Study design and participants
Data were derived from a large-scale cross-sectional study of social media-based National College Student Survey on Sexual and Reproductive Health (NCSS-SRH), which comprehensively measured youth SRH and related factors in China from November 2019 to February 2020. The study was commissioned by China Family Planning Association (CFPA) and collected by China Youth Network (CYN).
Higher education institution (HEI) study sites were selected using a multi-stage sampling technique considering school locations (eastern, central, or western regions), school types (vocational college, or university), grades, and sex, under which the probability of a university or a vocational college being selected was proportional to the number of students reported in the Educational Statistics Yearbook of China 2018 [29]. An anonymous self-administered questionnaire was distributed to students via the two most widely used social media platforms (WeChat and Sina Weibo) snowballing through the official societies or associations of CYN under each HEI site. To avoid repetitive responses, we used cookie-based duplication protection to automatically restrict each device from submitting the questionnaire more than once.
A total of 55,757 participants responded and submitted the questionnaire. Inclusion criteria for eligible participants in our analyses were (1) who provided informed consent, (2) aged between 17 and 24, (3) were enrolled in universities or colleges within China during the survey collection, (4) whose residence before college was in China and (5) who answered all questions and passed the consistency checks and logic verification. The final sample comprised 51,124 university and college students.
Ethical approval was obtained from the Institution Review Board of Tsinghua University.
Exposures
Previous studies have found that parents' marital status was associated with adverse SRH outcomes, such as sexual abuse, HRSB, and unintended health outcomes [9,10,11,12,13,14,15,16,17,18,19,20,21], which have been discussed in the background part. Divorce is the most directly objective measure of parents’ marital status, and perceived parental marital quality is the subjective measure of parents’ marital status. Considering every single measurement may have limitations from the validity or reliability of measurement or from theoretical aspects, we included these two measurements in our study. Perceived parental marital quality as an emotional dimension of the home environment and divorce as significant family events were identified as two exposures in this study. The parental marital quality was assessed by the following question: “How would you rate the overall marital quality of your parents when you were growing up on a scale of 0 to 10?”. Although another study used a 5-point Likert scale [8], we chose a scale of 0 to 10 in order to capture more subtle differences. Higher scores reflect a better relationship. Standardized Z scores of perceived parental marital quality were computed and used in all regression analyses.
All participants were asked “Did you experience the following situations in your family?” with parents’ divorce as one choice. For those who experienced parents’ divorce, they were required to choose their parents’ divorce timing from early and middle childhood (around 12 years old or younger), adolescence (around 13–18 years old), and young adulthood (around 19 years old or older). Participants who experienced parents’ divorce were defined as children from “divorced family” (coded as 1) whereas those who did not were defined as from “intact family” (coded as 0).
Outcomes
Based on the commonly used core concepts of SRH globally [30, 31], and previously mentioned SRH indicators measured in other studies examining the impact of parental marital status on young people’s SRH, we included sexual abuse, HRSB, and unintended health outcomes as the indicators of SRH in our study. The definition and variable selection of sexual abuse, high-risk sexual behaviors, and unintended health outcomes were based on current literature and China’s cultural background. Thirteen dichotomous variables (Yes/No) were generated to examine sexual experiences, HRSB, unintended health outcomes and sexual abuse. The respondents were asked to report whether they experienced any event separately. Sexual debut before the survey was used to indicate sexual intercourse experiences.
High-risk sexual behaviors
HRSB were evaluated by the first sexual intercourse with a non-intimate partner, casual sexual intercourse, multiple sexual partners and no contraceptive use in the last sexual intercourse [1, 13, 17, 19]. While first sexual intercourse with non-intimate and other casual sexual intercourse may not always have adverse consequences, these behaviors are more likely to pose higher SRH risks. This is the case primarily because these types of sexual behaviors tend to imply a lower sense of morality and responsibility in the Chinese context by contradicting loyalty to the partner or commitment to a relationship [32], and are more likely to be associated with unprotected sex [33, 34].
The first sexual partner of the respondents was dichotomized into “intimate” (defined as one’s boyfriend or girlfriend, wife or husband) and “non-intimate”. The accumulative numbers of sexual partners among sex-experienced respondents were classified into “one” and “multiple”. Information on casual sex (such as one-night stands) and no contraceptive use in the last sexual intercourse was obtained by asking “Have you ever had casual sex?” and “Did you use any contraceptives in the last sexual intercourse?”, with the answers “yes” or “no”.
Unintended health outcomes
Unintended health outcomes were assessed by unintended pregnancy (or pregnancy of the female partners in male respondents), induced abortion (or induced abortion of the female partners in male respondents), diagnosed STIs, and any genito-urinary system symptoms including urethral or vaginal discharge, painful urination, genital inflammation, genital ulcers, genital itching, genital herpes, and haematuria or vaginal bleeding [10, 11, 14]. Information on unintended health outcomes was obtained by asking, “Have you or your partner(s) ever had the following experiences?” with unintended health outcomes as options self-selected by participants. The experiences of sexual partners were only asked when measuring unintended pregnancy and induced abortion. Unintended health outcomes were categorized into “yes” and “no” as above.
Sexual abuse
Participants’ experiences of sexual abuse were measured on four fronts of progressive severity: forced to expose the genitals, forced to be kissed or fondled the genitals, forced to have oral intercourse, and forced to have penetrative vaginal or anal intercourse [35]. Respondents reported whether they had experienced these sexual abuse experiences and the results were recorded as “yes” and “no”.
Covariates
Socio-demographic covariates included age, household economic status, home district, home region and parental education levels. All variables were coded as categorical variables. Participants were divided by age into three groups: 17–18 years, 19–20 years, and 21–24 years. Household economic status was assessed based on a 7-point scale according to self-appraisal, and higher scores indicate better family financial status. It was classified as “low” for those who chose 1–3 points, “medium” for 4 points and “high” for 5–7 points. The home district was defined as the place of residence before college enrollment, which was categorized into “urban” and “rural”. The home region was used to describe the residential area before college enrollment, which was classified into “eastern,” “middle” and “western”. Parental education levels were measured by the highest education levels of the mother and father separately. They were categorized into “no formal school” “elementary school,” “junior and senior high school” and “college and above”.
Statistical Analysis
Although we use part of the data from NCSS-SRH which has a large sample size, we still conducted power calculations to determine whether the sample size for this study assures adequate power to detect statistical significance. We adopted commonly used assumptions, including a 5% margin of error, a power of 80%, and adding a 10% non-response rate to the minimum sample size estimated. Comparing all available studies, we found that the prevalence of sexual abuse among the general youth population in China is the relatively lowest prevalence compared to the prevalence of other outcomes chosen in this study. Since no study examined the prevalence of sexual abuse in adolescents from divorced families in China, we used the prevalence obtained from this study. As this figure is far below the Chinese average [36], it can ensure the sample size calculated will be large enough in Chinese context. We use the prevalence of 1.09% (from intact families) and 2.59% (from divorced families) of young people forced to have penetrative vaginal or anal intercourse which is the lowest prevalence in sexual abuse collected by this study, and the final sample size was calculated as 28,189, which was less than the sample size collected in this study. So, the final sample size of this study was large enough to provide sufficient power.
To improve the national representativeness of the sample, poststratification weights were added based on school locations, school types, grades, and sex according to the record of the Educational Statistics Yearbook of China 2018 [29]. This survey weight was added to all the statistical analyses of this study.
Descriptive statistics and regression analysis were performed. All demographic, socio-economic and SRH variables were presented in two groups: intact family and divorced family. Continuous variables were reported as the mean and standard deviation (SD) while categorical variables were described as percentage. We used multivariate logistic regression to assess the relationship of perceived parental marital quality and parents’ divorce with sexual intercourse, HRSB, unintended health outcomes and sexual abuse, stratified by sex. Since a statistically significant (P-value) result can be obtained if the sample is large enough, but the effect sizes are not directly affected by sample sizes [37]. We not only reported the P values, but also used the odds ratio (OR) and 95% confidence intervals (95% CIs) in our study as indicators of practical significance to show the size of the observed effect. Odds ratio (OR) and 95% confidence intervals (95% C.I.) were calculated to indicate the extent of association between the exposure and outcome variables. Considering the effect of a large sample size, P < 0.001 (two-sided) was considered as statistical significance. But we also reported the results when P < 0.05 and P < 0.01, given the fact that this is the level of statistical significance chosen for general studies. Age, household economic status, home district, home region and parental education levels were adjusted in all regression models. All data analyses were performed using Stata 16.