Antimicrobial Resistance Amplifies the Disease Burden Associated with Escherichia coli Contamination in Ready-to-Eat Foods: A Probabilistic QMRA-DALY Study
Description
We hypothesized that antimicrobial-resistant (AMR) Escherichia coli in ready-to-eat (RTE) foods causes a larger disease burden than non-resistant (non-AMR) E. coli, by raising both exposure and the severity of illness. We tested this for adults aged 15 years and older in Benin, Burkina Faso, Côte d'Ivoire, Ghana and Nigeria. Data were gathered through two quantitative systematic reviews following PRISMA 2020. The first included 10 studies giving 141 E. coli concentration records (CFU/g), each paired with the resistance of isolates from the same samples. AMR and non-AMR concentrations were derived as total count × resistant fraction and total count × (1 − resistant fraction). AMR meant resistance to at least one antibiotic. The second review included 24 studies and FAO/WHO GIFT datasets giving amounts eaten and eating days per year by age group. Every value was recorded as published, harmonized by documented rules, and traced to its paper and DOI. These inputs fed a probabilistic QMRA-DALY model in ProRisk. Dose combined concentration, amount eaten and the pathogenic fraction of E. coli (0.60). Infection risk followed a beta-Poisson dose-response for enteropathogenic E. coli. Cases, deaths and DALYs followed, with longer illness and higher mortality for resistant infections. Each run used 100,000 Monte Carlo iterations, with 11 sensitivity scenarios. Resistance was pervasive: in 36 of 42 food-country combinations, every isolate was resistant. AMR E. coli caused 784,708 DALYs a year (90% uncertainty interval 699,162-837,448), against 38,372 (6,736-84,157) for non-AMR E. coli. The DALY ratio was 20.4, with 4,220 versus 161 deaths a year. Nigeria carried 73% of the AMR burden. AMR DALYs exceeded non-AMR DALYs in all 1.2 million simulated iterations across the scenarios. The data are in four parts: Supplementary Data 1 (concentrations), 2 (consumption), 3 (ProRisk model files and inputs) and 4 (model outputs). Each has a README, and the workbooks link every model input to its source. The comparison between pathways is more reliable than the absolute counts. In Nigeria and Benin, predicted infection risk reached 1, so infection numbers are upper limits, not forecasts of reported cases. Results depend on how AMR is defined: under multidrug resistance, the DALY ratio fell to 1.6. Researchers can re-run the model, update inputs as new surveys appear, or apply the framework to other foods and regions.
Files
Steps to reproduce
1. Concentrations (Supplementary Data 1). Search PubMed, Scopus, and Google Scholar with the strings in Table S12, and screen using the guiding question (Supplementary Document 1, Section 1). Extract each food at each site as one record (Table S3). Derive AMR concentration = total E. coli (CFU/g) × resistant fraction, and non-AMR = total × (1 − resistant fraction), counting resistance to at least one antibiotic. Apply harmonization rules C00-C16 (Table S7; log in Table S14). For each food and country with three or more site values, fit a log10-normal distribution; otherwise use the food-group pool (Table S5). Ready_to_model_E_coli_concentrations.csv holds the results. 2. Consumption (Supplementary Data 2). Search PubMed with the strings in Table S14 and add GIFT datasets. Record every amount and frequency as published (Tables S4, S5). Harmonize by rules H01-H13 (Table S9): convert frequency categories to eating days per year, pool by sample size, map to age groups 15-19, 20-54, and 55+, and apply the age factor Uniform(0.82, 1.02) (Table S8). Match each food concentration to a consumption record (Supplementary Data 1, Table S6). Ready_to_model_food_consumption.csv holds the result; QMRA_inputs_joined_long.csv joins both reviews. 3. Model (Supplementary Data 3). Open WESTAFRECOLI_v21_Base.prorisk in ProRisk, or read it as JSON. Each file lists every input distribution and output formula. Dose = pathogenic fraction × concentration × amount eaten. Infection per eating occasion follows the beta-Poisson model (alpha 0.162, beta 1.403e6). Annual risk combines all foods eaten in a year. Cases of hemolytic uremic syndrome, end-stage renal disease, and death give YLD, YLL, and DALYs. Population: sheet Population. 4. Run. Set 100,000 Monte Carlo iterations and random seed 42. Run the base case and the 11 scenario files (S1-S10, with S3a and S3b). Outputs that are constant in every iteration are not model outputs; take their values from the sheet Fixed_value_outputs. 5. Check (Supplementary Data 4). Compare your medians and 5th-95th percentiles with the ProRisk exports and sheet Headline_results. Base case: AMR DALYs 784,708 (699,162-837,448); non-AMR 38,372 (6,736-84,157). A re-run in other software with a different random stream should agree within about 3% (sheet Cross_check).
Institutions
- Kwame Nkrumah University of Science and TechnologyAshanti, Kumasi