4.2.1 Background
The building of a simulation model was originally part of a research study (Hardy et al. 2007), funded by the Department of Health in the UK, to test whether screening MRSA using a rapid PCR test, which may provide results within a day, is more effective and cost-effective than the established culture test which may take up to four days to obtain results. The main research study was designed as a prospective randomised two-period cross-over study at Birmingham Heartlands Hospital, a typical large general teaching hospital in the UK which has more than 1,000 beds. The project involved seven surgical wards (denoted as ward A, B, C, D, E, F and G hereafter) and lasted for sixteen months. In the first eight months, four hospital wards
Agent-based Simulation Model of MRSA
there was a cross-over with the first four wards adopting the culture test and the other three the PCR test. Overall, fourteen scenarios were generated from the study with each ward being associated with two scenarios, one for the culture test period and the other for the PCR test period. In order to concentrate the sole effect of the rapid screening test, all controllable factors and intervention measures were kept the same between the culture and PCR test period for each study ward except the choice of screening test itself.
Apart from general infection prevention and control measures such as hand-washing, staff education and standard barrier precaution measures, active S&D strategies, including pre-emptive admission and repeat screening test, isolation and decolonisation treatment, were also implemented across all study wards. Detailed information regarding MRSA colonisation status, screening test, isolation, decolonisation treatment, ward movement, operation history, antibiotic treatment, demographics and other potential risk factors were collected for each patient. Observed data were kept in both an Access database and Excel spreadsheets. During the sixteen-month study period, 12,732 patients were admitted to the seven study wards with a total of 13,952 ward admission/re-admission episodes (a patient may be admitted to the ward several times). Altogether, 30,490 screening test samples were taken with 12,682 being admission screening tests and 17808 being repeat screening tests. Overall, 453 patients were identified as colonised with MRSA by admission screening tests (i.e., primary cases) and 268 patients were identified to have acquired MRSA colonisation while they stayed in the hospital by repeat screening tests (i.e., secondary cases).
Statistical analysis of the research study has demonstrated the effectiveness of the rapid screening test in reducing MRSA transmission in the hospital setting. Compared to study periods when culture screening tests were adopted, the transmission ratio, which is the ratio of the number of secondary cases to the number of primary cases, was significantly reduced in six out of seven study wards when rapid PCR screening tests were applied (Hardy et al. 2009).
Compared to statistical analysis, ABS model can describe the transmission dynamics of MRSA over time and evaluate the theoretical mean effectiveness of the rapid
Agent-based Simulation Model of MRSA
screening test when randomness is considered. Furthermore, the model can systematically explore scenarios that have not been implemented in the research study, such as the evaluation of other intervention policies (e.g., screening strategy, isolation and decolonisation treatment), and the sensitivity analysis of influencing factors (e.g., patients’ lengths of say, transmissibility of MRSA and the proportion of patients colonised with MRSA on admission).
4.2.2 Test Agent-based Simulation Model on Previous Studies
Before the full-scale ABS model was built, two preliminary tests were performed to use an ABS model to replicate the assumptions and results of two previous MRSA studies adopting mathematical compartmental models. The first study (Cooper et al. 1999), which focused on a single hospital unit, explicitly considered both patients and HCWs and classified them as either susceptible or colonised. The model assumed that the only transmission route of MRSA is patient-to-patient cross transmission via transiently colonised HCWs. Other main assumptions of the model include constant patient population, 100% bed occupancy, fixed detection rate, immediate removal of detected colonised patients and 100% hand-washing efficacy. An ABS model was built in Anylogic® to replicate the assumptions of the model. Applying the model, the same model experimentations were performed with the same input parameter values of the original model. In all scenarios tested, which include changing the transmissibility, changing the probability of colonisation on admission and changing the detection rate, the results of the ABS model closely matched that of the original mathematical model.
The second study (Robotham et al. 2006), which represented the whole hospital and its community, only considered patients and divided them into eight different compartments that included susceptible patients, undetected colonised patients, detected colonised patients who are isolated, detected colonised patients who are not isolated and other four compartments representing people in the community with different MRSA colonisation status (susceptible or colonised) and readmission rate (high or low). The model did not assume any explicit MRSA transmission route and infection transmission was assumed to follow the mass action assumption. Other
Agent-based Simulation Model of MRSA
and colonised patients can be decolonised with a natural recovery rate. An ABS model was built in Anylogic® to replicate the assumptions of the model. Due to the large community population size (around 170,000 people), only patients in the hospital (1000 patients) were represented individually as agents while the different groups of people in the community were represented as integer variables. The ABS model and the original model were compared with the same input parameter values. In all the scenarios tested, the results of the ABS model closely matched that of the mathematical model.
The main purpose of the two pilot models was to demonstrate that ABS can be at least as good as existing mathematical compartmental models to describe and study MRSA transmission dynamics even without further exploring the distinctive features and the identified relative advantages of ABS. Another aim was to test and get familiar with the software, Anylogic®, before it was used to develop the full-scaled MRSA model. Appendix B gives a detailed description of the two test ABS models and the comparisons with the mathematical models.