Risk and protective factors for anxiety during COVID-19 pandemic

We used Sojump, a professional online questionnaire survey platform to collect information on demographics, lifestyles and risk factors for COVID-19.This study was conducted among 20,102 participants from 34 provinces in China during January, 2020 by a web-based investigation. All participants included in this analysis were all recruited according to the following inclusion criteria: 1) residents aged 14–55 years, who can fully understand the questions; 2) those who could use a smartphone to complete the standardized questionnaires voluntarily participated in this study; 3) those who answered the questionnaire for more than 100 s. After excluding those who reported invalid information on date, such as the date before the outbreak of COVID-19 or beyond the date of filling out the form, whose data was unable to ensure its authenticity, or those with outliers on age (< 1% or > 99%), therefore 19,802 participants included in the final analysis. To fill in the form, the subjects were first asked to clearly state that they agreed to participant in the investigation. All participants provided written informed consent.

Variates were collected by a web-based investigation, covering information on demographic and socioeconomics, lifestyles including age, sex, body mass index (BMI), race, smoking status, drinking status, chronic diseases (including hypertension, hyperlipidemia, diabetes, asthma, chronic obstructive pulmonary disease (COPD), chronic bronchitis, heart disease, gout, thyroid nodules, thyroid cancer, and lung cancer), and present symptoms of SARS-CoV-2 infection (including fever, cough, runny nose, sore throat, shortness of breath, fatigue, nasal congestion, headache, vomiting and diarrhea), and regular physical activity, etc. Regular physical activity was defined as exercise regularly within the recent six months [14]. Current smoker was defined as an adult who has smoked 100 cigarettes in his or her lifetime and who currently smokes cigarettes, and former smoker was defined as an adult who has smoked at least 100 cigarettes in his or her lifetime but who had quit smoking at the time of interview. Individual who had never smoked, or who had smoked less than 100 cigarettes in his or her lifetime was defined as non-smoker [15]. Current drinker was defined as at least 12 drinks in past year; former drinking was defined as any one year in lifetime but no drinks in past year; while non-drinker was defined as fewer than 12 drinks in lifetime [16]. BMI was calculated by dividing self-reported weight in kilograms by height in meters squared. Each participant’s chronic disease history information was collected by asking the question “Have you ever been diagnosed with any diseases including hypertension, hyperlipidemia, diabetes, asthma, chronic obstructive pulmonary disease (COPD), chronic bronchitis, heart disease, gout, thyroid nodules, thyroid cancer and lung cancer?”

We used Self-Rating Anxiety Scale (SAS), a self-report scale developed by Zung [17], to assess anxiety symptom using 20 self-report items. There are 15 items worded symptomatically positive rated on a 4–1 scale (“a little of the time,” “some of the time,” “good part of the time,” and “most of the time”), and 5 items symptomatically negative rated on a 1–4 scale. A standardized scoring algorithm is used to define symptoms of anxiety, with an original raw score range of 20–80. The original raw score cut-off of 40 we used would be most appropriate when the SAS is used in research [18, 19]. The raw score is then converted to an index score by multiplying 1.25. In the Chinese public, the index score has the following 4 categories [4]: the index score of “< 50,” “50–59,” “60–69,” and “≥ 70” were defined as “normal,” “mild anxiety,” “moderate anxiety,” and “severe anxiety.” The scale also showed high internal consistency and good reliability (Cronbach’s alpha was .95).

We applied summary statistics to describe baseline characteristics and anxiety level of all participants. Ordinal multivariable logistic regression analyses were performed to estimate various risk and protective factors. We entered all variables in the first logistic regression model. Odds ratio (OR) and 95% confidence interval (CI) of anxiety associated with significant factors in the first model were estimated in the second logistic regression model. Points were attributed to the variables in the second model. The risk score was calculated by summing up the single score of each factor, which can also present the number of anxiety related factors for each participant (i.e. one who had the risk score of three means that they had three anxiety-related factors). The risk scores were grouped into scores of 0 (very low risk), 1–2 (low risk), 3 (moderate risk), and ≥ 4 (high risk).

An analysis of variance (ANOVA) and a chi-square test were used to compare the average anxiety index score and percentage of anxiety of the four groups. Both linear regression models and logistic regression models were used to estimate the association of developed risk score with anxiety index score and the risk of anxiety respectively. To further investigate whether the developed risk score can predict anxiety, we calculated the area under curve (AUC). P-values were 2-sided and considered statistically significant at less than .05. Analyses were performed by SPSS version 22.0 (https://​www.​ibm.​com/​products/​software, RRID: SCR_002865), and image analyses were conducted with R Project for Statistical Computing version 3.6.0 (http://​www.​r-project.​org/​, RRID:SCR_001905), RStudio version 4.0.2 (http://​www.​rstudio.​com/​, RRID:SCR_000432), Microsoft Excel version 2019 (https://​www.​microsoft.​com/​en-gb/​, RRID:SCR_016137) and ArcMap version 10.2 (https://​desktop.​arcgis.​com/​zh-cn/​arcmap/​).

Because this study used existing epidemiological data, it was not appropriate to involve patients or the public in the research.

All participants were from 34 provinces and approximately a half (n = 10,459) were from Guangdong Province, Hebei Province, and Shanxi Province (Fig. 1). The mean (SD) age of the 19,802 participants were 25.3 (8.1) years, ranging from 14 to 55, including 51.1% (n = 10,121) men, 4.9% (n = 964) front-line medical personnel, 10.6% (n = 2096) self-employed, 2.8% (n = 558) with chronic disease, 15.2% (n = 3004) with contact history, 87.5% (n = 17,334) are characterized as non-smokers, 73.9% (n = 14,629) as non-drinker, and 2.2% (n = 436) exposed to wild animals. Characteristics and anxiety levels of the participants were summarized in Tables 1 and 2. Overall, 15,277 (77.1%) participants were without anxiety, while 2157 (10.9%), 1268 (6.4%) and 1100 (5.6%) participants had mild, moderate, and severe levels of anxiety respectively. Characteristics of participants with different levels of anxiety are also presented in Table S1.

There are 17 factors with significance in the first multivariate logistic regression model (Table S2). In the second multivariate logistic regression analysis, we found that participants aged 25–35 years, males, former smokers or drinkers, or those who were front-line medical personnel, self-employed, exposed to wild animals, or had chronic disease, suspicion of SARS-CoV-2 infection, present symptoms of SARS-CoV-2 infection, regular physical activity, contact history, or met relatives or friends coming from Hubei in the past month were significantly associated with 48, 40, 21, 17, 112, 62, 31, 93, 66, 40, 37, 15, 23% respectively increased risk of anxiety, while those aged 14–24, wore masks or had knowledge about personal protective measures were associated with 33, 29, 75% respectively decline in risk of anxiety (Table 3).

We evaluated the contribution of each factor to the model by calculating the standardized coefficients, which can make the effect of factors on anxiety comparable (Fig. 2). Knowledge about personal protective measures (β = − 1.38) contributed most to the risk of anxiety, following by front-line medical personnel (β = 0.75), chronic disease (β = 0.66), suspicion of SARS-CoV-2 infection (β = 0.51), self-employed (β = 0.48), age (14–24: β = − 0.40; 25–35: β = 0.39), race (β = − 0.37), wearing masks (β = − 0.37), sex (β = 0.34), present symptoms of SARS-CoV-2 infection (β = 0.33), regular physical activity (β = 0.31), exposure to wild animals (β = 0.27), meeting relatives or friends coming from Hubei in the past month (β = 0.21), former smoking (β = 0.19), gatherings and meetings (β = − 0.19), drinking (β = 0.15), and contact history (β = 0.14).

The risk score was created using the single score from the statistically significant 17 factors in the final model (Table S3). All participants were further divided into four groups by the risk score of 0, 1–2, 3, and ≥ 4. Anxiety index score and the percentage of anxiety significantly increased with the risk score group of 0 (very low risk), 1–2 (low risk), 3 (moderate risk), and ≥ 4 (high risk) (average anxiety index score: 35.8, 35.4, 37.3 and 45.0 respectively; percentage of anxiety: 7.6, 9.3, 17.9 and 38.9%, respectively) (Figure S1 and Figure S2).

Continuous analysis by linear regression models showed that each one-point increase in risk score was associated with a 2.97 (95% CI: 2.86, 3.09) increase in anxiety index score. In categorical analysis, we also found the moderate risk group and high risk group had a 1.44 (95% CI: 0.27, 2.61) and 9.18 (95% CI: 8.04, 10.33) increase in anxiety index score, when compared with the very low risk group.

Compared with those in very low risk, participants in low risk, moderate risk, and high risk group had 26% (95% CI: − 7.4, 72%), 172% (95% CI: 100, 270%), and 733% (95% CI: 516, 1026%) higher risk of higher anxiety level respectively (Table 4). We further attempt to develop risk score as a potential predictor of anxiety by generating ROC (Fig. 3). We found that the AUC was 0.73 (95% CI: 0.72, 0.74) for model with the developed risk score, with the cut-off point of 3.5.