Fig 1.
Overview of clustering and prediction method.
nsCAI = condomless anal intercourse with non-steady partners.
Table 1.
Mandatory schooling in Switzerland runs from grade one through grade nine. An apprenticeship would typically follow after these nine years and include three or more years of on-the-job training in combination with part-time classroom schooling.
Fig 2.
Trends in nsCAI proportion in behavioural clusters.
The clusters contain the following proportions of the study population (from 0 through 4): 49%, 11%, 17%, 10% and 13%. Cluster 0 consists of participants who never reported nsCAI during the observation period (in this case mid-2001 until the cut-off date). The dashed vertical line on the 1st of May 2017 represents the cut-off date between the observation period (which corresponds to the period used to infer the clusters) and the outcome period. The shaded area shows the time in which nsCAI, nurse/physician-reported STIs and laboratory-confirmed syphilis were recorded as outcomes. nsCAI = condomless anal intercourse with non-steady partners.
Fig 3.
Bar plots for Bayesian information criterion values.
Regression models with different combinations of predictor variables to predict nurse/physician-reported STIs (left), laboratory-confirmed syphilis (center) and nsCAI (right) after cut-off. A smaller BIC represents a better prediction. Numbers to the right of the bars represent the p value of the likelihood ratio test between the two models in question. nsCAI = condomless anal intercourse with non-steady partners. BIC = Bayesian information criterion. pLRT = p value of likelihood ratio test comparing the respective models with and without clusters.
Table 2.
Summary of model comparison for predictions.
Columns from left to right: (1) Outcomes. nsCAI = condomless anal intercourse with non-steady partners, STI = sexually transmitted infection, nsP = sex with non-steady partners. (2) P value of likelihood ratio test (pLRT) of univariable model with behavioural clusters vs. null model. (3) pLRT of multivariable model including age, last nsCAI or nsP status, prior syphilis (or number of prior syphilis episodes for count outcomes) as predictor variables vs. the same model plus behavioural clusters. (4 through 9) Akaike information criteria (AIC), Bayesian information criteria (BIC), and area under the receiver operator characteristic curve (auROC) for multivariable model including age, last nsCAI or nsP status, prior syphilis (or number of prior syphilis episodes for count outcomes) as predictor variables and of the same model plus behavioural clusters. Clusters included in the models were inferred based on long-term nsCAI patterns for the top half of the table or long-term nsP patterns for the bottom half of the table. Green shading means a better performance for the model with clusters.
Table 3.
Ideal number of behaviour clusters to enhance predictions according to BIC.
A dash means adding behavioural clusters to prediction models did not improve BIC. BIC = Bayesian information criterion. nsCAI = condomless anal intercourse with non-steady partners. nsP = sex with non-steady partners.