This cross-sectional survey study recruited physicians in Eastern, Central, and Western China using stratified convenience sampling. To ensure sufficient representativeness, two provinces were selected in the Eastern, Central, and Western regions at the time of sampling, respectively, that is, a total of six provinces were selected. According to the standard for the division of China Eastern, Central, and Western regions from the current China Health Statistics Yearbook (7), with the consideration of the availability of sampling physicians in this survey study, Guangdong and Zhejiang provinces were selected in Eastern China, Hubei and Henan provinces were selected in Central China, and Chongqing municipality and Guangxi Zhuang autonomous region were selected in Western China. Typical sampling was then applied to select two tertiary public hospitals and two secondary public hospitals in each selected province. That is, a total of 24 public hospitals were mainly selected nationwide in China, including 12 tertiary and 12 secondary public hospitals. Among the selected hospitals, internal, surgical, obstetrics and gynecology, and pediatrics were further selected as main research departments, where targeted physicians were selected by random sampling.

Given that our survey study aimed to investigate the level of workload among Chinese physicians while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters, the setting of the research was confined to the consulting room in outpatient clinics. Therefore, the target population was physicians who provided medical services to outpatients in outpatient clinics, those who had to have been working for at least 4 months in the outpatient clinics, and those who had to be employed full-time for at least 1 year in their current medical institution, whereas physicians who provided medical services to outpatients in outpatient clinics for <4 months, those who only provided inpatient service, and those who were graduate students or trainees were excluded in this study.

To measure the workload tethered to the related paperwork during outpatient encounters, our previous research decomposed and further divided all of the work procedures performed by physicians themselves to provide complete medical services to outpatients into the following two large groups based on a real-time task analysis of 32 Chinese physicians during outpatient encounters: “physician-patient communication work tasks” characterized by direct patient interaction, and “non-physician-patient communication work tasks” characterized by paperwork (33); and these non-physician-patient communication work tasks mainly included recording medical history, issuing medical examinations, and issuing prescriptions (33). Therefore, above-mentioned non-physician-patient communication work tasks physicians themselves performed was considered as “paperwork” during outpatient encounters in this study. Given that different types of work tasks might result in different cognitive demands and resources demands, this survey study clearly explained the detailed work tasks involved with assessed workload to the targeted physicians before they filled in the questionnaire.

The Chinese physician mental workload scale developed by our research team in 2018 based on the combination of dimensions of NASA-TLX scale and Subjective Workload Assessment Technology (SWAT) frameworks (11) was the basis of our developed questionnaire survey in this study, which included six dimensions (mental demands, physical demands, temporal demands, perceived risk, frustration level, and performance), 12 items, and physician characteristics (e.g., gender, age, marital status, average monthly income, educational level, professional title, working years in the current medical institution, hospital level, hospital nature, personnel, department, working hours per week, number of outpatients serviced per day, self-rated health status) with good reliability and validity (Cronbach alpha = 0.81); and moreover, pairwise comparisons of these six dimensions constituted a total of 15 comparisons, and these comparisons were used to determine the weighting coefficient for each comparison, where the weight of each dimension was equal to the number of times that dimension was selected divided by 15 (11). In the questionnaire, we added several questions to collect other demographic information on working hours per week in outpatient clinics, amount of time spent per patient and self-rated outpatient satisfaction, reported by the participating physicians.

Then, we conducted a pre-survey on site in October 2020, to validate the developed measurement tool in 10 physicians who just finished the provision of the outpatient services in the outpatient clinic of a tertiary public hospital in Wuhan, Hubei. According to their comments or feedback, context-specific adjustments were then made to improve the accuracy and clarity of the questionnaire. Because of the impact of the COVID-19 epidemic in 2020, we further used wenjuanxing, a widely-used website for conducting surveys in China, to create an electronic questionnaire with which to survey physicians in this study.

This nationwide survey was conducted from November, 2020 to February, 2021. To improve the efficiency of data collection in the selected hospitals, a unique two-dimensional code of the electronic questionnaire was generated for each hospital. Prior to the beginning of the survey, an informed consent of the outpatient managers in each selected hospital was first requested and obtained, and they were then invited and volunteered to play the role of the project manager in their hospitals in this questionnaire survey. Subsequently, we sent the unique two-dimensional code of the electronic questionnaire to these outpatient managers of the corresponding hospital, and they then sent the two-dimensional code to the targeted department groups of physicians via WeChat or Tencent QQ group, where physicians who met the inclusion criteria for the targeted population were further invited to participate in this survey. Participants could scan the two-dimensional code of the electronic questionnaire via their phones to access and complete the electronic questionnaire. Before the formal survey, we introduced the purpose of the survey, provided the definition of physician workload and its involved non-physician-patient communication work tasks characterized by paperwork during outpatient encounters, and guaranteed that the survey data would not be used for other purposes. After an individual's consent was obtained, the survey was conducted accordingly. A WeChat or Tencent QQ account and mobile Internet Protocol address could be used to complete the electronic questionnaire only once. Given that the sample size should be recommended to be at least 10–15 times as many as the items of the scale (55) and should be also generally recommended to be at least 20 times as many as the variables which are considered to be included in the regression model, to improve the scale of the sample, these physicians who completed the questionnaire were also encouraged to share the survey website link to their Wechat Circle of Friends, WeChat or Tencent QQ group, where some physicians who met the inclusion criteria for the targeted population could participate in this questionnaire survey. The study was approved by the Ethics Committee of Tongji Medical College of Huazhong University of Science & Technology (No. IORG0003571).

Given that it's difficult to objectively quantify physicians' workload tethered to paperwork during outpatient encounters, we therefore used the Chinese physician mental workload scale to measure the physician workload while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters. That is, we only used the mental workload as the measure of physician workload while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters. The response to each of the 12 items was given based on a 10-point bipolar scale, ranging from 0 to 100; and for five of the six dimensions, i.e., mental demands, physical demands, temporal demands, perceived risk and frustration level, a score of 0 presents the lowest task load, whereas the dimension of performance is reverse-scored, with a score of 0 indicating the most successful performance of the task and the highest level of satisfaction with his/her performance (11). In this study, the calculation of physician workload followed the method from NASA-TLX scale (22); and therein the average score of all items of a corresponding dimension was the dimension score, whereas each dimension score was multiplied by the weight of the corresponding dimension and the sum of the scores was the total score of physician workload (11).

We performed exploratory latent profile analysis (LPA) based on the six dimension indicators of physician workload tethered to non-physician-patient work tasks characterized by paperwork during outpatient encounters in this study, where we explored homogenous subgroups in a heterogeneous group and then observed continuous variables in each subgroup. LPA, a “person-centered” statistical approach, belongs to finite mixture modeling, which can identify and describe “hidden groups” within a population (18, 54, 56). Data for the six dimension indicators of physician workload were input into the LPA, with one class initially and additional classes added incrementally, until a unique solution could not be determined; and therein the maximum parameter estimates with standard errors were applied. The model identification was checked using 200 initial stage starts and 200 final stage starts.

We tested different latent class models that categorized the physician workload patterns into one, two, three, four, five, and six groups. To determine the most appropriate latent class model, the best fit model was identified using the following key model indexes: Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size Adjusted BIC (ABIC), Lo-Mendell-Rubin (LMR), adjusted likelihood ratio test and bootstrap likelihood ratio test (BLRT) and Entropy. A lower value of AIC, BIC and ABIC represents better fitness of data into the estimated model (18, 54, 56); LMR and BLRT compare the model fit between two neighboring models (for example, k-1class model vs. k-class model), and a significant p value indicates that the k-class model fits the data better than the k-1-class model (54, 56). Entropy was used to assess the accuracy of classification in the estimated model, with a higher value indicating better classification, and the smallest group should have a minimum of 5% of the total sample in order to avoid over-stratification (56). A four-class model was identified in the LPA. Each participating physician was assigned into one of the physician workload subgroups with the highest probability.

Then, differences in physician workload scores among different workload subgroups were tested using the one-way analysis of variance (ANOVA) or Kruskal-Wallis rank tests. Subsequently, multinomial logistic regression analysis was performed to examine the potential relationship between the latent workload subgroups and demographic variables; and therein all demographic variables were set as independent variables since there was no collinearity problem between these demographic variables in this study, where the variance inflation factor was <10 (range: 1.07–2.65). The statistical analyses were performed using STATA (version 15.0) and Mplus (version 7.0).

In total, 2,038 online responses were received; of these, 104 responses were excluded because the time taken to answer the questionnaire was <60 s, or because they were not physicians, or they were physicians, but did not provide medical services to outpatients in outpatient clinics, and therefore, 1,934 eligible responses were remained. The detailed demographic characteristics of the 1,934 participating physicians are presented in Table 1. Among these physicians, 45.9% (887/1,934) were female, 44.1% (852/1,934) aged 31–40 years, 82.0% (1,585/1,934) were currently married, 63.8% (1,234/1,934) were from tertiary A hospitals, 38.0% (735/1,934) were from Eastern China, and 46.6% (902/1,934) rated health status as “moderate”. Moreover, the total mean physician workload score was 62.92 (SD = 14.70) while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters (Table 3).

In order to classify and identify the optimal model, this study extracted and compared the model solutions from the one-class to six-class models. According to model indexes, the best fitting LPA was the four-class model (Table 2), which had the lowest AIC (95,620.337), BIC (95,804.059), and ABIC (95,699.218). The p-values of the LMR test (<0.001) and BLRP test (<0.001) indicate that the four-class model was statistically significant. Moreover, the Entropy value (0.866 > 0.800), the proportion of physicians of the least class (8.8% > 5.0%) (Table 2) and the average profile probabilities of physicians in each category ascribed to each potential category (range: 0.920–0.930) also indicate a better classification in the four-class model.

Therefore, LPA identified four distinctive latent subgroups of physician workload tethered to the non-physician-patient communication work tasks characterized by paperwork during outpatient encounters. Figure 1 shows the latent subgroups of physicians (Classes 1, 2, 3, and 4), and their proportion (8.8, 34.0, 42.1, 15.1%, respectively), and the mean levels of six dimensions of physician workload, which can be distinguished as having relatively low (Class 1), medium (Class 2), high (Class 3) and very high levels (Class 4) of physician workload. That is, 8.8% (n = 170) were identified as low workload physicians (Class 1), 34.0% (n = 658) as medium workload physicians (Class 2), 42.1% (n = 814) as high workload physicians (Class 3) and 15.1% (n = 292) as very high workload physicians (Class 4). Table 3 shows comparisons of between different workload subgroups on physician workload scores, and these significant differences in total physician workload score and its dimensions scores were all found between different workload subgroups, indicating a reliable and valid grouping for physician workload tethered to the non-physician-patient communication work tasks characterized by paperwork during outpatient encounters.

Specifically, the “very high workload group” was characterized by the highest scores on the dimensions of mental demands, physical demands, temporal demands, perceived risk and frustration level, and the most successful performance of the task and the highest level of satisfaction with his/her performance, also named the “very high workload perception & very high self-evaluation group”. The “high workload group” further distinguished itself from the medium and low workload groups through higher scores on these five dimensions and more successful performance of the task and higher level of satisfaction with his/her performance, also named the “high workload perception & high self-evaluation group”, “medium workload perception & medium self-evaluation group”, and “low workload perception & low self-evaluation group”, respectively. Moreover, the gap in the physician workload scores reached 2.15 times between the very high and low workload subgroups [80.82 (SD = 12.08) vs. 37.55 (SD = 8.16), p < 0.001] (Table 3).

To further determine differences between different latent subgroups of physician workload across the demographic characteristics, multinomial logistic regression was performed to identify the significant determinants of the subgroups. Using “very high workload group” as the base outcome, we had following results (Table 4). Age, marital status, educational level, average monthly income, working years in the current medical institution, area, hospital level, hospital nature, department, working hours per week, outpatient working hours per week, number of outpatients serviced per day, amount of time spent per patient, self-assessed health status and self-assessed outpatient satisfaction were all significant factors that influenced the subgroups of physician workload tethered to non-physician-patient communication work tasks characterized by paperwork during outpatient encounters.

Specifically, compared to those aged 20–30 years, physicians aged 31–40 years or 41–55 years were more likely to belong to the “low (RRR (Relative Risk Ratio) = 2.498, p = 0.016; RRR = 2.923, p = 0.024, respectively) or medium (RRR = 2.061, p = 0.010; RRR = 1.977, p = 0.058 <0.10, respectively) workload groups” as compared with the odds of the “very high workload group”. Physicians being married were less likely to be assigned into the “low (RRR = 0.492, p = 0.047) or medium (RRR = 0.592, p = 0.056 <0.10) workload groups”. For educational level, physicians with higher educational levels were less likely to have a higher level of workload; compared to those with a PhD degree, physicians with undergraduate degrees were less likely to belong to the “medium (RRR = 0.386, p = 0.005) or high (RRR = 0.514, p = 0.045) workload groups”, and physicians with a postgraduate degree were less likely to be assigned into the “medium workload group” (RRR = 0.458, p = 0.015) as compared with the odds of the “very high workload group”. Physicians with an average monthly income of <5,000 RMB or 5,001–10,000 RMB were less likely than those with an average monthly income of 10,001–15,000 RMB to belong to the “medium workload group” (RRR = 0.601, p = 0.084 <0.10; RRR = 0.639, p = 0.046, respectively).

Compared to those working in the current medical institution for 1–5 years, physicians who worked 6–10 years or 16–20 years in the current medical institution were less likely to belong to the “low (RRR = 0.477, p = 0.026; RRR = 0.583, p = 0.030, respectively) or medium (RRR = 0.294, P = 0.009; RRR = 0.335, p = 0.002, respectively) workload groups”. For area, physicians who were from Eastern China were less likely than those from Central China to belong to the “low (RRR = 0.600, p = 0.058 <0.10) or high (RRR = 0.707, p = 0.073 <0.10) workload groups”. Physicians in tertiary A hospitals or secondary hospitals were more likely than those in tertiary B hospitals to be assigned into the “high workload group” (RRR = 1.645, p = 0.069 <0.10; RRR = 1.810, p = 0.043, respectively). Physicians in public specialized hospitals were more likely than those in public general hospitals to belong to the “low workload group” (RRR = 2.735, p = 0.047). Moreover, compared to those in Obstetrics and Gynecology, physicians in Internal or Pediatrics were more likely to belong to the “medium (RRR = 2.244, p = 0.004; RRR = 2.974, p = 0.004, respectively) or high (RRR = 2.091, p = 0.006; RRR = 3.061, p = 0.002, respectively) workload groups” as compared with the odds of the “very high workload group”.

For working hours per week, physicians who had longer working hours per week were likely to be assigned into the “very high workload group”; compared to those with more than 60 working hours per week, physicians who worked no more than 40 h or 41–60 h were more likely to belong to the “low (RRR = 6.243, p < 0.001; RRR = 2.510, p < 0.001, respectively), medium (RRR = 5.639, p < 0.001; RRR = 2.176, p < 0.001, respectively) or high (RRR = 3.160, p = 0.003; RRR = 1.632, p = 0.003, respectively) workload groups” as compared with the odds of the “very high workload group”. Compared to those who worked 16–24 h per week in outpatient clinics, physicians with 8–16 outpatient working hours per week were more likely to be assigned in to the “low (RRR = 2.001, p = 0.035), medium (RRR = 2.468, p < 0.001), or high (RRR = 1.881, p = 0.006) workload groups”, and physicians who worked no more than 8 h in outpatient practice were more likely to belong to the “medium workload group” (RRR = 1.799, p = 0.007) as compared with the odds of the “very high workload group”.

Physicians who saw more than 50 outpatients per day were less likely than those with 26–40 outpatients serviced per day to be assigned into the “low (RRR = 0.421, p = 0.008) or middle (RRR = 0.579, p = 0.022) workload groups”. For amount of time spent per patient, the odds of belonging to the “very high workload group” increased with the time that the participating physicians spent on per patient; compared to those with 10–15 min spent per patient, physicians with no more than 5 min or 5–10 min spent per patient were more likely to be assigned into the “medium (RRR = 1.991, p = 0.011; RRR = 1.773, p = 0.019, respectively) or high (RRR = 1.813, p = 0.023; RRR = 1.616, p = 0.040, respectively) workload groups” as compared with the odds of the “very high workload group”. For self-assessed health status, physicians with worse self-assessed health status were more likely to belong to the “very high workload group”; compared to those with good health status, physicians who rated health status as “fair” were less likely to be assigned into the “low (RRR = 0.703, p =.004), medium (RRR = 0.522, p = 0.001), or high (RRR = 0.658, p = 0.022) workload groups”, and physicians who rated health status as poor were also less likely to be assigned into the “medium workload group” (RRR = 0.373, p = 0.007). Moreover, physicians who rated outpatient satisfaction as “fair” were more likely than those who rated outpatient satisfaction as “high” to belong to the “low (RRR = 3.569, p < 0.001) or medium (RRR = 2.981, p < 0.001) workload groups” as compared with the odds of the “very high workload group”.

Therefore, according to the results of multinomial logistic regression analysis, compared with the other latent workload groups, physicians belonging to the “very high workload group” were more likely to be younger, married, those who had worse health status, lower educational levels and lower average monthly incomes, those who were from Eastern China, and worked more years in the current institution, more hours per week and longer outpatient hours per week, and those who worked in public general hospitals, tertiary B hospitals and Obstetrics and Gynecology, and saw more than 50 outpatients per week with more time spent on per patient, but with high outpatient satisfaction.

Overall, Chinese physicians reported medium levels of workload while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters. In this study, we identified four distinctive latent workload classes (that is, workload subgroups) among Chinese physicians: 15.1% were identified as very high workload physicians, compared with 8.8% as low workload physicians, 34.0% as medium workload physicians, and 42.1% as high workload physicians. This is a result of the combined effect of the six dimension indicators of physician workload. The “very high workload group” contributed disproportionally across all the six dimension indicators. Previous studies usually identified individuals with high workload among the evaluated physicians using single workload indicators (49) through the quartiles (50), or threshold values for workload [e.g., 50% of overall workload (51), >55 (12), or >60 (52) of NASA-TLX composite workload scores].

This study further showed that great variations in the latent workload subgroups among Chinese physicians across demographic characteristics exist. Compared with the other latent workload groups, physicians who were younger, married, those who had worse health status, lower educational levels and lower average monthly incomes, those who worked more years in the current institution, more hours per week and longer outpatient hours per week, those who worked in public general hospitals, tertiary B hospitals and Obstetrics and Gynecology, and those who saw more than 50 outpatients per day with more time spent on per patient were more likely to belong to the “very high workload group”, while performing non-physician-patient communication work tasks characterized by paperwork during outpatient encounters.