Relationship Between Sexual Behaviors with Non-committed Relationship Partners and COVID-19 Restrictions and Notification Rates: Results from a Longitudinal Study of Gay and Bisexual Men in Australia

Launched in 2014, the Flux Study is a national online, prospective observational study on GBM’s health. A detailed description of the study and its protocols has been published elsewhere (Hammoud et al., 2017). The study initially focused on use of illicit drugs, moving to attitudes and use of HIV pre-exposure prophylaxis (PrEP), and in 2020 it progressed to capture the experiences of the cohort in response to the COVID-19 pandemic. Part of this progression involved men responding to weekly surveys instead of surveys every 6 months. These surveys were sent every Sunday to capture their experience of COVID-19 from the preceding week in an ongoing basis in what were called weekly diaries beginning in May 2020. Information collected through a brief 5–10-min questionnaire captured essential information about sexual behavior, social connectedness, and access to healthcare. Every fourth week, participants responded to an extended 10–15-min survey to capture more detailed and contextual information including less common sex practices, such as group sex and sexualized drug use, sexual health checks, and practices employed to reduce transmission of COVID-19. Changes in behaviour from participants who responded to surveys prior to onset of the pandemic and the baseline of this stage of the study progression have been reported elsewhere (Hammoud et al., 2020).

Men were eligible if they were at least 16 years old, identified as gay or bisexual, or had sex with a man in the previous 12 months, and lived in Australia. Study promotion occurred online by advertising on a popular social media platform, Facebook, and LGBT and HIV community organizations across Australia, including recruitment of new participants in study weeks 8 (22–28 June 2020) and 37 (11–17 January 2021). Each week, participants who completed their weekly survey were entered into a raffle with three prizes: one $100 gift card and two $50 gift cards. Participants provided online informed consent, and enrolment was verified once participants activated a link via email. Ethical approval was granted from the Human Research Ethics Committee of UNSW Sydney.

All weekly surveys included questions about sexual behavior. This included the number of male partners in the previous 7 days, whether they lived in the same household with those partners, whether those partners included a main regular male partner (boyfriend or husband), or any regular fuckbuddies or “friends with benefits.ˮ All other partners were considered casual partners for this analysis. We defined “non-committed relationship partnersˮ as fuckbuddies/ “friends with benefitsˮ and casual partners collectively, and “relationship partnersˮ as partners with whom participants had an ongoing committed relationship (boyfriend or husband) (Bavinton et al., 2016). Participants were asked to recall their sexual partners/behaviors aggregated over the previous week.

Joinpoint regression using the grid search method and with p values computed using the permutation procedure based on Monte Carlo methods for model selection was used to identify changes in trends of mean numbers of all non-relationship partners (Kim et al., 2000; Lerman, 1980). Software used was Joinpoint Regression Program Version 4.8.0.1. (Statistical Methodology and Applications Branch, 2021).

We reported periods of lockdown for Australian jurisdictions and jurisdictional COVID-19 notification rates and described how these compare with changes in trends identified in the joinpoint model. Periods of lockdown were defined by the implementation of “stay-at-homeˮ orders, where leaving the home was allowed for only four reasons: to seek or provide medical or compassionate care, for essential shopping, for work or study where this is not possible from home, and for exercise (Prime Minister of Australia, 2020).

We used relatively narrow time intervals (weekly) for joinpoint analyses in this study given the relatively high response rate per unit time (week), where even smaller jurisdiction categories used in the analysis averaged over 50 responses per week. This potentially increases the capacity of joinpoint selection to reflect multiple changes in trend by jurisdiction, especially for Victoria following changes in lockdown status of that jurisdiction over the analysis duration. However, this also has the potential to affect model selection and fit, as well as to lead to spuriously correlated trends if models are overspecified. Methodological steps taken to limit potential overspecification of models included restricted maximum number of joinpoints per model (4), as well as the minimum number of observations between consecutive joinpoints (3) or between joinpoints and study start and finish dates (3). Generally, broad (zero joinpoint) trends were observed in jurisdiction categories for fitted models suggesting limited potential spurious correlation. For comparison, we also used broader time intervals (monthly) for joinpoint analysis. We have chosen to present data on a weekly basis because of increased fidelity to observed changes across all jurisdictions over the study period.

We also conducted a random effects Poisson regression, univariate time predictor variable (weekly) fit model for each jurisdiction for interpretability and comparability with the joinpoint output. The model was guided by the joinpoint program to indicate inflexion point locations. The model for Victoria was fit piecewise, where a separate model was used for each segment between joinpoints. The rate ratios were annualized and presented as percentage changes to ensure format comparability with the joinpoint output.

A sensitivity analysis was conducted for the joinpoint regression model to control for age and level of retention in the cohort (completed at least 50% of weekly surveys). Because differences in study follow-up associated with age might potentially correlate with the primary endpoint, we conducted sensitivity analyses to look at outcomes by jurisdiction by dichotomized age group (< 35/ >  = 35), as well as limited to those participants who had completed at least 50% of survey responses.

Sensitivity analysis using monthly time intervals were qualitatively similar to the weekly models. Random effects Poisson regressions outputs were also qualitatively similar to outputs presented in the primary analysis.

Results.

A total of 831 GBM completed at least one survey between weeks 1 and 52 (3 May 2020–2 May 2021). Completed responses to each weekly survey over the 52 weeks ranged from 326 to 556 (see Supplementary Material Table 1). On average, each participant completed the survey in 28.5 of 52 weeks (SD = 20.32). Participants had a mean age of 45.71 years (SD: 13.93), ranging from 19 to 82 years (Table 1). Most men (89.0%) identified as gay with small proportions identifying as bisexual (5.9%) or another sexuality (2.4%) or gave no response (2.6%). About one in ten men reported being HIV-positive.

One in five men (n = 168, 20.2%) reported no sex with male partners during the entire 52-week study period. About two-thirds (n = 559, 67.2%) reported sex with non-committed relationship partners, with over a quarter doing so in at least half of the surveys to which they responded. Among participants who reported sex with non-committed relationship partners, just over half (n = 470, 56.5%) of men reported no condomless anal intercourse with non-committed relationship partners (CLAI-NR). About two-fifths reported some CLAI-NR (n = 362, 43.5%), with over 10% doing so in at least half of the surveys to which they responded.

A mean of 49.0% men per week engaged in any sex with any type of male partner with minimal variation over the 52-week study period (Supplementary Material Table 1). The proportion of men having sex with committed relationship partners fell slightly over the study period from 29.5% (week 1) to 22.0% (week 52; p =  < 0.001). Although there was an overall increase in sex with non-committed relationship partners, this was not a consistent increase, differing from week to week and ranged from 10.1 (week 4) to 34.3% (week 47) (Fig. 2).

Sex with casual partners in the previous 7 days changed little over time from 19.9% in week 1 (3–10 May) to 19.4% in week 52 (26 April–2 May 2021). However, the lower proportions tended to be observed in the earlier weeks of the study, between weeks 2 (11.2%, 11–17 May 2020) and 16 (13.4%, 17–23 August). There was a gradual increase in sex with non-household partners (i.e., non-committed relationship partners that did not reside with participants) over the study period. Only about 3% of men each week reported living in the same household as non-committed relationship partners.

Prevalence of CLAI-NR was inconsistent in the first half of the study period, from its lowest at 7.1% in week 4 (25–31 May 2020) to a peak of 19.4% in week 21 (21–27 September 2020) but remained steady at about one in five in the second half of the study period.

The mean number of partners overall varied across the study period, ranging from its lowest at 0.82 in week 1 (3–10 May 2020) to a peak of 1.57 in week 39 (25–31 January 2021), with a more consistent increase occurring in the latter half of the study period. The mean number of non-committed relationship partners gradually increased over the study period from 0.53 (week 1) to 0.92 (week 52).

Group sex behavior was collected every 4 weeks and reported for the previous four weeks. The proportion of men reporting group sex was also inconsistent over the study period, peaking early in the study at 15.2% in week 4 (25–31 May 2020).

Among the 413 men who responded in week 52, 9.9% (n = 41) indicated they were avoiding all sexual contact to reduce their risk of COVID-19. These 41 men reported a mean of 0.02 non-committed relationship sex partners in that week, compared with 1.03 partners among the remaining respondents in week 52 (p = 0.002). Similar trends were found in earlier weeks.

Between 3 May 2020 and 2 May 2021, most Australian jurisdictions recorded no COVID-19 notifications for extended periods. When COVID-19 outbreaks did occur, they typically involved one, not all jurisdictions, and peak notifications in each jurisdiction varied, mostly for periods lasting no more than a few days. Joinpoint regression using permutation tests for the models found an overall continuous increase in the mean number of non-committed relationship partners over time (0.53–0.90, p < 0.001). Overall, the trends in the mean number of non-committed relationship partners varied across Australian jurisdictions (see Supplementary Material Table 2).

Over the study period, COVID-19 notifications in Victoria began to rise significantly from study week 8 (229 notifications, 22–28 June 2020) and peaked at 3125 during study week 13 (27 July–2 August 2020) (Fig. 1) which prompted a lockdown of metropolitan Melbourne and a neighboring local government area from study week 10 (6–12 July 2020) (Department of Health and Human Services, 2021) affecting approximately 83% of the state’s population (Australian Bureau of Statistics, 2021a, b; Mitchell Shire Council, 2021). This was followed by a state-wide lockdown, encompassing the remaining regional areas of the state, from study week 14 (3094 notifications, 3–9 August 2020). In week 19 (311 notifications, 7–13 September 2020), regional Victoria came out of lockdown, followed by metropolitan Melbourne in study week 22 (65 notifications, 28 September–4 October 2020) (Department of Health and Human Services, 2021). Although there were a number of weeks where metropolitan Melbourne was in lockdown and the rest of the state was not, the number of men outside of the lockdown during this time was too small to provide a reliable comparison of changes in behaviour. We observed three changes in the trend for the mean number of reported non-committed relationship partners from the joinpoint regression model. There was a trend increase from week 1 (85 notifications, 3–10 May 2020) to week 5 (15 notifications, 1–7 June 2020) (0.22–0.71, p 0.04), a trend decrease to week 16 (1,311 notifications, 17–23 August 2020) (0.71–0.18, p < 0.001), a trend increase to week 31 (0 notifications, 30 November–6 December 2020) (0.18–0.85, p 0.06), followed by a stable, slightly increasing, trend to week 52 (0 notifications, 26 April–2 May 2021) (0.85–0.82, p 0.61) (Fig. 3) (Department of Health and Human Services, 2021). A snap 5-day lockdown was also implemented for the entire state of Victoria across study weeks 41 (0.42, 13 notifications, 8–14 February 2021) and 42 (0.87, 5 notifications, 15–21 February 2021) (Department of Health and Human Services, 2021). The mean number of non-committed relationship partners remained much lower among Victorian participants compared to all other jurisdictions for the first half of the study period. However, mean partner number increased thereafter and peaked at 1.34 in week 48 (0 notifications, 18–24 January 2021) (see Supplementary Material Table 2) (Department of Health and Human Services, 2021) Fig. 4.

Over the study period, COVID-19 notifications in NSW varied but at no time reached similar levels to those observed in Victoria. Between study weeks 10 (16 notifications, 6–12 July 2020) to 21 (11 notifications, 14–20 September 2020), the state recorded a spike in notifications peaking at 92 notifications in week 13 (27 July–2 August 2020) (Fig. 1) (Data.NSW, 2021). No period of lockdown was introduced over this period. The mean number of non-committed relationship partners was at its lowest in week 13 (0.82, 92 notifications, 27 July–2 August) and peaked at 1.25 in week 18 (52 notifications, 31 August-6 September 2020) (see Supplementary Material Table 2) (Data.NSW, 2021). A localized lockdown of a specific local government area occurred from study weeks 33 (0.77, 88 notifications, 14–20 December 2020) to 36 (1.30, 19 notifications, 4–10 January 2021) (Data.NSW, 2021). A very small number of participants were living within the localized lockdown area, and, therefore, comparisons of changes in sexual behavior between those within the lockdown and outside were not reliable. We observed a continuously increasing trend over the study period in the joinpoint regression model for men in NSW (0.80–0.98, p < 0.001) (Fig. 5).

A snap 3-day lockdown occurred across study weeks 36 (0.77, 1 notification, 4–10 January 2021) and 37 (0.56, 1 notification, 11–17 January 2021) for the state’s capital, Brisbane, and surrounding local government areas due to cases in the community with the Alpha variant of the virus (Fig. 1) (Queensland Health, 2021a, b). The joinpoint regression model found little change in the mean number of reported non-committed relationship partners over the study period (0.45–1.12, p 0.50) (Fig. 5).

Two lockdowns occurred in two of the remaining jurisdictions grouped for this analysis. The mean number of non-committed relationship partners represented here are across all five jurisdictions (see Supplementary Material Table 2). The state of South Australia had a snap lockdown in study week 29 (0.43, 24 notifications, 16–22 November 2020) that lasted for 4 days (SA Health, 2020a, b, c, d, e). Western Australia’s capital, Perth, and the neighboring Peel region entered a 5-day lockdown across study weeks 39 (0.98, 1 notification, 25–31 January 2021) and 40 (0.66, 0 notifications, 1–7 February 2021) due to a case in the community with the Alpha variant (Department of Health, 2021). Mean number of non-committed relationship partners for these remaining jurisdictions remained fairly stable over the study period, although there were large fluctuations week on week. For these jurisdictions, we observed a gradual increase over the study period in the joinpoint regression model (0.25–0.60, p 0.00) (Fig. 5).

In analyses restricted to younger men (those aged under 30 years) and analyses restricted to men who completed at least 50% of weekly surveys, these jurisdictional trends remained unchanged.

Data on biomedical HIV prevention and treatment were captured as part of weekly surveys among this cohort of men; however, these data go beyond the capacity of this analysis focused on sexual behavior over biomedical HIV prevention. The impacts of COVID-19 outbreaks and resulting restrictions to curb increases and the impacts on biomedical HIV prevention and treatment should be explored in future research. Although it could not have been anticipated, comparable weekly data on the sexual behavior of Australian GBM did not exist before the COVID-19 pandemic, including in previous iterations of the Flux Study. Also, weekly data collection commenced after the initial restrictions were implemented and not captured before May 2020. This study utilized an online convenience sample and findings may not be representative of all Australian GBM. The calculations in this analysis relied on recall of sexual behavior with partners in the previous 7 days, and we believe that this short timeframe reduced the impacts of recall bias. These data may also be subject to social desirability bias. Given restrictions and periods of lockdown discouraged people from meeting others outside of their household, particularly for non-committed relationship partners, it is plausible that men underreported sexual encounters, particularly outside the household and during times of lockdown. Participants were not asked each week about whether restrictions that were in place or number of local infections directly impacted their sexual behavior or choice of sexual partner. Therefore, direct relationships between restrictions or local notifications on sexual behavior or sexual partner choice are not possible. As mentioned earlier, this is a relatively older cohort compared to other Australian community-based samples of GBM and a larger sample of younger men may not have exhibited the same trends in sex with non-committed relationship partners (Bavinton et al., 2020; Holt et al., 2019).