Impact of prior probabilities of MRSA as an infectious agent on the accuracy of the emerging molecular diagnostic tests: a model simulation

Objectives: Traditional microbiology identification takes 48-72 h to complete. This lag forces clinicians to rely on broad-spectrum empiric coverage. To address this gap, manufacturers are developing rapid molecular diagnostics (RMD). We hypothesised that RMD's accuracy is more dependent upon population risk of harbouring the culprit pathogen than to their sensitivity and specificity. Design: A mathematical model. Setting and participants: We used the range of risks (5-50%) for methicillin-resistant Staphylococcus aureus (MRSA) among patients hospitalised with complicated skin and skin structure infections (cSSSI), pneumonia or sepsis. Main outcome measures: We modelled the impact of changing a test's characteristics on its positive (PPV) and negative (NPV) predictive values, and hence the risk of overtreatment or undertreatment, within strata of an organism's population prevalence. MRSA diagnostics provided assumptions for the test sensitivity and specificity (95-99%). Scenarios with low sensitivity and specificity (90%), and best-case and worst-case scenarios normalised to the annual universe of populations of interest, were examined. Results: With a low prevalence (5%) and high test specificity, the PPV was 84%. Conversely, with 50% prevalence and 95% test specificity the PPV rose to >= 95%. Even when the test's specificity and sensitivity were both 90%, in a high-risk population both PPV and NPV were similar to 90%. In the worst-case scenario, 150 000 patients with cSSSI, pneumonia and sepsis annually were at risk for inappropriate treatment, 91% of these at risk for over-treatment. In the best-case scenario, 81% of 18 000 patients at risk for inappropriate coverage were subject to overtreatment. Conclusions: Although promising for limiting exposure to excessive antimicrobial coverage, RMDs alone will not solve the issue of inappropriate, and particularly overtreatment. Increasing pretest probability as a strategy to minimise antibiotic abuse results in more accurate patient classification than does developing a test with near-perfect characteristics. The healthcare community must build robust evidence and information technology infrastructure to guide appropriate use of such testing.