Applied conceptual framework
This study applied the International Health Partnership+ (IHP+) common monitoring and evaluation (M&E) framework, developed by the World Health Organization (WHO),15 to structure the evaluation of the NEPHSP on HTN and T2DM management (see Fig. 1). The IHP + framework offers a logical result chain that connects health system inputs and processes to outputs, outcomes, and ultimate impacts, facilitating a comprehensive assessment of policy effectiveness16. Several studies have used this framework to evaluate the impact of health policies17. The IHP + common M&E framework illustrates how systemic inputs—such as financial resources and physical infrastructure — along with operational processes, including policy execution and support mechanisms, lead to measurable outputs like the provision of health services and interventions. These outputs subsequently contribute to intermediate outcomes, for instance, increased intervention coverage and higher service utilization rates, and ultimately to long-term impacts, such as enhanced health status and strengthened financial risk protection. Arrows used in the framework denote the sequential and causal linkages between these indicators .
In this study, the “inputs and process” refer to the operational implementation of the NEPHSP across primary healthcare institutions. The “outputs” encompass direct service delivery indicators, such as hypertension and diabetes management rates, control rates, and medication adherence. The “outcomes” include intermediate effects such as increased utilisation of essential health services and improved patient compliance with follow-up protocols. The ultimate “impact” is measured by the change in mean annual total medical costs per capita, which reflects both financial risk protection and health system efficiency. By situating the evaluation within this structured framework, the study moves beyond intermediate process indicators to assess the program’s effectiveness in reducing economic burden — core goals of universal health coverage.
Study setting
We conducted a retrospective, province-wide study using electronic medical record data from Henan province (population: 97.9 million; area 167,000 km²), China. Henan has fully implemented NEPHSP and established 23 national demonstration zones for chronic disease prevention and control,18 providing a highly relevant context for evaluating cost-saving effects of NEPHSP management. We selected 11 counties/districts (see Fig. 2) across five distinct regions based on a stratified random sampling strategy: eastern (Zhecheng, Qixian), western (Mianchi, Yiyang), southern (Xinye, Luoshan), northern (Xun, Neihuang, Wen), and central (Jiaxian, Linying). Within these sites, a total of 339 healthcare institutions including community health centers/township health centers, Western medicine hospitals, and traditional Chinese medicine hospitals were recruited to extract medical record data.
Data sources and linkage
We extracted data from the front-page electronic medical record for the period from 339 healthcare institutions between January 1, 2022 and December 31, 2024. These records were sourced from the Henan Provincial Health Commission’s centralized database and included patients’ demographics (name, sex, date of birth), admission and discharge information, clinical diagnoses, and itemised cost data. The final dataset comprised 5,456,854 records from 339 healthcare institutions across the 11 selected counties and districts (see Supplementary Table 2).
We combined potential duplicate records for the same patient across multiple healthcare institutions and performed deterministic linkage between the medical record database and official NEPHSP registries using patient name, sex, and date of birth. Successfully matched individuals were classified as NEPHSP-managed, whereas all others formed the non-managed comparison group. Data underwent rigorous two-stage quality control: (1) Hospitals performed internal checks for completeness and accuracy, specifically excluding records lacking diagnoses or cost data before submission; (2) and the Provincial Health Commission conducted secondary quality control to ensure proper upload compliance. Records were included only if they contained complete demographic data (name, sex, date of birth with age > = 35).
In cases where a patient might have more than one record from different healthcare institutions, we consolidated records for the identical patient and finally identified 133,724, 163,989, and 161,911 HTN patients and 64,225, 77,765, and 79,053 T2DM patients in 2022, 2023, and 2024, respectively. We also found the number of patients diagnosed with both conditions was 34,208, 41,793, and 42,989 in each year. To identify patients managed under the NEPHSP, we performed a deterministic linkage between the medical record database and the official NEPHSP management registries from 11 districts/counties. The linkage was done based on a combination of patient name, sex, and date of birth. Individuals with a successful match were classified as the NEPHSP-managed group, while those without a match formed the non-managed comparison group. Post-linkage, the NEPHSP-managed group comprised 23,522, 29,680, and 29,871 HTN patients; 11,060, 13,452, and 13,833 T2DM patients; and 6,471, 7,997, and 8,242 patients with both conditions in 2022, 2023, and 2024, respectively. The participant flowchart can see Supplementary Fig. 1.
Definition of Disease and Outcomes
The primary diagnosis of HTN or T2DM was determined using an algorithm that combined ICD-10 codes with text mining of diagnosis fields from medical record front pages for relevant keywords (Supplementary Table 3). Comorbidities, identified from the same source using a predefined list of conditions based on clinical guidelines and expert consultation (Supplementary Table 4), were mapped to ICD-10 codes, and the total number per patient was calculated.
The primary outcome was the mean annual total medical cost per capita. For subgroup analysis, costs were disaggregated into eight mutually exclusive categories based on the nationally standardised expense classification system: (1) General Medical Services (basic care, therapeutic procedures, nursing, miscellaneous fees); (2) Diagnostics (pathology, laboratory, imaging, clinical assessments); (3) Treatments (non-surgical/surgical interventions); (4) Rehabilitation; (5) Traditional Chinese Medicine Therapies; (6) Western Medications; (7) Chinese Herbal Medicines (proprietary formulations/raw herbs); and (8) Blood Products (whole blood, albumin, globulin).
Statistical analysis
Population characteristics were summarised using proportions for categorical variables and means ± standard deviations for continuous variables. Group comparisons were conducted using Chi-square tests. To control for confounding, 1:1 propensity score matching (PSM) was performed via logistic regression based on age, sex, and number of comorbidities. After matching, the analysis included 23,515 to 29,853 matched pairs for HTN and 11,048 to 13,819 for T2DM per year from 2022 to 2024, with 6,471 to 8,242 patients having both conditions (see Supplementary Fig. 1). Covariate balance was assessed post-matching. Mean total medical costs were compared using a two-sample Z-test. Subgroup analyses were conducted by sex, age group, and number of comorbidities. Due to substantial missing data in insurance type and occupation, a sensitivity analysis was also conducted using a subset of the population with PSM controlling for sex, age, number of comorbidities, insurance type, and occupation to assess socioeconomic influences (Supplementary Table 5). All analyses were two-sided (p < 0.05) and performed using Python 3.14, R 4.3.2, and ArcGIS 10.8.1.