We pre-selected a high risk group for this cohort study in which HIV incidence and other risky sexual behaviour were shown to be high [11]. Contrary to the perception that fisher-folk may be difficult to retain in longitudinal studies, we have demonstrated a modest study dropout in fishing communities comparable to that reported in neighbouring inland non-fishing communities [9, 10]. In addition high risk individuals have been associated with similar loss to follow-up in the neighbouring fishing 23.1 % [8] and non-fishing community 16.3 % [14].

The results suggest a number of factors that are associated with dropout from the fisher-folk cohort including young age, male gender, ethnicity, having a new sexual partner, time spent in the fishing communities, being away from home for ≥2 nights and marijuana use. Consistent with other studies in the neighbouring fishing [8] and non-fishing communities [9, 10], USA [15] and Brazil [16], study dropout was associated with age, with those in the youngest age group (13–24) being most likely to dropout. Since there were nearly equal proportions of men (49 %) and women (51 %) in this age group, this could be due to both looking for work and new relationships.

Dropping out was associated with the ethnicity of the volunteers, with non-Baganda being more likely to dropout. This has been indicated in the previous study further North of Lake Victoria [8]. The majority of non-Baganda volunteers come from further afield, including neighbouring countries (Kenya, Tanzania and Rwanda) and some are traders and transporters who might spend relatively long periods away from the fishing communities. These were more likely to be involved in fishing or related activities compared to the Baganda who were likely to be engaged in small scale businesses. Furthermore, number of years spent in the fishing site was independently associated with study dropout, with those who have spent more than 1 year in the fishing communities being less likely to dropout. This further demonstrates that individuals who have lived in a community for a long time are more likely to be available for recruitment in prevention studies. Additionally these could easily be retained in study follow up. The finding that dropout was associated with use of marijuana is consistent with findings from cohorts in the USA [17] and Japan [18], where similar dropout rates were observed among illicit drug users.

In this cohort we ascertained the frequency of reporting a new sexual partner for both men and women and this was independently associated with cohort dropout. This is consistent with the results from the inland non-fishing community in which the reported number of lifetime partners was higher in those who changed residence to other neighboring villages and there was also more risky sexual behavior reported among those who changed residence [14]. The fact that higher risk volunteers were more likely to drop out could lead to underestimation of HIV incidence in a prevention trial.

A major strength in terms of HIV prevention studies for these fishing communities compared to the general inland population is the high HIV incidence [11] and comparable dropout rates [9, 10]. Depending on the interventions being evaluated and other proven prevention strategies available, future efficacy trials in this population could require relatively shorter follow up period and a smaller sample size. A sensitivity analysis largely showed similar factors associated with dropout using a more liberal definition of dropout. It is encouraging to note that individuals from fishing communities can travel away from their work to attend their visits at clinic located further afield. In fact, our study showed minimal difference in dropout between clinic attendees at the fishing site and off site. This allays fears that their fairly good retention was only attained at study clinics located within the fishing communities with poor or non-existent infrastructure for conducting efficacy trials.

The limitations of this study include having obtained sexual risk behaviour data by self-report in a face to face interview which could have led to underreporting by women or over reporting by men depending on who is interviewing them [19]. The use of mobile phone short messaging service (SMS) has been shown to be more reliable in collecting self-reported sexual activity [20]. However, the mobile phone services are not well developed in some fishing communities with intermittent mobile network connections likely to disrupt the messaging service.

The most common reason for missing study visits was migration (Fig. 1). This has been associated with an increased risk of HIV infection probably due to higher risk sexual behaviour among those who move [14]. Therefore using recruitment strategies that avoids known predictors of volunteer missing visits would facilitate high retention. Alternatively strategies that aim at retaining volunteers who are likely to miss visits are needed to maximise retention.