A systematic review and meta-analysis identify 22 studies that used various methods to predict colonisation with antibiotic-resistant bacteria at the time of hospital admission. The models were chosen to focus on MRSA and CPO colonisation. The “performance” of these tools varied widely, with a sensitivity of 15–100% and specificity of 46–98.6% for MRSA, and sensitivity of 30–81.3% and specificity of 79.8–99.9% for CPO. I think my main take-away from this that simple risk tools for predicting colonisation with MRSA and CPO (which are often used to determine who to test) are pretty blunt instruments. However, the more advanced tools making use of big datasets and machine learning can take us forward in predicting the risk of MRSA and CPO colonisation at the time of admission.

Most of the studies identified by the search criteria used in this article were MRSA (18/22), and mostly older studies, with 17 published before 2016. This seems a bit odd – you’d expect this field to be gathering pace not slowing down. So answers on a postcard to explain this one. Although it is true that the methods used have become more advanced over the timeframe of this review, as you’d expect.

The common risk factors across MRSA and CPO were previous admission (within 6 months) appearing in 77% of lists, recent antibiotic exposure (within 6 months) 68%, age 41%, sex 36%, diabetes 27%, intrahospital transfer 23%, and nursing home / long-term care facility 23%. And herein lies the problem. Additional risk factors identified in some studies were indwelling devices, critical care, and mechanical ventilation. The risk factors are really broad. Something like 1/3 of all hospital admissions will have had fairly recent hospitalisations, and somewhere between a 1/3 to a ½ of hospital inpatients are on antibiotics at any given time. So how do we turn these broad risk factors into really meaningful risk prediction tools?

The best performing models combined an electronic medical record (for good background demographics) with a questionnaire to capture patient-level risk-factor information, so these seem to be the best candidates to take forward for the kind of focussed prediction tools that we need to guide screening programmes and ultimately reduce the risk of transmission of antibiotic-resistant bacteria in our hospitals.