A targeted, systematic evidence review was conducted in accordance with the reporting logic of PRISMA 2020 (Page et al., 2021) to identify peer-reviewed studies relevant to PM2.5, particulate exposure, haze, smoke, biomass burning, respiratory injury, growth performance, hematology, immune response, microbiota, and indoor/outdoor exposure translation in broilers or closely-related avian systems. PubMed, Scopus, and the Web of Science were searched for articles published between January 2016 and March 2026. The search period was selected to comply with the journal’s request for up-to-date references and avoid reliance on outdated exposure-response assumptions. The search terms included combinations of broiler, chicken, poultry, PM2.5, fine particulate matter, particulate matter, haze, smoke, biomass burning, agricultural burning, maize stover burning, transboundary haze, tropical, open-sided housing, naturally-ventilated housing, respiratory pathology, lung injury, haematology, immune response, vaccination, microbiota, infiltration factor, indoor/outdoor ratio, acute exposure, and short-term exposure.

This review was designed to parameterize a predictive model rather than estimate the pooled effect size. A formal meta-analysis was not conducted because the eligible evidence differed substantially in terms of bird genotype, age, management, geography, particle source, PM fraction, co-pollutant mixture, exposure duration, outcome definition, and experimental design. The full-text screening retained direct broiler field evidence, controlled broiler exposure studies, tropical or open-sided housing evidence, and regional PM2.5 source-context studies. Eligibility criteria were applied to the records for full-text assessment, and the final model context set contained 18 peer-reviewed publications. Ten publications were used as primary model anchors and eight were used for context, mechanistic interpretation, or methodological support (Figure 1).

Eligible studies were original peer-reviewed publications in English that reported PM2.5, poultry-house particulate matter, smoke, haze, biomass-burning particulate exposure, or directly-relevant housing/exposure data linked to at least one veterinary endpoint, including growth performance, feed efficiency, hematology, respiratory function, respiratory pathology, microbiota, immune response, and clinically-relevant health indicators. These reviews were only used for mechanistic support or in regional contexts. Conference abstracts, theses, dissertations, grey literature, and non-peer-reviewed reports were excluded. Studies reporting only gaseous pollutants were excluded unless particulate data were available. Only studies with PM2.5 measurements were used quantitatively in the PM2.5 exposure modules. Studies reporting PM10, total suspended particles (TSP), or other proxies were not converted to PM2.5 unless the original publication provided a PM2.5-equivalent fraction or size-specific relationship. When no defensible conversion was available, those studies were retained only as qualitative or mechanistic evidence of the response direction and endpoint plausibility. This rule was intended to avoid unvalidated PM10/TSP-to-PM2.5 conversions and to prevent hidden uncertainty from being transferred into the Monte Carlo model.

For clarity, evidence was classified by the source and modeling role of PM2.5. Poultry-house PM/PM2.5 studies were used to support respiratory, inflammatory, microbiota, metabolic, and production-response directions. Maize stover and agricultural residue burning were used as the closest direct field analogs for smoke-associated hematological and respiratory pathology. Tropical/regional haze studies from Thailand and mainland southeast Asia were used to justify the environmental context, source categories, and transboundary haze relevance, but not to infer direct broiler dose–response coefficients. Open-sided housing studies support the assumption that naturally-ventilated broiler houses have strong indoor/outdoor environmental coupling.

A field-oriented, empirical model was developed to estimate the impact of a 3-d, acute PM2.5 exposure episode on production and health outcomes in open-sided broilers. Daily indoor PM2.5 was first estimated from daily outdoor PM2.5 using a simplified indoor/outdoor ratio approach adapted from established indoor particle infiltration and mass-balance frameworks (Bekierski & Kostyrko, 2021; Wallace & Ott, 2023) and from natural-ventilation studies using paired indoor/outdoor PM2.5 monitoring (Luo et al., 2021). The equation is not proposed as a new mechanistic airflow model; rather, it serves as a field-applied, exposure-translation equation for situations in which outdoor PM2.5 data are available but direct indoor monitoring is absent. The wind direction, distance from the burning source, curtain position, fan operation, and particle deposition were not explicitly simulated, and their effects were represented by uncertainty in the effective indoor/outdoor ratio.

C_in,t = r_IO × C_out,t (1)

where C_in,t is the estimated indoor PM2.5 concentration on day t, C_out,t is the 24-h mean outdoor PM2.5 concentration on day t, and r_IO is the effective indoor/outdoor ratio. In this study, r_IO was interpreted as an aggregate field coefficient that combines outdoor PM2.5 infiltration with potential indoor contribution from litter dust, feed dust, feather debris, bird activity, manure-derived particles, and the resuspension of settled particles. Therefore, r_IO should not be interpreted as a purely physical penetration coefficient.

The r_IO was modeled as r_IO ~ Uniform (0.6, 1.2) to obtain an effective indoor/outdoor ratio. Values of r_IO < 1.0 represent moderate attenuation from dilution, deposition, or short-term shielding within the home; values of r_IO > 1.0 indicate potential amplification of PM2.5 indoors from disturbances caused by litter, birds, feed dust, feather debris, manure-based particles, and resuspension of settled dust. The lower (0.6) and upper bounds (1.2) were selected based on the pragmatic uncertainty range for commercial open-sided poultry housing, which does not have a fully-sealed, mechanical filtration system because of the low sidewall/curtain height. The selected bounds were supported by information indicating that there could be both indoor and outdoor sources of PM2.5 exposure to birds in the home (Wallace & Ott, 2023); that indoor/outdoor ratios and penetration processes are widely used to estimate indoor exposure due to outdoor-generated PM2.5 (Bekierski & Kostyrko, 2021) and that passive, natural ventilation creates a high degree of coupling between outdoor and in-home concentrations of PM2.5 (Luo et al., 2021). Additionally, evidence from poultry-house studies indicates that bird behavior in these environments and how they disturb and resuspend the litter can contribute substantially to the generation of fine particulate matter(Sans et al., 2021); (Wang et al., 2023); (Jongbo, 2024).

The model included endpoint-specific modules for production performance, hematology, exploratory post-vaccination immune retention, general health impairment, and respiratory pathology. Production performance was expressed as a percentage reduction in ADG, percentage increase in FCR, and LW loss per bird during a 3-d episode. Because available broiler PM2.5 studies do not yet provide stable non-linear estimates for 3-d events, production responses were modeled as bounded linear functions above an indoor PM2.5 threshold of 50 µg/m3. The literature search identified 676 records; after title and abstract screening, 458 records were excluded and 218 full-text articles were assessed for eligibility. No full-text articles were excluded. Therefore, 218 studies were retained in the final qualitative synthesis and used as the evidence base for model parameterization.

Hematological responses were expressed as bounded percentage changes in hemoglobin and hematocrit because these were the most transferable physiological markers reported during field exposure to maize stover burning (Mahmoud et al., 2025). The postvaccination module was retained only as an exploratory retention function. Here, immune retention refers to the percentage of baseline post-vaccination immune responses assumed to remain during a PM2.5 episode; it does not predict antibody titer, protective threshold, or vaccine failure. Because no episode-matched field study has quantified short-term PM2.5-related vaccine titer decline in broilers, the βVac coefficient was assigned conservatively based on the mechanistic plausibility regarding PM-induced inflammation, oxidative stress, and immune perturbation. Therefore, the output should be interpreted as a trigger for enhanced serological monitoring, rather than as a diagnostic prediction.

General health impairment and respiratory pathology were modeled using logistic probability functions:

where C_in is the 3-d mean indoor PM2.5. The event definition for clinically-significant general health impairment was defined as a PM2.5-associated response that justified veterinary attention at the flock level, such as measurable growth loss, abnormal hematological shift, respiratory signs, or systemic pathology. The event definition for clinically-significant respiratory pathology was defined as the presence of lung injury reported in the source studies, including airway obstruction, epithelial injury, pulmonary inflammation, congestion, fibrosis, metaplasia, or histopathological lesions. As the source studies did not use a common lesion grading system, the model was not calibrated to a universal lesion grade. Instead, the logistic modules estimated the probability of a clinically-meaningful pathological event, and the outputs should be interpreted as decision support probabilities rather than formal diagnostic criteria.

A Monte Carlo simulation was used to propagate uncertainty through the indoor exposure and response modules. Simulations were conducted in Python 3.11 using NumPy for random sampling and numerical operations, pandas for data handling, SciPy for probability functions and rank-correlation analysis, and Matplotlib for figure production. A reference run of 1,000,000 iterations was compared with smaller candidate runs. The final analysis used 150,000 iterations because this was the smallest run size that maintained the maximum relative error of the scenario summary statistics below 0.5%. Monte Carlo simulation was selected because it permits uncertainty in exposure, infiltration, and biological response coefficients to be carried through to endpoint predictions, rather than being represented by a single deterministic value (Jiang et al., 2021).

Two 3-d scenarios were evaluated. The moderate, regional haze scenario used outdoor PM2.5 values of 150, 150, and 150 µg/m³ as 24-h means for days 1–3.The severe maize stover/biomass-burning scenario used outdoor PM2.5 values of 193, 400, and 611 µg/m³ as 24-h means for days 1–3, reflecting the field range reported during maize stover burning by (Mahmoud et al., 2025). These inputs define the environmental haze intensity; the biological exposure dose was the modeled indoor PM2.5, concentration after applying r_IO. The distance from the burning source was not explicitly included because the model assumes that the outdoor PM2.5, measurement is farm-relevant; therefore, the source distance is implicitly reflected in the measured or assumed outdoor concentration.

A fixed random seed (20260308) was established prior to sampling to ensure reproducibility. This number was a user-defined reproducibility identifier corresponding to the simulation date format and was not generated from the data. Model validation was limited to internal verification because no independent commercial broiler dataset with paired outdoor PM2.5, indoor PM2.5, production records, hematology, vaccine response, or lesion scoring was available. Internal verification included a convergence assessment, range checks against biological anchors, and Spearman’s rank correlation sensitivity analysis to identify the most influential uncertain parameters. External validation is explicitly required before the model can be used for regulatory or definitive diagnostic purposes.

The final model context evidence set consisted of 18 peer-reviewed publications published between 2016 and 2026. The set included poultry-house PM/PM2.5 studies addressing respiratory, inflammatory, microbiota, metabolic, and production endpoints (Shen et al., 2023); (Wang et al., 2023); (Wang et al., 2023); (Zhou et al., 2023); (Zhou et al., 2023); (Zhou et al., 2024); (Chen et al., 2024); (Shen et al., 2025), as well as direct field evidence from maize stover burning in broilers (Mahmoud et al., 2025). Tropical and regional source context studies supported the interpretation of biomass burning, agricultural burning, transboundary haze, PM2.5, chemical composition, and microbial components in Northern Thailand and mainland Southeast Asia (Jainontee et al., 2023); (Suriyawong et al., 2023); (Chantaraprachoom et al., 2024); (Inlaung et al., 2024); (Chansuebsri et al., 2024); (Yabueng et al., 2025). Additional studies support open-sided or naturally-ventilated housing context (Sans et al., 2021); (Jongbo, 2024), PRISMA-style reporting, and probabilistic uncertainty propagation (Jiang et al., 2021); (Page et al., 2021).

Exposure durations in biological studies ranged from controlled 7–14-d poultry-house particle exposure and 21-d PM2.5 exposure to longer-term PM2.5 inoculation or seasonal field exposure. These durations were longer than the 3-d haze episodes modeled in the present study. Consequently, long-duration studies were not converted into direct 3-d coefficients. Instead, they were used to define directionality, affected endpoint domains, and conservative bounds, whereas the 3-d predictions were generated from daily PM2.5 scenario inputs and bounded response functions.

Controlled broiler trials that were conducted to support this approach have examined how ADG, FCR, LW, and multiple systemic physiological modules may be integrated into the overall model of poultry health. Credible studies (Shen et al., 2023) have shown that exposure of poultry to particulates suspended in the air has negative effects on growth performance because of pulmonary damage, and many have found similar results with metabolic disease. (Chen et al., 2024) demonstrated that exposing chickens to long-term levels of PM2.5 lowered their growth performance and changed the sero-metabolomic pathway. Although these studies were helpful in establishing the direction and maximum limit of a response, they were not used to conclude that there would necessarily be the same magnitude of effect based on an exact timeframe of 3 d.

Exposure to PM2.5 consistently supports respiratory injury as a biologically plausible outcome. PM2.5, alters the pulmonary microbiota and increases inflammatory signaling in broilers (Zhou et al., 2023). Microbiota intervention reduced PM2.5-induced lung inflammation, supporting the causal contribution of dysbiosis to respiratory injury (Zhou et al., 2023). Inhaled PM2.5 also caused disrupted gut microbiota, increased gut-derived lipopolysaccharides, and impaired intestinal barrier markers (Zhou et al., 2024). A recent proteomic study identified dysregulated pulmonary proteins and potential injury pathways after exposure to poultry house particulates(Shen et al., 2025).

(Mahmoud et al., 2025) provided the most direct field anchor for clinical translation because broilers were evaluated during the burning of maize stover, a specific form of agricultural residue. The study reported a substantial elevation in PM2.5 carbon monoxide during the burning season, increased hemoglobin and hematocrit levels, decreased pulmonary function, pulmonary fibrosis, pulmonary and hepatic vascular embolism, and other systemic pathological changes. On this basis, the study was used as the principal field anchor for the hematological direction and respiratory pathology probability.

The indoor PM2.5 concentration was 135.0 µg/m3 under a moderate 3-d regional haze scenario with 95% prediction interval of 92.2–177.8 µg/m3. As reported by the model, the median ADG reduction was 5.2%, FCR increased by 2.8%, and LW loss was 9.4 g/bird/episode. Hemoglobin increased by 3.1%, and hematocrit increased by 4.4 %. The immunity measure retained 97.5% of baseline; there was an overall 17.1% probability of health-related impairments and 11.0% probability of respiratory pathologies occurring as a result of exposure. The median response curves and corresponding uncertainty bands constructed for the indoor PM2.5 concentrations are shown in Figure 5. The summary predictions for both moderate and severe scenarios are reported in Table 5.

The maize stover and biomass burning construction within a 3-d severity scenario yielded an indoor PM2.5 concentration of 361.2 µg/m3 with a 95% prediction interval of 246.8–475.5 µg/m3. The median ADG would be reduced by 19.3%, FCR would increase by 9.8%, and LW loss would reach 34.2 g/bird/episode. The median changes in hemoglobin and hematocrit were 11.8% and 16.0%, respectively. Exploratory immune retention decreased to 90.4% of the baseline, whereas the probabilities of general health impairment and respiratory pathology increased to 46.2% and 40.2%, respectively. Sensitivity analysis showed that the indoor/outdoor ratio and production-response coefficients contributed most strongly to growth-output uncertainty, whereas the pathology intercept and slope parameters dominated lesion-risk uncertainty. The median predictions and 95% prediction intervals are compared in Figure 6.