A stochastic mathematical model of methicillin resistant Staphylococcus aureus transmission in an intensive care unit: Predicting the impact of interventions

https://doi.org/10.1016/j.jtbi.2006.11.008Get rights and content

Abstract

Objectives: To estimate the transmission rate of MRSA in an intensive care unit (ICU) in an 800 bed Australian teaching hospital and predict the impact of infection control interventions.

Methods: A mathematical model was developed which consisted of four compartments: colonised and uncolonised patients and contaminated and uncontaminated health-care workers (HCWs). Patient movements, MRSA acquisition and daily prevalence data were collected from an ICU over 939 days. Hand hygiene compliance and the probability of MRSA transmission from patient to HCW per discordant contact were measured during the study. Attack rate and reproduction ratio were estimated using Bayesian methods. The impact of a number of interventions on attack rate was estimated using both stochastic and deterministic versions of the model.

Results: The mean number of secondary cases arising from the ICU admission of colonised patients, also called the ward reproduction ratio, Rw, was estimated to be 0.50 (95% CI 0.39–0.62). The attack rate was one MRSA transmission per 160 (95% CI 130–210) uncolonised-patient days. Results were not sensitive to uncertainty in measured model parameters (hand hygiene rate and transmission probability per contact).

Hand hygiene was predicted to be the most effective intervention. Decolonisation was predicted to be relatively ineffective. Increasing HCW numbers was predicted to increase MRSA transmission, in the absence of patient cohorting. The predictions of the stochastic model differed from those of the deterministic model, with lower levels of colonisation predicted by the stochastic model.

Conclusions: The number of secondary cases of MRSA colonisation within the ICU in this study was below unity. Transmission of MRSA was sustained through admission of colonised patients. Stochastic model simulations give more realistic predictions in hospital ward settings than deterministic models. Increasing staff does not necessarily lead to reduced transmission of nosocomial pathogens.

Introduction

Infections caused by antibiotic-resistant bacterial pathogens in the health-care setting are detrimental to patients and place a large burden on health-care institutions. Staphylococcus aureus is a common cause of hospital acquired blood stream infection and wound infection. Methicillin-resistant S. aureus (MRSA) leads to a higher mortality, morbidity (Engemann et al., 2003) and cost (Capitano et al., 2003) compared with methicillin-sensitive S. aureus (MSSA).

The proportion of isolates of S. aureus that are methicillin-resistant is increasing in many countries including Australia (Nimmo et al., 2003). It is likely that the increase in MRSA does not represent replacement of MSSA, but is an additional burden (Cooper and Lipsitch, 2004).

Methicillin resistance developed in S. aureus soon after this class of antibiotics was introduced (Ericksen and Erichsen, 1963). Most strains of health-care associated (HA) MRSA are also resistant to other classes of antibiotics including aminoglycosides and macrolides. Of even more concern is the recent observation that some MRSA isolates have been found to be resistant to glycopeptides (Bartley, 2002) and oxalidinones (Meka et al., 2004), the major alternative therapies for MRSA infection.

Antibiotic-resistant bacteria are believed to spread from patient to patient, principally via the hands of health-care workers (HCWs). Colonisation with MRSA frequently precedes infection. This transmissible, asymptomatic state will not be detected unless an active surveillance program is in place. Thus, halting the institutional spread of MRSA requires measures that affect colonised patients as well as those with overt infection.

Recommendations for the control of MRSA transmission include isolation (Garner, 1996) active surveillance cultures (Muto et al., 2003) and hand hygiene. While these guidelines are based on the best available evidence, few of the studies of hospital acquired infectious diseases use sound methodology (Cooper et al., 2003). The increase in the proportion of S.aureus isolates that are methicillin resistant in the face of infection control measures led to pessimism about their efficacy (Teare and Barrett, 1997). A recent study found that moving patients into single rooms or cohorted bays did not reduce MRSA acquisition (Cepeda et al., 2005), however this study screened for MRSA only weekly which may have led to long delays before colonised patients were removed from the general ward, diluting any benefit of isolation.

Mathematical models provide a means of predicting the likely impact of an intervention or the interaction of multiple interventions, capturing nonlinear transmission dynamics. Stochastic models have the additional advantage of predicting the expected variation in outcomes, which may be marked in small populations such as hospital wards. Statistical methods based on structured models provide a means of estimating transmission parameters from data.

In modelling community epidemics and emerging infectious diseases, the emphasis of model-informed infection control measures has been to achieve an effective reproduction ratio (the number of cases that occur due to the introduction of one infectious case, assuming a fully susceptible population) below unity. In the case of hospital associated pathogens such as MRSA, the mean number of secondary cases that arise within a ward during a single hospital admission (which we call the ward reproduction ratio, Rw) may be below unity, but colonised patients may go on to transmit MRSA in other wards and during subsequent hospital admissions leading to an overall reproduction ratio above unity (see Cooper et al., 2004 for full explanation).

In this study, we find a low ward reproduction ratio, Rw=0.5. Frequent re-introductions of MRSA maintain the endemic prevalence. We therefore use attack rate, defined as the number of MRSA transmissions per uncolonised patient day, as our outcome measure when predicting the impact of interventions.

This study differentiates imported cases of MRSA from those that occur during ward stay. All new cases are assumed to arise from other colonised patients via the hands of HCWs (cross transmission). We utilised a mathematical model to quantify MRSA cross transmission in an Australian intensive care unit. We collected data on admission, discharge and colonisation events as well as other critical model parameters, hand hygiene compliance and transmission per contact, to estimate the MRSA attack rate and the ward reproduction ratio. We overcame the challenge of unobserved events by using a Bayesian framework and considering the MRSA acquisition date as a latent variable. Stochastic and deterministic realisations of the model gave predictions of the likely impact of interventions including changes in HCW/patient ratio, patient cohorting, hand hygiene, length of stay, admission prevalence, decolonisation and ward size on the attack rate.

This study extends previous models because all parameters used to estimate transmission were derived through ward observation directly or fitted to acquisition data. Ward observations running in parallel to the data collection gave us realistic values for hand hygiene compliance and probability of MRSA transmission from a colonised patient to HCW. For the simulation component of the study, we incorporate ward size as a parameter, not previously considered, and predict the impact of increases in staff levels if this leads to increased contact rates. The study later considers the effect of decolonisation based on parameters derived from an experimental study (de la Cal et al., 2004).

Section snippets

Model

Our ward transmission model was a modification of the susceptible–infectious (SI) model with migration, described by Bailey (1975). Versions of this model have been used previously to analyse nosocomial transmission data (Sebille and Valleron, 1997, Sebille et al., 1997, Cooper et al., 1999, Austin et al., 1999, Grundmann et al., 2002, Raboud et al., 2005).

Patients and setting

This study included all patients admitted to the ICU of a 800 bed tertiary referral teaching hospital (Princess Alexandra Hospital, Brisbane, Australia) from 8th August 2001 to 3rd March 2004 (939 days inclusive). The ICU bed capacity varied during the study from 16 to 22.

Surveillance of colonisation

During the investigation period, all patient admissions were recorded in the ApacheIIITM database. The mean number of inpatients who met inclusion criteria for the study each day was 15 (median 16).

Ward policy was to swab all

Methods

To quantify cross transmission of MRSA in our study population, we estimated the attack rate (number of transmissions per uncolonised patient day) and the ward reproduction ratio, Rw.

We have no direct estimate of contact rate, c, or probability of transmission from HCW to patient, php. These two parameters are inseparable in the model, so we estimate the value of their product, the transmission parameter, φ=cphp. The admission and discharge dates of patients are directly observed in this study

Results

During the study period, 3329 patients met inclusion criteria. Of these, 100 patients were known to be colonised on admission and 77 met the criteria for new colonisation. Fig. 2 summarises the data.

Discussion

We used a Bayesian framework to quantify MRSA transmission and estimate the ward reproduction ratio of MRSA in an ICU in a large teaching hospital. The Bayesian methodology allowed us to incorporate unseen events, namely, the time of MRSA transmission.

This study used a four compartment modified SI model with migration. Ward observations of hand hygiene compliance and transmission probability per contact gave us estimates of all but one model parameter, which was readily fitted to the data.

We

Acknowledgments

This work was partially supported by a grant under the Australian Research Council Linkage Scheme (LP0347112) and NHMRC scholarship number 290541. The authors would like to thank Dr M. Whitby for providing advice and data. The authors would like to acknowledge the helpful comments of the anonymous referees.

References (36)

  • D.J. Austin et al.

    Vancomycin-resistant enterococci in intensive-care hospital settings: transmission dynamics, persistence, and the impact of infection control programs

    Proc. Natl Acad. Sci. USA

    (1999)
  • N. Bailey

    The Biomathematics of Malaria

    (1975)
  • J. Bartley

    First case of VRSA identified in Michigan

    Infect. Control Hosp. Epidemiol.

    (2002)
  • N. Becker

    Analysis of Infectious Diseases Data

    (1989)
  • Boyce, J., April 2005. Epidemiology; prevention; and control of methicillin-resistant Staphyloccus aureus in adults—I....
  • J.M. Boyce et al.

    Environmental contamination due to methicillin-resistant Staphylococcus aureus: possible infection control implications

    Infect. Control Hosp. Epidemiol.

    (1997)
  • B. Capitano et al.

    Cost effect of managing methicillin-resistant Staphylococcus aureus in a long-term care facility

    J. Am. Geriatr. Soc.

    (2003)
  • B. Cookson et al.

    Staff carriage of epidemic methicillin-resistant Staphylococcus aureus

    J. Clin. Microbiol.

    (1989)
  • Cited by (95)

    • Simulation models for transmission of health care–associated infection: A systematic review

      2020, American Journal of Infection Control
      Citation Excerpt :

      The models in this review used a number of calibration approaches: maximum likelihood estimation58,59; the least- square criterion60,61; Monte Carlo,39 Markov chain,49 and combinations of these methods.35,40,50,52,62-64 In particular, McBryde et al64 used a combination of Bayesian estimation, the Markov chain, and the Monte Carlo for model fitting. Similarly, Sadsad and colleagues40 combined a scatter search algorithm and a least squares criterion for model calibration.

    • Comparison of infection control practices in a Dutch and US hospital using the infection risk scan (IRIS) method

      2020, American Journal of Infection Control
      Citation Excerpt :

      The collection, interpretation, and cutoff points for the risk classification was used as described in Table 2. Risk classification was based on prevalence rates from (inter)national surveillance13-15 or expert opinion, and previous measurements if no reference values were available in the peer-reviewed literature.5,16-20 Compliance to guidelines and appropriateness of use was based on the local guidelines of the hospital.

    • Cost-effectiveness of interventions to improve hand hygiene in healthcare workers in middle-income hospital settings: a model-based analysis

      2018, Journal of Hospital Infection
      Citation Excerpt :

      Incidence of hospital-acquired and healthcare-associated BSI from this and other hospitals in the same region have also been reported in previous studies [22,23]. A previously described deterministic host–vector model was constructed to simulate MRSA transmission dynamics in an ICU (Figure 1) [24]. Patients can be admitted to the ICU either uncolonized or colonized with MRSA.

    View all citing articles on Scopus
    View full text