Development of a risk estimation model for condomless sex among college students in Zhuhai, China: a cross-sectional study

A cross-sectional survey was conducted among college students from six out of the seven universities in Zhuhai City, China. Inclusion criteria for participants included 1) college students enrolled in six universities in Zhuhai; 2) those who consented to participate in the survey and completed the questionnaire, and minors who had consent from their guardians; 3) self-reported history of sexual intercourse (including vaginal, anal, or oral sex).

Exclusion criteria for participants included 1) those with mental illness and intellectual disability; 2) those who did not fully understand the informed consent process and did not consent to the questionnaire, and minors who had no consent from their guardians; 3) those who self-reported no sexual experience.

The sample size estimation formula for a cross-sectional study was used to calculate the required sample size. With a proportion of condomless sex at last intercourse among college students in Zhuhai (P) of 7.3%, a precision error (d) of 0.1P, and a confidence level of 95%, the required sample size was calculated to be 5366, considering a nonresponse rate of 10% [24]. A multistage sampling method was used to recruit college students in Zhuhai City from September to October 2019. In the first stage, probability proportional to size sampling was used to determine the sample size of each university, where the probability of selecting a student was proportional to the total number of students in each university, ensuring the representativeness of the sample and improve the accuracy of the study results [25]. In the second stage, all departments/colleges in each university were classified into one of the four fields of major (literature, science, arts and medicine). One or two specific majors were then randomly selected from each field of major. In the third stage, random cluster sampling was used to select between one and five classes from each grade of the specific majors. All students in the selected classes were invited to self-administer the online questionnaire via a link. Using the sampling method described above, a total of 12,235 students were recruited, and 1430 students who had sex in the last year were included in this study. Supplementary Fig. 1 shows the schematic presentation of sampling procedure for this study. The study was approved by the Ethics Committee of the Zhuhai Centre for Disease Control and Prevention.

Data were collected using an electronic questionnaire. The survey was organized and coordinated by the Department of Epidemiology and Biostatistics, School of Public Health, Guangdong Pharmaceutical University. The questionnaire investigators received uniform training. Senior investigators examined the collected questionnaires for quality control to ensure the accuracy of the data.

The self-designed electronic questionnaire was developed based on the standardized National AIDS Sentinel Surveillance Questionnaire, specifically designed for college students [26], and the existing literature [6, 27‐31]. The questionnaire collected the following data: 1) Sociodemographic characteristics included sex, age, the field of major, grade, residence, monthly disposable income, and resident student. 2) Behavioral characteristics included sexual orientation, sexual partner seeking, ever having had vaginal sex, ever having had oral sex, ever having had anal sex, ever having had heterosexual intercourse, age at sexual debut (< 18 years, ≥18 years), condom use at first sex, experience of IPV [32, 33], the number of types of IPV experienced, ever having been tested for HIV, awareness of HIV-related knowledge and willingness to receive HIV-related education. IPV was measured with four items: ① Do you have an intimate partner (boyfriend/girlfriend, spouse, or other sexual partners)? ② Have you ever been threatened with violence, treated with violence (e.g. slapping, hitting, kicking, pushing, throwing things at you) or fought with your intimate partners? ③ Have your intimate partners ever insisted on having sex with you or force you to have sex when you don’t want to? ④ Have your intimate partners ever verbally threatened, demeaned in front of others, ridiculed for your appearance, forced to get high or drunk, or stalked, or having property destroyed or damaged? [32, 33]. The types of IPV experienced were physical violence, sexual violence, and psychological violence. There were five items ‘no intimate partner, no experience, one type, two types, three types’. 3) Substance use before sex included smoking before sex, using e-cigarettes before sex, drinking alcohol before sex, and using club drugs before sex. 4) Psychosocial factors included attitudes, norms, and self-efficacy for condom use which were measured by the Sexual Risk Behavior Beliefs and Self-efficacy (SRBBS) scale [28] with a Cronbach’s α of 0.943. A 10-item short version of the Big Five Inventory [29], Form V of the Sensation Seeking Scale for adolescents [30], and the 10-item Sexual Compulsivity Scale [31] were used to measure the five-factor model of personality, sensation seeking, and sexual compulsivity, and their internal consistency estimates (i.e., Cronbach’s α) were 0.652, 0.857, and 0.941, respectively. Responses to the SRBBS and sexual compulsivity were given on four-point scales labeled ‘1’ (strongly disagree), ‘2’ (disagree), ‘3’ (agree), and ‘4’ (strongly agree). Responses to the five-factor model of personality and sensation seeking were given on five-point scales labeled ‘1’ (strongly disagree), ‘2’ (disagree), ‘3’ (unknow), ‘4’ (agree), and ‘5’ (strongly agree). 5) Condomless sex, which was the outcome variable in this study, was defined as not using a condom during the last sexual intercourse in the last year [6].

In the univariate analysis, non-normal continuous variables are expressed as the median (M) and interquartile range (IQR) and were compared using the Mann-Whitney test. Categorical variables were compared using the χ² test.

The dataset was randomly split into a derivation cohort (70%) and a validation cohort (30%). We obtained all variables from the questionnaires and used the least absolute shrinkage and selection operator (LASSO) regression to select potential variables associated with condomless sex with 10-fold cross-validation. LASSO regression can be used to screen variables and adjust for complexity while fitting a generalized linear model. It eliminates the weaker factors with greater penalties, whose coefficient shrinks towards zero, and keeps the most vital factors in the model [20, 23]. Stepwise multivariable logistic regression analysis was then used to determine the final independent predictors of condomless sex to construct the nomogram [23, 34]. The nomogram proportionally converts each regression coefficient in the multivariable logistic regression into a score. Finally, the total score for each participant was used to calculate the predicted probability of condomless sex by functional transformation [19]. We used 1000 bootstrap resamples as internal validation to estimate the accuracy of the model. The area under the receiver-operator characteristic curve (AUROC) and the calibration curves were used to evaluate the discrimination and calibration of the model, respectively. All analyses were performed using R, version 4.0, and P < 0.05 was considered statistically significant.

A total of 1430 students from six universities in Zhuhai who had sex in the last year were included in this analysis, of whom 18.2% (260/1430) had condomless sex at last intercourse. As shown in Table 1, there were 57.6% males and 42.4% females with a mean age of 20.98 (1.43) years. Most of them had an urban residence (73.1%), had a disposable income of more than 2000 yuan per month (73.6%) and lived in school dormitories (96.8%) (Table 1). As shown in Table 2, the most of the students identified themselves as heterosexual (85.5%) and had had heterosexual intercourse (94.6%). The proportions who had ever had vaginal sex, oral sex and anal sex were 91.3%, 64.7% and 8.9%, respectively. Approximately one in five (20.3%) students had their first sexual intercourse before the age of 18, of whom 23.4% did not use a condom the first time they had sex. The proportion of students who had ever experienced IPV was 18.7%. Only 7.6% of the students had been tested for HIV, and 83.4% had acquired HIV-related knowledge. However, 6.4% of students were reluctant to accept HIV-related education (Table 2). As shown in Table 3, The proportions of smoking, using e-cigarettes, drinking alcohol and using club drugs before sex were 24.3%, 8.4%, 35.1% and 2.1%, respectively. The median scores for attitude towards condom use, condom use norms, and condom use self-efficacy were 15.0 (IQR, 12.0–15.0), 15.0 (IQR, 12.0–15.0), and 14.0 (IQR, 11.0–15.0), respectively. The median score for sexual compulsivity was 18.0 (IQR, 11.0 ~ 21.0) (Table 3).

Compared to the condom-using group, the condomless group had higher proportions of students who were non-heterosexual (P = 0.002), had oral (P = 0.002) and anal sex (P < 0.001), had the first sex before the age of 18 (P < 0.001), did not use condoms at first sex (P < 0.001), and had experienced more than two types of IPV (P < 0.001) (Table 2). In addition, the condomless group was more likely to smoke (P = 0.002), use e-cigarettes (P < 0.001), drink alcohol (P < 0.001) and use club drugs (P < 0.001) before sex than the condom-using group (Table 3), while the condomless group was less likely to have vaginal sex (P < 0.001), have heterosexual sex (P = 0.001) and be willing to receive HIV-related education (P = 0.002) than in the condom-using group (Table 2). In terms of scores, attitude towards condom use (P < 0.001), condom use norms (P < 0.001), and condom use self-efficacy (P < 0.001) were lower in the condomless group than in the condom-using group, while sexual compulsivity (P < 0.001) was higher in the condomless group than in the condom-using group (Table 3).

All the potential factors associated with condomless sex were included in the LASSO regression. After selection by LASSO regression (Supplementary Fig. 2), eight variables were retained, including ever having vaginal sex, ever having anal sex, having heterosexual intercourse, condom use at first sex, more types of IPV, attitudes towards condom use points, condom use self-efficacy points and sexual compulsivity points. After multivariable analysis, six variables remained that were independently statistically significant predictors of condomless sex. As shown in Table, students who had experienced more types of IPV (aOR, 1.58; 95% CI, 1.31 ~ 1.92) and had anal sex (aOR, 1.75; 95% CI, 1.06 ~ 2.84) were more likely to have condomless sex. However, students who had heterosexual intercourse (aOR, 0.37; 95% CI, 0.21 ~ 0.70), used condoms at first sex (aOR, 0.20; 95% CI, 0.14 ~ 0.27), had high scores on attitudes towards condom use (aOR, 0.87; 95% CI, 0.80 ~ 0.95) and self-efficacy for condom use (aOR, 0.84; 95% CI, 0.78 ~ 0.90) were less likely to have condomless sex (Table 4).

These independent predictors were used to construct a nomogram to estimate the individualized risk of condomless sex (Fig. 1). In the nomogram, the options for each variable correspond to a particular point on the top row. The points of each variable are added together to give a total point, which corresponds to the probability on the bottom row. In addition, we have provided a list of the specific total point and corresponding probability of condomless sex in Supplementary Table 1. Based on the specific total point and the corresponding probability of condomless sex, we divided the college students into three subgroups. The possibility of condomless sex was less than 5% in the low-risk group (total points < 30), between 5 and 50% in the moderate-risk group (30 ≤ total points ≤ 170), and more than 50% in the high-risk group (total points > 170). We also compared the actual proportion of condomless sex with the predicted possibility in the three classified subgroups (Table 5). The proportion of condomless sex was 2.6% in the low-risk group which accounted for 29.9% of all students. The proportion of condomless sex was 19.1% in the moderate-risk group which accounted for 60.9% of all students. The proportion of condomless sex was 63.4% in the high-risk group which accounted for 9.2% of all students. The actual possibility of condomless sex differed significantly (P < 0.001) between the three subgroups.

In the derivation and validation cohorts, the AUROC of the nomogram prediction model was 0.83 (95% CI, 0.80–0.85) and 0.85 (95%CI: 0.80–0.90), respectively (Fig. 2). The calibration plots showed graphically that the predicted estimate was in good agreement with the ideal value (Fig. 2).