Assessing the effect of patient screening and isolation on curtailing Clostridium difficile infection in hospital settings

Background Patient screening at the time of hospital admission is not recommended as a routine practice, but may be an important strategy for containment of Clostridium difficile infection (CDI) in hospital settings. We sought to investigate the effect of patient screening in the presence of asymptomatic carriers and in the context of imperfect patient isolation. Methods We developed and parameterized a stochastic simulation model for the transmission dynamics of CDI in a hospital ward. Results We found that the transmission of CDI in the hospital, either through asymptomatic carriers or as a results of ineffective implementation of infection control practices, at the time of hospital admission. The results show that, for a sufficiently high reproduction number of CDI, the disease can persist within a hospital setting in the presence of in-ward transmission, even when there are no asymptomatically colonized patients at the time of hospital admission. Conclusions Our findings have significant public health and clinical implications, especially in light of the emergence and community spread of hypervirulent CDI strains with enhanced transmission rates and toxin production. Rapid detection of colonized patients remains an important component of CDI control, especially in the context of asymptomatic transmission. Screening of in-hospital patients with potential exposure to colonized patients or contaminated environment and equipment can help reduce the rates of silent transmission of CDI through asymptomatic carriers. Electronic supplementary material The online version of this article (doi:10.1186/s12879-017-2494-6) contains supplementary material, which is available to authorized users.

When considering laboratory testing with time-delay, we included two compartments of D + and D − to account for time-interval between sample collection and the release of laboratory results for colonized patients with and without immune responses, respectively. During this time-interval, patients are neither isolated nor treated for CDI, and therefore the possibility of in-ward transmission exists. In this case, the model can be expressed as Screening is implemented at the time of hospital admission ( Figure S1) or for in-hospital patients with exposure to C. difficile. Parameters of the model are described in Table 1

Basic reproduction number
To calculate the basic reproduction number, we applied the next generation method (van den Driessche and Watmough, 2002), by representing F and V as the matrices for new infections and transitions in the infection subclasses and Taking the Jacobian of F and V at the infection-free equilibrium, we obtain and where, at the infection-free equilibrium, According to the next generation method, the reproduction number is given by the spectral radius (dominant eigenvalue) of J F J −1 V . This gives We used this expression to calculate the transmission rate of C. difficile for a given R 0 , fixing all other parameter values. We note that in the absence of interventions, the model with time-delay reduces to the model without time-delay, and therefore has the same R 0 . Prevalence of CDI for R 0 = 1.07 Figure S2 shows the prevalence of C. difficile for the scenarios described in the main text.    Figure S3: Prevalence of C. difficile in the model with rapid laboratory testing, with R 0 = 2.6 over 200 days without screening (A-C) and with screening (D-F) 92.5% of patients at the time of hospital admission. Curves represent the prevalence of undiagnosed colonized patients (black), and isolated patients (grey). The total prevalence is the sum of black and grey curves. Effectiveness of isolation for CDI patients was 100% (A,D), 90% (B,E), and 80% (C,F).
Simulation results for R 0 = 2.6

Model with rapid laboratory testing
For the mean value of R 0 = 2.6, Figure S3 shows the prevalence of C. difficile for three scenarios in which the effectiveness of patient isolation in preventing in-ward transmission is 100% ( Figure S3A,D), 90% ( Figure S3B,E), and 80% ( Figure S3C,F). In these simulations, the baseline scenario without screening (θ = 0) was compared with the scenario of 92.5% screening at the time of hospital admission. When the effectiveness of patient isolation is 100%, the prevalence of C. difficile reduces from 18 cases without screening to 13.4 cases (on average) with 92.5% screening of patients at the time of hospital admission. This corresponds to approximately 25.8% (95% CI: 25.6% -25.9%) reduction of prevalence 50 days after the start of screening.  Figure S4: Prevalence of C. difficile in the model with a time-delay in laboratory testing, with R 0 = 2.6 over 200 days without screening (A-C) and with screening (D-F) 92.5% of patients at the time of hospital admission. Curves represent the prevalence of undiagnosed colonized patients (black), and isolated patients (grey). The total prevalence is the sum of black and grey curves. Effectiveness of isolation for CDI patients was 100% (A,D), 90% (B,E), and 80% (C,F).

Model with a time-delay in laboratory testing
Compared with the results for rapid testing, Figure S4 shows a significantly lower effect of patient screening on reducing the prevalence of CDI when an average of 2 days is considered for the time-delay between sample collection and the release of laboratory results. Regardless of the effectiveness of patient isolation, we observed that the reduction of CDI prevalence remains below 1% with screening of patients at the time of hospital admission. . In panels (A) and (C), curves represent the reduction achieved with screening 92.5% of patients at the time of hospital admission, where the effectiveness of patient isolation in preventing in-ward transmission is: 100% (black); 90% (red); and 80% (grey). In panels (C) and (D), curves represent the reduction achieved with screening 92.5% of patients at the time of hospital admission, where the effectiveness of patient isolation in preventing in-ward transmission is: 90% (red); and 80% (grey). Screening 90% of in-hospital patients with exposure to CDI started on day 100 (shaded area). Figure S5A (black curve) shows that when the effectiveness of patient isolation is 100% in preventing in-ward transmission in the model with rapid laboratory testing, the daily incidence of C. difficile is reduced by over 25.2% on average (95% CI: 24.6% -25.8%) as a result of 92.5% screening at the time of hospital admission. With lower effectiveness of isolation, this reduction drops below 4% ( Figure S5A, red and grey curves). For the model with an average of 2 days between sample collection and the release of laboratory results, the relative reduction of incidence is negligible (below 1%) regardless of the effectiveness of patient isolation ( Figure S5C).

Relative reduction of CDI incidence
We then implemented screening of inpatients with exposure of C. difficile in addition to screening of patients at the time of admission. For the model with rapid laboratory testing, Figure S5B shows the temporal percentage reduction of the CDI incidence, when screening 90% of in-hospital patients started on day 100. This resulted in an increasing trend in the percentage reduction of C. difficile incidence over time ( Figure S5B, red and grey curves), reaching levels comparable to those achieved with screening patients only at the time of hospital admission when the effectiveness of patient isolation was assumed to be 100% ( Figure  S5A, black curve). However, simulating the model with a delay of 2 days in the release of laboratory results indicate marginal benefits, achieving at most 3% increase in the relative reduction of CDI incidence by inpatients screening ( Figure S5D).

Sensitivity analysis and PRCC
We carried out a sensitivity analysis using the Latin Hypercube Sampling (LHS) technique and calculated Partial Rank Correlation Coefficients (PRCCs) to investigate the effects of variation in parameter values on the model outcomes (Marino et al., 2008), and specifically on the prevalence of C. difficile as the response variable. LHS is a stratified (near-random) sampling technique without replacement. In this method, the random parameter distributions are divided into a number of equal probability intervals, and are then sampled.
The goal of this analysis was to identify key parameters whose uncertainties contribute to prediction imprecision, and to rank these parameters by their relative importance in contributing to this imprecision. To allow for the simultaneous variations of the parameters, samples of size 1000 were generated in which each parameter was treated as a random variable and assigned a probability function. These parameters were sampled using LHS method (near-random sampling) within their respective ranges. To calculate PRCCs, we assumed that there is no correlation between the input parameters (Marino et al., 2008). The parameters with large PRCC values (close to 1 or −1) and their corresponding p-values smaller than the significance level (0.05) have the largest influence on the model outcomes (Taylor, 1990). We examined scatter plots to verify the existence of monotonic relationships between the parameters used in the LHS sampling and the response variable ( Figures S6-S9). The PRCC values and their associated p-values are presented in Table S1 and S2 for reproduction numbers of R 0 = 1.07 and R 0 = 2.6, respectively. The relative influence of model parameters with R 0 = 1.07 is summarized in Table 1 of the main text. The relative influence of model parameters with R 0 = 2.6 is summarized in Table S3.

References
Pauline van den Driessche and James Watmough. Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission. Mathematical biosciences, 180 (1) Figures S6-S9 shows the scatter plots of partial residual of parameters used in the sensitivity analysis, corresponding to reproduction numbers R = 1.07 and R 0 = 2.6, and models with rapid testing and with a time-delay in the release of laboratory results.