Objective: To use transmission dynamic models to assess the cost-effectiveness of alternative MRSA infection control strategies in intensive care units (ICUs), with a particular emphasis on comparing currently available screening technologies.
Methods: We developed a dynamic, stochastic, individual-based model of MRSA transmission in ICUs to enable comparison of screening and isolation policies (screening technologies evaluated included: culture on conventional agar, culture on chromogenic agar and polymerase chain reaction (PCR)-based systems). We synthesized evidence from multiple sources (data, literature and expert opinion) to parameterize the model, including uncertainty in each of our parameter estimates. Incremental changes in resource use and health benefits for each of the alternatives were evaluated in different settings (e.g. prevalence, ICU size).
Results: In an ICU with an MRSA admission prevalence of 5%, a strategy of targeted screening of high risk patients using chromogenic agar, followed by isolation of patients identified as MRSA positive, was the most cost-effective option with a cost of £4800/quality adjusted life year (QALY) gained (compared to the baseline strategy). NHS decision makers tend to consider programmes cost-effective if the incremental cost-effectiveness ratio (ICER) is beneath a £20,000-£30,000 threshold. Screening all patients on admission using PCR, despite reducing the number of un-isolated MRSA positive bed days by more than half (compared to chromogenic agar), had an ICER of ~£22,000/QALY. Effectiveness of isolation was, by far, the most influential parameter. Using 5 percentile and 95 percentile estimates of effectiveness gave monetary net benefits of implementing PCR screening for all admissions coupled with isolation ranging from a loss of £400 per admission to a benefit of £100 per admission. Moreover, when uncertainty in all parameter estimates was propagated through the model, the probability of making the right decision (when choosing between 12 strategies) was low, with an error probability of ~0.9.
Conclusions: These models allow comparison of many interventions, taking into account all available evidence, and thus provide a useful guide for policy makers. This research demonstrates the need for more data to reduce uncertainty in the parameter estimates, such as the effectiveness of interventions, to enable cost-effective decision making. By allowing quantification of the importance of key uncertainties, these models provide a rational basis for setting future research priorities.