Serum trace metals across environmental-pressure strata in Campania, Italy (SPES): a population-based cross-sectional biomonitoring study Authors: Pellegrino Cerino, Antonio Pizzolante, Carlo Buonerba, Paolo Montuori Published: 2026 ABSTRACT AbstractBackground and ObjectivesWe characterized fine-scale geographic variation in serum trace metals across Campania (southern Italy).Materials and MethodsIn 2016–2017, the SPES programme sampled residents aged 20–49 years from municipal registries across Campania. Municipalities were stratified using an environmental-pressure index into three broad impact strata: high-impact, medium-impact and low-impact. For higher spatial resolution, municipalities were also grouped into 21 predefined municipal clusters. We measured 20 serum trace metals by ICP-MS and estimated age- and sex-adjusted geometric mean ratios (GMRs) comparing each stratum or cluster with the study-wide geometric mean.ResultsOf 4205 enrolled participants, 4188 were included. Medium-impact areas showed higher mercury (GMR 1.87, 95% CI 1.73–2.02) and antimony (1.95, 1.71–2.23) relative to the cohort average. At the municipal-cluster scale, the Irno Valley clusters around the Fonderie Pisano foundry (Baronissi-Pellezzano and Salerno) had about six-fold higher mercury (GMR 5.71 and 6.10) and about three-fold higher antimony (3.45 and 2.93), whereas the Valle del Sabato cluster showed about ten-fold higher antimony (GMR 10.5) and about seven-fold higher nickel (7.16) and cadmium (7.67).ConclusionsOverall serum metal concentrations were broadly comparable with European general-population ranges, but pronounced small-area heterogeneity indicates localized hotspots that provincial or regional averages can obscure.Keywords: Human biomonitoring, trace metals, environmental exposure, geospatial analysis, Campania Introduction Pollution and chemical exposures remain a leading, preventable driver of morbidity and premature mortality globally and in Europe, and their impacts are often concentrated where environmental and social stressors co-occur [1–2]. Among chemical hazards, metals and metalloids warrant sustained public health attention because several are persistent in environmental media, may bioaccumulate through food chains, and are linked to cardiovascular, renal, neurological, reproductive, and carcinogenic outcomes [3–4]. Recent ecotoxicological literature also shows that cadmium, lead, and agrochemical contaminants can accumulate in aquatic organisms and produce measurable growth, behavioral, haematological, biochemical, and morphological effects in sentinel fish species, reinforcing the broader environmental relevance of persistent metal and contaminant pollution [5–7]. In Europe, emissions of selected heavy metals have declined over recent decades, yet legacy contamination, diffuse sources, industrial point sources, and complex exposure pathways can sustain population exposure and community concern [8]. A central challenge for prevention is that environmental monitoring-while essential for source identification-does not quantify the internal dose that results from combined routes of exposure and individual behaviors. Human biomonitoring (HBM) directly measures chemicals or biomarkers in human specimens, integrating inhalation, diet, occupation, lifestyle, and other pathways into a biologically relevant metric for surveillance, risk assessment, and evaluation of interventions [9]. Europe has invested in advancing and harmonizing HBM through WHO/Europe’s work and coordinated initiatives such as HBM4EU, which demonstrated the feasibility and value of generating comparable exposure data across countries and populations [10–11]. However, large, population-based HBM datasets remain limited in settings where community concern is high, exposures are heterogeneous, and risk narratives are contested. The Campania region in southern Italy illustrates this challenge. For decades, parts of Campania-particularly in the provinces of Naples and Caserta-have been associated with the “Land of Fires” (Terra dei Fuochi) phenomenon, encompassing illegal trafficking, dumping, and burning of waste alongside pressures from urbanization and industrial activities [12–13]. Epidemiological surveillance and ecological analyses have reported modest but recurring excesses in selected outcomes in some areas, but causal attribution remains difficult because contamination is spatially heterogeneous, exposures occur as complex mixtures, and behavioral and occupational factors can confound or modify associations [14]. Previous biomonitoring efforts in Campania provided important signals while also underscoring the need for broader, population-based follow-up capable of supporting geospatial contrasts and adjustment for individual determinants of exposure [15–16]. To address these gaps, the Exposure Study on Susceptible People (SPES) was designed as a population-based HBM survey embedded in an explicit “source-pathway-target” framework and informed by multi-matrix environmental monitoring in Campania [17–19]. Municipalities were geo-stratified using a pressure index derived from a comparative risk assessment model combining georeferenced information on contamination sources and environmental matrices, and then grouped into clusters (domains) intended to enable interpretable comparisons of internal dose across gradients of environmental pressure [18,20]. SPES recruitment was designed around random sampling from municipal registries of residents aged 20–49 years, with collection of biospecimens and structured information on residential history, occupation, lifestyle, and diet, including use of validated dietary assessment instruments aligned with EPIC methodology [19,21]. SPES was promoted and sponsored by the Istituto Zooprofilattico Sperimentale del Mezzogiorno (IZSM) and was implemented as a monocentric investigation coordinated and conducted by the Istituto Nazionale Tumori IRCCS “Fondazione G. Pascale” (Naples) under the scientific leadership of the late principal investigator Dr Maurizio Montella [17,19]. Under a regional agreement, the University of Naples Federico II (Department of Public Health) provided technical and methodological support for selected data-processing and statistical components, but did not take an active role in participant investigation [20]. The principal dissemination output to date has been an online technical report (grey literature) that did not undergo full journal peer review [20]. Against a backdrop of long-standing community concern about environmental contamination in parts of Campania, we conducted a secondary analysis of the SPES serum trace-metals module using a prespecified, fully scripted pipeline and a de-identified analytical dataset. Our objectives were to quantify geographic differences in serum trace metals across impact-area strata and municipal clusters while appropriately handling < LOQ values, assay nonperformance, and covariate missingness; provide a parsimonious, interpretable set of contrasts adjusted for age group and sex to support future analyses; and provide a transparent, fully scripted analysis and missing-data pipeline with prespecified QC checks. Although SPES fieldwork was conducted in 2016–2017, dissemination has been delayed by organizational and system-level disruptions, including reprioritisation during the COVID-19 period. We prespecified the following analytic targets: the primary objective was to estimate differences in serum trace-metal concentrations across impact-area strata (high, medium, and low) and municipal clusters. Primary endpoints were serum trace-metal concentrations (20 analyte channels), summarized as geometric means and compared using geometric mean ratios (GMRs) with 95% CIs from log-linear models adjusted for age group and sex. The point-source proximity analysis using residential distance and direction was exploratory and hypothesis-generating. Materials and methods Study design, setting, and participants We conducted a secondary cross-sectional analysis of data from the SPES population-based human biomonitoring programme implemented in the Campania region (southern Italy) in 2016–2017 within the “Campania Trasparente” monitoring framework [16–20]. This report follows the STROBE Statement and the STROBE-ME extension; statistical reporting follows SAMPL recommendations [22–24]. Field operations (sampling and recruitment, informed consent, clinical examination, biospecimen collection and storage, and laboratory determinations) were completed within the parent SPES programme under a prospectively approved protocol; the present work used a curated, de-identified, frozen analytical extract and involved no new participant contact, specimen collection, or laboratory testing [17,19,20]. Municipalities were geo-stratified using the Municipality Index of Environmental Pressure (MIEP), previously developed by some of the coauthors of this manuscript, and aggregated into municipal clusters. MIEP is a municipality-level composite geostratification tool rather than an individual exposure metric. It was developed from the Campania Trasparente comparative-risk framework by combining georeferenced information on potential contamination sources, territorial pressures, and environmental matrices at the municipality scale [18,20]. Its purpose in this analysis was to support sampling and broad stratum-level contrasts; it does not encode participant-specific residential history, time-activity patterns, diet, occupation, smoking, proximity to local sources, or metal-specific toxicokinetics. Consequently, high-, medium-, and low-impact strata should be interpreted as contextual pressure categories, not as ordered individual-dose categories expected to show monotonic gradients for every serum metal. Recruitment targets were allocated across high-, medium-, and low-impact strata in a 4:2:1 ratio [18,20]. Participants were randomly sampled from municipal registries using study software and invited via mailed contact with call-center follow-up, supported by territorial recruitment units and communication campaigns [17,20]. The protocol eligibility criteria were age 20–49 years, both sexes, negative history for HCV/HBV/HIV, and residence in the municipality for ≥ 10 continuous years; later SPES outputs describe closely related criteria (20–50 years and ≥ 5 years residence) [17,19,20]. The analytical extract included a small number of enrolled participants outside the protocol age range (observed 19–55 years; Web Appendix Table S3). Ethics approvals and written informed consent were obtained within the parent SPES programme (Ethics Committee, IRCCS “Fondazione G. Pascale”, Commissioner Deliberation n.590, 03 Aug 2016). This secondary analysis was conducted under SPES data-governance procedures [19–20]. The parent SPES programme enrolled 4,205 participants. Recruitment, clinical examinations, and biospecimen collection were conducted between November 2016 and August 2017 [17,20]. For this manuscript, 17 participants recruited in a municipality outside the study inclusion area were excluded, yielding an analytic dataset of n = 4188 participants assigned to 21 municipal clusters [20]. Clusters 20–21 correspond to the Irno Valley area around the Fonderie Pisano foundry; in the dataset used here, cluster assignment is based on municipality of residence (cluster 20: Baronissi and Pellezzano; cluster 21: Salerno). The planning-stage protocol describes 15 territorial domains; the implemented SPES dataset and technical report define the 21 clusters used here. The analysis sample size was fixed by the parent programme; no separate power calculation was undertaken for this reanalysis. Participant inclusion is summarized in Web Appendix Figure S1. Fifty-one participants provided biospecimens but did not complete questionnaire/diet modules (“non-responders”). These records were retained; multiple-imputation model fitting was restricted to responders (n = 4137), with non-responders carried in prediction-only mode (Web Appendix Figure S2). Participants were not excluded for missing covariate data (Figures 1 and 2). Campania (Italy) study municipalities stratified by environmental-pressure impact area (high, medium, and low impact) used for SPES geostratification. n indicates the number of municipalities in each stratum; other Campania municipalities are shown in grey. Administrative boundaries: ISTAT “Limiti amministrativi correnti” (1 January 2025), CC BY 4.0. Geometric mean ratios (GMRs) for selected serum trace metals by impact area (high, medium, and low impact) relative to the overall cohort mean, from adjusted AREA-route models (adjusted for age group and sex). Points show GMRs and horizontal bars show 95% confidence intervals. An exploratory point-source subanalysis evaluated gradients in serum metal concentrations with residential distance and direction among participants with geocoded residential coordinates within a 10 km buffer around the Pisano foundry centroid (WGS84: 40.7097° N, 14.7785°E). Of 398 addresses in the buffer, 360 participants were retained for modeling after geocoding and availability checks (Web Appendix Figure S3). This buffer was used only for the exploratory distance and direction models and did not affect municipality-based cluster assignment in the main analyses. Data sources, variables, and laboratory procedures Questionnaire/clinical data and laboratory results were captured in the SPES web-based data-management platform and linked via a unique participant identifier [19–20]. A de-identified analytical extract linking questionnaire and clinical modules and laboratory results was provided with documentation of variable definitions and conventions; the extract was frozen before analysis. Dataset construction and derivations were fully scripted to provide an audit trail and did not overwrite original laboratory concentrations (Web Appendix Supplementary Methods Sections 1.1, 1.3–1.4, and 1.9; Tables S2, S17, S22, and S27–S28). Primary geographic predictors were impact-area stratum (high, medium, and low environmental pressure) and municipal-cluster membership (21 clusters) [17,18,20]. Municipality identifiers were retained for descriptive mapping. Primary endpoints were serum concentrations of 20 trace-metal analyte channels measured by ICP-MS; endpoints are referred to by element name in the main text (Li, Be, Cd, As, Cr, Co, Cu, Fe [two channels], Hg, Mn, Mo, Ni, Pb, Sb, Se, Sr, Tl, V, Zn). Concentrations were recorded in ug/L, except Fe and Zn (mg/L), as provided in the analytical extract. Isotopic channel labels, measurement units, distributional family assignments, working scales, and left-censoring imputation models are listed in Web Appendix Table S17. Biospecimens were collected and processed using standardized SPES procedures [17,19–20]. In brief, blood was collected and serum and plasma were separated by refrigerated centrifugation; serum was aliquoted, stored at −80 degrees C, and tracked in the IZSM biobank under continuous temperature monitoring. Before analysis, samples were thawed at room temperature, mixed, diluted 1:10 in nitric acid, and analyzed in duplicate by ICP-MS (NexION 350, PerkinElmer) using 89Y, 103Rh, and 175Lu as internal standards; the mean of duplicate measurements was used for statistical analyses. Calibration curves met R2 > 0.99, and quality control included internal controls and certified reference materials, following SPES documentation [20]. Detection-limit reporting in this secondary analysis is based on the analyte-specific LOQ fields contained in the frozen analytical extract; instrument-level limits of detection were not available in the archived dataset. For each metal, the extract includes indicators for assay performance, whether values were below the analyte-specific LOQ, the LOQ value, and prespecified QC flags. Metal-specific assay nonperformance and < LOQ counts/proportions are reported in Web Appendix Tables S7 and S20. Values < LOQ were retained and treated as left-censored [25–26] at the LOQ for censoring-aware imputation, rather than being replaced by zero or LOQ/2. Using standard analytical-method terminology [27,28], the available documentation supports single-laboratory QA/QC reporting but does not provide an inter-laboratory comparison or proficiency-testing validation report for the serum trace-metals module. Because metal distributions are typically right-skewed, descriptive summaries use the original concentration scale, whereas inferential models use natural-log concentrations; standardized z-score log concentrations were also analyzed for cross-metal comparability. Primary downstream models were purposefully parsimonious and included age group and sex as individual-level covariates, together with the relevant geographic factor in each model route. A broader prespecified set of socioeconomic, lifestyle, occupational, anthropometric, residential-history, and dietary variables was available and was used for descriptive summaries and as auxiliary information in the imputation stage (Web Appendix Table S2), but these variables were not included in the primary geographic contrast models. Main statistical analysis Participant characteristics are summarized using counts (%) and mean (SD) or median (IQR). For metals, we report geometric means with 95% confidence intervals (computed on the log scale), alongside medians and IQRs. Sampling weights and post-stratification (raking) weights from the parent programme were not available in the de-identified analytical extract; therefore, descriptive summaries are unweighted and describe the analytic sample. To reflect the sampling framework in inferential analyses, we report results by impact-area stratum and include either impact-area stratum (AREA route) or municipal-cluster indicators (CLUSTER route) as design variables in regression model [19–20]. Potential sources of bias include differential participation across strata, questionnaire nonresponse, and selection into the geocoded sub-cohort. Detailed counts of individuals invited or approached and reasons for nonparticipation were not available in the de-identified extract; we therefore report the available module-completion counts (Web Appendix Figure S1) and geocoding workflow (Web Appendix Figure S3) and apply regression adjustment and multiple imputation under the prespecified missing-data framework. Deterministic preprocessing preserved questionnaire skip logic (explicit “Unknown” and, where applicable, “Not applicable” levels), parsed numeric fields, and enforced coherent handling of < LOQ values; the corresponding recoding rules, HI20 plausibility thresholds/counts, and metal measurement/censoring summaries are documented in Web Appendix Tables S7, S19, S20, and S22. Primary inferential analyses quantified differences in serum metal concentrations across impact-area strata and municipal clusters, adjusting for age group and sex. These models were designed to estimate interpretable spatial contrasts rather than fully adjusted source-specific causal effects; therefore, additional determinants of internal metal dose, such as diet, smoking, occupation, and socioeconomic context, were not included in the main contrast layer. Because clusters are nested within strata by design, we fit two parallel model routes rather than including both factors simultaneously: Adjusted geometric means and geometric mean ratios were obtained from log-linear models using estimated marginal means. For each categorical factor, each level was contrasted against the model-based overall mean (computed with proportional weights reflecting the observed sample distribution). Estimates were pooled across imputations using Rubin’s rules [29]. False discovery rate was controlled within each modeling route using the Benjamini-Hochberg procedure (target FDR 5%); both nominal p values and FDR-adjusted q values are reported [30]. For the point-source sub-analysis, eligible addresses were geocoded (automated forward geocoding [31] with confidence-based acceptance followed by manual verification when needed; Web Appendix Figure S3). Distance to the foundry centroid and initial bearing were computed on WGS84; bearings were discretized into eight 45-degree sectors (N as reference), and distance was scaled per kilometer. Within each imputed dataset, we fit metal-specific linear regressions with log concentration as outcome and age group, sex, distance, and direction sector as covariates; estimates were pooled across imputations. The joint direction effect was assessed with a multivariate Wald test for pooled MI estimates. Multiplicity across the 20 metals in this exploratory analysis was addressed using Benjamini-Hochberg FDR. Missing data and software Missingness was classified by mechanism: left-censoring (1.25-fold to ≤1.50-fold, and large shift >1.50-fold. Agreement summaries were calculated across all geographic contrasts and, separately, among geographic contrasts significant in the matched primary-analysis rebuild, to distinguish stability of the primary geographic signals from variation in null or near-null contrasts. Detailed implementation rules are provided in Web Appendix Supplementary Methods Section 1.10; metal-specific matched-set denominators and quality-control checks are provided in Web Appendix Tables S29–S30. Route-level agreement and contrast-level comparison outputs are reported in Web Appendix Tables S31–S33. Results Of 4205 participants enrolled in SPES, 17 recruited in Battipaglia were outside the study inclusion criteria and were excluded, leaving 4,188 participants in the analytic dataset (2395 [57.2%] in the high-impact area, 1,196 [28.6%] in the medium-impact area, and 597 [14.3%] in the low-impact area). Mean age was 34.9 years (SD 8.8) and 2,258 (53.9%) participants were women; age-group distribution was <30 years: 1,389 (33.2%), 30–39 years: 1,293 (30.9%), and ≥ 40 years: 1,506 (36.0%). Questionnaire and dietary data were available for 4,137 (98.8%) responders (Table 1). Stage-specific reasons for nonparticipation at recruitment stages were not available in the de-identified analytical extract. Table 1.Baseline characteristics of the study population, overall and by area.CharacteristicOverall (n = 4188)High-impact area (n = 2395)Medium-impact area (n = 1196)Low-impact area (n = 597)Participants, n418823951196597Age, years34.9 (8.8)34.6 (8.8)35.5 (8.8)35.1 (8.9)Age group, n (%) <301389 (33.2%)824 (34.4%)371 (31.0%)194 (32.5%)30–391293 (30.9%)753 (31.4%)350 (29.3%)190 (31.8%)≥401506 (36.0%)818 (34.2%)475 (39.7%)213 (35.7%)Female sex, n (%)2258 (53.9%)1299 (54.2%)638 (53.3%)321 (53.8%)Socioeconomic deprivation index (0–100)33.3 [8.3–37.5]33.3 [8.3–37.5]29.2 [8.3–39.6]33.3 [8.3–37.5]Body mass index, kg/m²25.4 (4.6)25.3 (4.5)25.4 (4.6)25.5 (4.6)BMI category, n (%) Underweight (<18.5 kg/m²)72 (1.7%)36 (1.5%)21 (1.8%)15 (2.5%)Normal weight (18.5–24.9)2198 (52.5%)1281 (53.5%)631 (52.8%)286 (47.9%)Overweight (25.0–29.9)1333 (31.8%)747 (31.2%)376 (31.5%)210 (35.2%)Obesity (≥30.0)584 (13.9%)331 (13.8%)167 (14.0%)86 (14.4%)Occupational exposure (yes), n (%)826 (19.7%)447 (18.7%)214 (17.9%)165 (27.6%)Current cigarette smoker, n (%)1276 (30.8%)733 (30.6%)357 (31.1%)186 (31.2%)Alcohol intake, g/day1.25 [0.15–7.07]1.25 [0.15–6.32]1.83 [0.15–7.25]1.25 [0.15–11.97]Total physical activity (met_total_week), MET-h/week36.0 [15.8–66.8]35.2 [14.8–64.0]36.2 [16.8–64.5]38.2 [15.8–80.2]Total energy intake, kcal/day1891 [1462–2427]1875 [1461–2423]1841 [1440–2313]2075 [1541–2642]Rice, g/day4.8 [2.0–10.0]4.9 [2.0–10.7]4.8 [2.4–9.4]4.5 [1.9–8.7]Fish, g/day35.7 [19.9–57.1]37.9 [21.8–60.6]33.7 [20.2–52.2]28.5 [14.2–53.1]Crustaceans & molluscs, g/day11.2 [3.7–16.1]11.2 [3.8–16.1]10.8 [3.7–16.1]9.2 [3.5–16.1]Leafy vegetables (raw + cooked), g/day18.0 [9.1–32.9]18.8 [9.4–34.1]16.5 [8.1–29.0]18.1 [9.6–35.4]Fruit (citrus + other), g/day155.5 [76.4–250.2]154.3 [77.0–252.4]153.4 [72.3–239.3]163.0 [81.2–272.2]Processed meat, g/day30.6 [18.8–47.1]30.6 [18.3–46.9]30.6 [19.8–46.6]31.5 [19.3–48.6]Values are n (%) for categorical variables, mean (SD) for age and BMI, and median [IQR] for all other continuous variables. Percentages are calculated among participants with non-missing data for that variable. BMI categories follow WHO cutoffs (<18.5 underweight; 18.5–24.9 normal weight; 25.0–29.9 overweight; ≥30.0 obesity). Socioeconomic deprivation index and physical activity (met_total_week) were available for n = 4,137 overall (high: 2,393; medium: 1,147; low: 597). Dietary variables and alcohol intake were also available for n = 4,137 overall (high: 2,393; medium: 1,147; low: 597). The deprivation index ranges from 0 (least deprived) to 100 (most deprived). met_total_week denotes the total weekly metabolic equivalent (MET) hours per week (MET-h/week) derived from the SPES physical activity questionnaire. Occupational exposure indicates self-reported employment in occupations with potential exposure to chemical hazards, coded as described in Web Appendix 1. Baseline characteristics of the study population, overall and by area. Values are n (%) for categorical variables, mean (SD) for age and BMI, and median [IQR] for all other continuous variables. Percentages are calculated among participants with non-missing data for that variable. BMI categories follow WHO cutoffs (<18.5 underweight; 18.5–24.9 normal weight; 25.0–29.9 overweight; ≥30.0 obesity). Socioeconomic deprivation index and physical activity (met_total_week) were available for n = 4,137 overall (high: 2,393; medium: 1,147; low: 597). Dietary variables and alcohol intake were also available for n = 4,137 overall (high: 2,393; medium: 1,147; low: 597). The deprivation index ranges from 0 (least deprived) to 100 (most deprived). met_total_week denotes the total weekly metabolic equivalent (MET) hours per week (MET-h/week) derived from the SPES physical activity questionnaire. Occupational exposure indicates self-reported employment in occupations with potential exposure to chemical hazards, coded as described in Web Appendix 1. Data completeness and missingness Assay nonperformance was ≤5% for 18 of 20 analyte channels but substantial for mercury (785 [18.8%] unperformed assays) and nickel (1,861 [44.4%] unperformed assays); analyses for these metals used the measured subsets (mercury n = 3403; nickel n = 2327; Web Appendix Table S7). Among performed assays, left-censoring at the LOQ ranged from 0.02% (copper, selenium, strontium, and zinc) to 71.0% (lead), with high censoring also for beryllium (69.2%), cadmium (44.7%), chromium (41.1%), and antimony (36.2%) (Web Appendix Table S20). Because mercury and nickel were analyzed on measured subsets, estimates for these metals apply to participants with performed assays; differential assay nonperformance by batch or calendar time cannot be excluded. Multiple-imputation diagnostics are summarized in Web Appendix Tables S23–S28 and Figures S4–S5. These measurement features have different implications for inference. High proportions of < LOQ results mainly reflect left-censoring, as in Pb-208, Be-9, Cd-111, Cr-52, and Sb-121, whereas Hg-202 and Ni-60 were chiefly affected by assay nonperformance and were therefore analyzed on measured subsets. For highly censored analytes, estimated geometric means and GMRs are necessarily more dependent on the assumed censored-data distribution and should be interpreted with greater caution, particularly when contrasts are small, borderline, or isolated. Overall metal distributions Metal concentrations were right-skewed and varied over several orders of magnitude (Web Appendix Table S8). Units vary by analyte (iron and zinc in mg/L; all other analytes in ug/L; Web Appendix Table S17). As exemplars (all in ug/L), overall geometric means were 0.0294 for lead (71.0%