Associations of social media and health content use with sexual risk behaviours among adolescents in South Africa

Abstract Increasing rates of mobile phone access present potential new opportunities and risks for adolescents’ sexual and reproductive health in resource-poor settings. We investigated associations between mobile phone access/use and sexual risks in a cohort of 10–24-year-olds in South Africa. 1563 adolescents (69% living with HIV) were interviewed in three waves between 2014 and 2018. We assessed mobile phone access and use to search for health content and social media. Self-reported sexual risks included: sex after substance use, unprotected sex, multiple sexual partnerships and inequitable sexual partnerships in the past 12 months. We examined associations between mobile phone access/use and sexual risks using covariate-adjusted mixed-effects logistic regression models. Mobile phone access alone was not associated with any sexual risks. Social media use alone (vs. no mobile phone access) was associated with a significantly increased probability of unprotected sex (adjusted average marginal effects [AMEs] + 4.7 percentage points [ppts], 95% CI 1.6–7.8). However, health content use (vs. no mobile phone access) was associated with significantly decreased probabilities of sex after substance use (AMEs –5.3 ppts, 95% CI –7.4 to –3.2) and unprotected sex (AMEs –7.5 ppts, 95% CI –10.6 to –4.4). Moreover, mobile phone access and health content use were associated with increased risks of multiple sexual partnerships in boys. Health content use was associated with increased risks of inequitable sexual partnerships in adolescents not living with HIV. Results suggest an urgent need for strategies to harness mobile phone use for protection from growing risks due to social media exposure.


Introduction
Sub-Saharan Africa has the largest and fastest-growing youth population, with multiple overlapping sexual and reproductive health (SRH) needs. 1 In South Africa, the youth population is projected to reach more than 11 million by 2030. 2 South African youth experience heightened sexual health risk: most new HIV (human immunodeficiency virus) infections occur during adolescence, 3,4 and one in three girls become pregnant before age 20. 5 Adolescents and young people have been early and enthusiastic adopters of digital technologies.In 2019, 71% of South African households had a mobile phone user, and 64% had access to the internet, 6 with youth aged 15-24 years comprising 71% of internet users. 7][10][11] Mobile health (m-Health) initiatives may be a key pathway to improve SRH knowledge and HIV prevention among young people in resource-limited settings. 9,12Previous studies have shown that m-Health might help prevent adolescents from engaging in risky sexual behaviours as a result of improved SRH knowledge. 9,13Findings from a cluster-randomised control trial among 756 females aged 14-24 years in Accra, Ghana, indicated that text messaging improved their SRH knowledge, which in turn led to decreased risks of pregnancy. 14Most of m-Health interventions were implemented, taking into account the intersections with adolescents' sexual and reproductive health and rights in both policy and practice.
However, m-Health interventions remain limited in scope and coverage, without proper evidence of large scaling-up process. 15A study on how 4500 young people in Ghana, Malawi and South Africa used mobile phones revealed that the majority of them had never heard of m-Health interventions, let alone participated in them. 16Searches via social media and websites have proven to be an innovative way to engage adolescents. 11,17Qualitative evidence supports user-driven health content use (rather than campaign-driven health content use) as an effective means of improving health behaviours, [16][17][18] but quantitative research, especially related to sexual risk behaviours, is lacking.
This study attempted to close this gap by focusing on the "informal" uses of m-Healthnamely guided by creative and strategic use of mobile phonesfor safer sexual risk behaviours. 16The paper did not overlook the fact that adolescents are also susceptible to exposure to risks in the online space.A recent meta-analysis showed that frequent use of social media among adolescents is associated with increased risks of drug use, risky sexual practices and violent behaviours. 19e aimed to examine the associations of sexual risk behaviours with access to and use of mobile phones among a cohort of adolescents in South Africa, and to assess these associations by sex and HIV status.

Study design
We conducted a prospective cohort study "Mzantsi Wakho" amongst 1563 adolescents living with and without HIV in South Africa.This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement for cohort studies (Appendix 1). 20

Study setting and participants
Research was conducted in 180 communities in a health sub-district in the Eastern Cape Province, South Africa.The Eastern Cape has high levels of poverty and HIV. 21The study recruited adolescents living with HIV from 52 primary healthcare clinics in 2014-2015 and 15 additional clinics in 2016-2018, and then recruited neighbouring adolescents not living with HIV in local communities.In the first wave of Mzantsi Wakho, which commenced in 2014-2015, 1519 adolescents were successfully interviewed, including 1046 living with HIV and 473 not living with HIV (Figure 1).Wave 2 (1454 interviewed, 1030 of whom were living with HIV; 1410 complete cases) was conducted in 2015-2017 and wave 3 (1429 interviewed, 1010 of whom were living with HIV; 1353 complete cases) in 2017-2018.In waves 2 and 3, participants resided in Eastern Cape, Free State, Gauteng, KwaZulu-Natal, North-West and Western Cape, due to high levels of migration.Ethical protocols were approved by the University of Cape Town (Cape Town, South Africa; CSSR 2013/ 4, approved 14 April 2013), Oxford University (Oxford, UK; CUREC2/12-21, approved 20 December 2012), Provincial Departments of Health and Education, and all participating healthcare facilities.All adolescents and their primary caregivers provided written informed consent at all survey rounds in Xhosa or English, and POPIA-compliant data management instruments were used across the data lifecycle to protect the personal information of participants. 224][25][26] All were selfreported for the past 12 months: (1) sex after substance use defined as sexual intercourse when the participant was drunk or used drugs (available in waves 2 and 3); (2) unprotected sex defined as no or infrequent condom use with partners (available in all waves); (3) multiple sexual partnerships defined as two or more sexual partners (available in all waves) and ( 4) inequitable sexual partnerships defined as sex in exchange for material support of any kind or sexual partner at least 5 years older than participant (available in all waves).

Mobile phone access and use
Self-reported data on current access to and use of mobile phone were available in waves 2 and 3. Adolescents were asked if they had a mobile phone (including a smartphone, Apple iPhone, Blackberry, basic phone and SIM card) and if it was their own or shared with someone.In this study, mobile phone access refers to self-reported Figure 1.Mzantsi Wakho cohort study flow chart.Abbreviations: T1, wave 1; T2, wave 2; T3, wave 3 ownership of a mobile phone with a functional SIM card.Adolescents who owned a mobile phone reported what they used it for and the frequency of use.Mobile phone use options included SMS, WhatsApp, Facebook, Mixit, health information, information about sexual health and HIV-related information.Frequency of use was reported in the questionnaire as "1.Never", "2.once a month", "3.once a week", "4.once a day" and "5.two or more times a day".Hereafter, we refer to daily or multiple times daily use as (frequent) use of mobile phone (recoded as binary).Mobile phone use measure was then classified as "0.no access", "1.social media alone" (referring to frequent mobile phone use for SMS, Facebook, Mixit or WhatsApp only) and "2.health content" (referring to frequent mobile phone use for both social media and sexual health or HIV-related information).Access to and use of mobile phone were measured in waves 2 and 3.

Covariates
We included eight covariates in our models based on associations identified in existing literature 27 : rural residence; informal housing; household poverty measured as access to the eight highest socially perceived necessities in the nationally representative South African Social Attitudes Survey (enough food, money for school fees, to see a doctor when needed, school uniform, basic clothing, soap, school books and shoes) 28 ; being married or in a consensual relationship; age; sex; living with HIV; school enrolment and survey wave.All covariates were categorical variables, except for the participant's age, which was continuous, and mean-centred before analysis to facilitate interpretations.Covariates changed for very few individuals in the data.Therefore, all covariates, except for the survey wave, were measured at baseline (between March 2014 and September 2015).

Data and bias
0][31] Little's χ 2 test, with the presence of covariates and unequal variances between different missing-value patterns, indicated that data were not missing completely at random (Appendix 3). 32Frequencies showed that more missing data occurred in wave 2 than in wave 3. The proportion of missing data on sex after substance use was 13.5% in wave 2 and 6.5% in wave 3; the proportion of missing data on inequitable sexual partnerships ranged from 3.8% in wave 2 to 6.5% in wave 3.There was no missing data on mobile phone access and use (neither in wave 2 nor in wave 3).The amount of missing data for covariates at baseline was 0.1% for household poverty and informal housing.There was no missing data on sex, age, rural residence, relationship status, school enrollment and HIV status.4][35] Imputation models included all aforementioned covariates and auxiliary variables that were predictive of sexual risk behaviours.

Statistical methods
We first described sociodemographic characteristics, mobile phone access and use, individual sexual risk behaviours in the full sample, then by adolescent sex and HIV status.The correlations between individual outcomes were weak or moderate.We reported Spearman's bivariate correlation coefficients between outcome variables in Appendix 4. We reported the descriptive statistics for the original sample (i.e.data with missing values).Second, we used mixed effects logistic regression models on imputed data to account for repeated measures. 36A summary of the models is reported in Appendix 5. We estimated the overall association of individual sexual risk behaviours with mobile phone access (Model 1) and mobile phone use for social media alone and health content (Model 2), adjusting for baseline covariates.We reported average marginal effects from the models to estimate associations of individual sexual risk behaviours with mobile access and use.We also estimated the separate associations of individual sexual risk behaviours with mobile phone access (Appendix 5, Model 3) and mobile phone use (Appendix 5, Model 4) for two sub-groups: boys and girls, and for adolescents living with and without HIV.To prevent underpowered (stratified) analyses, we estimated average marginal effects (AMEs) in all four combinations (of boys or girls, living with HIV or not) using three-way interaction terms (between participant's sex, HIV status and mobile phone access/use).Third, we conducted sensitivity analyses, comparing the results in an analysis that included only complete cases (from data with no imputations) to results from imputed data.All analyses were done using Stata 17 37 .Statistical significance was defined a priori as p < 0.05.

Results
Table 1 shows baseline socio-demographic characteristics of participants (n = 1410), access to and use of mobile phones (in waves 2 and 3), and sexual risk behaviours (in waves 2 and 3).69% of participants recruited at baseline were living with HIV.The baseline mean age of respondents was 13.8 years.57% were female, 27% resided in a rural area, 18% were living in informal houses and 67% were in poor households.Nearly onethird of the sample was in a relationship and 94% were enrolled in school.In wave 2, about half of participants (50%) had access to a mobile phone (including 43% for boys, 56% for girls, 57% for adolescents not living with HIV and 48% for adolescents living with HIV), 44% used mobile phones for social media alone (including 24% for boys, 41% for girls, 49% for adolescents not living with HIV and 36% for adolescents living with HIV) and only 17% for health content (including 20% for boys, 15% for girls, 8% for adolescents not living with HIV and 21% for adolescents living with HIV).In wave 3, there was a slight increase in the proportion of mobile phone access (53%) including 47% for boys, 65% for girls, 63% for adolescents not living with HIV and 55% for adolescents living with HIV; 34% used mobile phone for social media alone (including 25% for boys, 43% for girls, 55% for adolescents not living with HIV and 45% for adolescents living with HIV) and 23% used mobile phone for health content (including 22% for boys, 24% for girls, 7% for adolescents not living with HIV and 30% for adolescents living with HIV).
Sexual risk behaviours were similar across waves (Figure 2), except for unprotected sex which increased from 15% in wave 2 to 22% in wave 3 (Figure 2B).For boys, there was no change in the prevalence of any of the sexual risk behaviours between waves 2 and 3.For girls, the prevalence of unprotected sex increased from 19% in wave 2 to 29% in wave 3 (Figure 2B).The proportion of self-reported unprotected sex and sex after substance use was significantly higher among adolescents not living with HIV than among adolescents living with HIV in waves 2 and 3 (Figure 3A and Figure 3B).There was a significant increase in the proportion of selfreported unprotected sex between waves 2 and 3 among adolescents not living with HIV (from 21% in wave 2 to 32% in wave 3) and adolescents living with HIV (from 12% in wave 2 to 17% in wave 3) (Figure 3B).In wave 3, the proportion of selfreported multiple sexual partnerships was higher among adolescents not living with HIV (20%) than among adolescents living HIV (14%) (Figure 3C).
We fit two multivariable models to determine the association of individual sexual risk behaviours with mobile phone access (Table 2, Model 1) and mobile phone use (Table 2, Model 2).We found no evidence of an association between mobile phone access and sexual risk behaviours after adjusting for covariates.
Social media use alone (as compared to no mobile phone access) was associated with a significantly increased probability of unprotected sex (AMEs +4.7 percentage points [ppts], 95% CI 1.6 to 7.8, p = 0.003).Health content use (as compared to no mobile phone access) was associated with a significantly decreased probability of sex after substance use (AMEs -5.3 ppts, 95% CI -7.4 to -3.2, p < 0.001) and decreased probability of unprotected sex (AMEs -7.5 ppts, 95% CI -10.6 to -4.4, p < 0.001).
Then we estimated separate effects of mobile phone access (Table 3, Model 3) and mobile phone use (Table 3, Model 4) for boys, girls and adolescents living with and without HIV.Mobile phone access was associated with a significantly increased probability of self-reported multiple sexual partnerships among boys only (AMEs 5.5 ppts, 95% CI 1.1-9.6,p = 0.013).In all four categories, adolescents who used mobile phones for health content had significantly lower probabilities of reporting unprotected sex (for boys: AMEs -6.8 ppts, 95% CI -11.4 to -2.2, p = 0.004; for girls: AMEs -7.9 ppts, 95% CI -12.5 to -3.2, p = 0.001; for adolescents not living with HIV: AMEs -10.0 ppts, 95% CI -17.0 to -3.1, p = 0.005; for adolescents living with HIV: AMEs -6.0 ppts, 95% CI -9.4 to -2.6, p = 0.001) than adolescents with no access to a mobile phone.In all four categories except for adolescents not living with HIV, health content use (vs.no mobile phone access) was associated with significantly decreased probabilities of reporting sex after substance use (for boys: AMEs -6.8 ppts, 95% CI -11.5 to -2.2, p = 0.004; for girls: AMEs -3.4 ppts, 95% CI -6.2 to -0.6, p = 0.018; for adolescents living with HIV: We conducted a sensitivity analysis with complete cases and compared to the imputed dataset.The results from the sensitivity analysis and main analysis were similar, confirming overall results (Appendix 6a and 6b).

Discussion
This study provides valuable insight into the rates of mobile phone use alongside sexual risk behaviours of young people who are taking the initiative to "use m-Health" informally, in a high-poverty context in South Africa.We found that 57% of adolescents in our study had access to a mobile phone in 2018; this is consistent with the national average of 55% of mobile phone access in 2019. 7Social media use alone was almost ubiquitous, but less than a quarter of adolescent mobile phone owners had accessed health content in the past 12 months.
Our findings showed no association between mobile phone access and self-reported sexual These findings support and advance the existing literature.Two recent reviews found no associations between access to a mobile phone and adolescent sexual risk, suggesting that access alone to mobile phones is insufficient to improve adolescents' health outcomes. 11,38Our study supports these conclusions, and additionally examines the relationship amongst a large group of adolescents living with HIV, finding that mobile phone access in itself neither reduces nor increases sexual risks for this group.
Our study found that exclusive use of social media (without concurrent use of health content) is associated with increased unprotected sex.0][41][42] In this South African sample, the increase in unprotected sex amongst adolescents living with HIV and using mobile phones for social media alone suggests that this group may be particularly vulnerable to damaging effects of social media (alone and without access to any health-related information).
Two meta-analyses and a recent systematic review of mobile health interventions report overall benefits for adolescent health behaviours, Figure 3. Sexual risk behaviours by participant's HIV status across waves.Abbreviations: Cl, confidence interval.HIV, human immunodeficiency virus .HIV-, adolescents not living with HIV.HIV+, adolescents living with HIV including sexual health. 13,43,44This study adds to the evidence base by demonstrating beneficial associations of real-world use of health content, in almost all cases alongside social media, amongst adolescents living with and without HIV in South Africa.However, across both waves and in all subgroups, less than a quarter of adolescents had accessed any health content at all on their mobile phones in the past year.This suggests that while health content may be valuable, its uptake amongst this very high-risk group remains low, with consequent need for large-scale interventions that increase access and use of ageappropriate up-to-date sexual and reproductive health intervention.This also includes access to health content via social media.Data are average marginal effects (and 95% CI) from models with interaction between participant's sex, HIV status and mobile phone access (Model 3) or mobile phone use (Model 4), adjusting for rural residence, informal housing, household poverty and marital status.Data were obtained from mixed-effects logistic regression models on ten imputed data (n = 1353 including 587 boys, 766 girls, 933 living with HIV and 420 not living with HIV).
We note several limitations.First, the use of retrospective self-report measures of sexual risk behaviours may increase bias related to social desirability and recall biases.However, self-report is currently the only feasible way to measure most adolescent sexual risk behaviours.To mitigate measurement errors the study used widely validated measures in previous adolescent sexual health research in South Africa.Second, despite the longitudinal design, causality between the access to and use of mobile phones and sexual risk behaviours cannot be confirmed.Third, although we found no systematic differences between participants who completed both survey rounds and those who dropped out, we cannot fully rule out possible biases from unmeasured sources of confounding and attrition.Other caveats include potential bias due to data not completely missing at random.However, we used multiple imputations to account for biases in missing data.We also repeated the analysis in a sensitivity analysis using complete cases analysis and again found similar results.Fifth, we fit multivariable mixed-effects logistic regression for each outcome separately, therefore, assuming that the covariances among random effects across all sexual risk behaviours outcomes and the covariances among the residuals equal zero.Nevertheless, our findings from sensitivity analyses do not deviate from estimates from multivariable models.
The study also has a number of strengths.It adds to evidence from formal m-Health interventions, by examining a real-life sample of adolescents who use (and do not use) mobile phones informally to improve their SRHR knowledge and how they engage in safer sexual practices.In the absence of large-scale formal m-Health programmes in resource-limited settings, data from informal m-Health can inform development of adolescents' SRHR and HIV prevention programmes.However, the measure of informal m-Health used in this study is insufficiently detailed to assess the complexity of its associations with adolescents' sexual risk behaviours.Future surveys should attempt to collect data on mobile phone use in relation to sexual risk behaviours.We collected longitudinal data from adolescents living with and without HIV and analysed a sample of adolescent girls and boys, living with HIV or not, adding to our understanding of informal use of social media and health content amongst and comparing between these important groups.
This study suggests that mobile phones can be a medium of both risk and resilience for adolescents in Southern Africa.Social media use onlywithout concurrent health content usewas associated with increased sexual risk.Health content use was protective, even in the context of concurrent social media use, but underutilised.This highlights important next steps for programming: to identify approaches that increase informal m-Health use amongst adolescents and to deliver and assess these in realworld settings.For example, UNICEF and partners have recently proposed a set of toolkits to address sexual and reproductive health and HIV prevention needs. 45As mobile and internet access increases exponentially in Africa over the next decade, it is essential that we minimise associated risks, and capitalise on the potential of mobile phones to improve adolescent sexual and reproductive health.

Data sharing and data availability statement
Prospective users, policymakers/government agencies/researchers (internal/external) will be required to contact the study team to discuss and plan the use of data.Research data will be available on request subject to participant consent and having completed all necessary documentation.All data requests should be sent to the Elona Toska (elona.toska@uct.ac.za) or William Rudgard (william.rudgard@spi.ox.uk).).De plus, l'accès à un téléphone portable et l'utilisation de contenus liés à la santé étaient associés à des risques accrus de partenariats sexuels multiples chez les garçons.L'utilisation de contenus liés à la santé était associée à des risques accrus de partenariats sexuels inéquitables chez les adolescents non séropositifs.Les résultats suggèrent un besoin urgent de stratégies visant à exploiter l'utilisation du téléphone portable pour se protéger contre les risques croissants dus à l'exposition aux médias sociaux.

Figure 2 .
Figure 2. Sexual risk behaviours by participant's sex across waves.Abbreviations: Cl, confidence interval

Table 2 .
Multivariable association between mobile phone access and use and selfreported sexual risk behaviours.Results from mixed-effects logistic regression models on ten imputed datasets (n = 1353)