Recreational Water Risk from Extended-Spectrum Beta-Lactamase-Producing Escherichia coli of Broiler Origin: A Quantitative Microbial Risk Assessment

Abstract

Extended-spectrum beta-lactamase (ESBL)-producing E. coli from broiler farms can reach watersheds used for recreational swimming. We assessed short-term swimmer exposure by extending a modular quantitative microbial risk assessment (QMRA) to the recreational water pathway linking land manure application to in-stream fate and transport with dilution and decay. We modeled single-event exposure doses and estimated loss of disability-adjusted life years (DALYs). We ran sensitivity analyses on several parameters and compared outputs to published recreational water assessments that include ESBL E. coli. Assuming a worst-case scenario, single-event doses were lower for adults (2.95 CFU; UI 0.14–6.11) and higher for children (8.78 CFU; UI 0.56–17.20) on day 1 after land application, then dropped below 0.01 CFU by day 200, with DALY losses from 10⁻⁷ to 10⁻¹⁰. Uncertainty was dominated by fate and transport. Stronger particle binding, faster in-stream decay, and larger effective volumes lowered exposure, while higher shedding, greater flow, and larger wash-off raised it. Estimates fell at the low end of prior studies. Swimmer exposure appears to be extremely low and short-lived. The modular QMRA links farm contamination to bathing-site risk and supports risk-based monitoring (after spreading or storms) and short-term forecasts that focus advisories on short, higher-risk windows after litter application.

1. Introduction

Antimicrobial resistance (AMR) is not only a clinical problem but also an environmental one, with soils, sediments, and surface waters acting as reservoirs [1,2]. In broiler chicken production, antimicrobial use increases selection pressure for resistant Enterobacterales, including extended-spectrum β-lactamase (ESBL)-producing Escherichia coli (E. coli) [3,4]. Broiler flocks can acquire ESBL E. coli via several routes, including contamination early in the production pyramid and occasional vertical carryover from breeders [5,6,7]. At the same time, biosecurity gaps enable introductions, and once present, within-flock spread is typically rapid [8]. After depopulation, land application of litter may transfer large bacterial loads to fields where ESBL E. coli may persist for weeks to months, depending on temperature, moisture, and vegetation cover [9,10]. Hydrologic events then mobilize bacteria to streams and rivers; stormflow and runoff often produce short-lived peaks in riverine E. coli [11]. Catchments with greater agricultural land use usually show higher E. coli concentrations than forested or mixed-use sub-catchments, indicating proximity and intensity of animal production as risk factors for surface waters [10,12]. Manure management can mitigate environmental loading: poultry-litter composting reduces antibiotic-resistant E. coli and antibiotic residues, short-term storage lowers ESBL-producing E. coli loads, and anaerobic digestion can further reduce viable resistant bacteria, although effectiveness appears to be context-dependent [13,14,15]. Because survival, mobilization, and in-stream fate depend on local drivers such as solar radiation, temperature, dilution and volume, particulates, and residence time, site-specific environmental assessment is necessary. This is particularly relevant for bathing waters where exposure occurs via incidental ingestion. Evidence shows that ESBL E. coli occur in recreational waters and that swimming exposure is plausible, especially downstream of wastewater discharges or livestock areas. The dose at the bathing site is shaped by dilution and decay [16,17,18,19,20].

Building on our prior integrated QMRA model linking farm, soil, river, occupational, and foodborne pathways [21,22], we extend the framework to the recreational-water pathway by estimating single-event ingestion doses and DALYs for recreational swimmers at a hypothetical downstream site, alongside sensitivity analysis of key environmental drivers.

2. Materials and Methods

2.1. Study Design and Model Scope

We extended our previously published modular QMRA, adding a recreational-water swimming exposure module (Figure 1). The upstream modules (farm, soil, river) follow the structure and parameterization described in Sarnino et al. (2025b) [21] and are only summarized here. The swimming module and the dose–health–burden mapping are detailed below, with ingestion behavior and in-stream decay choices aligned with the framework of O’Flaherty et al. (2019) [18]. The most important parameters of the farm, soil, and swimming modules are listed in Table 1.

2.2. Farm Module

We simulated a 36-day conventional broiler cycle in discrete daily steps using an SI transmission model and mass-balance of shedding to litter. The main output is the end-cycle ESBL E. coli litter concentration (CFU/g), which feeds the soil module. The litter concentration $C_{\mathrm{l}\mathrm{i}\mathrm{t}}$ is calculated as follows:

$$C_{\mathrm{l}\mathrm{i}\mathrm{t},i}\hspace{0.25em} =\hspace{0.25em} {\displaystyle \frac{M_{\mathrm{e}\mathrm{n}\mathrm{v},i}\left(T_{\mathrm{e}\mathrm{n}\mathrm{d}}\right)}{M_{\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{t}\mathrm{e}\mathrm{r}}+M_{\mathrm{f}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{s},i}}}$$

(1)

where $M_{\mathrm{e}\mathrm{n}\mathrm{v}}\left(T_{\mathrm{e}\mathrm{n}\mathrm{d}}\right)$ is the cumulative ESBL E. coli mass excreted to the barn environment by day 36 (CFU), after within-flock transmission, growth, shedding, and environmental decay; $M_{\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{t}\mathrm{e}\mathrm{r}}$ is the initial litter mass placed (g); and $M_{\mathrm{f}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{s}}$ is total feces deposited (g) by the flock, in each Monte Carlo simulation $i$.

Full equations, distributions, and assumptions are published in Sarnino et al. (2025b) [21].

2.3. Soil Module

Assuming a worst-case scenario (conservative approach), untreated broiler litter is applied to a field immediately after depopulation and the load in the soil at day $t$ is calculated as follows:

$$L_i\left(t\right)\hspace{0.25em}\hspace{0.25em} =\hspace{0.25em} C_{\mathrm{l}\mathrm{i}\mathrm{t},i}\hspace{0.25em}\hspace{0.25em} \times \hspace{0.25em} R_{\mathrm{a}\mathrm{p}\mathrm{p}}\hspace{0.25em} \hspace{0.25em} \times \hspace{0.25em} {10}^3\hspace{0.25em} \hspace{0.25em} \times \hspace{0.25em} e^{-{\lambda }_s}$$

(2)

where $R_{\mathrm{app}}$ is the mass of litter applied to the field per m², ${10}^3$ converts kilograms to grams so units match the concentration term, and $e^{-{\lambda }_s}$ is the first order decay.

The surface load (CFU/m²) declines by first-order die-off fitted to experimental soil data, as implemented in Sarnino et al. (2025b) [21]. We adopt the same fitted decay constant and application-rate distributions; daily loads are propagated to the river via the runoff scheme below.

2.4. River Transport and Decay

Daily wash-off of the mobile fraction was routed to the receiving water and updated with the same first-order in-stream loss used in Sarnino et al. (2025b) [21], which follows the Mancini approach [23]. In brief, the daily decay rate is a baseline loss modified by temperature, salinity, and light exposure that depends on incident radiation, the light-extinction coefficient, and depth. We use a 1-day step:

$$C_{t+1}^{\left(L\right)}\hspace{0.25em} \hspace{0.25em} =\hspace{0.25em} C_t^{\left(L\right)}\hspace{0.17em}{e}^{-k_t}\hspace{0.25em} +\hspace{0.25em} {\displaystyle \frac{L_i\left(t\right)\hspace{0.17em}A\hspace{0.17em}\left(1-K_d\right)\hspace{0.17em}{f}_{\mathrm{w}\mathrm{a}\mathrm{s}\mathrm{h}}}{V}}$$

(3)

Each day, the previous water concentration decays by the Mancini-type rate $k_t$ and the new wash-off input, computed from the current soil areal load $L_i\left(t\right)$, field area $A$, mobile fraction $(1-K_d)$ and wash-off fraction $f_{\mathrm{w}\mathrm{a}\mathrm{s}\mathrm{h}}$, is added and diluted in the effective mixed volume $V_{\mathrm{e}\mathrm{f}\mathrm{f}}$. $C$ (CFU/L) is then converted to CFU/mL dividing it by 1000. Parameter ranges for temperature, salinity, radiation, light extinction, depth, and effective volume follow O’Flaherty et al. (2019) [18] for a Tyrrhenian-coast (Rome) setting; for the effective mixed volume we use their smaller bathing site (Site 1), matching Sarnino et al. (2025b) [21]. For the swimming step, local mixing is summarized by the daily exchange fraction, which is the ratio of daily inflow to effective mixed volume. This fraction is applied to the river concentration to obtain the bathing-site concentration used for the dose.

Table 1. Key input variables for the farm, soil, and river modules.

Module	Parameter	Unit	Value	Source
Farm	Excretion rate of ESBL E. coli ($ϵ$)	fraction	0.3	[24]
Farm	Initial load per ESBL positive broiler	CFU/broiler	100	[24]
Farm	Growth rate for ESBL E. coli in the broiler’s intestine ($r$)	log10 CFU	Uniform (0–5)	[24]
Farm	Fraction of ingested contaminated feces along with feed (${\rho }_{\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{e}\mathrm{s}\mathrm{t}}$)	fraction	0.014	[24]
Farm	Transmission rate ($\beta )$	/day	0.31	[25]
Farm	Initial prevalence of colonized broilers ($p_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}}$)	fraction	0.01	Expert opinion
Farm	Litter quantity per square meter (L)	g/m2	1000	[24]
Soil	First-order decay rate (${\lambda }_s$)	/day	0.0362	[26]
Soil	Broiler litter applied to field surface ($R_{\mathrm{a}\mathrm{p}\mathrm{p}}$)	kg/m2	2	User defined
Soil	Field area (A)	m2	4046.86	User defined
River	Bacteria partition coefficient ($K_d)$	fraction	0.95	[27]
River	Mobile-phase cells ($W_i$)	fraction	0.50	[27]
River	River-water temperature (T)	°C	21–28	[18]
River	Salinity ($EC)$	fraction	Uniform (0.035–0.075)	[18]
River	Global solar irradiance (IA)	Ly/h	Triangle (17.3–25.4)	[18]
River	Light-extinction coefficient (et)	/m	Uniform (0.26–0.31)	[18]
River	Mean water-column depth (H)	m	Uniform (0.5–6)	[18]
River	Effective pool volume at water site ($V)$	L	Uniform (6.75 × 107–8.25 × 107)	[18]

Table 1. Key input variables for the farm, soil, and river modules.

Module	Parameter	Unit	Value	Source
Farm	Excretion rate of ESBL E. coli ($ϵ$)	fraction	0.3	[24]
Farm	Initial load per ESBL positive broiler	CFU/broiler	100	[24]
Farm	Growth rate for ESBL E. coli in the broiler’s intestine ($r$)	log10 CFU	Uniform (0–5)	[24]
Farm	Fraction of ingested contaminated feces along with feed (${\rho }_{\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{e}\mathrm{s}\mathrm{t}}$)	fraction	0.014	[24]
Farm	Transmission rate ($\beta )$	/day	0.31	[25]
Farm	Initial prevalence of colonized broilers ($p_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}}$)	fraction	0.01	Expert opinion
Farm	Litter quantity per square meter (L)	g/m2	1000	[24]
Soil	First-order decay rate (${\lambda }_s$)	/day	0.0362	[26]
Soil	Broiler litter applied to field surface ($R_{\mathrm{a}\mathrm{p}\mathrm{p}}$)	kg/m2	2	User defined
Soil	Field area (A)	m2	4046.86	User defined
River	Bacteria partition coefficient ($K_d)$	fraction	0.95	[27]
River	Mobile-phase cells ($W_i$)	fraction	0.50	[27]
River	River-water temperature (T)	°C	21–28	[18]
River	Salinity ($EC)$	fraction	Uniform (0.035–0.075)	[18]
River	Global solar irradiance (IA)	Ly/h	Triangle (17.3–25.4)	[18]
River	Light-extinction coefficient (et)	/m	Uniform (0.26–0.31)	[18]
River	Mean water-column depth (H)	m	Uniform (0.5–6)	[18]
River	Effective pool volume at water site ($V)$	L	Uniform (6.75 × 107–8.25 × 107)	[18]

2.5. Swimming Exposure Assessment

To quantify human ingestion of ESBL E. coli during recreational swimming, we apply a two-step downstream mixing followed by an ingestion-dose calculation for each simulation $i$ and day t. All the parameters used for building the module can be found in

Table 2. Key input variables for the swimming module.

Variable	Description	Source	Unit	Value
$C_i$	ESBL E. coli concentration in the watershed	[21]	CFU/mL	Stochastic
$Q_i$	Daily inflow to bathing reach (dilution)	[18]	m3/day	Triangular (67,500, 75,000, 82,500)
${Ve}_i$	Instantaneous water volume for dilution	[18]	m3	Uniform (67,500, 82,500)
$I_R^{\left(A\right)}$	Ingestion volume per adult swim	[18]	mL/event	Uniform (0–70.67)
$I_R^{\left(C\right)}$	Ingestion volume per child swim	[18]	mL/event	Uniform (0–205.33)
$r_{\mathrm{DR}}$	Exponential dose–response parameter	[28]	/CFU	2.18 × 10−6
$F_{\mathrm{U}\mathrm{P}\mathrm{E}\mathrm{C}}$	Fraction of E. coli that are uropathogenic	[28]	fraction	0.1
$p_{\mathrm{c}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{n}\mid \mathrm{g}\mathrm{u}\mathrm{t}}$	P (urinary colonization |gut colonization)	[28]	probability	Uniform (0.35, 0.46)
$p_{\mathrm{U}\mathrm{T}\mathrm{I}\mid \mathrm{c}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{n}}$	P (symptomatic UTI| urinary colonization)	[28]	probability	0.067
$\delta$	DALY burden per UTI case	[28]	DALY/case	Uniform (3.7, 12.84)

.

2.6. Downstream Mixing

$$\begin{array}{cccc}& C_{\mathrm{s}\mathrm{w}\mathrm{i}\mathrm{m},\mathrm{i}}\left(t\right)=\hspace{0.25em} C_i\left(t\right)\hspace{0.25em} \times \hspace{0.25em} {\displaystyle \frac{Q_i}{{Ve}_i}}& & \end{array}$$

(4)

Here, $C_i\left(t\right)$ (CFU/mL) is the river-model concentration, $Q_i$ (m³/day) is the randomly drawn inflow to the downstream bathing pool (e.g., from a tributary), and $V_i \left({\mathrm{m}}^3\right)$ is the pool’s effective volume. The ratio $Q_i/V_i$ represents the additional dilution experienced en route to the swimming site.

2.7. Ingestion Dose

$$\begin{array}{cccc}& D_{k,i}\left(t\right)=C_{\mathrm{s}\mathrm{w}\mathrm{i}\mathrm{m},\mathrm{i}}^{\left(\mathrm{m}\mathrm{L}\right)}\left(t\right)\hspace{0.25em} \times \hspace{0.25em}{I}_{R t,i}^{\left(k\right)}& & \end{array}$$

(5)

where $D_{k,i}\left(t\right)$ is the dose for age group $k$ (A = adult and C = child) in simulation $i$ on day $t$.

$I_{R t,i}^{\left(k\right)}$ (mL per event) is the randomly sampled ingestion volume taken from a uniform distribution, different for each age group

2.8. Dose–Health Endpoints and DALY

Following the framework provided by Heida et al., 2025 [28], we translate dose $D_{t,i}^{\left(k\right)}$ to gut (GI) colonization with $r$ as the exponential dose–response parameter and $f_{\mathrm{U}\mathrm{P}\mathrm{E}\mathrm{C}}$ as the fraction of ESBL E. coli that are uropathogenic:

$$P_{\mathrm{G}\mathrm{I},t,i}^{\left(k\right)}=1-\exp \hspace{0.17em}\left(-r\times f_{\mathrm{U}\mathrm{P}\mathrm{E}\mathrm{C}}\times D_{t,i}^{\left(k\right)}\right)$$

(6)

Progression to urinary-tract colonization and then symptomatic UTI uses two conditional probabilities that are independent of exposure route and age in our base case:

$$P_{\mathrm{U}\mathrm{T}\mathrm{I},t,i}^{\left(k\right)}=P_{\mathrm{G}\mathrm{I},t,i}^{\left(k\right)}\times P\left(\mathrm{UC}∣\mathrm{GI}\right)\times P\left(\mathrm{symptomatic} \mathrm{UTI}∣\mathrm{UC}\right)$$

(7)

To summarize, the swimming module gives the dose via accidental swallowing; $P(\mathrm{UC}\mid \mathrm{GI})$ is the probability that someone who is GI colonized also has urinary colonization (UC); and $P(\mathrm{symptomatic} \mathrm{UTI}\mid \mathrm{UC})$ is the probability that UC leads to clinical UTI. Finally, we estimate the DALY per event as follows:

$$\mathrm{DALY}/\mathrm{event}=P_{\mathrm{U}\mathrm{T}\mathrm{I},t,i}^{\left(k\right)}\times {\mathrm{DALY}}_{\mathrm{per} \mathrm{case}}$$

(8)

2.9. Uncertainty and Sensitivity Analysis

To capture variability and uncertainty, we conducted 1000 Monte Carlo iterations. Model outputs are reported as the mean and 95% uncertainty interval (UI), defined as the 2.5–97.5th percentiles across iterations. Sensitivity analysis was conducted using Partial Rank Correlation Coefficients (PRCCs) to quantify drivers of variability in exposure dose (95% CIs).

2.10. Software

The QMRA was built in R (version 4.3.1) using RStudio (version 2024.09.1 build 394). Data handling and visualization were supported by the tidyverse [29]. Distribution sampling was performed with mc2d [30], while nonlinear regression was fitted using nls2 [31]. Parallel processing was enabled through furrr [32]. Sensitivity analyses were conducted with the sensitivity package in combination with lhs [33,34]. The complete QMRA model is openly available on ENVIRE GitHub, https://github.com/ENVIRE-JPIAMR (accessed on 20 October 2025).

3. Results and Discussion

Exposure at a hypothetical bathing site yielded mean adult doses of 2.95 CFU per swim event (SD 1.76, UI 0.14–6.11) on day 1 after litter application (Figure 2), corresponding to a 5.75 × 10⁻⁷ risk of gut colonization (GI) (UI 2.69 × 10⁻⁸–1.19 × 10⁻⁶), a 1.56 × 10⁻⁸ risk of urinary tract infection (UTI) (UI 7.16 × 10⁻¹⁰–3.26 × 10⁻⁸), and 1.29 × 10⁻⁷ disability-adjusted life years (DALYs) loss per swim (UI 5.98 × 10⁻⁹–3.36 × 10⁻⁷) (

Table 3. Mean probability of gastrointestinal colonization (GI) and urinary tract infection (UTI), and corresponding disability-adjusted life years (DALYs) lost per swimming exposure event at selected days after manure application, shown separately for adults and children, 1000 iterations.

Day	Mean Risk (GI)	Mean Risk (UTI)	Mean DALY	Age Group
1	5.75 × 10−7	1.56 × 10−8	1.29 × 10−7	Adult
10	4.21 × 10−7	1.14 × 10−8	9.76 × 10−8
50	9.84 × 10−8	2.66 × 10−9	2.20 × 10−8
100	1.59 × 10−8	4.33 × 10−10	3.70 × 10−9
150	2.67 × 10−9	7.26 × 10−11	6.13 × 10−10
200	4.39 × 10−10	1.19 × 10−11	9.71 × 10−11
1	1.71 × 10−6	4.65 × 10−8	3.83 × 10−7
10	1.23 × 10−6	3.35 × 10−8	2.80 × 10−7
50	2.89 × 10−7	7.84 × 10−9	6.54 × 10−8	Child
100	4.66 × 10−8	1.27 × 10−9	1.07 × 10−8
150	7.78 × 10−9	2.12 × 10−10	1.78 × 10−9
200	1.27 × 10−9	3.43 × 10−11	2.80 × 10−10

).

Children, ingesting on average 8.78 CFU per swim event on day 1 after litter application (SD 4.94, UI 0.56–17.20), faced a 1.71 × 10⁻⁶ gut-colonization risk (UI 1.09 × 10⁻⁷–3.35 × 10⁻⁶), a 4.65 × 10⁻⁸ UTI risk (UI 2.77 × 10⁻⁹–9.39 × 10⁻⁸), and 3.83 × 10⁻⁷ DALYs per event (UI 2.04 × 10⁻⁸–9.77 × 10⁻⁷). Full simulation outputs (summary statistics across 1000 Monte Carlo iterations) are provided in Table S1 (Supplementary Materials).

By day 200, doses fell below 0.01 CFU, yielding negligible risks for colonization, UTI, and minimal DALYs lost per swim. The steepest drop occurred within the first week, after which exposure decreased gradually, reflecting early dilution and decay.

The sensitivity analysis (Figure 3) indicates that the soil–water partition coefficient is the dominant driver (PRCC ≈ −0.90, 95% CI −0.91 to −0.89), implying that stronger sorption markedly lowers planktonic and thus ingestible E. coli at the bathing site. The on-farm decay rate is the second strongest and negative influence (PRCC ≈ −0.31, 95% CI −0.38 to −0.24). Among positively associated factors, on-farm shedding rate (ε; PRCC ≈ +0.20, 95% CI +0.13 to +0.25), broiler gut carrying capacity (K; PRCC ≈ +0.19, 95% CI +0.13 to +0.26), flow rate (PRCC ≈ +0.19, 95% CI +0.13 to +0.25), wash-off fraction (PRCC ≈ +0.14, 95% CI +0.08 to +0.22), and the maximum adult ingestion volume (PRCC ≈ +0.15, 95% CI +0.08 to +0.20) increase exposure. Negative effects are observed for the soil decay rate (PRCC ≈ −0.19, 95% CI −0.25 to −0.12), bathing-site volume (PRCC ≈ −0.18, 95% CI −0.23 to −0.11), litter mass (PRCC ≈ −0.14, 95% CI −0.21 to −0.08), and a small but significant effect of soil bulk density (PRCC ≈ −0.07, 95% CI −0.14 to −0.01). The ingestion-rate parameter shows little influence (PRCC ≈ −0.03, 95% CI −0.09 to +0.03). Parameters linked to flock size, target weight, initial prevalence, transmission coefficient, and manure application rate appear negligible under the tested scenarios [21]. Overall, exposure is most sensitive to hydrodynamic and fate processes, particularly partitioning, decay, and dilution.

Our simulated single-event ingested doses at the bathing site fall below upper-bound exposures reported for sites influenced by wastewater or mixed sources. They are, however, of the same order as extended-spectrum beta-lactamase (ESBL)-specific QMRAs. For antibiotic-resistant (AR) E. coli in wastewater treatment plant (WWTP)-impacted bathing waters, O’Flaherty et al. predicted mean concentrations of 0.45–345 CFU per 100 mL; using typical adult ingestion volumes, medians of about 13–20 mL per event, and 90th percentiles of at least 100 mL, that range maps to roughly 0.06–3.5 CFU per event at median ingestion and about 3–36 CFU per event at upper-percentile ingestion, which brackets our central estimates [18]. For ESBL-producing E. coli specifically, van Heijnsbergen et al. (2022) estimated up to 61 CFU per event in the Dutch-German Vecht catchment, while Heida et al. (2025) projected low disability-adjusted life year (DALY) burdens from ESBL exposure under US recreational-water criteria, consistent with our per-event DALYs [28,35].

Two factors likely explain why our doses sit toward the lower end of the literature span. First, our scenario couples broiler origin with early downstream dilution, whereas the highest reported concentrations often reflect near-field WWTP influence or constrained circulation in small embayments. Broiler litter typically carries limited ESBL E. coli loads, so the mass available for off-field transport—and thus dose—is low compared with wastewater inputs [10,28,35]. Second, we parameterize first-order photoinactivation under realistic daylight, which accelerates early decline; laboratory and field studies report sunlight decay constants for E. coli of about 1–6 per day, depending on optical properties and depth [36,37].

The temporal pattern we observe, a steep drop in the first week followed by a slower decline, matches expectations for a system where initial loads are quickly diluted and photo-inactivated, then give way to a lower, persistence-controlled tail. High-flow events can temporarily reverse that decline. Multi-catchment studies show that stormflow and overland runoff produce short-lived peaks in E. coli, and event-based sampling demonstrates that routine monitoring can miss these peaks [38,39]. Recent models can forecast rainfall-driven spikes and their arrival times at bathing sites, which suggests a path toward event-responsive advisories in agricultural and mixed catchments [40].

The negative PRCC we find for the water–soil partition coefficient indicates that stronger sorption lowers planktonic and, therefore, ingestible E. coli. This aligns with evidence that a substantial fraction is associated with particulates and streambed sediments, where cells persist longer and can re-enter the water column during resuspension [41].

Even where indicator criteria are met, ESBL-specific exposures may persist under certain hydrologic or source-mix conditions, particularly near livestock or downstream effluents. Reviews of AMR in recreational waters highlight the need to complement classic indicators with contextual information, for example, event triggers, source tracking, and time-of-day light exposure, and to target high-risk windows for public health messaging [16,19].

Our results show front-loaded exposures, highest just after land application, then a steady decline as first-order decay and dilution act. Monitoring should be seasonal and targeted, focusing on the weeks after spreading and during rainfall or flow anomalies. combining event-responsive sampling with predictive tools, for example, rainfall-triggered alerts and near-real-time enzymatic proxies, can better align warnings with actual risk, especially early in the season or after manure application in the catchment [42]. Upstream controls, including storage or curing, composting, anaerobic digestion, delayed spreading relative to peak recreation, vegetated buffers, and incorporation where feasible, can materially reduce early peaks. Because uncertainty clusters in partitioning, decay, and dilution, managers will gain most by refining local light attenuation, depth, and event hydrology, and by collecting a small number of event-based samples to calibrate triggers.

Future models should test interventions at two points: in-flock controls to reduce shedding and farm contamination (e.g., competitive exclusion, bacteriophages) [25,43] and manure treatment/management (e.g., storage, composting, or anaerobic digestion) [13,15,44] to lower field loads and mobility. In addition, validating the model with real-world data and including a wider range of contamination sources would enhance its practical relevance and general applicability.

Limitations

Several limitations merit emphasis. In our model, the only source of contamination is broiler litter, and this likely underestimates exposure in mixed-source catchments where wastewater effluent often dominates near-field microbial loads at bathing sites [18,35,45]. Even though broiler litter contributes to elevated concentrations of resistant E. coli in watersheds, it represents only a minor share of the overall contamination sources [46].

In addition, our assumption of the immediate application of untreated litter represents an upper bound. Second, our fate module assumes simple first-order decay; adding lag and repair dynamics, including photo-repair or regrowth, and accounting for salinity and light interactions could refine k-estimates and reduce uncertainty in different seasons or low-turbidity conditions [36,47,48].

As in Heida et al., our model uses enteropathogenic E. coli (EPEC)-derived data as a proxy for ESBL/UPEC, and assumes that gastrointestinal to urinary tract infection progression is independent of exposure route, despite known biological complexity in UTI pathogenesis [49].

The biggest limitation is probably our worst-case source approach: we assume fresh, untreated litter applied immediately after depopulation. This likely overestimates initial loads, as common management practices such as storage, composting, or anaerobic digestion can reduce E. coli loads by several orders of magnitude before runoff occurs [14,50]. Therefore, results should be interpreted as upper-bound estimates.

For environmental parameterization, we treat the bathing reach as a single well-mixed control volume. Mixing and dilution use a reach-averaged volume and a daily inflow drawn from simple site-specific distributions, rather than resolving near-field mixing zones, rainfall–runoff pulses, or sediment resuspension. We apply key fate parameters as site-level constants or narrow ranges (for example, partitioning or sorption and wash-off fraction as constants; irradiance, light extinction or attenuation, temperature, and depth as ranges), so spatial heterogeneity within the bathing area is not represented. The daily time step also smooths sub-daily travel times and short peaks after rain or wind-driven mixing.

Swimmer behavior is another uncertainty. Ingestion is modeled with broad age-specific Uniforms, but not activity-dependent variation (for example, diving or time in water) or correlations, so a small high-ingestion subgroup could disproportionately influence the upper tails.

To conclude, our model lacks field data for validation and comparison with similar approaches, as it is, to our knowledge, the first study to model the spread of ESBL-producing E. coli from broiler flocks to recreational bathing waters.

4. Conclusions

This study extends a modular QMRA to the recreational-water pathway. It shows that under a worst-case scenario, broiler-origin ESBL-producing E. coli may yield very low, short-lived swimmer exposures. Single-event doses were small for adults and higher for children on day 1 after land application, then fell below 0.01 CFU by day 200, with per-event risks of gut colonization and UTI in the 10⁻⁷ to 10⁻¹¹ range and DALY losses in the 10⁻⁷ to 10⁻¹⁰ range. Sensitivity results point to fate and transport as the primary levers: stronger particle binding, faster in-stream decay, and larger effective volumes reduce exposure, whereas higher shedding, greater flow, and larger wash-off increase it. These patterns sit at the lower end of published recreational-water assessments that include ESBL E. coli, which appears consistent with early dilution and photoinactivation in our scenario. Two limits remain essential. We did not include mixed fecal sources or consider any manure treatment, and assumed first-order decay, which could bias seasonal k-values. Still, the framework offers a tractable way to connect farm controls to bathing-site risk. It supports event-responsive monitoring and simple predictive tools that focus advisories on short, high-risk windows after litter application.