Spatial association of surface water quality and human cancer in China

Surface water is a vital resource, and deterioration of water quality threatens availability, sustainability, and public health. Carcinogenic risk is central to health assessments of surface water, yet the relationship between surface water quality and cancer remains insufficiently understood in China, which bears nearly a quarter of global cancer cases. Individual-level studies have linked drinking water sources and disinfection by-products to several cancers, but assessing exposure to surface water pollutants is challenging due to multiple exposure routes and complex environments. Ecological analyses enabled by extensive water monitoring and cancer registry networks can offer broader insights. This study integrates nationwide surface water and cancer data to investigate spatial associations between multiple pollutants and multiple cancer types, identify key pollutants within river basins, and forecast the cancer burden potentially attributable to poor surface water quality.

Prior research has associated drinking water characteristics and by-products with risks of oesophageal, gastric, colorectal, and renal cancers in cohort/case-control designs, but such individual-level approaches are difficult to apply to surface water due to complex exposure pathways (ingestion, dermal, inhalation) and lag times. Evidence accumulation on carcinogens is slow, with few additions to IARC Group I in recent years. Ecological studies leveraging monitoring data have linked water pollution frequency to digestive cancer mortality in China’s Huaihe River Basin. China’s nationwide networks for water quality monitoring and cancer registries provide a unique opportunity for broader assessment. However, spatial heterogeneity in water bodies, pollutant levels, and pollutant co-existence complicates appraisal. The coexistence and joint effects of multiple pollutants had not been comprehensively evaluated at a national scale prior to this work.

Study design and data integration: The study integrated 2021 surface water quality data from 3632 monitored sections of the National Surface Water Environmental Quality Monitoring Network and cancer incidence data from the 2019 Annual Report of the China Cancer Registry (487 registries; 381.6 million population). Water sections were matched to cancer registries via a buffer-based spatial linkage: each water monitoring section served as the center of a 30 km radius buffer, reflecting decreasing influence with distance from rivers. This retained 2331 of 3632 sections and 486 of 487 registries, covering populations in all 31 mainland provinces. River basins: Analysis considered nine basins (Songhua & Liaohe, Haihe, Huaihe, Yellow, Continental, Southwest, Yangtze, Southeast, Pearl). Yangtze was subdivided into upstream, midstream, downstream; Continental and Southwest were combined where appropriate. Pollution indicators and thresholds: Twenty-one routinely monitored pollution indicators were included. Inferior water quality for each indicator was defined using the national 75th percentile of its measured value as the threshold (stricter than EQSSW Level III for all but total nitrogen). Spatial autocorrelation and clustering: Global Moran’s I tested spatial autocorrelation for each cancer-related indicator; Local Moran’s I identified high-high (HH) and low-low (LL) clusters and high-low (HL) and low-high (LH) outliers. CAMS grading: The Cluster Analysis of Multi-pollutants in Space (CAMS) design quantified multi-pollutant co-existence by counting, for each section, the number of indicators classified as HH or HL (denoted H). Grades: 0 (H=0; 26.9%), 1 (H=1–2; 38.4%), 2 (H=3–5; 24.0%), 3 (H≥6; 10.7%). Cancer outcomes: Eleven cancers were analyzed (ICD-10 codes): oesophagus (C15), stomach (C16), colorectum (C18–C21), liver (C22), gallbladder (C23–C24), pancreas (C25), lung (C33–C34), bone (C40–C41), breast (C50), kidney (C64–C66, C68), brain (C70–C72, C32–C33, D42–D43), selected by incidence, mortality, and 5-year survival relevance in China. Statistical analyses: - Negative binomial regression assessed associations between individual indicators and cancer incidence within river basins where >20% of sections exhibited inferior quality for that indicator (to avoid dilution). Incidence and rate ratios (RRs) with 95% CIs across CAMS grades were estimated. - Pairwise cross-indicator correlations used Spearman’s rho. - Machine learning: XGBoost models (learning rate 0.04, max depth 3, subsample 0.8) with SHAP values quantified the relative importance of indicators for total cancer incidence overall and within each river basin. - Population attributable fraction (PAF) and excess cases (EC): PAF = Σ p_i (RR_i − 1) / Σ p_i RR_i, where p_i is prevalence of exposure at CAMS Grade i, RR_i the relative risk at Grade i. EC = incidence × PAF. Exposure prevalence by grade was evaluated via Thiessen polygon analysis. - Projections: Linear regression projected annual total incidence of selected cancers for 2017–2030 based on 2010–2016 trends; combined with PAFs to estimate future water-quality–attributable cases assuming no water quality improvement. Software: Data integration and negative binomial regression were conducted in SAS 9.4. Spatial analyses used standard GIS/statistical methods; SHAP values were derived from XGBoost outputs.

- Eleven of 21 indicators were positively associated with at least one cancer: total nitrogen, petroleum, total phosphorus, permanganate index, COD, volatile phenol, fluoride (F−), ammonia nitrogen (NH3-N), arsenic, selenium, and zinc. All cancer-related indicators exhibited significant spatial autocorrelation (global Moran’s I, p<0.05). - CAMS dose-response: A graded increase in cancer incidence with higher CAMS grades was observed for multiple cancers. Examples (RR vs Grade 0; 95% CI): • Stomach: Grade 1 RR 1.10 (1.06–1.13), Grade 2 RR 1.17 (1.13–1.21), Grade 3 RR 1.17 (1.13–1.22). • Pancreatic: 1.09 (1.03–1.16), 1.26 (1.19–1.34), 1.26 (1.19–1.34). • Kidney: 1.88 (1.63–2.17) at Grades 1–3. • Oesophageal: 1.17 (1.14–1.22) at Grade 2; 1.46 (1.40–1.55) at Grade 3. • Lung: 1.09 (1.07–1.11) at Grade 2; 1.24 (1.19–1.30) at Grade 3. • Brain and bone showed significant increases at Grade 3 only (RR ~1.18 and ~1.09, respectively). - Overall incidence across 11 cancers rose with CAMS grade (e.g., Total: Grade 0 incidence 197.62 per 100,000; Grade 2 220.15; Grade 3 222.95; corresponding RRs 1.11 and 1.13). - Co-exposure patterns: Higher CAMS grades corresponded to poorer quality across most pollutants (e.g., F− inferior-quality rates: Grade 0 1.0%, Grade 1 12.5%, Grade 2 52.8%, Grade 3 90.8%) and increasingly complex inter-pollutant correlations. - River basin patterns: • Huaihe Basin: high incidence (top for oesophageal; third for gallbladder, bone); key indicators: permanganate index, petroleum, COD. • Haihe Basin: elevated stomach and oesophageal cancer; fluoride implicated. • Songhua & Liaohe Basin: highest liver and kidney, high pancreatic, lung, breast, colorectal; selenium implicated. • Yangtze downstream: highest combined incidence (288.50 per 100,000); key indicators: total nitrogen, petroleum, permanganate index, zinc, fluoride, volatile phenol. Up- and midstream showed little or no association. - Attributable burden and projections: Average 5.0% of new cancers in registry areas attributable to poor surface water quality (highest in Huaihe 9.3%, Songhua & Liaohe 7.3%, Yellow River 5.8%). Assuming no water quality improvement, total new cancer cases projected to increase by 7.8% to 874,218 by 2030, with 388,431 cumulative cases during 2022–2030 attributable to poor water quality.

The study demonstrates a nationwide spatial association between surface water quality and multiple cancer incidences in China, addressing the gap in understanding multi-pollutant, multi-cancer relationships at population scale. The CAMS framework captures coexisting pollutants and spatial dependencies, revealing dose-response relationships for several cancers and underscoring joint effects beyond single-pollutant assessments. River basin analyses identify basin-specific key pollutants and contextual factors (e.g., economic development gradients, agricultural and industrial activities), informing targeted interventions. Findings suggest that current standards (e.g., EQSSW) are not tailored for cancer risk assessment; data-driven, dynamic thresholds could better guide policy and continuous improvement. While surface water is a critical driver, other environmental exposures (e.g., air pollution) likely contribute to observed patterns, particularly in some basins, emphasizing the need for integrated environmental health strategies. The projected increase in water-quality–attributable cases underlines urgency for pollution source control (fertilizer management, industrial and domestic wastewater treatment) and sustainable development policies.

This work introduces an integrated, river-basin–oriented framework linking nationwide surface water quality and cancer incidence, and proposes the CAMS design to quantify multi-pollutant spatial co-exposures. It provides consistent evidence of dose-response relationships for multiple cancers, identifies basin-specific key pollutants, and estimates current and future cancer burdens attributable to poor water quality. Policy implications include revising water quality standards for cancer risk relevance (e.g., adopting annually updated percentile-based thresholds), prioritizing basin-specific pollution control (agricultural nitrogen management, industrial discharge regulation), and fostering cross-sector data harmonization. Future research should refine exposure thresholds, investigate pollutant interactions and forms (e.g., inorganic selenium), evaluate causality with complementary designs, and extend the CAMS approach to other environmental domains and health outcomes.

- Ecological design limits causal inference and is susceptible to unmeasured confounding (e.g., air pollution, socioeconomic factors); findings reflect population-level associations. - Multiple pairwise tests between indicators and cancers warrant cautious interpretation. - Inferior-quality thresholds set at national 75th percentiles may not be optimal for health protection; negative findings for some indicators may reflect suboptimal cut-offs (results similar with 80th percentile). - Annual averaging of water quality smooths temporal variability and may mask short-term peaks; cancer registry incidences are relatively stable but time lags between exposure and cancer remain. - Buffer-based exposure assignment (30 km) and spatial interpolation (Thiessen polygons) introduce assumptions about influence distance and population exposure prevalence. - Reliance on available monitoring indicators and registry coverage; other pollutants or exposure pathways were not assessed.