Study participants
We recruited 2 814 individuals, of whom 1 436 (51%) were men [24]. The age of the respondents was 20 to 40 years, with a mean age for the women 27.0 (± 4.3) years and a mean age for the men 27.8 (± 4.4) years. Two thirds of the men and women were single [24].
Latent Classes
Based on the nine selected manifest variables, we fitted models with two to five LCs for the men and up to three LCs for the women (because the models with more than three classes did not converge). The AIC and BIC indicated optimal models, with five classes for the men and three classes for the women (Figure 1, Online supplement Table 2S). Additionally, the highest entropy value (0.85) indicated that the optimal model for the men was a four-class model (Online supplement Table 2S).
We interpreted, labelled, and ordered the LCs based on the item-response probabilities (Online supplement Table 3S-4S), with class 1 being the least risky behaviour, and class 4 for the men and class 3 for the women, the riskiest behaviour. We present the frequency of the highest risk category of each variable in Figures 1 and 2. For the men, 8% (n= 110) fell into class 1, labelled “Mixed steady and non-steady partnerships, low substance use” (Table 3S). Thirty percent (n= 441) of the men fell into the class 2, labelled “Steady partnership with/without concurrent partners”, They were considered medium risk, with a low probability of not having a steady current relationship and a higher probability of many sexual partners with and without concurrent relationships. For the men, we could further separate LCs of high-risk behaviour: “Non-steady partnerships with many partners, condom users” (class 3, n = 601) and “Non-steady partnerships with many partners, condom non-users” (class 4, n= 284). These contained 42% and 20% of the men, respectively (Table 3S).
Amongst the women, 10% (n= 134) fell into class 1 of low-risk sexual behaviour, labelled “Mixed steady and non-steady partnerships, low substance use” (Online supplement Table 4S). Thirty-two percent (n= 441) of the women fell into the class 2, labelled “Steady partnership with/without concurrent partners” (medium risk). The largest class 3, containing 58% (n= 803) of the women, was labelled “Non-steady partnerships with many partners”, and was characterised by a high probability of having a non-steady current partner and a higher probability of having 6 or more sexual partners during the previous 12 months compared with the other female LCs. The probability of frequent alcohol use before sex was high amongst the women and men across all LCs, with the exception of class 1.
Demographic and sex-specific variables across latent classes
Class membership was similar amongst the men and women across the age groups and marital status (Online supplement Table 5S – 6S). Notably, the younger (20–29 years of age) men (76%) and women (77%), and single men (88%) and women (96%) were more likely to belong to high-risk classes (4 and 3, respectively). The men in class 4 were also more likely (40%) to impregnate women unintentionally than in other LCs. The absolute majority (80-87%) of the women used some type of contraception across LCs. However, the women in class 3 were more likely to use the barrier method (35%). There was no major difference in the use of emergency contraceptive pills or a history of induced abortion across LCs.
Distal outcomes
Short-term outcome: repeated testing during past 12 months and current chlamydia infection
For repeated testing for chlamydia during past 12 months, we found significantly higher odds of 1.72 (95%CI: 1.16 – 2.54) in high-risk behaviour class 3 compared with class 1 (Figure 3, Online supplement Table 7S) amongst the women. There was a borderline association with high-risk behaviours class 3 amongst the men, adjOR = 1.60 (95%CI: 0.97 – 2.65), Figure 3, Online supplement Table 8S.
Amongst the men, class 4 had a 3.03 (95%CI 1.32 – 6.93) times higher odds than class 1 of testing positive for the current chlamydia infection (Figure 4, Online supplement Table 9S). Class 3 in the men had borderline significant increased odds as well: adjOR = 2.16 (95%CI: 0.97 – 4.83). None of the associations were statistically significant for this outcome amongst the women (Figure 4, Online supplement Table 10S).
Long- term outcome: repeated lifetime testing and repeated lifetime chlamydia infection
Both high-risk male classes and the high-risk female class were all significantly associated with at least a two-fold increased odds of repeated lifetime testing (Figure 3, Online supplement Table 7S-8S). Amongst the men, class 3 had an adjOR = 2.26 (95%CI: 1.50 – 3.40), while class 4 had an even higher association: adjOR =3.03 (95%CI: 1.93 – 4.74), Figure 3 and Online supplement Table 8S. Amongst the women, we estimated a 1.85 (95%CI: 1.24 – 2.76) higher odds of repeated lifetime testing in high-risk class 3 than class 1. For this outcome, in both the men and women, we estimated a significant linear trend, which indicated a dose-response relationship: increasing levels of risk behaviour LCs were associated with an increased odds of repeated lifetime testing for chlamydia.
Furthermore, all LCs were associated with having had repeated lifetime chlamydia infection amongst the men, and the odds increased with each level of the outcome (never, once, twice or more). Class 3 had a 1.84 (95%CI: 1.03 – 3.26) higher odds than class 1 of having had chlamydia once during their lifetime compared with never. Class 4 had a 2.54 (95%CI: 1.39 – 4.64) higher odds than class 1 of one chlamydia infection during their lifetime compared with never. The odds of having chlamydia twice or more compared with having had it once varied across LCs amongst the men, though they were not statistically significant (Online supplement Table 9S). Amongst the women, none of the associations were statistically significant for lifetime repeated chlamydia infection (Figure 4, Online supplement Table 10S).