Online news monitoring as a complementary sentinel tool for occupational heat-related adverse events: a six-year descriptive study in Italy (2020–2025)

Climate change is increasing the frequency and intensity of heatwaves, with significant impacts on occupational health. Work-related heat illnesses and injuries are often underestimated by traditional surveillance systems. This study aimed to assess the use of national and local daily newspapers as a sentinel surveillance tool for monitoring occupational adverse events associated with heat stress in Italy. Online news articles published between May and September from 2020 to 2025 were analyzed and selected according to the PECO framework. Events involving heat-related illnesses and injuries potentially associated with high temperatures in occupational settings in Italy were included. Demographic and occupational information was collected for each event. Climate exposure was assessed using the Universal Thermal Climate Index (UTCI) based on the ERA5 Human Thermal Comfort (ERA5-HEAT) dataset. A total of 120 media-identified events were included; annual counts ranged from 4 in 2020 to 33 in 2025. Overall, 50.8% occurred in July. Most workers were male (87.5%; average age 50 years). The sectors most frequently represented were construction (35.0%) and agriculture (20.0%), predominantly involving outdoor activities (79.2%). Maximum UTCI values, when available, indicated strong or very strong heat stress in 98.1% of media-identified cases. Fatal outcomes accounted for 55.0% of events. In univariable analysis, advancing age was associated with an increased risk of fatal outcome (OR: 1.05; 95% CI: 1.01–1.09). The association between maximum UTCI category and outcome was not statistically significant. Systematic analysis of online news media may provide a complementary sentinel approach for detecting occupational events potentially associated with heat exposure.

In the context of climate change, exposure to extreme temperatures is emerging as one of the major public health challenges (1). In recent years, the frequency, intensity, and duration of heatwaves have increased significantly (2). Heat is now recognized as one of the leading environmental risk factors for health, with substantial effects on both the general population and exposed workers, contributing to a significant increase in mortality and morbidity (3, 4). In urban areas, the impact of heatwaves is further amplified by the urban heat island effect and air pollution, resulting in excess preventable deaths (5, 6).

Heat stress, by compromising thermoregulatory mechanisms, can cause the onset of various heat-related illnesses, ranging from mild forms to potentially fatal conditions such as heatstroke (7). Heat stress also contributes to acute events such as stroke and myocardial infarction, as well as to the worsening of several chronic conditions, particularly cardiovascular, respiratory, and renal diseases (8). Exposure to high temperatures is associated with increased healthcare demand, including higher rates of emergency department visits and hospital admissions (9).

Thermal stress in workplaces poses a risk to workers’ health and safety, with potential consequences for productivity (4). The risk of occupational heat-related illnesses can be influenced by many factors, including environmental conditions, physical activity, individual characteristics of the worker, and the personal protective equipment (PPE) worn (5). An international systematic review reported that the risk of workplace accidents rises by approximately 1% for each 1 °C increase in ambient temperature and by 17% during heatwaves (10). Heat waves were defined as periods of at least six consecutive days with maximum temperatures above the 90th percentile of the reference climatological values (2). Exposure to high temperatures at work is associated with an increased risk of injuries, reduced cognitive and physical performance, and the onset or exacerbation of various diseases (11–13), particularly renal disorders. For example, heat stress has been hypothesized to contribute to the development of Chronic Kidney Disease of undetermined cause (CKDu), a form of chronic kidney disease observed in young workers engaged in physically demanding tasks, likely of multifactorial origin (14).

This study aims to describe occupational heat-related adverse events identified through systematic monitoring of online Italian news sources over a six-year period (2020–2025), and to explore the potential of press-based surveillance as a sentinel approach complementary to traditional surveillance systems. The Universal Thermal Climate Index (UTCI) was used to characterize thermal conditions on the days and locations of reported events. In Italy, occupational heat-related events may be identified through several institutional sources, including the Italian Workers’ Compensation Authority (INAIL) occupational injury and disease archive, hospital discharge records, and cause-of-death data. However, these sources do not systematically identify heat as a causal or contributing factor, particularly in the case of injuries indirectly associated with heat exposure. Moreover, insurance-based systems do not cover all categories of workers, while informal employment is not represented. Institutional data may also become available only after a considerable delay. Systematic monitoring of online news sources may therefore provide a complementary and more timely sentinel approach, capable of identifying events and contextual information that may not promptly emerge from traditional surveillance systems.

This research is part of the Worklimate and Worklimate 2.0 projects, funded by INAIL, aimed at improving knowledge on the impact of extreme temperatures in occupational settings and at developing preventive tools and strategies (15). A preliminary analysis was published in 2023 by the same research group; the extension of the observation period allowed the inclusion of additional events and the updating of results (16).

Data collection

A descriptive analysis of articles published online by Italian national and local daily newspapers, between 2020 and 2025, during the period May–September, was conducted. Newspapers were selected based on circulation data certified by ADS (Accertamenti Diffusione Stampa S.r.l.), the company responsible for monitoring press circulation in Italy (17). To gather, verify, and consolidate all relevant information for each worker involved, multiple articles from different newspapers referring to the same event were reviewed through additional manual Google searches using targeted keywords. Because multiple articles could refer to the same occupational event, the event rather than the individual news article was considered the unit of analysis.

Data collection was carried out from 2020 to 2024 using VALIRIA, a web-based application developed by the Joint Laboratory of Technological Solutions for Clinical Pharmacology, Pharmacovigilance and Bioinformatics at the University of Florence. The agreement under which the application had been provided included technical maintenance until the end of 2024. After the maintenance period expired, VALIRIA was no longer maintained and could not be used reliably for the 2025 monitoring period. Google Alerts was therefore adopted in 2025 as an operational alternative. Both tools allowed predefined queries to be run through the Google search engine and results to be delivered automatically to researchers. The same conceptual search strategy was retained, and periodic manual searches using targeted keywords were performed to improve completeness.

The search strategy was structured according to the PECO (Population—Exposure—Comparator—Outcome) model to identify heat-related adverse events potentially associated with occupational heat stress. Media-identified events occurring in Italy among workers exposed to high temperatures in outdoor or non–air-conditioned indoor environments were included. Reported cases with insufficient information regarding the event or its possible link to heat exposure were excluded. The following search strategy was used: [(“cambiamento climatico” OR “caldo killer” OR “caldo torrido” OR temperatur* OR “riscaldamento globale” OR afa OR caldo OR siccità) AND (lavoro OR lavorator* OR “sito di costruzione” OR “bracciante agricolo” OR agricoltore OR azienda OR trattore OR agricoltura OR giardinaggio) AND (“colpo di calore” OR incidente OR infortunio OR “malattia improvvisa” OR morto OR morire OR caduta)], which can be translated as: [(“climate change” OR “killer heat” OR “scorching heat” OR temperature* OR “global warming” OR sultriness OR hot OR drought) AND (work OR worker* OR “construction site” OR “day laborer” OR farmer OR company OR tractor OR farming OR garden) AND (“heat stroke” OR accident OR injury OR “sudden illness” OR dead OR die OR fall)].

For each media-identified event, when available, information was collected on sex, age, nationality, occupational sector and job activity (coded according to Classification of Economic Activity ATECO 2025) (18), outcome (fatal/non-fatal) and date and location of the event. At the macro-area level, regions were grouped as follows: Northern Italy included Piemonte, Valle d’Aosta, Liguria, Lombardia, Trentino-Alto Adige, Veneto, Friuli-Venezia Giulia, and Emilia-Romagna; Central Italy included Toscana, Umbria, Marche, and Lazio; Southern Italy and the Islands included Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicilia, and Sardegna.

During full-text review, we also assessed whether the articles provided information on the worker's degree of acclimatization, previous occupational experience, and heat-related safety training. Because these data were not reported consistently, they were not included as analytical variables. Events were eligible when the news report referred to heat, high temperatures, hot weather conditions, or a heat-related condition as part of the event context or as a possible contributing factor. UTCI values were assigned only after event inclusion and were not used as an eligibility criterion or as confirmation of a causal association.

Automated alerts were reviewed on a daily basis and complementary manual searches were conducted for each identified event, in order to verify the data and ensure that information relating to each individual worker involved was as complete as possible. Screening was performed by a team of eight investigators (GI, EM, AB, LF, MBr, CC, AF, and ML). Reports were initially assessed at article level and subsequently consolidated at event level. Exact duplicate articles and syndicated reproductions were removed. Articles were considered to describe the same event when they referred to the same date, location, worker or group of workers, occupational activity, and event circumstances. When multiple articles described the same event, the information was merged into a single event record.

Assessment of heat stress

For events with reported date and location, the daily average, minimum, and maximum values of the UTCI were calculated. The ERA5-HEAT (Human thErmAl comforT) dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) within the Copernicus Climate Change Service (C3S) program was used to compute UTCI values (19). ERA5-HEAT provides hourly UTCI estimates on a regular 0.25° × 0.25° latitude–longitude grid, corresponding to a grid spacing of approximately 28 km (and finer in the west–east direction at Italian latitudes). For each media-identified event, the geographic coordinates of the centroid of the reported municipality were used to identify the ERA5-HEAT grid cell containing that point, and the corresponding daily mean, minimum, and maximum UTCI values were extracted programmatically for the reported event date through the Copernicus Climate Data Store Application Program Interface (CDS-API) in Python. Given this spatial resolution, small neighbouring municipalities may fall within the same grid cell and be assigned identical UTCI values; moreover, each grid-cell value represents an area-averaged reanalysis quantity rather than a point measurement, whose representativeness decreases with increasing terrain complexity (20).

UTCI is one of the most comprehensive methods for assessing human thermo-physiological comfort under climatic conditions (21). It expresses an equivalent temperature (°C) that summarizes complex thermal conditions determined by air temperature (°C), thermal radiation (W m −2), relative humidity (%), and wind speed (m s −1), assuming a moderate metabolic rate (135 W m −2, equivalent to walking at 4 km/h on level ground) and typical urban clothing (22). It is derived from an advanced multi-node thermophysiological model coupled with a clothing model, in which the reference individual's clothing adapts to air temperature according to the typical behaviour of the general population. This operational implementation, however, does not allow the metabolic rate, the physical workload, or the clothing and PPE to be personalized (21). The articles did not provide sufficiently detailed and consistent information on task-specific metabolic workload or clothing and personal protective equipment to allow these factors to be analysed.

Based on criteria derived from simulated physiological responses, UTCI values are classified into ten thermal stress categories, divided into five cold stress and five heat stress levels (21). In this study, we refer exclusively to the heat stress categories identified in our records: “slight cold stress” (0.0 °C < UTCI ≤ 9.0 °C); “no thermal stress” (9.0 °C < UTCI ≤ 26.0 °C); “moderate heat stress” (26.0 °C < UTCI ≤ 32.0 °C); “strong heat stress” (32.0 °C < UTCI ≤ 38.0 °C); “very strong heat stress” (38.0 °C < UTCI ≤ 46.0 °C); “extreme heat stress” (UTCI > 46.0 °C).

Statistical analysis

Analyses were conducted using IBM SPSS Statistics version 30.0.

Descriptive analyses were performed for qualitative and quantitative variables. Group comparisons were conducted using ANOVA (with Tukey post hoc test). Before conducting statistical analyses, normality and homogeneity of variances were assessed, respectively, using Shapiro–Wilk tests (P > 0.05) and Levene's tests (P > 0.05). Associations between qualitative variables were assessed using the chi-square test.

Exploratory inferential analyses were performed to assess associations with fatal outcome within the selected sample of media-identified events. Univariable binary logistic regression was used to examine age, expressed per one-year increase, and sex, with female sex as the reference category. Because the assumptions of Pearson's chi-square test were not met for the 3 × 2 table comparing maximum UTCI category with fatal/non-fatal outcome, the two-sided Fisher–Freeman–Halton exact test was applied. Given the limited sample size and missing information, all inferential analyses were considered exploratory. A p value ≤0.05 was considered statistically significant.

From May to September over the 2020–2025 period, 120 events potentially associated with occupational heat exposure were identified through online national and local newspaper articles and reported in the annual project reports (15), available for free consultation.

The identified events were treated as media-identified occupational events potentially associated with heat exposure. The available news reports did not allow confirmation of a causal role of heat or a consistent distinction between heat illness and injury. It was not possible to verify all possible related risk factors, such as the workers’ level of acclimatization, based on news sources. Details collected for each event are reported in the Supplementary Material.

Demographic characteristics of workers

Among the 120 workers identified, 87.5% (n = 105) were male. The average age was 50 years (range: 17–81; SD: 12.7). More than half of the events (n = 66; 55.0%) involved individuals aged 40–70 years. In 19 cases (15.8%), workers were younger than 40 years; in particular, 5 cases (4.2%) were under 30 and 3 cases (2.5%) older than 70 years. In 26.7% of cases (n = 32), age was not reported in press articles. Nationality was reported as Italian in 34.2% (n = 41) of cases and foreign in 24.2% (n = 29); in the remaining 41.7% (n = 50), nationality was not specified (Table 1). Information was missing for age in 32 events (26.7%), nationality in 50 events (41.7%), and occupational sector in 1 event (0.8%).

Temporal and geographical distribution of events

The annual number of media-identified events varied during the study period, from 4 reports in 2020 to 33 in 2025, with the highest number observed in 2025 (Figure 1). These descriptive counts should not be interpreted as an epidemiological time trend because the monitoring platform changed in 2025 and media attention may have varied over time.

The monthly distribution shows a marked concentration in July across all years considered (n = 61, 50.8%). Over the 6 years, only two media-identified events were reported in May (in 2022 and 2023) and none in September, while June and August showed substantially lower numbers compared with July (31.7% and 15.8%, respectively).

Events identified through media reports were reported in almost all Italian regions, except for Valle d’Aosta, Molise, Basilicata, and Sardegna. Lombardia recorded the highest number of cases (n = 19, 15.8%) (Figure 2). At the macro-area level, 50.0% (n = 60) of events occurred in Northern Italy (Piemonte, Liguria, Lombardia, Trentino-Alto Adige, Veneto, Friuli-Venezia Giulia Emilia-Romagna), 22.5% (n = 27) in Central Italy (Toscana, Umbria, Marche, Lazio), and 27.5% (n = 33) in Southern Italy (Campania, Abruzzo, Puglia, Calabria, Sicilia).

Occupational sectors

When reported, workers were employed in eight occupational sectors (Figure 3).

The construction sector was the most represented (n = 42; 35.0%), followed by agriculture (n = 24; 20.0%). In construction, 31.0% (n = 13) of media-identified events involved workers performing tasks at height (roofs or scaffolding), and more than half of the events (n = 24, 57.1%) occurred in Northern Italy.

In agriculture, most cases were recorded in Southern Italy (n = 13, 54.2%); in this sector, nearly half of the workers were of foreign origin (n = 11, 45.8%) and 50.0% (n = 12) were day laborers.

Overall, 79.2% (n = 95) of all media-identified events occurred in outdoor settings; 15 workers (12.5%) were driving a vehicle at the time of the event. No women were employed in the agricultural or construction sectors.

Thermal stress conditions (UTCI)

For 103 media-identified events, the biometeorological index UTCI (daily average, minimum, and maximum values) was calculated (Table 2).

Differences between groups were assessed using one-way ANOVA to compare UTCI values. Associations between categorical variables were analyzed using the chi-square test, complemented by Cramer's V to measure the strength of association. Significant differences emerged between years for daily maximum UTCI, with higher values in 2021 compared with the following three-year period (p ≤ 0.05) (Figure 4).

Based on monthly values, July and August showed higher daily averages, minimum and maximum UTCI values than May and June; ANOVA revealed significant differences for daily average UTCI (p < 0.001) and daily minimum UTCI values (p = 0.005). The highest average values were recorded in Southern Italy, with significant differences between macro-areas observed only for daily minimum UTCI (p = 0.045).

According to the thermal stress classification, no media-identified event was associated with “extreme heat stress.” Daily average UTCI most frequently fell within the “moderate heat stress” category (n = 78, 75.7%), while daily maximum UTCI was more often compatible with “strong heat stress” (n = 45, 43.7%) or “very strong heat stress” (n = 56, 54.4%). Minimum UTCI generally did not indicate heat stress conditions (n = 96, 93.2%). For 7 events, observed mainly in the mountainous areas of northern Italy, the index showed values indicative of “slight cold stress” (Figure 5).

Fatal events

Fatal outcomes were reported in 55.0% of the media-identified events (n = 66).

Exploratory analyses were conducted to describe factors potentially associated with fatal outcomes. Aong media-identified events with known age, fatal events involved older workers on average than non-fatal events (52.3 vs. 44.2 years; p = 0.011). In univariable logistic regression, each additional year of age was associated with higher odds of fatal outcome among the reported events (OR = 1.05; 95% CI: 1.01–1.09; p = 0.011; Table 3). After stratifying workers into four age groups (<40, 40–55, 56–70, >70 years), chi-square analysis showed a statistically significant difference (p = 0.048), with a higher frequency of fatal events observed among workers in the two oldest age groups. Male sex was associated with higher odds of fatal outcome compared with female sex in univariable logistic regression (OR = 6.00; 95% CI: 1.60–22.55; p = 0.008; Table 4).

Maximum UTCI values were higher among fatal events, and fatal outcomes were more frequent in the highest thermal stress categories. Among media-identified events with available maximum UTCI data, fatal outcomes occurred in 1 of 2 events classified as moderate heat stress, 23 of 45 events classified as strong heat stress, and 39 of 56 events classified as very strong heat stress. Although the proportion of fatal outcomes was descriptively higher in the very strong heat stress category, the overall association between maximum UTCI category and outcome was not statistically significant according to the two-sided Fisher–Freeman–Halton exact test (p = 0.101).

No significant differences emerged by year or month, although July accounted for the highest number of deaths.

The agricultural sector showed a higher frequency of fatal events compared with other sectors in this media-identified sample (p ≤ 0.05; Cramer's V = 0.33).

During the 2020–2025 period, surveillance based on the analysis of online press sources enabled the identification of occupational heat-related illnesses and injuries potentially associated with thermal stress across Italy. This approach shares conceptual similarities with epidemic intelligence frameworks widely used in infectious disease surveillance. The most prominent current implementation is the World Health Organization (WHO) Epidemic Intelligence from Open Sources (EIOS) initiative, established in 2017 and now used by more than 110 WHO Member States, which applies automated scanning of open digital sources—including news media, social networks and informal reports—to detect public health threats in near real time, often before formal notification systems capture them (23, 24). Analogously, press-based monitoring of occupational events potentially related to heat exposure can function as a weak-signal detection tool, complementary to institutional data sources, and particularly valuable for capturing events in sectors characterised by informal labour arrangements, irregular employment, and limited occupational health coverage.

Heat-related adverse events and associated risk factors

The analysis showed a higher frequency of media-identified events in Northern regions of Italy and a marked concentration in July, when higher average thermal stress values were recorded. Events identified from newspaper reports occurred mainly on days characterized by moderate to strong heat stress conditions. The absence of extreme levels suggests that risk does not depend only on the intensity of environmental exposure, but also on organizational and individual determinants, such as physical workload, duration of exposure, clothing, and work practices (25). Under conditions of mild or moderate thermal stress, workers may continue activities during the hottest hours of the day, thereby increasing exposure time and the likelihood of adverse effects (26). The greater number of cases identified in northern regions, despite generally less extreme temperatures than in parts of southern Italy, suggests that the geographic distribution observed in media-based surveillance may be influenced not only by differences in exposure, but also by differential visibility, reporting practices, and the probability that occupational heat-related events are brought to public attention. For example, no media-reported events were identified in Basilicata, a relatively small and less densely populated region. This absence should not be interpreted as evidence that occupational heat-related events did not occur. Rather, it may illustrate how population size, the availability and reach of local news sources, and differences in media visibility can affect the likelihood that an event is identified through press-based surveillance.

The distribution of the identified events should not be interpreted as evidence that the corresponding groups had a higher population risk. The number of workers employed in each sector or demographic group was not available, and the probability of inclusion depended on occupational exposure, workforce size, employment conditions, recognition of the event, and media visibility. The results therefore describe the composition of media-identified events rather than sector-, sex-, or nationality-specific incidence. Nonetheless, consistent with the literature, most cases involved men employed in the construction and agricultural sectors, often in outdoor settings (27, 28). The overrepresentation of these sectors reflects high exposure to solar radiation, physically demanding tasks, and potential gaps in preventive measures. Moreover, it is essential to deepen the analysis of the construction sector by highlighting that 31.0% of media-identified events (13 out of 42) involved workers engaged in tasks at height, such as activities on roofs or scaffolding. Exposure to high temperatures, besides causing direct heat-related pathologies, is associated with a reduction in cognitive performance and a decline in attention levels (11). In inherently dangerous operational contexts like working at heights, these effects of thermal stress may act as triggers for serious injuries, including those indirectly related to heat (such as falls) which are often not fully captured as such by traditional insurance-based surveillance systems.

Among the identified events, heat-related adverse events were more frequently reported among migrant workers, possibly reflecting a combination of structural and individual factors, including concentration in physically demanding, low-wage, outdoor jobs with limited shade or rest facilities; lack of acclimatization at the start of the season; language and cultural barriers that reduce comprehension of safety instructions; fear of reporting symptoms or requesting breaks due to precarious employment status; and reduced access to occupational health services and preventive training (29, 30). This context warrants specific targeted preventive measures, including multilingual communication campaigns, culturally adapted training, and enhanced surveillance in sectors with high migrant labour concentration. In the Italian setting, these dynamics are especially relevant in the agricultural sector, where informal labour arrangements and seasonal migrant workers are particularly common.

Numerous studies report a higher risk of developing heat-related illnesses among younger workers (10–12) and those who are not adequately acclimatized (27, 31); however, in our study, only five workers were under the age of thirty, and the newspaper articles analyzed did not report information on the degree of acclimatization, the professional experience of the workers involved, or their level of safety training.

A substantial proportion of workers (12.5%) were driving a vehicle at the time of the event, with potential implications for road safety. Prolonged exposure to high temperatures inside vehicles may increase the likelihood of errors and injuries; an Italian study reported a positive association between work-related road traffic accidents and high temperatures (Relative Risk, RR = 1.06; 95% CI: 1.01–1.11) (32).

In this study, media-reported event outcomes were analyzed by collecting information to identify factors and characteristics associated with fatal cases. Fatal outcomes were reported in 55.0% of the identified events. This proportion describes the selected media-reported sample and is likely to be strongly influenced by the greater likelihood that fatal events receive press coverage. Fatal outcomes were observed more frequently among men, agricultural workers, and individuals over 55 years of age. These patterns are broadly consistent with previous studies (33, 34); however, in the present study they should be interpreted as descriptive associations rather than evidence of independent effects, given the limited sample size, missing information, and the absence of denominators by sector, age, or workforce composition.

Fatal outcomes were more frequent among press-reported events occurring under very strong heat-stress conditions. However, the association between maximum UTCI category and outcome did not reach statistical significance in the exact analysis. Fatal media-identified events occurred more frequently in July and August, which are characterized on average by higher daily maximum UTCI. The results suggest that peaks of thermal stress (daily maximum UTCI) are more relevant to the severity of the event, particularly fatal outcomes, than the average daily exposure to thermal stress.

In univariate analysis, the odds of fatal outcome increased with advancing age by 5% for each additional year (OR: 1.05; 95% CI: 1.01–1.09),Recent systematic evidence based on locally defined extreme heat events found an association with all-cause mortality, with substantial heterogeneity across geographical settings, definitions, and population groups. These findings underscore the importance of adapting heat thresholds and preventive responses to local climatic conditions and population vulnerability rather than relying exclusively on universal temperature cut-offs (35).

Importance of sentinel surveillance systems in occupational health prevention

Current surveillance systems, such as the INAIL archive or hospital discharge forms, do not allow systematic identification of occupational events associated with thermal stress. Heat-related illnesses are not always recognized by workers or physicians, and sometimes not even at autopsy as the primary cause of death, partly due to the lack of standardization in diagnostic definitions (33). Although the death certificate of National Institute of Statistics (ISTAT) provides for “heat waves” to be indicated among the circumstances that caused the trauma, this information does not systematically emerge in practice.

Analysis of the INAIL database (2010–2021) identified 569 cases directly attributable to heat exposure, including 25 fatalities, representing 0.014% of all accidents at work and 0.35% of deaths. However, these figures do not capture events indirectly associated with heat (16). Other studies have estimated that between 2006 and 2010 extreme temperatures account for an attributable fraction of 0.14% of occupational injuries in Italy, particularly among outdoor workers (11). A recent Italian study reported a relative risk of 1.12 (95% CI: 1.09–1.14) for workplace accidents attributable to an increase in occupational heat exposure in the 2014–2019 period (36).

It should be noted that INAIL mandatory insurance coverage does not extend to all workers in Italy: for example, self-employed professionals, riders and merchants are structurally excluded under current legislation (37). Furthermore, according to ISTAT estimates, approximately 3.1 million full-time equivalent work units were employed irregularly in Italy in 2023, representing 12.7% of total employment, and are therefore entirely outside institutional recording systems, including INAIL coverage. Irregular employment is particularly prevalent in the construction (12.8%) and agriculture (17.6%) sectors—the two sectors most represented in our dataset—suggesting that the actual burden of occupational heat-related events in these contexts is likely substantially underestimated by official insurance-based surveillance systems (38). These structural gaps further support the value of complementary surveillance approaches capable of capturing events that formal systems systematically miss, enabling the immediate collection of data on adverse events related to exposure to high temperatures and providing a non-standardized but timely picture of the phenomenon and its associated burden. Integration of insurance databases, health information systems, activities of the Epidemiology services, occupational health surveillance services, and high-resolution meteorological data could enhance the detection of such events and support targeted preventive interventions for vulnerable groups and high-risk periods. For example, workers exposed to risks from severe microclimates and ultraviolet radiation are reported by the company occupational physician to INAIL, in the form of aggregated data (39).

A prospective pilot system could operate through a staged workflow. Automated queries would generate daily alerts during the warm season. Trained occupational or public health personnel would perform initial screening, remove irrelevant records, and merge multiple reports referring to the same event. Potential events would then be classified as probable, possible, or uncertain using standardized criteria. Verification would involve additional searches and, where legally and operationally feasible, consultation of occupational health, emergency care, local health authority, mortality, or insurance sources. Meteorological information would subsequently be linked to the verified date and location. Climate-driven predictive methods may also strengthen early-warning systems. Recent graph-based machine-learning research demonstrated the feasibility of short-term heatwave forecasting in a data-sparse arid environment. Although such models predict meteorological hazards rather than occupational health events, their integration with workforce distribution, occupational exposure, and sentinel surveillance data could be explored as part of future locally validated prevention systems (40). High-priority events, including fatalities, clusters, or events occurring in unexpected sectors or thermal conditions, could be reviewed promptly by occupational prevention services. Periodic record linkage with INAIL, hospital, mortality, and meteorological databases would allow evaluation of completeness, false positives, timeliness, and added value. Appropriate data-governance and privacy procedures would be required before operational implementation.

Preventive measures should rely on an integrated approach combining technical, organizational, and training interventions, promoting acclimatization and health surveillance for more susceptible workers (41). According to a recent European study, the adoption of heat prevention plans, including early warning systems and actions to protect workers’ health, has been associated with a 25% reduction in excess deaths attributable to extreme heat over the last 30 years (42). Since 2021, several municipalities and regions in Italy have issued contingent ordinances to limit or prohibit outdoor work during the hottest hours of summer days (June-August), based on forecasts from the Worklimate platform (43). The measures included restrictions between 12:30 pm and 4:00 pm on days identified as high-risk for sun exposure and intense physical activity, particularly for non-acclimatized workers. Nevertheless, these measures may also have unintended consequences. Where wage-support mechanisms are unavailable or not activated, restrictions may result in reduced or delayed income. Work may also be shifted to very early morning or evening hours or compressed into more intensive schedules. Informal, precarious, and piece-rate workers may receive less effective protection than formally employed workers. Heat-related work restrictions should therefore be accompanied by adequate income protection, advance planning, and enforcement measures designed to avoid transferring occupational risk to other working hours or to the most vulnerable workers. In summer 2024, the ordinances were extended to 15 regions, including much of Southern, Central, and some areas of Northern Italy, affecting 1.7 million workers (44, 45). A recent Italian study demonstrated a decline in injury rates in regions that implemented the ordinances (21.9% decrease in the construction sector), thus highlighting the potential effectiveness of the measures in reducing heat-related injuries (45). However, in this study we observed an increase in reported events over time, a finding that should be interpreted considering both the impact of the COVID-19 pandemic during the early years of observation and growing media attention to climate change and health (1, 4). A wider media coverage over the years could indicate a greater sensitivity of the press towards occupational health. It should also be noted that journalistic coverage tends to emphasize the most severe cases, potentially leading to underestimation of non-fatal and indirectly heat-related events (46). Moreover, in 2025 the highest number of cases was reported, despite no corresponding increase in UTCI values. According to the European Earth observation program Copernicus, summer 2025 was the fourth warmest in the ERA5 record (47). The discrepancy between the rise in reported cases and biometeorological trends may reflect selective media coverage—where heightened public attention can distort the relationship between reported and actual event frequency (48)—as well as differences in scale and measurement, since a local daily biometeorological index is not directly comparable with seasonal climate averages at the European level.

Strengths and limitations

The widespread availability of digital sources and the partial automation of article retrieval make this approach potentially reproducible with limited resources. Automated alerts reduced the time required for initial retrieval, whereas manual verification, deduplication, and collection of additional information required up to approximately 30 min per event. Multi-year monitoring and linkage with a spatially consistent, temporally continuous reanalysis-based thermal comfort dataset allowed ambient thermal conditions to be characterized for events with available date and location. The PECO framework also supported a structured search strategy and the collection of sociodemographic, occupational, and contextual information that is not consistently available in traditional surveillance databases.

Despite this, it was not possible to monitor all newspapers, and other media outlets were excluded. Only freely accessible articles were included. The exclusion of paywalled content may have reduced both the number of detectable events and the availability of detailed information on event circumstances.

Future research could extend this approach beyond online newspapers by integrating social media, trade union reports, emergency service communications, local institutional websites, and other open digital sources. Automated text-mining or natural-language processing methods could assist with screening large volumes of content, although human validation would remain essential to verify events, remove duplicates, and assess the plausibility of the occupational and heat-related association. Such an extension would also require careful consideration of misinformation, privacy, source representativeness, systematic comparison between local and national news sources and geographical differences in digital media use.

A further limitation concerns the change in the monitoring tool used during the study period. News collection was performed through the VALIRIA platform from 2020 to 2024 and through Google Alerts in 2025. Although the same conceptual search strategy was maintained and manual searches were used to improve completeness, differences in retrieval sensitivity, indexing, ranking, and notification algorithms may have influenced the number of events detected. For this reason, the increase observed in 2025 should be interpreted with caution and may partly reflect methodological discontinuity and greater media attention, rather than a true increase in event occurrence. A direct parallel comparison between the two tools in 2025 was not possible as VALIRIA was no longer operationally available.

Furthermore, there is a strong media selection bias, with the most serious and sensational cases receiving the most coverage. The daily reality of the problem is underestimated, and the information is often incomplete and not technical. Collaboration with journalists and media organizations could improve the consistency and usefulness of news reports for sentinel surveillance. Whenever available and compatible with privacy and ethical requirements, reports of suspected occupational heat-related events should specify the date and approximate time of the event, municipality or worksite location, occupational sector and task, indoor or outdoor setting, activity being performed, circumstances and sequence of the event, clinical outcome, whether heat was considered a possible contributing factor by healthcare professionals or authorities, environmental conditions, work-rest arrangements, preventive measures, personal protective equipment, and available information on acclimatization and heat-related safety training. Such a minimum reporting framework could facilitate event verification and comparison without disclosing personal identifiers.

There is also the possibility of potential climate classification errors due to possible inaccuracies in the location and date of the event reported in the newspapers, and the UTCI index does not consider fundamental elements such as the PPE worn, the intensity of physical activity or degree of acclimatization. The UTCI values used in this study characterize modelled ambient outdoor thermal conditions at the reported location and date and should not be interpreted as measurements of individual occupational exposure. The spatial resolution may not capture local workplace microclimates, shade, radiant heat from surfaces or machinery, or indoor conditions. In particular, the ∼28 km resolution of the reanalysis, combined with the use of the municipality centroid as the sampling point, means that small neighbouring municipalities may share the same grid-cell value, and that local conditions may diverge from the area-averaged reanalysis estimate, especially in orographically complex or coastal settings (20). Moreover, the standard UTCI assumptions regarding metabolic rate and clothing may not represent physically demanding construction, agricultural, or industrial tasks or the use of occupation-specific personal protective equipment.

Finally, the statistical analyses were limited by the relatively small number of media-identified events, the absence of appropriate denominators, incomplete information for several variables, and the exploratory nature of the study. For these reasons, regression analyses should be considered hypothesis-generating and useful for describing potential patterns, rather than for estimating causal relationships or developing predictive models of fatal outcome.

Systematic monitoring of online news sources identified media-reported occupational events potentially associated with heat exposure and provided contextual information that is often unavailable or delayed in institutional databases. The method appears particularly suited to detecting severe sentinel events and early signals, but it cannot estimate incidence, overall burden, population risk, or case fatality. Prospective pilot studies are needed to evaluate case definitions, sensitivity, specificity, timeliness, resource requirements, and added value compared with existing surveillance systems before considering formal integration into occupational health surveillance frameworks.