Researchers: Mical Paul¹, Noa Eliakim Raz²

Background: Antibiotic treatment at end of life (EOL) is futile at times and contributes to antibiotic resistance development.

Objectives: To introduce EOL prediction into a decision support system for antibiotic treatment (TREAT) and assess its effects on the system’s advice.

Method: A prediction score for non-infection-related 30-day mortality (NIRM30) was developed using a retrospective database of elderly medical inpatients in Rambam Health Care Campus (H1). The score was validated on two prospectively collected cohorts of medical inpatients suspected of or diagnosed with bacterial infections in H1 and Rabin Medical Center, Beilinson Hospital (H2). NIRM30 was used to tailor the survival benefit of antibiotic treatment in the DSS to patient’s life expectancy. We compared TREAT’s baseline advice to its advice with NIRM30 and to physicians’ prescriptions. We assessed HCWs’ attitude and opinions through a survey, the actual practice of antibiotic prescription at EOL on the prospective cohorts, and acceptability of TREAT’s advice through a survey of infectious diseases experts.

Findings: NIRM30 performed well with areas under the receiver-operating-characteristics curve of >0.8 in the derivation (N=11,884 patients) and both validation cohorts (N=1,974 in H1 and 3,817 in H2). Applied to TREAT, NIRM30 resulted in treatment restriction for 46/463 (9.9%) of patients. Mean ecological antibiotic costs, increasing with the spectrum of coverage of the antibiotic, were 920$ for physician’s prescriptions, 346$ for TREAT at baseline and 284$ for TREAT with NIRM30. While antibiotics were prescribed to 368/413 (89.1%) of patients who died within 30 days, HCWs expressed an opinion that antibiotics at EOL should be reserved as invasive interventions. Experts were more conservative than TREAT with NIRM30.

Conclusions: TREAT recommends significantly narrower-spectrum and less antibiotics than physicians. Treatment restriction through a formal model has the potential to assist clinicians.

Recommendations: We recommend TREAT’s use and a public debate on antibiotic treatment at EOL.

Research end date: 11/2017