrictive assumptions, including shared time windows and identical event definitions, which should generally be confined to structural sensitivity analyses rather than base-case models.

To illustrate these concepts, the authors analyzed empirical parameters from the China Stroke Primary Prevention Trial, which compared enalapril-folic acid against enalapril alone among Chinese adults with hypertension without prior stroke or myocardial infarction. Over a median follow-up of 4.5 years, the trial demonstrated a significant reduction in first stroke with a reported hazard ratio of 0.79 and a 95 percent confidence interval of 0.68 to 0.93. In the control group, 355 of 10,354 participants developed a first stroke. Converting this crude proportion under a constant-hazard assumption yielded an annual baseline first-stroke probability of approximately 0.00772, while applying the 0.79 hazard ratio on the hazard scale generated an intervention probability of approximately 0.00611. The authors emphasized that because the trial measured first-stroke occurrence, this hazard ratio maps validly only to the first-event transition pathway and cannot be directly assigned to post-stroke case-fatality parameters without introducing substantial bias.

To resolve these recurring parameterization failures, the authors formulated a seven-step evidence-to-pathway audit framework designed to evaluate parameter suitability prior to model computation. The sequence mandates identifying the original measure name, defining the denominator and risk set, specifying the time window, matching the target pathway, verifying mathematical conversion feasibility, explicitly handling non-convertibility, and reporting uncertainty transparently. Crucially, the authors argued that declaring an epidemiological measure non-convertible is a legitimate and scientifically credible modeling outcome. Rather than forcing incompatible clinical data into unevidenced transition nodes, modelers should adopt conservative base-case assumptions, such as assuming no direct drug effect on post-event case fatality, and explore alternative structural assumptions through scenario analysis. Reporting should follow comprehensive audit tables documenting original sources, numerators, denominators, conversion formulas, and competing risk constraints.

The authors outlined important methodological limitations inherent to their review and the resulting framework. Because the review focused specifically on primary stroke prevention among adults with hypertension, findings cannot be extrapolated directly to secondary prevention or other chronic diseases without redefining pathways, risk sets, and time windows. Furthermore, because the primary objective was mapping reporting practices and methodological concepts across heterogeneous literature, no formal critical appraisal or risk-of-bias evaluation was conducted on included sources. The proposed audit framework remains a conceptual methodology that has not yet been empirically validated in completed health-economic models. Search coverage was constrained by the exclusion of Embase and CENTRAL due to institutional subscription restrictions, restriction to English-language full-text articles, and retrospective protocol registration on OSF Registries on 11 August 2026, which occurred after source screening and data extraction had concluded.

Subject of Research: Epidemiological parameter alignment and pathway auditing in pharmacoeconomic stroke models

Article Title: Epidemiological measure-to-pathway alignment in pharmacoeconomic models for primary stroke prevention among adults with hypertension: a scoping review and methodological framework

Article References: Zhou, J., Zhang, H., Liu, Y., Li, S., Chen, X., & Jiang, J. (2026). Epidemiological measure-to-pathway alignment in pharmacoeconomic models for primary stroke prevention among adults with hypertension: a scoping review and methodological framework. Frontiers in Medicine, 13, Article 1969971. https://doi.org/10.3389/fmed.2026.1969971

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1969971

Keywords: pharmacoeconomics, primary stroke prevention, hypertension, state-transition models, Markov models, transition probabilities, hazard ratios, case-fatality risk, disease-process decomposition, frequency-scale conversion, competing risks, parameter migration, China Stroke Primary Prevention Trial, decision-analytic modeling, PRISMA-ScR, scoping review, health technology assessment