Cost-effectiveness-oriented management (CEOM) of cardiovascular risks at primary healthcare settings in Anhui, China: a protocol for a cluster randomised controlled trial

Abstract

Introduction

Cardiovascular disease (CVD) represents a significant health and economic burden in China. Despite extensive research on CVD management in primary care, the cost-effectiveness of current practices remains suboptimal. The overall aim of this study is to test the efficacy of an integrated, cost-effectiveness-oriented management (CEOM) intervention to improve CVD risk and harm management in primary healthcare settings in Anhui, China.

Methods and analysis

This open-label, multi-centre, cluster-randomised controlled trial will be conducted in 32 village clinics in Anhui Province, China. Clinics will be randomised (1:1) to the CEOM intervention or usual care. The CEOM intervention integrates a prospective cost-effectiveness analysis from a societal perspective into the clinical workflow. It assesses patient eligibility and prioritises intervention themes and specific items based on predicted incremental cost-effectiveness ratios (ICER), guiding clinicians to deliver tailored management procedures. Implementation is supported by standardised training, automated performance feedback and peer support. Participants (n=1920) are permanent residents aged ≥35 years with diagnosed hypertension, diabetes and/or CVD. The primary outcome is the ICER, using Quality-Adjusted Life Years (QALYs) as the primary effect measure. The ICER will be evaluated based on cumulative QALYs and direct and indirect costs assessed at 12 and 24 months. Secondary outcomes include changes in knowledge, attitudes and practices, major adverse cardiovascular events, QALYs as estimated using the EQ-5D-5L ratings and direct and indirect costs. Data will be analysed using linear mixed models and generalised estimating equations following intention-to-treat principle.

Ethics and dissemination

The study was approved by the Medical Ethics Committee of Anhui Medical University (83230358). Results will be disseminated via peer-reviewed journals, conferences and policy briefs.

Trial registration number

ISRCTN registry, ISRCTN87887485. Registered on 28 January 2026.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This cluster randomised controlled trial incorporates cost-effectiveness considerations into clinical decision-making for cardiovascular disease management.

• The intervention integrates digital decision support with supervision and incentives to address implementation barriers in rural primary care.

• The trial uses a large, representative sample from rural Anhui, enhancing generalisability to similar low-resource settings.

• A main limitation of the trial is its inability to blind participating clinicians and patients due to the nature of the intervention.

Background

Cardiovascular disease (CVD) presents a growing threat to global public health, with prevalence and mortality rates rising over the past three decades.¹ The situation is particularly difficult in China, which now bears the burden of an estimated 330 million cases and 4.58 million annual deaths.^{2 3} This epidemic imposes not only a substantial burden on quality of life, accounting for 98 million disability-adjusted life years annually in China,⁴ but also a heavy economic burden, with direct healthcare costs over RMB 540 billion (approximately 75 billion US$) in China.⁵

Primary care plays a key role in combating CVD. Yet CVD-related care in primary settings falls far below expected due to resource and ability limitations, low compliance with guidelines, inadequate supervision and support and others.⁶,⁸ Although more and more guidelines are recommending a cost-effectiveness perspective in prevention and treatment decision,^{9 10} there is a lack of pragmatic measures to realise the recommendation in routine primary care practice, especially in resource-limited areas. Most clinical trials focus on major adverse cardiovascular events (MACE) as primary endpoints, relegating economic value to a secondary, auxiliary role.¹¹,¹³ Cost-effectiveness is currently¹⁴,¹⁶ used almost exclusively as a retrospective evaluation metric rather than a pragmatic tool for clinical decision-making.

The lack of cost-effectiveness thinking in routine practices may be attributed mainly to the practical difficulty of identifying which specific interventions yield the highest health gain.¹⁷,²⁰ To bridge this gap, we developed a cost-effectiveness-oriented management (CEOM) model. This model uses ensemble learning to identify and rank the most efficient intervention ingredients for individual patients. Coupled with a user-friendly support and supervision system, the model aims to streamline high-priority interventions and maximise the economic efficiency of CVD-related services in primary care settings. This study protocol describes the design and implementation details of a cluster randomised controlled trial (cRCT) aimed at evaluating the cost-effectiveness of the CEOM model in primary care settings for CVD management.

Aim/objectives

The overall aim of this study is to test the efficacy of a CEOM to reduce CVD risks and harms at primary healthcare settings in Anhui, China. The primary hypothesis of the study is that, compared with usual care, CEOM will result in a better incremental cost-effectiveness ratio (ICER) over 24 months. In addition, the study is also expected to benefit participants in the intervention arm in terms of: physiological measurements; CVD-related knowledge, attitudes and practices (KAP); MACE and quality of life.

Methods and analysis

Trial design

This study is designed as an open-label, multi-centre, cRCT with two parallel arms (intervention vs usual care), in which village clinics will be treated as the clusters or units of randomisation. The study will be conducted in accordance with a pre-specified protocol (V.2.1, finalised on 15 January 2026), which has been registered with the International Standard Randomised Controlled Trial Number (ISRCTN) registry under the registration number ISRCTN87887485. The study is planned to run from 1 March 2026 to 30 March 2028. Statistical analyses will be performed in strict adherence to the Statistical Analysis Plan, which will be published separately.

Participants and recruitment procedures

Selection of village clinics

A total of 32 village clinics will be selected from Anhui Province. The sampling will adopt a multi-stage stratified strategy implemented by an independent statistician. First, one county will be selected from each of the 16 prefectures in Anhui. Second, within each selected county, two geographically non-adjacent townships will be randomly chosen; this geographical separation is explicitly designed to reduce the risk of intervention contamination between the study arms. Finally, one village clinic will be randomly selected from each township.

Implementation population

The CEOM intervention is a cluster-level health management model integrated into routine primary care. Therefore, all eligible CVD patients registered in the Essential Public Health Service Registry within the 16 intervention village clinics will be subject to the CEOM management protocol.

Recruitment of the evaluation cohort

In each of the selected village clinics, the primary care clinicians responsible for chronic disease management will be recruited after informed consent. Patient recruitment will proceed in four steps. Step 1: assisted by our research staff, one participating clinician from each site clinic will screen records in the Essential Public Health Service Registry to identify eligible patients in their clinic. Step 2: from the pool of eligible patients, 60 individuals per clinic will be randomly selected as potential candidates for the evaluation cohort. Step 3: selected patients will be invited for informed consent specifically regarding personal data collection and long-term follow-up. Those who agree will undergo a face-to-face interview for baseline data collection. Step 4: if the selected patient declines to participate in the evaluation, another patient will be randomly selected from the remaining eligible pool (repeating Steps 2 and 3) until the target of 60 evaluation participants per cluster is achieved (figure 1).

Figure 1. Trial flow chart. CCA, complete case analysis; CEOM, cost-effectiveness-oriented management; CVD, cardiovascular diseases; ITT, intention-to-treat.

Eligibility criteria

Clinician inclusion and exclusion criteria

Eligible clinicians are individuals who: (1) Hold a valid licence of practice and work full-time at the participating clinics, (2) Are directly responsible for the routine management and clinical care of CVD-related patients and (3) Agree to participate in the study and sign the informed consent form. Those who will retire or plan to leave their current practice within the next 24 months are excluded.

Patient inclusion and exclusion criteria

Eligible patients are permanent residents (residing >6 months/year) aged 35 years or older, diagnosed with hypertension, type-2 diabetes and/or established CVD (specifically coronary heart disease, stroke or transient ischaemic attack), who are willing and able to complete follow-up assessments. Patients will be excluded if they have experienced acute cardiovascular events within the past 3 months (such as acute myocardial infarction, unstable angina or acute stroke), exhibit severe cognitive, mental or sensory impairments that preclude smooth follow-up assessments, or plan to move out of the study area.

Randomisation, allocation and blinding

An independent statistician will conduct the randomisation using the Sealed Envelope web-based system. The randomisation will be stratified by county: within each participating county, the two selected village clinics will be randomly assigned in a 1:1 ratio to either the intervention or the control group. To ensure allocation concealment, the randomisation result will only be generated and revealed to the study coordinators after the baseline survey of all participants within a cluster has been completed. Due to the nature of the intervention, blinding of village clinicians and patients is not possible. However, to minimise detection bias, follow-up data collectors and the study statistician will remain blinded to the group allocation throughout the data collection and analysis phases.

Control

Participants in the control group will receive standard usual care delivered by primary care clinicians in accordance with China’s National Basic Public Health Services guidelines. This routine management includes maintaining electronic health records (EHR) and conducting mandatory quarterly face-to-face follow-up visits. During these encounters, clinicians will monitor basic physiological indicators (specifically blood pressure and fasting blood glucose), check medication compliance, and provide generic lifestyle counselling.

Intervention

As shown in figure 2, for any primary healthcare encounter in the intervention arm, the patient will be first assessed as to whether they need CEOM using an eligibility algorithm. If yes, a list of intervention themes (ordered by priority weights) will be produced (using a priority model) for the attending clinician to choose. Here, the attending clinician will retain full clinical autonomy and will hold the absolute right to decline or modify the recommended themes based on their professional judgement and the patient’s immediate clinical status. Then, a further list of intervention items (ordered in the same way) under the selected theme will be generated. After that, a detailed intervention procedure (including a take-away self-management leaflet; see online supplemental appendix 1 for the English version template) for the current patient will be produced using refinement algorithms for the clinician to practise accordingly.

Figure 2. Clinical implementation pathway of the CEOM model. (CEOM, cost-effectiveness-oriented management; CVD, cardiovascular diseases; ICER, incremental cost-effectiveness ratios; QALYs, Quality-Adjusted Life Years; ICERi,k refers to the ICER of participant i for intervention theme k; ΔCi,k and ΔEi,k indicate potential incremental changes in costs and effectiveness after applying intervention theme k.

The 12 intervention themes are designed to improve: diet and nutrition; tobacco and alcohol abstinence; physical exercise and activity; self-monitoring and checks; psychological problems coping; use of professional services; blood pressure management; blood sugar management; blood lipid management; body weight management; CVD complications management and involvement of family members. Each of these themes consists of a set of specific information items.

Clinician adherence to the generated recommendations will be systematically and automatically logged by the online support system. CEOM also will include automated performance assessment and feedback. The assessment will use routinely collected medical records as the main data source verified via minimum telephone surveys.

Trial implementation and safety

In order to facilitate implementation of the above intervention, the participating clinicians will receive a 1 day training session on the CEOM, alongside access to a user-friendly online support system, a dedicated WeChat group for peer-to-peer problem sharing, and automated performance feedback. The CEOM intervention will be discontinued if participants withdraw consent, relocate or experience severe clinical complications requiring specialised care. Participant adherence will be verified via periodic telephone surveys. Routine concomitant care for chronic conditions will be permitted in both arms.

Priority model

A quantitative model has been designed to prioritise both patient and intervention themes based on their predicted individual-level cost-effectiveness over a specific period (12 or 24 months). The model will estimate the individualised ICER_i,k for each participant i across each of the 12 intervention themes k. This will be done by predicting the potential incremental changes in costs ($\mathrm{\Delta }{C}_{i,k}$) and effectiveness ($\mathrm{\Delta }{\text{E}}_{i,k}$) if theme k were applied, relative to a non-intervention baseline. The calculation follows the formula as given below:

$ICE{R}_{i,k}=\frac{\mathrm{\Delta }{C}_{i,k}}{\mathrm{\Delta }{E}_{i,k}}=\frac{C_{i,k}-C_{i,0}}{E_{i,k}-E_{i,0}}$

where${\text{C}}_{\text{i}\text{,}\text{k}}$ and ${\text{E}}_{\text{i}\text{,}\text{k}}$ are predicted costs and Quality-Adjusted Life Years (QALYs) for individual i under theme k intervention; while ${\text{C}}_{\text{i}\text{,0}}$ and ${\text{E}}_{\text{i}\text{,0}}$ represent outcomes under the routine baseline care. Here, all costs will be assessed from a societal perspective, covering both direct and indirect costs with detailed components presented in the outcome measures section. Baseline costs (${\text{C}}_{\text{i}\text{,0}}$) will be estimated by inflating baseline observed costs using appropriate Consumer Price Index inflation rates. Baseline QALYs (${\text{E}}_{\text{i}\text{,0}}$) under usual care will be estimated by assuming the patient’s baseline EQ-5D-5L health utility score remains constant over the follow-up period. Intervention-specific costs (${\text{C}}_{\text{i}\text{,}\text{k}}$) and predicted QALYs (${\text{E}}_{\text{i}\text{,}\text{k}}$) under each intervention theme will be generated using a random forest model,²¹ built on 19 features including indices of the 12 intervention themes and seven other covariables (for more information about features please refer to online supplemental appendix 2). Given this formula, the intervention themes for each individual patient will be prioritised according to the rules as summarised in table 1.

Table 1. Rules for prioritising intervention themes.

Recommendation category	Criteria	Interpretation
Unconditional acceptance (UA)	$\mathrm{\Delta }{E}_k$>0 and $\mathrm{\Delta }{\text{C}}_k$<0	Theme k intervention is dominant (improves health and reduces costs) and receives the highest priority.
Conditional acceptance	$\mathrm{\Delta }{E}_k$>0 and $\mathrm{\Delta }{\text{C}}_k$>0	Theme k intervention is cost-effective only if ${\text{ICER}}_{\text{i}\text{,}\text{k}}$ is below the threshold (1–3 times GDP per capita per QALY);25 these are ranked after UA themes.
Unconditional rejection	$\mathrm{\Delta }{\mathrm{E}}_k$<0	Theme k intervention is predicted to reduce health outcomes and is excluded regardless of cost.

ICER, incremental cost-effectiveness ratios; QALY, Quality-Adjusted Life Year.

Outcome measures

The primary outcome measure will be ICER, which will be derived from cumulative QALYs and total societal costs calculated over the 24-month follow-up periods. QALYs will be estimated using the linear interpolation method (trapezoidal rule) based on health utility scores derived from the EQ-5D-5L questionnaire, which will be calculated using the established Chinese-specific value set developed by Luo et al.²² Consistent with the societal perspective, total costs will comprise: (1) Direct medical expenditures (including outpatient visits, inpatient hospitalisations and medication usage) extracted from the regional EHR and supplemented by patient self-reported survey data to fully capture out-of-system retail pharmacy purchases, online drug procurement and non-local medical encounters, (2) Intervention implementation costs (including physician training, health management materials and system maintenance) tracked and calculated via a project-specific micro-costing approach,²³ (3) Direct non-medical costs (including transportation and lodging expenses incurred during medical encounters) collected via self-reported follow-up questionnaires and (4) Indirect costs determined by valuing self-reported productivity losses due to patient absenteeism using the human capital approach, with daily wage rates derived from the corresponding occupational sectors in the latest Anhui Statistical Yearbook.

Secondary outcomes will include the ICER evaluated at the 12-month follow-up period, as well as changes in physiological, behavioural and clinical metrics assessed at baseline, 12 months, and 24 months. Physiological indicators include systolic and diastolic blood pressure, fasting plasma glucose (FPG), glycated haemoglobin (HbA1c), low-density lipoprotein cholesterol, total cholesterol and body mass index. Behavioural changes will be evaluated using KAP scores across 12 specific domains: improving nutrition and diet, tobacco and alcohol abstinence, maintaining physical activities, performing self-monitoring, psychological problems coping, use of professional service, facilitating family engagement, controlling blood pressure, controlling blood sugar, controlling blood lipid, controlling body measures, controlling CVD complications (details of the scores will be published separately in our paper about the statistical analysis plan of this trial findings). Additionally, binary secondary outcomes include the occurrence of non-fatal myocardial infarction, non-fatal stroke and cardiovascular death reported in the past year. Finally, individual components of costs (direct medical costs, intervention implementation costs, direct non-medical costs and indirect costs) will also be analysed separately as secondary economic endpoints.

Data collection

Patient-level data will be collected via face-to-face household interviews at baseline, 12 months and 24 months. During these on-site visits, trained investigators will record participant responses using an electronic questionnaire developed in EpiData (V.3.1). At baseline, trained investigators will record participants’ data, including sociodemographic factors (age, sex, ethnicity, marital status and education level) and clinical history (CVD duration, prior acute cardiac/stroke events and comorbidities). At all three assessment time points, the questionnaire will capture variables including health-related quality of life (EQ-5D-5L), disease management behaviours across the 12 intervention themes (assessed via KAP scales) and past-year economic burdens (comprising direct medical costs, direct non-medical costs and indirect costs due to productivity losses). Concurrently, anthropometric and physiological parameters, specifically systolic and diastolic blood pressure, height, weight and waist circumference, will be measured on-site by investigators using standardised instruments and protocols. In contrast, haematological indicators (including FPG, HbA1c and lipid profiles) recorded within a maximum of 3 months prior to enrolment or follow-up will be extracted directly from the regional resident EHR. For each assessment time point, these data will be derived from the participant’s most recent health examination record available within this window.

Cluster-level data, characterising the participating clinicians, will be collected by the research staff immediately following the acquisition of written informed consent. Collected variables will include gender, age, medical education, years of practice and number of CVD-related visits in the previous year.

Sample size

The CEOM intervention is implemented at the cluster level for all managed CVD patients within the intervention clinics; the sample size was specifically calculated for the evaluation cohort to ensure sufficient statistical power for the primary outcome. While the ICER is the primary outcome, its ratio-based nature poses significant challenges for direct power calculation due to statistical instability and non-linearity. Therefore, the sample size was determined based on the net monetary benefit (NMB), a linear rearrangement of cost-effectiveness data that allows for robust statistical inference. Following the approach suggested by Glick,²⁴ the NMB was calculated as NMB=WTP×ΔQALY−ΔCost, assuming a willingness-to-pay (WTP) threshold of one time the annual per capita GDP of China.²⁵ Using the SD of NMB from our pilot data, we assumed a baseline NMB of 0 and an expected absolute increase of 1000 Chinese Yuan (approximately 139 ¥) in the intervention group, with an SD of 4200 Chinese Yuan (approximately 583 ¥). With a two-sided alpha of 0.05 and 90% power, the initial sample size required for a simple randomised trial was estimated to be 371 participants. To adjust for the cluster randomisation at the village clinic level, we assumed an ICC of 0.05 and a conservative average cluster size of 50 participants. This resulted in a design effect of 3.45, and the required total sample size for the primary analysis was determined to be 1280. To ensure sufficient statistical power for subgroup analyses, the final recruitment target was expanded to 1728. Considering an anticipated 10% attrition rate,²⁶ the final recruitment target was set at 1920 individuals (averaging 60 per cluster).

Data analysis

Statistical analyses will be conducted using a generalised linear mixed model (GLMM) framework to estimate intervention effects while rigorously accounting for hierarchical clustering at the village clinic level. The primary analysis will strictly adhere to the intention-to-treat principle; thus, participants in the intervention arm who do not meet the active CEOM criteria will still be retained and analysed within their allocated group. Subsequently, a per-protocol sensitivity analysis will be conducted, restricted to participants who were successfully screened, deemed eligible by the algorithm and initially managed under the recommended intervention themes, to examine the maximum potential benefit of the active CEOM framework. Missing values for primary and secondary outcomes will be addressed using Multiple Imputation by Chained Equations. The imputation model will include baseline characteristics (eg, age, sex and county), baseline values of the outcome variables and available follow-up data to predict missing values under the missing at random assumption. To resolve the inherent instability of ratio-based metrics, cost-effectiveness will be primarily assessed through the NMB across a WTP range of 1–3 times China’s per capita GDP.²⁵

For continuous outcomes, including NMB, physiological indicators, the 12 domain-specific KAP scores and EQ-5D-5L utility scores, will be analysed using GLMMs specifying a Gaussian distribution and an identity link function. These models will include fixed effects for treatment arm, follow-up time (12 and 24 months) and a treatment-by-time interaction term, while adjusting for baseline values and relevant covariates. To address the skewed distribution of medical costs, a Gamma distribution with a log link will be used. Binary clinical events, such as non-fatal myocardial infarction, stroke and cardiovascular death, will be modelled using a binomial distribution and logit link to estimate adjusted ORs.

To account for the study’s multi-level design, all models will incorporate nested random intercepts (individual participants within village clinics). This structure controls for both the correlation between repeated measures from the same individual and the clustering effect at the clinic level. Safety data will be summarised descriptively, with formal GLMM-based testing conducted for adverse events (AEs) where data density will allow for model convergence.

Interaction terms between the intervention group and key baseline characteristics subgroup variables will be included in GLMMs to statistically test for subgroup differences.

Data handling and repository

All participant data will be de-identified and assigned a unique study identification code. The correspondence between the ID codes and personal identities will be stored in a separate, password-protected file accessible only to the principal investigator. All electronic data (including the EpiData database and analytical datasets) will be stored on encrypted, access-controlled servers. Paper-based records (eg, consent forms) will be kept in locked cabinets in a secure office. An independent Data Monitoring Committee, appointed by the Scientific Research Department of Anhui Medical University, will oversee the trial’s progress and participant safety. Given the low-risk nature of the CEOM intervention, no formal interim analyses or specific stopping rules for futility or efficacy are planned.

Trial monitoring

Trial monitoring will be conducted every 3 months to ensure the study is performed in strict accordance with the protocol. The Monitoring Committee, comprised of independent experts in primary healthcare and clinical trials, will perform these audits.

Adverse events

As this is a management optimisation trial, a favourable safety profile is anticipated. AEs and serious AEs are defined in accordance with ICH-GCP guidelines.²⁷ Pre-specified study endpoints (eg, myocardial infarction, stroke and CVD death) will be documented as efficacy outcomes and not included in AE reporting, unless they are directly caused by the intervention (eg, errors in clinical decision-making systems). Participating clinicians will conduct safety monitoring during routine clinical visits. Any AE suspected to be associated with the intervention (eg, psychological distress) must be reported to the research team within 24 hours to facilitate causality assessment. An independent Data and Safety Monitoring Board, established at Anhui Medical University, will conduct regular safety reviews.

Patients and public involvement

While patients were not directly involved in the formulation of the research question, their feedback was integral to the intervention design. During the pilot phase, representatives from the target population and village practitioners were consulted to assess the usability of the CEOM components and the burden of the data collection process. Their insights led to refinements in the questionnaire length and the interface of the intervention tools. Results of the study will be disseminated to participating villages through plain language summaries provided to the clinics.

Discussion

This trial is among the first to translate cost-effectiveness from a retrospective reporting metric into a prospective clinical tool. Through integration of the CEOM model, we go beyond generic risk stratification to deliver a personalised intervention sequence. This approach takes full account of the variability in different patients’ needs for various interventions, with the aim of maximising health benefits under the fixed resource constraints of rural primary care.

A major methodological advantage of this study lies in the application of GLMMs with nested random effects. We specifically selected this analytical framework to address the complex structure of our study data, as it accounts for both the clustering effect at the village clinic level and the correlation between repeated measures collected at 12 and 24 months.²⁸ Evaluating the treatment-by-time interaction will also reveal how the intervention’s value changes as it progresses from initial implementation to sustained management.

A potential limitation of this study is its 24-month follow-up period, which may not be long enough to capture all definitive cardiovascular events or fully assess lifetime cost-effectiveness.²⁹ That said, this duration is sufficient to evaluate sustained behavioural changes, physiological control and early trends in healthcare utilisation, all of which are key predictors of long-term outcomes. Moreover, findings from the 24-month follow-up will provide empirical evidence for modelling the long-term economic trajectory of the CEOM framework.

If proven effective, the CEOM framework has the potential to redefine chronic disease management especially in resource-constrained settings. It provides a scalable strategy for health systems to prioritise high-value care, ensuring that limited public health resources are allocated to the individuals and interventions that generate the greatest return on investment.

Ethics and dissemination

The study protocol has been reviewed and approved by the Medical Ethics Committee of Anhui Medical University (Approval Number: 83230358). Written informed consent (online supplemental appendix 3) will be obtained from all participating clinicians and patients prior to enrolment. Participants will be explicitly informed of their right to withdraw from the study at any time without reprisal or effect on their routine medical care. Any significant protocol modifications will be submitted to the ethics committee for approval and communicated to the trial registry.

The findings of this trial will be disseminated through publication in peer-reviewed journals and presentations at international and national academic conferences. Given the health economic focus of the study, we will also prepare policy briefs for local health authorities to inform decision-making regarding CVD management strategies.

Data sharing

De-identified individual participant data will be made available beginning 3 months and ending 5 years following the publication of the primary manuscript. Data will be shared with researchers who provide a methodologically sound proposal to achieve the aims in the approved proposal. To gain access, data requestors will be required to sign a data access agreement. Proposals and requests for clarification should be directed to shenxr@ahmu.edu.cn. The steering committee will consider requests for specific data points provided that such disclosure does not jeopardise future planned publications by the research team.