Late HIV Diagnosis Clusters in Rural Yinzhou, With Older Adults at Highest Risk

More than three-quarters of people diagnosed with HIV in one eastern Chinese district over a 25-year period learned of their infection late, according to a retrospective cohort study of the region’s full case registry. Among 1,404 people diagnosed with HIV or AIDS in Yinzhou District, Ningbo, between April 2001 and June 2025, 1,072 — or 76.4 percent — were classified as late diagnoses, meaning they arrived at diagnosis with a CD4 cell count below 350 cells per microliter or with an AIDS-defining illness regardless of their count. Only 332 patients, 23.6 percent, were diagnosed before reaching that threshold. The figure, reported in Frontiers in Public Health, exceeds recent estimates from other parts of China and underscores how much undetected infection may persist even where new case numbers appear to be falling.

The study drew on the China National HIV/AIDS Comprehensive Response Information Management System, a national registry into which hospitals and disease control centers report HIV-positive cases and follow-up information in real time. The researchers extracted the full registry of HIV-positive cases from Yinzhou District for 2001 through 2025, along with a separate, non-overlapping random sample of 400 cases from the wider Ningbo area for 2007 through 2025 to serve as an external validation dataset. After excluding 101 Yinzhou cases and 76 Ningbo cases that lacked a recorded CD4 count, the primary analysis covered 1,404 patients. The analyzed Yinzhou cohort included 1,215 men (86.5 percent) and 189 women (13.5 percent), with a median age at diagnosis of 34 years.

The proportion of late diagnosis has not simply held steady. Supplementary analysis of annual figures from 2002 to 2024 showed the proportion fluctuating without a clear monotonic trend but generally rising from 64.9 percent to 74.7 percent over the past decade. Over the same period, the number of new HIV cases declined overall, from 117 to 85 per year. The authors caution that this divergence could mislead interpretation of epidemic control: a falling case count may mask the fact that large numbers of people living with HIV remain undiagnosed, and those who are eventually found are increasingly presenting late.

To identify who was most at risk of late diagnosis, the team used Poisson regression with robust standard errors, reporting relative risks rather than odds ratios because the outcome was common. After adjustment for covariates, each additional year of age at diagnosis was associated with a 0.3 percent higher risk of late diagnosis (adjusted relative ratio 1.003, 95 percent confidence interval 1.001 to 1.006). Two factors were protective. Patients who tested voluntarily had an 8.2 percent lower adjusted risk than those diagnosed through opportunistic testing (adjusted relative ratio 0.918, 95 percent confidence interval 0.849 to 0.993), and patients living in the northwest of the district had a 6.7 percent lower adjusted risk than those in the northeast (adjusted relative ratio 0.933, 95 percent confidence interval 0.876 to 0.994). In the external validation dataset of 324 Ningbo patients — of whom 249, or 76.9 percent, were diagnosed late — older age again emerged as a significant predictor, with an adjusted relative ratio of 1.007 (95 percent confidence interval 1.002 to 1.012), although voluntary testing was not statistically significant there, a difference the authors attribute to differing testing patterns between the district and the wider city.

Age shaped more than just statistical risk; it marked a distinct clinical and behavioral profile. Among the 136 patients aged 60 or older, 119 — 87.5 percent — were diagnosed late, well above the district-wide figure. Older patients generally had lower educational attainment, worked in manual or agricultural labor or were unemployed, and were far more likely to be tested compulsively rather than voluntarily: 62.5 percent of older adults were tested compulsively, compared with 49.3 percent of younger patients who were diagnosed voluntarily. Older adults also reported a higher proportion of extramarital sex history, particularly commercial sex. The authors connect this to documented patterns of health-seeking behavior among older Chinese adults, especially in rural areas, who may favor informal self-treatment when symptoms arise, partly out of a perceived responsibility not to consume family income or disrupt daily routines, and may only be tested after pressure from family members — by which point optimal opportunities for earlier detection have passed.

Using a stratified grid method with empirical Bayes shrinkage to stabilize estimates in smaller strata, the researchers crossed age groups with testing behavior to locate the highest-risk subgroup. Patients aged 60 and over who were tested non-voluntarily formed the priority group, with a posterior relative ratio of 1.097 (95 percent posterior interval 1.013 to 1.174) relative to the overall study population; 103 of 119 patients in this stratum were late presenters, a raw rate of 86.6 percent. The lowest-risk stratum was patients under 30 who tested voluntarily, with a posterior relative ratio of 0.89 (95 percent posterior interval 0.822 to 0.958). The gap between the highest- and lowest-risk strata was roughly 23.3 percent, and the estimates remained stable before and after shrinkage adjustment.

Geography added a second dimension to the risk picture. The district’s HIV services — three voluntary counseling and testing clinics, one designated HIV/AIDS hospital, and one post-exposure prophylaxis clinic — are all located in the northwestern part of Yinzhou, which is entirely urban. The northeast is also fully urban, while the rest of the district mixes urban and rural settlements. A global Moran’s I statistic of 0.65 (p < 0.001) indicated strong spatial clustering of late diagnosis across the district’s 25 sub-districts. Local Moran’s I analysis identified High-High clusters — hotspots of late diagnosis — in the central and southeast areas, and Low-Low clusters, or cold spots, in the northwest. Sub-districts in the northwest had notably more HIV cases but a lower proportion of late diagnoses, while the southeast, especially, had few cases but a relatively high late-diagnosis proportion. The authors note the southeast is predominantly rural with almost no HIV-related medical resources, a pattern consistent with earlier studies linking rurality to late presentation, including work in Tennessee, United States, where the association with rurality disappeared after adjusting for system-level factors such as poverty, education, and health insurance.

Late diagnosis also carried a survival penalty. During the study period there were 61 deaths (4.3 percent of the cohort), 58 of them among late presenters and 3 among non-late presenters. Late presenters had a median CD4 count at diagnosis of 206 cells per microliter and a median delay of 8 days. Kaplan–Meier curves over 120 months of follow-up showed significantly lower survival among late presenters (log-rank p = 0.002). In a Cox proportional hazards model adjusted for sex at birth, educational level, marriage status, testing behavior, and occupation, late diagnosis was associated with a 3.03-fold higher hazard of death (95 percent confidence interval 1.04 to 8.84, p = 0.04), compared with a crude hazard ratio of 3.20 (95 percent confidence interval 1.14 to 9.03). Because Schoenfeld residual tests flagged violations of the proportional hazards assumption for age and marital status, the authors stratified by marital status and used attained age as the primary time scale after sensitivity analyses showed age-based models were more stable.

The authors frame their findings within a multi-level model of health behavior, in which individual, interpersonal, institutional, and structural factors jointly determine outcomes. In Yinzhou, age, testing behavior, and spatial disparities in medical resources overlap: the clinics that could enable voluntary testing are concentrated in the urban northwest, while the hotspots for late diagnosis lie in rural areas far from those services. They point to prior evidence that multi-level, community-based interventions — including training, community advisory committees, and expanded voluntary counseling and testing sites — can raise HIV testing and prevention service utilization, and argue that interventions designed for younger adults may not transfer to older rural populations.

The study has limitations the authors acknowledge. The late-diagnosis definition based on CD4 count could misclassify patients in seroconversion as late because of temporarily low counts; routinely collected surveillance data omitted some potentially relevant variables, such as partners’ HIV status, with missing rates as high as 51.85 percent for the number of non-marital heterosexual partners and 38.68 percent for same-sex partners, addressed through multiple imputation by chained equations; baseline characteristics differed between the derivation and validation cohorts, with standardized mean differences above 0.1 for several variables including age and education, limiting direct comparability; and the 25-year span covers a period in which public awareness and testing strategies evolved substantially, while incorporating calendar year into the models caused instability. Even so, a sensitivity analysis using the stricter CD4 threshold of 200 cells per microliter preserved the direction of the associations, with age, student status, northwest residence, and voluntary testing remaining significant. The authors conclude that policymakers should attend to the high proportion of late diagnosis rather than focusing solely on new case counts, and that people aged 60 and over — particularly those reached through non-voluntary testing — should be prioritized for testing promotion in Yinzhou.

Subject of Research: Spatial clustering and risk factors for late HIV diagnosis in Yinzhou District, Ningbo, China

Article Title: Spatial clustering and risk factors for late HIV diagnosis in the Yinzhou, China: a population-based retrospective cohort study

Article References: Chen, Y., Yu, W., Wang, B., Liu, B., Jiang, H., Gu, Y., Lin, L., & Gong, D. (2026). Spatial clustering and risk factors for late HIV diagnosis in the Yinzhou, China: a population-based retrospective cohort study. Frontiers in Public Health, 14, Article 1888278. https://doi.org/10.3389/fpubh.2026.1888278

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1888278

Keywords: late HIV diagnosis, Yinzhou District, Ningbo, spatial autocorrelation, Moran's I, CD4 count, older adults, voluntary testing, retrospective cohort, survival analysis, UNAIDS 95-95-95, China