Machine says go, doctor says no: an ecological momentary assessment analysis examining clinicians' perceptions of, and their antibiotic prescribing behaviour when using rapid molecular diagnostic tests in intensive care.

BACKGROUND: Rapid molecular diagnostics such as the BioFire FilmArray Pneumonia Panel (the Pneumonia Panel) can improve antibiotic stewardship by supporting doctors to make more targeted antibiotic prescribing decisions faster compared to routine microbiology. However, factors influencing how these test results translate to individual prescribing decisions are poorly understood. The INHALE randomised controlled trial (RCT) evaluated the application of the Pneumonia Panel to manage suspected hospital-acquired and ventilator-associated pneumonias (HAP/VAP) in English intensive care unit (ICU) patients. This behavioural study examines clinicians perceived and actual antibiotic prescribing behaviour, within the INHALE RCT.

METHODS: Clinicians treating ICU patients completed brief questionnaires within 24 h of their prescribing decision for intervention-arm cases (N = 159), exploring factors influencing their decision and perceptions about the test results. Actual prescribing behaviour was extracted from the trial database. A 4-block hierarchical logistic regression identified predictors of prescriptions being consistent with Pneumonia Panel results.

RESULTS: 65% (N = 104) of prescribing decisions were consistent with Pneumonia Panel results. The test result itself was a dominant factor: 88% (N = 98) of decisions were consistent when results were positive (pathogens found). However, only 13% (N = 6) of decisions were consistent when no pathogens were detected. Consequently, clinicians were often reluctant to eschew initial antibiotics or de-escalate early where appropriate, 'erring on the side of caution'. Clinicians perceptions, specifically the speed of results, concurrent antibiotic treatment, the patient having additional confirmed evidence of infection, and believing the patient is unlikely to have a non-respiratory infection predicted prescribing decisions being aligned with test results (all p < .05).

CONCLUSIONS: Findings have implications for the roll-out of rapid diagnostics in practice, particularly regarding the management of negative results. Implementation strategies need to be behaviourally intelligent, connecting with how clinicians think and behave.