in which Uα/2=1.96 when α = 0.05, P represented the prevalence rate of hypertention (which was 14.2% in this study), and d was the admissible error (which was 0.15P). In the survey, the theoretical sample size was 1290 as determined by a multistage stratified sampling procedure, which included an extra 20% to allow for lost of participates during the investigation. 645 patients with hypertension were randomly selected from the PI group and the CM group in 5 communities, respectively. Eligibility criteria were: (a) aged 20 years or older; (b) willing to participate in the survey; (c) lacal household-registered and living in communities for most of the time; (d) diagnosed as hypertension according to 2010 Chinese Guidelines for Prevention and Treatment of Hypertension; (e) patients from PI group had received intervention for a year since July, 2015; (f) patients from CM group had never received “1 + 1 + N” Phycisians intervention.
This study was approved by the ethical review committee of the School of Public Health, Xiamen University (Xiamen, China). Written informed consent was obtained for each patient and questionnaires were completed by face-to-face interview. Interviewers read the questions exactly as they appeared on the questionnaire. Options were verbally provided by the patients and the corresponding codes were then filled in the questionnaires by the interviewers. A total of 1207 questionnaires were recovered, with a response rate of 93.57%.
In the present study, demographic characteristics included age, gender, body mass index (BMI), locality, duration of hypertension, family history, and referral experience. Early life health behavior included smoking history, drinking history, and amount of sleep.
Measures Of Health-related Quality Of Life And Self-management
To investigate the association between self-management with the HRQOL of hypertensive patients, Quality of Life Instruments for Chronic Diseases of Hypertension(QLICD-HY) and self-management Behavior Rating Scale were developed by ourselves based on literature reviews and nominal/focus group discussions. Cronbach’s coefficient for QLICD-HY and the self-management scale were 0.972 and 0.937 respectively, indicating adequate reliability and validity. They were reliable and appropriate in this study.
HRQOL was used as the health outcome and was measured by QLICD-HY.  QLICD-HY was created by combining the general module (QLICD-GM) and the specific module for hypertension. The QLICD-GM included 32 items, consisting of three domains: physical function (e.g. mobility, sensory, appetite, sleep, and energy), psychological function (e.g. emotion, cognition, and stress), and social function (e.g. social support and adaptation, work, family). Meanwhile the specific module included 17 items reflecting the symptoms, side effects and mental health conditions specific to hypertension.
Self-management was established by the hypertension patients self-management Behavior Rating Scale. The scale consists of 33 items assessing behaviors related to medication adherence, monitoring, sport, dietary, and emotional management. Each item was scored on a 5-point Likert scale where the respondents could choose from one of five responses ranging from ‘strongly disagree’ to ‘strongly agree’. Domain scores were converted using the following formula for comparison: SS=(Rs-Min)*100/R, where SS, Rs, Min, and R represented standardized score, raw score, minimum score of the dimension, and range of scores in the domain, respectively. The higher the HRQOL/self-management score, the higher the quality of life/self-management level.
The database used was established using EpiData Version 3.1 (The EpiData Association, Odense, Denmark). All questionnaires were coded and double-entered by two independent professional data-entry staff members. SAS software version 9.4 for Windows (SAS Institute, Inc., Cary, NC, USA) was used to compute the descriptive analysis of patient demographics and disease-related information.
First, descriptive statistics were calculated for basic demographic variables, the mean and standard deviation (SD) were calculated for continuous variables, and frequencies and percentages were calculated for categorical variables. Chi-square tests and Kruskal–Wallis tests were used to test the significance among variables. P values less than 0.05 were considered significant.
Second, analyses were performed using SPSS version 22.0 software including AMOS 20.0 (IBM Corp, Chicago, IL, USA). Structural equation modeling (SEM) was performed using maximum likelihood estimation to test single-mediator models, in which the relation between the treatment groups and HRQOL was mediated through self-management. Boxes indicated manifest measurement variables, ovals indicated latent variables operationalised by manifest indicators (Fig. 1(A)). A binary indicator for the study treatment groups predicted a potential mediator self-management (path a) and the mediator self-management predicted HRQOL (path b). The model allowed for obtaining regression coefficients and the product of paths (ab) represented the specific indirect effect for a mediator, which was referred to as the mediated effect. The study treatment groups indicator predicted that a HRQOL that was not explained by the mediated path ab as path c’. The total unmediated effect of the intervention was path c (Fig. 1(B)), when no mediators were included in the model), which, in this case represented the between-group differences in HRQOL. The proportion of the total effect of the intervention due to mediating effects was computed as (c-c’)/c. [19, 20] To enable the analysis of hypertension self-management and HRQOL as a composite measure, modelling was performed as latent variable operationalized by the Hypertension Patients Self-management Behavior Rating Scale and QLICD-HY, and included self-management with five observed variables: management of medication adherence, monitoring management, sport management, diet management and emotional management, and HRQOL with four variables: physical function, psychological function, social function and the specific module for hypertension. Ovals represented latent factors, whereas rectangles represented observed variables. Moreover, arrows represented path coefficients for regression of an observed variable onto a latent factor or of one factor onto another. Various fit indexes associated with the SEM technique were assessed and included chi-square, goodness of fit index (GFI), comparative fit index (CFI), increasing fit index (IFI), normed fit index (NFI), and root mean square error of approximation (RMSEA). [21–23]
Third, path analytic models [24, 25] were used to assess multiple-mediator models, in which all five potential mediators were utilized, each with a path from the study treatment groups variable to the mediator (paths a1-5, Fig. 2(A)), and from the mediator-predicted HRQOL (paths b1-5).  The total unmediated effect of the treatment was path c (Fig. 2(B)).
Mediation by a specific variables was deemed significant when zero was not included as a specific indirect effect (ab) of 95% CI.  To account for potential confounding factors by socio-demographic variables, tested models were adjusted for BMI, family history, duration of hypertention, and life styles by modelling associations between these covariates and the three main variables. [10, 28]