Transmission model
We developed an agent-based model of person-to-person transmission in a hypothetical household using NetLogo (v 6.0.2) software [15]. We introduced in a family a single ESBL-EC colonised person and simulated the transmission dynamics of ESBL-EC and control interventions over a one-year time horizon. We studied four households: two adults, two adults and a child, two adults and a baby, and two adults and a child and baby (so families of 2-4 persons).
The oral-faecal route was indicated as the most frequent route of human-to-human ESBL-EC transmission [9]. Thus, in the model, we hypothesised that hand contamination with ESBL-EC most likely occurs when: 1) a colonised person is using the toilet or 2) a person is changing the diapers of a colonised baby. Non-human sources may also represent a reservoir of ESBL-EC for humans and contribute to the spread of resistance in the community (e.g. by contaminated meat/vegetables, pets or the environment). By simplification we included in the model a single parameter representing the background acquisition, based on a recent study [3].
Cross-transmission among individuals occurred via contaminated hands. We assumed that feeding a baby or eating with contaminated hands could lead to gut colonisation. If HH was performed after using the toilet/changing diapers and before eating/feeding, it prevented ESBL-EC contamination and colonisation, respectively.
The structure of contact patterns is highly associated with age and gender [16,17]. In order to infer the contact network in a modelled household and routes of human-to-human transmission, we considered four profiles of household members: adult woman, adult man, child (≥ 3 years old) and baby wearing diapers (< 3 years). For simplification, other household members (relatives, visitors, etc.) were not considered in the model. We modelled each profile explicitly; it had its own contact frequency with household members, level of HH compliance, and exposure to antibiotics.
Each individual could be in one of four infectious states: 1) susceptible (negative for ESBL-EC), 2) contaminated (hands), 3) colonised (in the digestive tract) or 4) colonised and contaminated. The probability of changing the state for each person was updated daily, and depended on the nature and frequency of contacts, ESBL-EC infectious state of household members, HH compliance, and antibiotic exposure.
We hypothesised that exposure to antibiotics may facilitate the transmission of ESBL-EC in two ways: by increasing the probability of colonisation in contaminated persons receiving antibiotics, and by increasing the probability of transmission from a colonised person treated with antibiotics [18].
We derived parameter estimates including daily contacts (Figure 1), HH practices (Table 1), and other model inputs from the literature (Supplementary Table S1).
Table 1. Probability of handwashing with soap in most critical situations for ESBL-EC transmission. Based on data from [17]. *(%) of HH opportunities.
|
HH after using the toilet (%)*
|
HH before meals (%)*
|
HH after changing diapers (%)*
|
HH before feeding (%)*
|
Woman
|
40
|
36
|
60
|
0
|
Man
|
17
|
33
|
50
|
0
|
Child
|
29
|
50
|
-
|
-
|
Based on French data on antibiotic use in the community, the daily probability of antibiotic prescription was higher for children and babies [19].
There is a lack of data about the probability of hand contamination with ESBL-EC after changing diapers or using the toilet in households. We hypothesised that the probability of hand contamination was higher after changing diapers than after person-to-person contact. For the probability of contamination after using the toilet, we undertook a conservative assumption that it would be the same as for the contact with contaminated hands. In a supplementary analysis, we studied the impact of our assumptions on main results.
An unknown parameter, the daily probability of gut colonisation in a contaminated person (pcol), was calibrated in order to reproduce the transmission rate estimated in the study of Arcilla et al. [6].
A detailed description of the model, the main model parameters and details on parameters calibration can be found in the Supplementary Text S1.
Infection control strategies
In the base case scenario, with no intervention, we considered compliance with HH and antibiotic exposure reported in the literature (Table 1 and Supplementary Table S1). Then, we assessed two scenarios with implementation of control strategies: (1) 50% improvement in compliance with HH for all HH opportunities, (2) antibiotic restriction, with a 50% reduction of patients receiving antibiotics and 25% reduction of treatment duration.
Model simulations and outcomes
The main outcome of interest was the probability of ESBL-EC transmission to a household member during one year following the introduction of an index case. We also estimated the mean time of persistence of ESBL-EC colonisation, defined as the time it takes to get rid of bacteria from all household members. The outcomes were estimated from 30 000 simulations of the stochastic model for each set of parameter values.
Univariate uncertainty analysis
The confidence intervals presented in the main analysis reflect the uncertainty due to the stochastic processes of the model and not that associated with uncertainty in parameter estimates. To assess the impact of the latter on our results, we performed several uncertainty analyses. We tested the impact of a lower duration of intestinal colonisation, a lower duration of hand contamination with ESBL-EC, a higher base case level of HH after using the toilet and before eating, a higher probability of hand contamination after using the toilet, and a lower probability of hand contamination after changing diapers (Supplementary Text S2).
We then studied the impact of a higher probability of acquisition from external sources (Supplementary Text S2).
We also investigated the model in which the daily frequency of contacts between man and woman was higher than the reported 1.2 contacts/day to take into account that contacts between man and woman may be less frequent than reported but may last longer (e.g., by sleeping in the same bed).
We tested a less than 50% improvement in HH to take into account the difficulty of improving HH in the community or that the quality of HH needed to eradicate bacteria from hands may be lower than the assumed 100%.
Finally, we investigated the impact of higher, 62% reduction in antibiotic use to reach the lowest European level of antibiotic use in the community, observed in the Netherlands [20].
Sensitivity analysis
We ran a multivariate sensitivity analysis to quantify the impact of input parameters on the model output. These parameters included epidemiological parameters (probability of contamination after using the toilet/changing diapers, probability of colonisation, duration of colonisation etc.); and transmission control parameters, i.e. compliance with HH and probability of antibiotic prescription. For this analysis, we considered a household composed of two persons or four persons and a woman as an index case. We used Latin Hypercube Sampling (LHS) to generate N=100 parameter sets from our parameter ranges (Supplementary Table S1). For each set of model parameters, we calculated N model outputs (over 30 000 simulation replicates). Then, we used the Partial Rank Correlation Coefficient (PRCC) to quantify and rank the impact of input parameters on the probability of ESBL-EC transmission in a household (Supplementary Text S1).