Prevalence and risk factors of COVID-19-related generalized anxiety disorder among the general public in China: a cross-sectional study

Study design and participants

A cross-sectional web-based survey was conducted in China from March 16 to April 2, 2020. It adopted a combination of stratified random sampling and convenience sampling. First, according to China’s geographical and economic development level, it was divided into three sub-regions (eastern, central, and western China) (National Bureau of Statistics, 2021). Second, the Questionnaire Star electronic questionnaire platform (website: https://www.wjx.cn/) was used to complete the terminal output of the questionnaire, that was thereafter distributed to more than 30 regions in eastern, central, and western China through WeChat, QQ sharing, and web links. The sample size was expanded by snowball sampling during the survey. Participation was voluntary and anonymous. Inclusion criteria were: (a) respondents who consented to participate in the study after being informed of the survey’s purpose and procedure, (b) respondents having access to the Internet via a computer or smartphone, and (c) respondents who had stayed in China during the COVID-19 pandemic. To reduce selection bias, the exclusion criteria were: (a) not completing the questionnaire and missing data, (b) answering time of more than 20 min or less than 2 min, and (c) the questionnaire having logic errors (e.g., answers to all items of the questionnaire were the same). The target sample size was calculated according to the recommendation of 5–10 observed values per studied variable (Jiang, 1997). In this study, there were 14 variables, and the expected sample size was 140 (14 × 10). The study was conducted with the approval of the Ethics Committee of Southern Medical University (Ref ID: NFYKDX002).

Patient and public involvement

The participants and the general public were not involved in the design, recruitment, and conduction of this study.

Calculation of anxiety scores

The Chinese version of the Generalized Anxiety Disorder Scale-7 (GAD-7)—a valid self-reported psychometric scale—was used to measure individuals’ anxiety levels in the past 2 weeks during COVID-19. The questionnaire comprised seven items using a four-point Likert scale ranging from 0 (not at all) to 3 (nearly every day), with a total score between 0–21 (Ruiz et al., 2011; Johnson et al., 2019). The levels of anxiety were classified as follows: normal (0–5), mild (5–9), moderate (10–14), and severe (15–21) (Kroenke et al., 2007). In a previous study (Liu et al., 2020), a total score higher than five was deemed as anxiety. This instrument has been widely used in China and also been confirmed to have good retesting reliability and validity (Nyongesa et al., 2020; Zhang et al., 2021). Similarly, in this study, the reliability was found to be very good (Cronbach’s alpha = 0.94).

Covariates for analysis

Socio-demographic variables in the analysis included sex, age, marital status, family registration, educational attainment, occupational background, and annual family income. In addition, the following health status data were collected and included: (1) good self-reported health (yes/no) and (2) chronic diseases (yes/no). Other factors included COVID-19-related variables: (1) Have you been diagnosed with COVID-19? (yes/no), (2) Have your family or friends been diagnosed with COVID-19? (yes/no), (3) Have you been in quarantine owing to COVID-19? (yes/no), and (4) What are your perceptions of COVID-19 (uncontrolled/controlled).

First, univariate analysis was performed to analyze the effects of socio-demographics and COVID-19-related factors on anxiety. Variables associated with the anxiety score through univariate analysis were included for subsequent multiple linear regression analysis. Furthermore, ordinal logistic regression analysis was used to assess the hierarchical relationship of the results.

Ordinal logistic regression

Ordinal logistic regression was conducted to investigate the relationship between the risk factors and severity of anxiety, which was classified on the basis of participants’ GAD-7 score outcomes: (1) A (minimal level of anxiety if the scores were <5), (2) B (mild level of anxiety if the scores were ≥5 but <10), and (3) C (moderate and severe level of anxiety if the scores were ≥10). To test the robustness of the results, the ordinal logistic regression model was further adjusted for demographics, including details relating to age, sex, and family, but as these did not change the results, only the analysis results, without adjusting for these variables, have been listed below.

Interaction effects

The interaction was evaluated as a test of the coefficient of the product term formed by the factors involved. In this study, two interaction effects were assessed: (1) quarantine and self-reported health, and (2) perceptions of COVID-19 and self-reported health.

To understand the interaction effect between quarantine and self-reported health, the respondents were categorized into four groups: (a) quarantined respondents with poor self-reported health, (b) quarantined respondents with good self-reported health, (c) non-quarantined respondents with poor self-reported health, and (d) non-quarantined respondents with good self-reported health. The GAD-7 scores of these groups were then compared.

To analyze the interaction between perceptions of COVID-19 and self-reported health, the respondents were divided into four categories: (a) those who perceived that COVID-19 cannot be controlled with poor self-reported health, (b) those who perceived that COVID-19 cannot be controlled with good self-reported health, (c) those who perceived that COVID-19 can be controlled with poor self-reported health, and (d) those who perceived that COVID-19 can be controlled with good self-reported health. Thereafter, the GAD-7 score values of these groups were compared.

Statistical analysis

In the present study, categorical variables were summarized as frequencies/percentages and continuous variables as means ± standard deviations (SD). Chi-squared tests were performed to compare categorical data. Univariate analysis was used to extract influence factors and multivariate regression analysis to examine the independent predictors of anxiety. To determine interaction effects, multiple group comparisons were conducted, and significant differences between the groups were evaluated by the Least Significant Difference test assuming equal variances or Tamhane’s T2 (M) test assuming unequal variances. A two-tailed p < 0.05 was considered as statistically significant. Statistical analysis was done with SPSS for Windows, version 25.0 (SPSS Inc., Chicago, IL, USA).

Participants’ characteristics

A total of 10,980 respondents participated in the survey, but only 10,824 are included in the analysis, after excluding those who do not meet the inclusion criteria. The variation in the number of respondents is depicted through a flow chart (Fig. S1). The response rate to the questionnaire is 98.58%. The number of people in each province have been listed in Table S1. Among the 10,824 respondents, with a cutoff score of five, the overall prevalence of generalized anxiety disorder is 37.69%. Socio-demographic variables are presented in Table 1. Among the respondents, 53.83% are female (p < 0.001), 70.82% are aged <40 years (p < 0.001), 74.8% are married (p = 0.048), 74.8% come from urban areas (p < 0.001), and 68.1% have an annual family income less than RMB 100,000 (p = 0.043). Health status and COVID-19-related variables are presented in Table 2. While 17.3% respondents have poor self-reported health (p < 0.001), 5.64% have chronic diseases (p < 0.001), 10.6% have been quarantined (p < 0.001), and 88.8% perceive that COVID-19 can be controlled (p < 0.001). p < 0.05 indicate that GAD-7 scores are statistically different among these groups.

Multiple regression analysis of factors associated with anxiety during COVID-19

Multiple linear regression analysis reveal that sex, age, family register, self-reported health, chronic diseases, quarantine, and perceptions of COVID-19 are significantly correlated with generalized anxiety disorder (Table 3). Specifically, respondents who are female, are aged <40 years, and lived in rural areas are more likely to suffer from anxiety. In terms of health status, self-reported unhealthy participants and those with chronic diseases have higher levels of anxiety. Regarding COVID-19-related variables, quarantine is a risk factor for anxiety, while perceiving that COVID-19 can be controlled is a significantly protective factor for anxiety. Additionally, eastern, central, and western China data are separately analyzed (Tables S2 and S3). Furthermore, sensitivity analysis, namely logistic regression analysis, further confirm these results (Table S4).

Association between risk factors with the severity of anxiety

For this analysis, participants are categorized as A, B, and C, as described in the Materials and Methods section. The number of participants with minimal (GAD-7 score <5), mild (GAD-7 score ≥5 but <10), and moderate and severe anxiety (GAD-7 score ≥10) are 6,744, 3,355, and 725, respectively. Ordinal logistic regression analysis reveal that self-reported health, chronic diseases, quarantine, and perceptions of COVID-19 are all significantly associated with the severity of anxiety (Table S5).

Interaction effect for GAD-7 scores

Three different analyses are undertaken to evaluate the main effects of quarantine and self-reported health, as well as their interactions with the GAD-7 scores. First, both quarantine (β = 0.062, 95% CI [0.556–1.037], p < 0.001) and self-reported health (β = 0.107, 95% CI [0.921–1.321], p < 0.001) are significantly associated with GAD-7 scores, when both are included in the linear regression model. Second, in the formal interaction test between quarantine and self-reported health, the GAD-7 scores are statistically significant different among groups (p < 0.001). Finally, in the four groups, the results show that in the non-quarantined group, the relationship between GAD-7 scores and self-reported health is weak. Moreover, the difference in mean values of GAD-7 scores between the good and poor self-reported health groups is only 1.19. This difference increase to 1.34 in the quarantined group. Similarly, in the group of respondents with good self-reported health, the relationship between quarantine and GAD-7 scores is weak. Moreover, the difference in mean value of GAD-7 scores between the non-quarantined and quarantined groups is only 0.82. This difference increase to 0.97 in the group of respondents with poor self-reported health (Table 4). These results indicate that the interaction between quarantined and self-reported health status has a significant effect on anxiety (Fig. S2).

To evaluate whether there is a significant interaction effect between perceptions of COVID-19 and self-reported health on GAD-7 scores, the above analyses are duplicated using controlled/uncontrolled perceptions of COVID-19 as a dichotomous variable instead of quarantined/non-quarantined. The results suggest that perceptions of COVID-19 (β = −0.076, 95% CI [−1.198 to 0.729], p < 0.001) and self-reported health (β = 0.107, 95% CI [0.921–1.321], p < 0.001) are significantly associated with GAD-7 scores. Additionally, the interaction test between self-reported health and perceptions of COVID-19 yield a statistically significant result (p < 0.001). Multiple group comparisons show that in the group of respondents with perceptions of COVID-19 being controlled, the relationship between GAD-7 scores and self-reported health is weak. Moreover, the difference in mean value of GAD-7 scores between the good and poor self-reported health groups is only 1.11. This difference increase to 1.57 in the group of respondents with perception that COVID-19 is uncontrolled. Likewise, in the group of respondents with good self-reported health, the relationship between perceptions of COVID-19 and GAD-7 scores is weak. Moreover, the difference in mean values of GAD-7 scores between the groups with perceptions of COVID-19 being controlled and uncontrolled is 0.92. This difference increased to 1.38 in the group with poor self-reported health (Table 4). These results suggest a significant effect on anxiety by the interaction between different levels of perceptions of COVID-19 and self-reported health status (Fig. S3).