Health Utility and Its Risk Factors in University Staff in China: A Cross-Sectional Survey from an Occupational Health Perspective

Backgrounds: The health of university staff is a major occupational health concern worldwide. Studies have reported low health-related quality of life (HRQOL), low job satisfaction and poor mental health in this occupational group. However, none of previous studies have measured health utility and compared it to a national norm. Therefore, this study was conducted to gain a deeper understanding of the HRQOL of university staff in China and to identify risk factors inuential to their health. Methods: This was a cross-sectional survey conducted in a public university in China. Participants were interviewed face-to-face for demographic and socioeconomic information and health conditions. The Chinese version of the EQ-5D-5L instrument was used to measure HRQOL for calculating health utility. The relationship between health utility and sample characteristics was rst examined using t-test and correlation analysis. Multivariate generalized linear models were further applied to evaluate the signicance of these associations while adjusting for other variables. Results: The sample (n=154) had a mean age of 40.65 years and slightly more females (51.30%). The overall prevalence of diseases or symptoms was 81.17%. Participants attained the means (SDs) of 0.945 (0.073) and 83.00 (11.32) for the health utility and visual analogue scale respectively. The most affected domain was the anxiety/depression with 40.26% of participants reporting problems and 37.66% of the sample reported problems in the pain/discomfort domain. There were less than 5% participants reported problems in the mobility, self-care or daily activity domains individually. Multivariate models revealed that psychological/emotional conditions were associated with the largest utility loss of -0.067 (95%CI: -0.089, -0.045) followed by having a Master’s degree or higher (-0.048, 95%CI: -0.09, -0.005) and pain in body parts other than head, neck and back (-0.034, 95%CI: -0.055, -0.014). Conclusions: University staff in China may have worse HRQOL than the general population, which manifested mainly with the pain/discomfort and anxiety/depression domains. The signicant factors for utility loss were having a Master’s degree or higher, psychological conditions and pain in body parts other than the head, neck and back. Targeted health promotion policies and programs should be created to benet this occupational group and society overall.


Introduction
The health of university staff is a major occupational health concern in China and worldwide (1,2).
Academics are tasked with cultivating competent graduates and progressing science and culture to better society, and thus ill-health in this population has extensive social and national implications. Studies have found that burnout is a common occurrence in university staff which has contributed to poor health (3,4). Mental health is also impaired in this occupational group (5). This situation may be worse in China where the "996" (working 9 a.m. to 9 p.m. six days a week) work culture is prevalent (6, 7). Studies have already reported that college teachers suffered low health-related quality of life (HRQOL) (8), poor mental health and low job satisfaction (9,10). It has been further suggested that the HRQOL of university staff is worse than the general population.
The Chinese government has paid attention to the HRQOL of its people with the "Healthy China 2030" national strategy (11). Since 2003, the National Health Services Survey (NHSS) conducted every 5 years has started to measure the HRQOL of the Chinese population (12). The NHSS HRQOL data aims to provide a landscape of the health status of Chinese populations, and create national norms of health utility for clinical and economic use (12)(13)(14). However, national norms can serve as an authoritative reference system for comparison purposes, but they are limited in providing valid HRQOL estimates for subpopulations, such as various occupational groups.
University staff are a unique occupational group in that they are highly skilled and highly educated, and their jobs demand intensive physical and mental work. To date, few studies have directly investigated the HRQOL of university staff with most previous studies centered on health care personnel (3,15). The results from such studies cannot be applied to university staff in other academic disciplines. Furthermore, health utility was rarely reported and compared with a national norm. Thus, the health status of the population of university staff remains largely unknown.
Recognizing this gap, we investigated the HRQOL of a sample of university staff using the newer version of the EQ-5D (European Quality of Life 5 Dimensions) instrument from an occupational health perspective. The study set two objectives: 1) To gain a deeper understanding of the HRQOL of university staff and its speci c characteristics in China; 2) To identify risk factors in uential to their health utility.
From this study, we hope to provide scienti c evidence to healthcare authorities and university management for health policy-making and the design of health promotion programs.

Study setting and population
This was a community-based cross-sectional survey conducted in a public university, the University of Shanghai for Science and Technology (USST) in Shanghai, which is the economic engine in China. The study population was de ned as all staff, either academic or nonacademic, currently working in the university. Academia in China is a specialized profession characterized by high education attainment, job security and good welfare. Conventionally they are considered to be in the middle or upper social class.
Due to convenience and resource limitations we chose the staff of the USST Business School.
During the study period from 1 November 2020 to 15 January 2021, the Business School had a workforce of 216 personnel. There were 17 personnel away for reasons such as studying overseas, long-term sick or parental leave, hospitalization, resignation or near-retirement. Nine personnel o cially rejected the invitation and a further 35 were out of contact. Finally, a total of 155 people participated in the study -a response rate of 78%.

Data Collection
The eligibility criteria were: 1) o cially employed at USST; 2) not hospitalized or immediately after hospital discharge; 3) not handicapped or disabled; 4) able to carry out day-to-day work normally; 5) able to give personal consent.
All staff were personally invited through email, WeChat or face-to-face contact. The recruitment advertisement was also announced at school-level meetings and internal WeChat workgroups. All participants signed written consent forms before joining the study. The study was approved by the IRB committee of the Air Force Medical Center in Beijing.
Data were collected through face-to-face interview. Participants completed the questionnaires in the presence of the interviewer. One postgraduate and four undergraduate students conducted the interviews.
The ve interviewers attended three training sections each lasting two hours. This pre-interview training aimed to ensure equivalent task understanding, procedures and interactions with respondents. Interviewers were trained to give brie ngs and answer questions only, and not to promote a "right" or "wrong" answer as a measure to minimize the social desirability effect.

Health Utility Measurement
We used the Chinese version of the EQ-5D-5L instrument to measure HRQOL (16). The EQ-5D is a preference-based HRQOL instrument asking participants to rate their present-day health. Compared to the old version of the EQ-5D-3L instrument, the EQ-5D-5L instrument is more favorable for HRQOL measurement due to its greater discriminatory power and lower ceiling effect (17,18). Its validity and reliability have been previously validated in various Chinese populations (17,19,20).
The EQ-5D instrument has two parts. The rst part is the descriptive system which classi es 3,125 health states in ve health domains (mobility, self-care, usual activities, pain/discomfort and anxiety/depression), each with ve ordinal severity levels (no problems, slight problems, moderate problems, severe problems, and extreme problems/unable to). A respondent rates his/her health subjectively against the most appropriate statement in each of the ve dimensions based on their health on the day of interview. The scores of the ve dimensions were used to calculate health utility according to the Chinese speci c value set (21). Health utility by de nition has a range of 0 (death) to 1 (full health).
The second part is called the visual analogue scale (VAS), a 10-cm vertical bar anchored at 0 (worst imaginable health) and 100 (best imaginable health). The VAS measures the overall health status rated by the respondents themselves.

Health Determinants Investigation
We also collected information on health risk factors including demographic (age, gender, height, weight), lifestyle or behavioral (smoking, drinking) and socioeconomic (education, marital status, number of children) variables (22). Clinical information was also collected including current diseases, symptoms and other health conditions using the WHO Health and Work Performance Questionnaire (23)..

Statistical analysis
One participant was removed due to missing data. The nal sample size for analyses was 154 participants. Categorical variables were described as counts and proportions. The prevalence of health conditions was estimated as the proportion of the sample presenting with certain health conditions. Continuous variables were presented as the mean and standard deviation (S.D.). The skewness and kurtosis of continuous variables were also explored to choose the appropriate models for multivariate analysis.
BMI was categorized into three groups -underweight/normal weight, overweight and obese using a gender-speci c standard for Chinese populations (24,25). Marital status was dichotomized into alone (single, divorced or widowed) and married. A total of 40 health conditions were reported. This was too many to be each treated as independent variables in the multivariate analysis for the sake of statistical power given the sample size. To carry out valid multivariate analyses, the six most prevalent conditions (≥20%) were treated as independent variables. These conditions were back/neck pain, pain in other areas, fatigue, high blood cholesterol, headache and sleeping problems. The remaining 34 conditions were combined into 7 groups by anatomical and/or physiological system, i.e. allergy, high blood pressure/cardiovascular, digestive system, respiratory system, muscular-skeleton and psychological/emotional conditions and others.
Univariate analysis involved correlation analysis between health utility and individual continuous variables, and t-test and ANOVA comparing the utility of different risk level of categorical variables. A generalized linear model (GLM) was chosen to perform multivariate analyses as the dependent variable, EQ-5D utility, followed a negatively skewed distribution (skewness= -1.89) to which GLMs were immune.
Multivariate analysis evaluated the associations of health utility with sample characteristics. To remove the effect of collinearity and to produce consistent coe cients, the GLMs were set to use robust methods for parameter estimation and pro le likelihood methods for con dence intervals. Unlike ordinary linear regression that models raw data of the dependent variable, GLMs model the means of the dependent variable. Thus, in our study the parameter coe cients of the GLMs represented the mean utility change associated with the speci c variables.
The multivariate analysis had two steps. The rst step was to build 12 condition-speci c GLMs for individual health conditions when controlling for demographic and socioeconomic variables. In the second step, multiple conditions were evaluated simultaneously in one model together with demographic and socioeconomic variables. These conditions were selected for inclusion in the model on the condition that they achieved borderline signi cance (p<=0.01) in their own models. The analysis took p<0.05 as being statistically signi cant. SPSS version 19 (SPSS Inc) was used for the analysis.

Results
The characteristics of the sample are presented in Table 1. Our sample belongs to the working-age population and therefore had a mean age of 40.65 years with an upper age limit of the retirement age at 60 years. The mean BMI was 23.49 kg/m2 indicating that enrolled staff were generally considered overweight for males and females alike (24). Only seven (4.55%) persons working at the school had a Bachelor's degree only. The majority were married and living with their families. For the staff with children, the majority had one child only which was consistent with the One-Child policy in China. Up to 74% of the sample never smoked whereas half had the experience of drinking alcoholic beverages. In total, 40 diseases or symptoms were reported showing that each staff member suffered on average 3.81 health conditions ( Table 2). The overall prevalence of health conditions was 81.17%. The most prominent condition was pain (headache, back and/or neck pain, pain in other body parts) affecting 98 (63.64%) participants among whom nearly half were in icted with back and/or neck pain. There were six conditions with a minimum prevalence of 20%. They were, in a descending order, chronic back and/or neck pain, pain in other body parts, chronic fatigue or low energy, high blood cholesterol, headache and sleeping problems. Despite the symptoms and diseases reported, the overall mean utility was 0.945, with only 0.055 (5.5%) utility loss compared to the full health of 1 ( Figure 1). The mean VAS score was 83.00, which was 17% lower than 100 for the best health. Both health indices followed a negative distribution and were highly correlated (r=0.595, p<0.001).
The health pro le of the sample was re ected by the 5 EQ-5D domains in Figure 2. There were 74 (48.1%) participants who did not report any problems in the 5 domains, ie. they rated themselves in full health.
This proportion also indicated the ceiling effect of the EQ-5D-5L instrument in measuring HRQOL of our sample. The most affected domain was anxiety/depression. Overall, 40.26% of participants reported anxiety or depression problems, one subject had severe problems, 7 subjects had moderate problems and 54 subjects had slight problems. The second most affected domain was pain/discomfort in which 37.66% of the sample reported problems including 7 subjects with moderate severity. On the other hand, much less participants reported problems in the mobility, self-care and daily activity domains. None of these domains captured a case with problems more severe than slight problems.
As shown in Table 3, age and number of health conditions were negatively correlated with utility. Univariate analysis did not nd signi cant comparisons of health utility for demographic and socioeconomic factors. For health conditions, all conditions except respiratory system conditions were associated with signi cant utility loss. Those subjects living with any condition reported lower utility than those without. The largest utility loss was related to psychological/emotional conditions. The subjects presenting with psychological/emotional conditions had a mean utility of 0.88, which was 0.09 lower than those without. The condition-speci c multivariate GLMs were combined and summarized in Table 4. In all 12 models, age was a consistently signi cant factor as one year older was related to a mean utility loss between -0.003 and -0.002. Likewise, education level of a Master's degree or higher was signi cantly associated with a mean utility loss ranging from -0.071 to -0.044 when compared with having a Bachelor's degree only. Other demographic or socioeconomic variables were either not signi cant or had unstable signi cance levels in these models. While the GLM controlling for other variables, high cholesterol, back and/or neck pain, insomnia, fatigue, pain in other body parts, severe headache, digestive system conditions and psychological/emotional conditions were signi cant indicators for utility loss. The mean utility impairment associated with these health conditions ranged from -0.088 for psychological/emotional conditions to -0.030 for severe headache. In the GLMs evaluating multiple health conditions (Table 5), psychological/emotional conditions were associated with the biggest mean utility loss of -0.067, followed by education level of a Master's degree or higher and pain in other body parts, which were associated with a utility loss of -0.048 and -0.034 respectively. Notably male gender, marriage, smoking or having children were associated with utility gain. However, these factors did not achieve statistical signi cance in predicting health utility.

Discussion
Our study investigated the HRQOL of a sample of university staff and discovered that the mean health utility was 0.945 and mean VAS was 83.00. The utility loss was mainly caused by problems with two health domains, anxiety/depression and pain/discomfort in which 40.26% and 37.66% of participants respectively reported some problems. Multivariate analyses identi ed three risk factors related to utility loss. They were psychological/emotional conditions, higher education level and pain in body parts other than back, neck and head, where each factor was associated with a utility loss of -0.067, -0.048 and -0.034 respectively. Additionally, all self-reported health conditions were more or less related to lower utility regardless of their statistical signi cance. To the best of our knowledge this is the rst study reporting on the health utility of university staff in China.
It has not been con rmed that the health of university staff is worse than the general population although previous studies have suggested so (3,5). The direct comparison of HRQOL between university staff and the general population is lacking and the difference, if it ever existed, has not been analyzed quantitatively. Health utility measured in our study can be compared to population norms directly and quantitatively in order to gain an in-depth understanding of the HRQOL of university staff. A study describing Chinese HRQOL norms was recently published and reported on the health utility of EQ-5D-5L scores for 1,296 dwellers in ve China cities (26). In this study, the subjects in the same age range (n=965) as our sample achieved a mean utility of 0.961, which is higher than the mean utility of 0.945 of our sample. Additionally, the mean VAS score (86.28) of this cohort was also higher than that of our sample (83.00). The differences in both indices did not reach statistical signi cance re ecting that the two study populations are generally considered healthy. For individual EQ-5D health domains, our sample did worse by reporting more problems in all ve domains than the age-matched participants in that study.
The differences in the anxiety/depression and pain/discomfort domains were statistically signi cant.
There were 40.26% and 37.66% of subjects in our study that reported problems in the anxiety/depression and pain/discomfort domains respectively, compared to 27.56% and 28.7% of age-matched subjects reported in the population norm study (26). The statistical signi cances were 0.0013 and 0.0155 respectively.
Considering the social determinants of health, we compared our sample to subjects with similar socioeconomic characteristics such as employment status, health insurance status, geographical area and education level in the population norm study (26). Our university staff reported (a) signi cantly lower VAS scores and a larger proportion of pain/depression problems than those who were fully employed or with health insurance, and (b) lower VAS scores than the city dwellers regardless of other socio-economic factors. Our sample also had more problems in the pain/discomfort domain when compared to people with university degrees or higher (26). These ndings suggest that university staff may have poorer HRQOL than comparable general populations.
The above observation was further supported by comparing our study with the NHSS 2013 enrolling 188,720 Chinese across mainland China (13). Our sample had a signi cantly lower utility than the national population (0.945 vs 0.985). Given that the national sample was heterogenous with respect to health determinants, our sample was further compared to subpopulations with similar socioeconomic characteristics, especially to people with university degrees or higher considering that education level is the strongest characteristic of university staff and is closely relevant to other factors such as dwelling area, employment status, health insurance etc. Compared to the education-matched cohort, the larger utility and VAS gaps were revealed. Our school staff were 0.049 and 2.44 points lower in health utility and VAS respectively -both of which were statistically signi cant. Moreover, our sample was 7.68 times more likely to have pain/discomfort and 16.10 times more likely to have anxiety/depression.
It could be inferred that the poor HRQOL of university staff was speci cally related to the pain/discomfort and anxiety/depression domains. Our staff were three times and 7.56 times more likely to have pain/discomfort and anxiety/depression compared to the national population. Compared to other socioeconomically similar cohorts of working age, employment status or living in an urban area, the likelihood of pain/discomfort and anxiety/depression in our sample were at least three and eight times higher respectively, in addition to a signi cantly lower utility (13). However, the probability of having problems with mobility, self-care and usual activity are comparable. Our ndings were consistent with previous reports. University lecturers are highly stressed and tend to have poor mental health (5,9,27).
Lecturers with Doctoral and Master's degrees tended to have impaired mental health in comparison with their peers with Bachelor's degrees only (9).
In the NHSS, people categorized into the highest education level, university degree or higher, enjoyed the best HRQOL (12,13). This is understandable given that higher education usually translates to a good work environment, job security, insurance and living in economically developed areas. The similar association has been reported in a variety of populations (28). We found con icting results that having a Master's degree or higher is related to reduced health utility relative to those with a Bachelor's degree only.
This contradiction may reveal the limits and risks when applying the ndings from a national survey or general population to a speci c population given its unique set of occupational and socioeconomic characteristics. Actually, our sample would be categorized into the highest education group if enrolled in the NHSS (12,13,26). We analyzed one of the highest educated populations in the NHSS and provided a detailed picture of their HRQOL. The value of our study is that we have generated necessary supplementary evidence. There were several advantages in our study. The ceiling effect of the EQ-5D instrument in our study was only 48%, much lower than the 84.2% in the NHSS and other similar studies in Asia, speci cally China (33)(34)(35). It was also lower than those reported in US and European studies (36, 37). The ceiling effect of a HRQOL instrument limits its ability to measure relatively suboptimal health when subjects are generally healthy such that health utility tends to be overestimated. A modest ceiling effect in our study secures the reliability and validity of our ndings. This should be attributed to the EQ-5D-5L instrument, which has strong discriminatory power (17,18). Another advantage is that we employed a GLM for multivariate analysis, which is superior to more commonly used linear regression models, and produces more stable estimates for populations.
Some limitations with our study are notable. The sample size is small which may affect its representativeness of the population of university staff in China. However, university staff are homogenous in terms of age, education level, job duty and socioeconomic characteristics. This means that, statistically, a small sample may have good representativeness of the study population (38). Health conditions of our participants are self-reported indicating the presence of symptoms and known diseases. This may be prone to information bias. However, studies have shown that self-reported chronic conditions are accurate and even more accurate than clinical diagnosis (39,40). Moreover, asking participants to report their own health problems in some way captures their subjective perception of the severity of health conditions. The conclusion drawn from our study that university staff have a worse health status than the general population might be subject to measurement bias. Our study used the EQ-5D-5L instrument to measure HRQOL whereas the NHSS studies used the EQ-5D-3L instrument (12,13), which due to its lower discretionary power, systematically generated higher health utility than the EQ-5D-5L instrument (17). Our conclusion might also be explained by the trend that the health status of the Chinese has been decreasing over time (14,41). Following this trend, the health utility in the present study would naturally be lower because our study came up seven and 12 years later than the NHSS studies (12,13). However, our ndings were less likely to be caused by measurement bias or the natural trend.
Compared with the more recent study that also used the EQ-5D-5L instrument (26), university staff had lower health utility and signi cantly more problems in the pain/discomfort and anxiety/depression domains.

Conclusion
The present study found that that university staff in China have worse HRQOL than the general population. The health utility loss was mainly manifested with impairment in two health domains, the pain/discomfort and anxiety/depression domains. The signi cant factors associated with utility loss were having a Master's degree or higher, psychological conditions and pain in body parts other than the back, neck and head. Our ndings raise an important occupational health issue concerning university staff and call for targeted health promotion policy and programs by university management and government to bene t university staff and society overall.
awarded to HJZ, (grant number: 10-20-303-601). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Authors' contributions HJZ conceptualized and designed the study. Then he was in charge of data analysis and interpretation and nally drafting the manuscript. GFL conceptualized the study, re ned the questionnaire and collected the data. NN interpreted the results and was a major contributor in writing the manuscript. JS and WXC designed the study and provided important administrative support. MMG and YLD interviewed the participants and complied the data. JW conceptualized and supervised the study and collected important scienti c materials. All authors read and approved the nal manuscript. Distribution of health utility and visual analogue scale