Geospatial Investigations in Colombia Reveal Variations in the Distribution of Mood and Psychotic Disorders

Geographical variations in the prevalence and distribution of mood and psychotic disorders have been extensively documented and studied in high-income countries (HICs). These variations often reflect disparities in access to mental healthcare, a phenomenon observed and quantified as early as 1852. Studies consistently demonstrate a distance-decay effect, where healthcare utilization decreases proportionally with distance or travel time to the facility. Research in HICs also suggests that geographical variations may be influenced by localized social, environmental, and genetic factors affecting both treatment utilization and disease risk. The availability of comprehensive population registries and public databases in HICs has facilitated geospatial investigations of mental health treatment. However, such data have been historically scarce in low- and middle-income countries (LMICs), limiting similar research opportunities. Recently, the emergence of electronic health record (EHR) databases in some LMICs offers new potential for population-level analyses of individuals seeking mental healthcare. In Colombia, a middle-income country, the primary sources of population-level mental health data have been national surveys. While these surveys provide valuable information on prevalence and perceptions of access to care, they lack crucial details, such as the distribution of severe and acute presentations of disorders (particularly psychotic disorders), and actual utilization of services. Moreover, the sparse sampling in these surveys limits geospatial investigations. This study addresses these limitations by analyzing the EHR database of the Clínica San Juan de Dios Manizales (CSJDM), a comprehensive mental healthcare provider for the one million inhabitants of Caldas, Colombia. The CSJDM’s data offer a detailed picture of geospatial variation in treated mental illness across the entire department, providing insights unavailable through previous data sources.

Existing literature extensively documents geographical variations in mental health disorders within high-income countries. Early studies, such as Jarvis's 1852 work on the supposed increase of insanity, and later research by Faris and Dunham on mental disorders in urban areas, established the foundation for understanding these spatial patterns. These studies, and subsequent investigations, employed statistical models to demonstrate a clear relationship between mental healthcare utilization and distance, a phenomenon known as distance-decay. This effect indicates that the likelihood of an individual accessing healthcare diminishes as the distance to the healthcare facility increases. More recent studies have expanded upon this, demonstrating the role of socioeconomic position and distance on mental health care utilization. Research also indicates the influence of neighborhood environmental factors, such as green spaces and urban density, on depression risks. In addition to environmental factors, studies using geospatial big data analysis have explored the links between immigrant concentration and mental health service utilization, highlighting the complex interplay of social and geographical factors. However, research on geographical variations in LMICs has been limited due to data scarcity. This study contributes to filling this gap by providing a geospatial analysis of mental health data from a comprehensive mental health facility in a middle-income country, allowing a comparison with findings from HICs and providing a deeper understanding of the contextual factors influencing access and distribution of mental health disorders in LMIC settings.

This retrospective cross-sectional study utilized electronic health records (EHRs) from the Clínica San Juan de Dios Manizales (CSJDM) in Caldas, Colombia, spanning the years 2005–2018. The EHRs contained demographic information, residential addresses, and diagnostic codes using the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10). The Institutional Review Boards (IRBs) at CSJDM and UCLA approved the study, which involved anonymized data uploaded to a HIPAA-compliant server. The study focused on 16,295 adult patients (aged 18–90) residing in Caldas with an initial diagnosis of bipolar disorder (BPD), schizophrenia (SCZ), or major depressive disorder (MDD). Initial diagnoses were used to avoid bias from right censoring. Addresses were geocoded using OpenCage's geo-coder, with addresses formatted to improve geocoding accuracy. Addresses with OpenCage confidence scores below 6 were excluded, ensuring accurate geographic coordinates. Annual incidence rates per 100,000 individuals were calculated for each municipality and at a finer 5km x 5km grid level, using population estimates from the WordPop database. A friction surface map was created using AccessMod5, incorporating data on topography, land cover, water bodies, and road networks, to quantify travel time to the CSJDM from each grid cell. Zero-inflated negative binomial regression was used to model the relationship between travel time and the expected number of cases, considering inpatients and outpatients separately, and for each diagnosis. Spatial scan statistics (SaTScan) identified hotspots of inpatient residences, using a 5km x 5km grid and a maximum window size encompassing 25% of Caldas' population. Oliveira's F function quantified uncertainty in hotspot borders. Sensitivity analyses explored the impact of gender and variations in maximum hotspot size.

The study revealed significant inequities in geographic accessibility to mental healthcare in Caldas. Approximately 50% of the population lives within 1 hour's drive of the CSJDM, while about 10% live more than 4 hours away. A strong distance-decay effect was observed for outpatients, with a 20% decrease in expected outpatient cases for every hour increase in travel time to the CSJDM. This effect was primarily driven by outpatients with MDD. In contrast, no significant distance-decay was found for inpatients, regardless of diagnosis. Instead, inpatient cases clustered in nine distinct hotspots identified using spatial scan statistics. These hotspots were geographically dispersed throughout Caldas, suggesting that factors beyond distance to the CSJDM influence inpatient care utilization. One hotspot, located approximately 1.5 hours from the CSJDM, showed a nearly six-fold overrepresentation of inpatient BPD cases, significantly exceeding the BPD incidence observed in other regions or in similar studies in other countries. This hotspot also displayed a significant overrepresentation of MDD cases, but not SCZ. Another hotspot showed a striking nearly nine-fold overrepresentation of SCZ cases despite being located five hours from the hospital. The overall incidence rate for inpatients was highest in Aranzazu, a municipality about 55km from the CSJDM, reflecting high concentrations of inpatient mood disorders.

The findings highlight significant inequities in access to mental healthcare in Caldas, Colombia. The observed distance-decay effect for outpatients aligns with previous studies in HICs, suggesting that many individuals requiring outpatient treatment may live too far from the CSJDM to access care. Conversely, the absence of distance-decay and the presence of geographically dispersed hotspots for inpatients indicate that other factors beyond geographic accessibility are crucial determinants of inpatient utilization. The identification of hotspots, particularly the remarkable overrepresentation of BPD in one cluster, suggests a need for further research to explore potential underlying sociodemographic, environmental, or genetic factors contributing to these spatial variations. These could include the impact of socioeconomic inequality, exposure to violent conflict and displacement, or genetic predispositions specific to the region. Future studies integrating detailed sociodemographic data and potentially genetic information could provide a more comprehensive understanding of these factors.

This study demonstrates significant geographic variations in the distribution of mood and psychotic disorders in Caldas, Colombia, highlighting inequities in access to mental healthcare. The distance-decay observed for outpatients underscores the need for improved accessibility, while the identified inpatient hotspots suggest the influence of local factors beyond distance. Further research is needed to investigate these local factors, integrating data on socioeconomic status, exposure to conflict, and genetic predispositions to inform strategies for more equitable access to mental healthcare and improve the understanding of mental illness etiology in this population.

This study focused on first-visit data, potentially neglecting longitudinal variations in patient trajectories or diagnoses. The smaller inpatient sample size might limit the power to detect subtle distance-decay effects. Travel time estimations were based on average speeds, neglecting seasonal or daytime variations. Data on factors like stigma and treatment-seeking behaviors were unavailable, limiting a comprehensive understanding of care utilization patterns.