We report a population-scale profile of sexual violence within the Colombian armed conflict using complete data from the statutory RUV and the NCHM database of armed conflict victimization. Victims of sexual violence were overwhelmingly female and disproportionally from minority ethnic groups. Sexual violence was more likely to co-occur with types of armed-conflict-related violence that more directly targeted individuals. Geospatial analysis revealed that sexual violence may be more widely present than is apparent from raw event counts and exhibited significant geographic clustering with distinct temporal patterns across conflict periods. For the entire conflict period, event density was highly clustered within relatively confined geographic ranges, with locations within approximately 2.4 km of one another sharing similar underlying probability of events. By contrast, during 2014–2024 there was reduced spatial intensity and broader geographic clustering. Analysis of temporal contagion indicated that apparent event dependence was largely attributable to variation in the background rate. After adjustment, residual event-to-event contagion was minimal, decaying in less than one day, suggesting that temporal clustering is driven by structural and contextual conditions rather than incident-specific characteristics. These findings suggest that sexual violence patterns in the Colombian armed conflict have become geographically more dispersed in recent years with reduced temporal contagion effects, indicating potential changes in conflict dynamics and perpetration over time.

Research on the perpetration of sexual violence by armed groups indicate that it is not an inevitability of armed conflict but subject to group norms, leadership, conflict dynamics and broader political conditions20. These may have changed significantly in recent years. Since the height of the Colombian armed conflict, the reduction in violence has been accompanied by a fragmentation of armed groups. There is now less concentrated areas of control with influence spread more thinly across wider territories21,22. This may explain the transition from higher-intensity, geographically focused patterns of sexual violence, potentially more associated with it occurring as part of combat operations, to broader clustering with less intense and shorter-lived temporal contagion, potentially reflecting its relationship with longer-term territorial control and community subjugation, as well as a reaction to public recognition and condemnation12,23.

Theoretically, accounts of conflict-related sexual violence differ in whether they emphasize event-level triggering or social norms as causal factors. Accounts in which sexual violence is deployed strategically as a response to specific events24, whether as part of coordinated operations or in retaliation against rival groups, would be more consistent with event-to-event contagion. By contrast, ‘practice’ account25 and socialization framework26, in which sexual violence is tolerated or condoned within armed groups for non-strategic reasons, would predict that rates are sustained by group-level conditions rather than primarily triggered by prior events. Our findings suggest that the latter account is more consistent with the data from the Colombian armed conflict. Across both periods, attenuation of temporal contagion after adjustment for background rate indicates that temporal clustering is largely explained by structural factors rather than prior events triggering subsequent ones. However, comparison between time periods suggests that event-level effects have become weaker still in the 2014–2024 period, consistent with violence persisting as a normalized group-level practice rather than as deployed in pursuit of specific objectives.

Geographic clusters of armed-conflict-related sexual violence indicate priority areas for provision of services for victims. Clinical response to sexual violence requires emergency and post-acute care27 as well as long-term support services28,29. Barriers relating to stigma, fear and health service preparedness may also delay service engagement30, and we note that initiatives to promote access to existing services are themselves likely to lead to increase in the number of successfully supported victims. The geographic clustering of sexual violence covers large cities, where health services are concentrated in Colombia, as well as rural areas and small towns, including those where control is disputed between armed actors and health services may be limited. This makes the equitable provision of healthcare a significant challenge due to problems with access, resources and active interference by armed groups that particularly affect services related to women’s health and sexual violence31,32. Consequently, while specialist sexual violence pathways may be feasible in areas of high concentration of health services, better integration of sexual violence treatment and support pathways in primary and community care may be an alternative option for rural areas33.

Colombia’s national PAPSIVI programme (an acronym from its Spanish language title Programa de Atención Psicosocial y Salud Integral a Víctimas, in English ‘Psychosocial Care and Comprehensive Health Program for Victims’) provides healthcare and psychosocial support to victims of the armed conflict nationally. Analysis of service data indicates it successfully provides priority access with appropriate support provision to victims of the conflict registered as experiencing sexual violence34,35, although clinical and social outcomes, and the extent to which victims of sexual violence who have not registered as such are being successfully supported, are currently unknown.

By comparing Hawkes process models with and without baseline adjustment, we identified that temporal contagion is largely driven by slower-moving trends in sexual violence probably driven by structural factors associated with broader conflict dynamics, rather than short-term event-to-event contagion more associated with incident-specific characteristics. Immediate intervention to prevent further incidents of sexual violence may be practical in some situations. However, these results suggest that depending solely on a policy of reactive intervention is unlikely to be the most successful strategy for prevention of armed-conflict-related sexual violence in the Colombian context because the temporal window for preventing local contagion effects may be very small—estimated at below 1 day in our study. To address these, social and structural interventions focused on prevention may be more appropriate. Indeed, interventions to reduce sexual violence by armed groups in conflict settings have focused on exactly these processes—including reducing potential victims’ exposure to situations that would allow opportunistic violence, increasing women’s autonomy and reducing economic dependence, reducing impunity and the perception of impunity through improved legal accountability, while deploying public campaigns aimed at increasing awareness and reducing acceptance of sexual violence36. More recent initiatives have developed and deployed specific behaviour change interventions for both state and non-state armed actors aimed at reducing the perpetration of armed conflict-related sexual violence37, including in Colombia38, although the effectiveness of these interventions has yet to be determined.

We note some limitations of this study. With regard to the demographic data, the profile of victims is relatively consistent across data sources, with some notable differences. Victims in the national RUV were older than victims in the NCHM database. This is probably due to the fact that, while the NCHM database records specific events, the national RUV is a record of victims’ registrations, who may register several years after experiencing sexual violence. There is a higher proportion of victims from minority ethnic groups in the RUV, potentially reflecting its sole reliance on self-registration rather than a third-party reports, the latter of which may be more limited in areas with large populations of primarily Indigenous and Afrocolombian populations.

We also note that the diversity of sexual violence is not well accounted for by the single coding of ‘sexual violence’ used in this study. Sexual violence may variously include rape, forced prostitution, sexual slavery, forced abortion, forced nudity or sexual assault, to name but a few relevant criminal acts39, and it is likely there may be different geographical spread and dynamics that depend on the specifics of these acts.

In addition, conflict-related sexual violence is probably underreported and underdetected, with good evidence that this is the case in Colombia40. Nevertheless, different national data sources, including crime statistics, armed conflict victim registrations and multisource identification, tend to show similar relative changes in incidence in sexual violence that track changes in the intensity of the conflict indicating they may be sampling a similar population19. Nonetheless, the extent to which there are specific groups who are more likely to have incidents of sexual violence remain unrecorded is not yet fully understood. Indeed, we note that individuals living in areas of active conflict41, displaced women10, sexual and gender minorities42, and people from minority ethnic groups43 have been identified as facing particular structural barriers in reporting sexual violence and accessing care.

However, there is not a simple relationship between underreporting and underrecognition in the geospatial models reported in this study. LGCP modelling can identify areas of likely high prevalence despite low raw counts through inferences from spatial clustering and population density, so mapped patterns should be read as estimates of inferred risk rather than a complete representation of all victimization. Nevertheless, selective underrecognition may occur where barriers to reporting are unevenly geographically distributed in a way that is not accounted for differences in population density, for example, where populations of similar densities and similar underlying exposure to sexual violence show different levels of reporting due to differences in stigma, chance of reprisals or accessibility of services. Although differences in reported and inferred prevalence may indicate areas at higher risk of undetected sexual violence, underreporting is a widely recognized feature of sexual violence and addressing this challenge requires dedicated initiatives. Although this issue remains understudied in Colombia, evidence from a range of conflicts suggests that sexual violence reporting and detection may be improved by destigmatising victims; reducing risks faced by survivors; strengthening trust in institutions; ensuring the availability of multiple, accessible reporting and detection pathways; systematically curating data; guaranteeing that reports lead to meaningful responses; and implementing gender- and culture-sensitive services44,45,46,47, and these are probably important initiatives to apply in Colombia.

In conclusion, the trajectory of sexual violence in the Colombian armed conflict reflects a shift from geographically concentrated, high-intensity incidents towards more diffuse patterns, with clear areas of high historical prevalence where victim support remains essential. The temporal dynamics of sexual violence suggest that interventions that proactively address structural and long-term contextual determinants may be an important component of prevention.