This longitudinal observational study was conducted in Dehong (Figure 1A), which has a population of approximately 1.3 million and an area of 11,526 km². The study region is divided into 5 districts (Ruili, Mangshi, Lianghe, Yingjiang, and Longchuan), encompassing 51 township-level districts (6 in Ruili, 12 in Mangshi, 9 in Lianghe, 15 in Yingjiang, and 9 in Longchuan).

Data on newly diagnosed HIV cases at the township level in Dehong were obtained from the national HIV/AIDS Comprehensive Response Information Management System (CRIMS) by the Dehong Center for Disease Control and Prevention. The data cover the period from January 1, 2010, to December 31, 2022, and include HIV cases that met the following criteria: (1) reported in Dehong; (2) resided in the specific townships within Dehong; and (3) of Chinese nationality. Township-level population data were obtained from the 2020 China Population Census [10]. All data were collected at the township level. The distance from each township to the China-Myanmar border was defined as the shortest distance between the township administrative boundary and the international border. Based on the calculated distances, all townships were classified into 4 distance-based groups according to proximity, which served as the stratification variable in the GeoDetector analysis. The rate of newly diagnosis HIV cases was obtained by dividing the number of newly diagnosed cases by the local resident population.

As reported in previous studies, the HIV transmission in Dehong may be associated with cross-border population movements between China and Myanmar. We investigated the spatially stratified heterogeneity (SSH) of newly diagnosed HIV case rates, stratified by the distance of the township to the China-Myanmar border. The GeoDetector q statistics were calculated to measure the SSH of newly diagnosed HIV case rates across different township-level units in Dehong, and the formula is as follows [11]:

In the formula (Eq. 1), N is the total number of townships in Dehong Prefecture, σ2 denotes the variance of newly diagnosed HIV case rates across all townships, Nh refers the number of townships in stratum h (1, 2, …, L), and σh2 represents the variance of newly diagnosed HIV case rates within the stratum h. The q value ranges from 0 to 1, with a value of 0 indicating complete randomness and 1 representing perfect stratified heterogeneity.

The Poisson and log link regression functions used to perform analyses were as follows:

In the Poisson likelihood function (Eq. 2), yit represents the number of newly diagnosed HIV cases in township i at year t, ni denotes the total population in township i, and rit is the modeled risk of newly diagnosed HIV/AIDS cases in township i at year t.

In the log link regression function (Eq. 3) that describes the risk of newly diagnosed HIV infection, α is the overall log disease risk for the entire study period, and (Si+Ui) is the disease risk in township i. Specifically, Si captures the spatially structured component, and Ui represents the spatially unstructured component. The term t* denotes the median time of the entire study period, with (b0t*+vt) and b1it* specifying the overall and local temporal trend in township i, respectively. Specifically, the term b0t* captures the linear trend, while vt accounts for the nonlinearity in the overall trend pattern. Given the observed decrease in the newly diagnosed HIV cases rate of over the study period, if b1i < 0, it indicates that the decreasing trend in township i is faster than the overall decreasing trend. Conversely, b1i > 0 suggests a slower trend relative to the overall decrease. The Gaussian noise was captured by εit.

The study area was classified into 9 categories (3 risk categories × 3 trend categories) based on the posterior probability of P (exp (Si) >1 |data) and P (b1i >0 |hi, data), following a specific 2-stage criteria [12]. In the first stage, townships were classified into 3 categories: hotspot, coldspot, and not cold/hot spot. A township is classified as a hotspot (hi=1) if P (exp (Si) >1 |data) is greater than 0.8, and as a coldspot (hi=2) if this probability is less than 0.2 [9]. The remaining townships were classified as not cold/hot spots (hi=3). In the second stage, temporal changes were categorized into 3 classes (faster decrease, common decrease, and slower decrease) according to P (b1i >0 |hi, data). A township was defined as exhibiting a slower decreasing trend compared to the overall decreasing trend if P (b1i >0 |hi, data) >0.8, a faster decreasing trend if P (b1i >0 |hi, data) <0.2, and a common decreasing trend if 0.2 <P (b1i >0 |hi, data) <0.8 [9]. Consequently, a total of 9 categories were established.

Hyperparameters for prior distributions were chosen following the approach of Li et al [9], which has been widely applied in BSTHM. Convergence of the Markov Chain Monte Carlo chains was evaluated using the Gelman-Rubin statistic, with values below 1.05 indicating satisfactory convergence. The variance partition coefficient was measured to assess model performance. All BSTHM analysis were conducted using WinBUGS 1.4.3.

This study was approved by the Institutional Review Board of the National Center for AIDS/STD Prevention and Control, Chinese Center for Disease Control and Prevention (approval number: X241028828). The data used in this study were from the CRIMS. No personally identifiable information was present in the data analyzed for this study.

Between 2010 and 2022, a total of 5045 newly diagnosed HIV cases were reported in Dehong. Of these cases, 3235 (64.1%) were male, 2570 (50.9%) were aged between 31 and 49 years at the time of diagnosis, 2120 (42.0%) were of Han ethnicity, and 2937 (58.2%) had a primary school education or less (Table 1). More than half of the cases (n=2713, 53.8%) were married or cohabiting with a partner. In terms of the transmission routes, 4211 (83.5%) were infected through heterosexual contact, followed by 562 (11.1%) through injecting drug use, and 164 (3.3%) through homosexual contact.

The rate of newly diagnosed HIV cases in Dehong decreased over time, from 57.1 cases per 100,000 population to 13.3 cases per 100,000 population, representing a 76.7% reduction over the past 13 years (Figure 1B). Furthermore, the proportion of townships with a rate of newly diagnosed HIV cases exceeding 60 cases per 100,000 population decreased from 52.9% (27/51) in 2010 to 3.9% (2/51) in 2022. Among the 51 townships in Dehong, the number of townships with more than 20 newly diagnosed cases decreased from 10 in 2010 to 1 in 2022 (Figure 2). GeoDetector analysis revealed notable SSH in the rate of newly diagnosed HIV case when stratified by the distance of townships to the China-Myanmar border (q=0.27; P=.004).

In this study, using township-level data on newly diagnosed HIV infection in Dehong from 2010 to 2022, we quantitatively characterized fine-scale spatiotemporal patterns of HIV risk using BSTHM. The number of newly diagnosed HIV cases has declined significantly over the past 13 years, accompanied by notable SSH. Substantial spatiotemporal heterogeneity was observed in the risk of newly diagnosed HIV infection. By jointly modeling spatial structure and temporal dynamics, several townships were identified as areas that are likely to become or remain hotspots in the future.

In our analysis of newly diagnosed HIV infection from 2015 to 2022, IDUs accounted for 11.1% of cases, while heterosexual contact primarily dominated HIV transmission, responsible for 83.5% of cases. Since HIV was first identified among IDUs in Dehong in 1989 [4], the Chinese government has allocated massive resources to curb HIV transmission. At the provincial level, the Yunnan government has initiated multiple campaigns against HIV and illicit drugs, with remarkable achievements. A study incorporating HIV incidence assays to estimate the HIV incidence among IDUs in Dehong estimated that the HIV incidence among Chinese IDUs in Dehong decreased from 3.11% (95% CI 1.59%‐4.64%) in the 2009‐2010 period to 0.58% (95% CI –0.06% to 1.04%) in the 2015‐2017 period [13]. Furthermore, the proportion of IDUs among newly diagnosed male HIV cases in Dehong also showed a downward trend, decreasing from 46.7% in 2012 to 32.0% in 2016 [6]. These shifts reflect a structural transition in transmission dynamics, with heterosexual contact becoming the leading cause of local HIV infections, accounting for 84.7% of newly diagnosed HIV cases during the 2015‐2017 period [14]. In 2017, the Yunnan government issued an implementation plan aimed at achieving the 90-90-90 HIV targets by the end of 2020. These unprecedented efforts to strengthen HIV prevention, testing, and treatment have resulted in a significant reduction of 76.7% in the rate of newly diagnosed HIV cases over the past 13 years.

Notable SSH was observed in newly diagnosed HIV cases rate across Dehong (q=0.27; P=.004). The spatial relative risk was apparently higher in townships closer to the China-Myanmar border and declined with increasing distance (Figure 3B). Consistent with the area classification analysis, most hotspots were concentrated in the cross-border region. Cross-border mobility and high HIV prevalence of Burmese individuals may play a significant role in HIV transmission. Due to lower income levels in Myanmar, many Burmese individuals migrate to China in pursuit of better economic opportunities, resulting in frequent interactions between the 2 populations. Previous studies have noticed the high HIV prevalence among Chinese-Burmese mixed couples [15,16]. Chen et al [17] recruited 617 Yunnan-Myanmar IDUs in 3 counties of Yunnan and conducted molecular transmission analysis on 117 (19.0%) people living with HIV. Their findings indicate that Yunnan-Myanmar IDUs play a pivotal role in the bidirectional cross-border HIV transmission [17]. Hu et al [18] further demonstrated the strong association between Burmese immigrants and HIV transmission in Dehong. According to a survey of 393 Burmese immigrants in Dehong in 2020, they had low levels of HIV knowledge, condom usage, and willingness to undergo HIV testing [19]. Consequently, the convergence of high background prevalence, frequent population mobility, and constrained access to HIV services likely contributes to the persistent elevated risk observed in border townships. Furthermore, in districts such as Ruili and Longchuan, where ports are in close proximity to main residential and commercial areas, the frequency of cross-border mobility and interactions is increased.

To control the cross-border transmission of HIV, several measures have been implemented in Dehong. These include strengthening comprehensive management of cross-border populations, providing free ART to Chinese-Burmese mixed couples or long-term resident immigrants living with HIV, and expanding HIV testing among immigrants engaged in long-term employment in Dehong. Nevertheless, the persistence of high-risk hotspots identified through BSTHM suggests that additional, more precisely targeted interventions are needed. The identification of these townships provides important epidemiological clues for prioritizing resource allocation. Further prevention and control measures are required for hotspot areas of cross-border HIV transmission. First, cross-border HIV prevention and control mechanisms should be established by strengthening bilateral collaborations between China and Myanmar. Second, programs should be implemented to improve health education, promote condom use, and increase HIV testing rates among Burmese immigrants. Third, appropriate HIV treatment programs should be explored for short-term Burmese immigrants, as they are currently ineligible for China’s free ART program. Lastly, efforts should be made to explore ways to improve treatment adherence among HIV-infected immigrants to prevent treatment failure and the development of drug resistance.

In this study, we further estimated the local temporal trend of the spatial pattern using BSTHM, which can predict the spatial pattern shift in the future. The local trends were estimated by comparing their trend patterns with the overall trend pattern. In this study, we found that 2 hotspots exhibited a slower declining trend than the overall trend, while 4 coldspots with a slower declining trend showed a tendency to become “hot” (Table 2). These townships exhibited higher inclination of hotspots in the future and should be prioritized for targeted interventions. Further in-depth epidemiological investigations are warranted to explore the local drivers of transmission in these areas and to inform more precise and context-specific intervention strategies. To our knowledge, few studies in China have investigated the combined spatial patterns and temporal trends in the rate of newly diagnosed HIV infections. Previous spatial analysis studies primarily employed spatial autocorrelation analysis methods to examine the spatial distribution characteristics separately across different time periods [20-24] and used spatial scan statistics to detect cluster areas [21,25]. None of these studies incorporated spatiotemporal correlations. This study modeled spatial relative risk, overall temporal trends, and local temporal trends at the township level, thereby constructing a multifaceted picture of the evolving risk of newly diagnosed HIV cases across Dehong, representing a methodological advantage over traditional spatial analyses. This small area analysis has important implications for the allocation of resources in controlling HIV transmission in similar settings, as interventions and measures can be directly implemented in hotspots or potential hotspots to reduce new infections.

This study is subject to several limitations. First, since HIV diagnoses are not made immediately after HIV infection, the incidence rate could not be calculated in this study. Nevertheless, in the context of generally increasing HIV testing coverage in Dehong, the observed decline in newly diagnosed cases likely reflects, at least to some extent, a reduction in new HIV infections, indicating the effectiveness of local HIV control measures. Second, the analysis in this study was conducted at the township level; the macro-level factors potentially associated with the rate of newly diagnosed HIV infections were not investigated due to the data availability. Third, all analyses in this study were restricted to newly diagnosed HIV cases of Chinese nationality, as data on foreign nationals were not available. Consequently, the findings primarily reflect HIV infection patterns among the local resident population and may underestimate the overall contribution of cross-border populations to the regional HIV epidemic.

In conclusion, the rate of newly diagnosed HIV cases in Dehong has significantly decreased from 2010 to 2022. By applying a BSTHM framework to routine surveillance panel data, this study reveals substantial variations in the spatial distribution and temporal trends of the risk of newly diagnosed HIV infections within Dehong, with high-risk areas concentrated in townships near the China-Myanmar border. Future efforts should focus on exploring effective measures to reduce new infections in these high-risk areas. This analytical approach provides a new perspective for assessing local needs, guiding resource allocation, and developing targeted interventions to control HIV infection in other high-burden or highly heterogeneous settings.