The effects of family physician–contracted service on health-related quality of life 1 and equity in health in China

Background Family physician – contracted service (FPCs) has been recently implemented in Chinese 27 primary care settings. This study was aimed at measuring the effects of FPCs on residents’ 28 health-related quality of life (HRQoL) and equity in health among the Chinese population. 29 Methods The study data was drawn from the 2018 household health survey (Shaanxi Province, 30 China) using multistage, stratified cluster random sampling. We measured HRQoL using EQ-5D-3L 31 based on the Chinese-specific time trade-off values set. Coarsened exact matching (CEM) technique 32 was used to control for confounding factors between residents with and without a contracted family 33 physician. The concentration index (C) was calculated to measure equity in health. 34 Results Individuals with a contracted family physician had significantly higher HRQoL than those 35 without, after data matching (0.9355 vs. 0.8995; P <0.001). Additionally, the inequity in HRQoL 36 among respondents with a contracted family physician was significantly lower than those without a 37 contracted FP (Cs of EQ-5D utility score: 0.0084 vs. 0.0263; p<0.001). 38 Conclusions This study highlights the positive effects of FPCs on HRQoL and 39 socioeconomic-related equity in HRQoL. Future efforts should prioritize the economically and 40 educationally disadvantaged groups, the expansion of service coverage, and the competency of 41 family physician teams to further enhance health outcome and equity in health.

6 employment status (unemployed/employed), commercial medical insurance (with/without), 115 minimum travel time to the nearest health-care facility (within 15 minutes/more than 15 minutes), 116 basic medical insurance (Urban Employee Basic Medical Insurance [UEBMI]/Urban-Rural Resident 117 Basic Medical Insurance [URRBMI]), chronic conditions (yes/no), and residential areas (urban or 118 rural areas). The household consumption expenditure per equivalent adult was used as a proxy 119 measure of economic status. We divided the household consumption expenditure per equivalent adult 120 into five groups, the first quintile represents the poorest economic group (i.e., the lowest 20%) while 121 the fifth quintile represents the wealthiest economic group (i.e., the highest 20%). 122 We inquired whether responders had contracted with a family physician by asking "Are you 123 contracted with a family physician?'', which they could respond with "yes" (scored as 1) and "no" or 124 "I've heard nothing of this services" (scored as 0). 125 Health-related quality of life (HRQoL) combines physical and mental health into a summary 126 score, which ranges from 0 (death) to 1 (perfect health). Notably, certain scores can also measure 127 states worse than death [27], such as an index score of EQ-5D ranges -0.59 (worst possible health 128 state) to 1 (best possible health states) according UK tariffs, and by construction, the value of 0 is 129 equal to death and negative values represent HRQoL worse than being dead [28]. In our study, 130 HRQoL was measured by the health utility values for the validated Chinese version of the EuroQol 131 five-dimensional questionnaire-three-level version (EQ-5D-3L) [29,30]. The questionnaire includes 132 five questions and consists of five dimensions, i.e., responder's mobility, self-care, usual activities, 133 pain or discomfort, and anxiety or depression; each dimension has three response levels (1 = no 134 problems, 2= moderate problems and 3 = extreme problems). We employed the Chinese time 135 trade-off values for EQ-5D-3L to measure the utility values of the EQ-5D-3L [31], which has also 136 been widely used in other studies [32,33]. The overall utility values of EQ-5D-3L ranges from -0.149 137 (having extreme problems) to 1 (no problems). 7 The pretreatment covariates differ between the groups with and without FPCs for the observational 141 data, the Coarsened Exact Matching (CEM) technique is designed to improve the estimation of causal 142 effects via a powerfully matching method to keep a better balance of distributions of the covariates 143 between groups, and thereby reducing the bias [34]. The idea of CEM is to temporarily coarsen each 144 variable into substantively meaningful groups, match on these coarsened data exactly, and then only 145 retain the original (uncoarsened) values of the matched data [35]. We applied the CEM method to 146 control confounding variables between the two groups to investigate whether a contracted family 147 physician could improve HRQoL and promote health equity. The detail of technique is well described 148 in detail in previous literatures [34,35] Insurance, and living in rural region in group with FPCs was greater than group without FPCs.

198
After matching using CEM method, covariates imbalances were eliminated between the two groups.

Description of EQ-5D and its concentration index 205
The utility values of EQ-5D and its each dimension are presented in Table 3. After data matching, 206 the mean EQ-5D utility values between respondents with FPCs were 0.9355 (95% CI: 207 0.9302-0.9409), and without FPCs were 0.8995 (95% CI: 0.8926-0.9063). The Cs of EQ-5D score 208 for the respondents with a contracted family physician was significantly lower than those without a  Table 4 presents the decomposition results of Cs of the overall EQ-5D scores. The partial effects, 215 absolute contribution and percentage contribution of each determinant to the inequality of the 216 overall EQ-5D score are presented.

217
Partial effect estimates indicated that among respondents with a contracted family physician, 218 those aged 45 years and above, having chronic conditions, and married, divorced and widowed status 219 were more likely to have lower overall HRQoL. Among those who did not contract with a family 220 physician, the following factors were negatively associated with the overall HRQoL: aged 60 years among those with a contracted family physician was lower (0.003) than those without (0.022).

Discussion 233
Benefiting from the CEM technique and a representative dataset (the 2018 household health survey in 234 Shaanxi Province, northwest China), our study found that individuals with a contracted family 235 physician had significantly better HRQoL that was measured with EQ-5D-3L than those without.

236
Moreover, it also suggested that the inequities in HRQoL were lower among those who contracted 237 with a family physician than those who did not.

257
Our study found that the pro-rich bias was slightly reduced by family physician-contracted and care services which as we found no significant effects in HRQoL in the dimension of self-care 290 and extreme anxiety. Well-implemented primary healthcare services tend to bring benefits to the 291 most vulnerable in the communities as well as to those with complex healthcare needs [44]. 13 One of our study merits is that we examined the effects of FPCs on health and performed the 293 CEM technique to guarantee better balance of covariate distribution between individuals with and 294 without FPCs. The second is that we used the EQ-5D-3L instrument that was validated among the 295 Chinese general population [30] as a health outcome measure to assess the effect of FPCs on our 296 sample. Some limitations should be noted. Firstly, some unobservable or unmatched factors, such as 297 health literacy and beliefs, may have potential effects on the results. Secondly, the results should be 298 explained as associations rather than causal effects as we used cross-sectional data, thus further 299 longitudinal studies should be conducted. In addition, the severe ceiling effect of EQ-5D-3L cannot 300 be fully eliminated when measuring HRQoL among the general population. Specifically, although 301 71.7% of the residents in our study rated themselves as in full health, it was slightly smaller than 302 results in another Chinese study [51]. Lastly, the samples were obtained from one province, and there 303 were heterogeneities regarding FPCs policies and socioeconomic environments between regions, 304 which might limit the study generalizability. Ethics approval and consent to participate 320 Informed consent was obtained from household members before the interview. Approval for this 321 study was obtained by the Ethics Committee of Health Science Center, Xi'an Jiaotong University 322 (approval number 2020-1256).

Consent for publication 324
Not applicable.

Availability of data and material 326
The data used in this study belong to the Health Commission of Shaanxi Province and contain the 327 personal information (e.g., name, personal communication information, property status, health 328 condition, physical defect etc.) of participants. The authors were involved in data collection. Due to 329 the sensitive nature of these data and restrictions imposed by the Health Commission of Shaanxi 330 Province, the authors cannot make these data publicly available. Other researchers who want to use 331 the data may submit requests for data access to the Health Commission of Shaanxi Province at 332 sxwjwwz@126.com.

Competing interests 334
The authors declare that they have no competing interests.       Statistically differences (P < 0.05) in each dimension of EQ-5D between two groups based on Multinomial logistic regression ("No problem" was set as the base outcome); statistically differences (P < 0.05) in utility values of EQ-5D between two groups based on Tobit regressions; all regression adjusted for sex, age group, chronic conditions, economic status, education level, marital status, working status, basic medical insurance, commercial medical insurance, minimum travel time to the nearest health-care facility and residential areas.