Study area
Ruili is the southwestern-most city in China, has an area of 1020 km2 and shares a 169.8 km border with Myanmar. Twenty out of a total of 30 administrative villages in the city are located on the border, with 2 border land ports at Jiegao and Xinhe village (Fig. 1). The population of the city is approximately 210,000, of which nearly 60% were reported as ethnic minorities in 2017. Due to environmental continuity, and closed cultural, historical, and linguistic ties, cross-border marriage and trade are common in the area [9].
Data collection and management
In 1985, the HIV/AIDS case reporting system was established in China. All medical institutions, institutions for disease prevention and control, blood donation and supply organizations report HIV/AIDS cases meeting the diagnostic criteria directly to the reporting system. HIV-positive cases are considered on the basis of blood sample screening by enzyme-linked immunosorbent assays (ELISA) and confirmed by Western blot (WB), radioimmunoprecipitation (RIPA) and immunofluorescence assays (IFA). Case’s information was collected by face-to-face interviews using a national standard questionnaire at HIV/AIDS confirmatory laboratory. The high-risk behaviors recorded for all subjects were drug use, heterosexual or homosexual contact, blood products access, surgery exposure, and mother-infant transmission. Possible transmission route for each case was assessed by epidemiologist at the confirmatory laboratory.
This study involved HIV/AIDS patients who were Chinese nationals, Ruili residents, and who had a detailed address at the administrative village level recorded in the reporting system from 1989 through 2016. HIV/AIDS referred to the presence of either HIV infection, the development of AIDS, or both at the time of reporting.
General epidemiological analysis
Case information was entered into the R program (version 3.2.1) for data exploration and analysis. The time trend was described by year. Demographic characteristics, including age, gender and occupation are presented in graphs. The annual average case prevalence rate by administrative village was explored in MapInfo (version 15, serial number: MINWCA 1500000240).
Cluster analysis
The clusters of HIV/AIDS cases infected by either drug use or heterosexual contact were evaluated by global spatial autocorrelations, local spatial autocorrelations, and space-time scanning analyses, respectively. Population data were derived from the Information Management Section of the Yunnan Center for Disease Control and Prevention. The population of each village was employed as denominator in the analyses for HIV/AIDS cases associated with either drug use or heterosexual contact.
Global spatial autocorrelations were employed to assess the spatial distribution patterns (cluster/disperse/random) of HIV/AIDS cases associated with either drug use or heterosexuality throughout Ruili city. Values for Moran’s Index (I) close to − 1 or + 1 indicated a strong positive or negative spatial autocorrelation, respectively. When the P-value was lower than 0.05 in the Z test for the value, the pattern of distribution was considered to be clustered. Otherwise, the pattern was considered to be random.
Local spatial autocorrelations were used to identify clusters at the village level, which helped to understand heterogeneities within the global pattern or driving the overall clustering pattern. Local indicators of spatial autocorrelation (LISA) were examined by the Z-test. When the P-value was lower than 0.05, a local autocorrelation existed. The association patterns were divided into four categories: High-High, High-Low, Low-Low, and Low-High. Both the global and local spatial autocorrelations were performed using GeoDa software (version 1.4.6) [10].
Spatial autocorrelation analysis can determine the spatial heterogeneity of a disease; however, it has limited function in detecting the specific aggregation range of the disease and its clustering over time. A discrete Poisson model was employed to detect the distribution of HIV/AIDS clusters over space and time. The method scanned areas of high HIV incidence with a time allocation of 1 year, a maximum size of a spatial cluster equal to 50% of the at-risk population, a maximum size of a temporal cluster equal to 50% of the study period and a maximum of 999 Monte Carlo simulations. A log likelihood ratio (LLR) and relative risk (RR = observed number of cases/expected number of cases) were calculated. A cluster was considered statistically significant if the P-value was lower than 0.05. Space-time scanning was composed by SaTScan software (version 9.6). All of the results of the spatial analyses were visualized by using MapInfo (version 15, serial number: MINWCA 1500000240) [11].