Reviewed by: David Cluck, PharmD; East Tennessee State University
Bottom Line: Prediction models that stratify patients with sepsis and risk of resistant gram-negative bacilli (GNB) infections often drive treatment guidance however this stratification poorly extrapolates to other institutions. This study encourages institutions to develop their own risk stratification models to better identify patients at risk for resistant GNB.1 The findings are juxtaposed with a recent publication utilizing artificial intelligence-based software (AI) for selection of empiric and targeted antimicrobial therapy.2 In short, use of AI will expand, at the very least providing real-time integrated clinical guidance at the facility level.
Vazquez Guillamet and colleagues hypothesized that differing patient case mixes and gram-negative bacilli (GNB) resistance rates would explain the variable performance of resistant GNB prediction models across hospitals.1 Case mix indices represent the average diagnosis relative weight or proportion for each included hospital. To test these hypotheses, investigators developed deep learning models based on artificial neural networks to model complex medical data and to project levels of antimicrobial resistance in GNB responsible for community-onset and hospital-onset sepsis. Adult patients admitted from January 2016 and October 2021 with sepsis diagnosis from 10 acute-care hospitals in rural and urban areas across Missouri and Illinois (Barnes-Jewish Hospitals) were retrospectively analyzed. Inclusion criteria included blood cultures indicating sepsis, having received 4 days of antibiotic treatment, and having organ dysfunction (vasopressor use, mechanical ventilation, increased creatinine or bilirubin levels, and thrombocytopenia). A total of 39,893 patients were included representing 85,238 sepsis diagnoses.
The outcome of interest was GNB antimicrobial resistance level in community-onset and hospital-onset sepsis episodes categorized to replicate choices made in clinical practice. Culture results were stratified for ceftriaxone-susceptible GNB (SS), ceftriaxone-resistant but cefepime-susceptible GNB (RS), and ceftriaxone- and cefepime-resistant GNB (RR). The models were tested across hospitals and patient subgroups. Models were assessed using area under the receiver operating characteristic curve (AUROC) and area under precision recall curve (AUPRC). The use of AUPRC evaluated the positive predictive value over all sensitivity thresholds allowing assessment of the model to project resistance.
Findings from this study were not entirely different than what most institutions are likely to encounter. RS and RR isolates were more common in hospital-onset sepsis compared with community-acquired sepsis (RS: 2389 episodes [5.7%] vs 1667 episodes [3.9%]; RR: 1626 [3.9%] vs 796 [1.8%]). History of previous infections in conjunction with history of resistant isolates were more common in the RS and RR groups in community-onset sepsis. RR isolates were cultured from urinary specimens in community-onset sepsis, and from respiratory specimens in hospital-onset sepsis. Moreover, in community-onset sepsis, 375 RR episodes (47.1%), 420 RS episodes (25.2%) and 3483 of 40,744 (8.5%) SS episodes were among patients with resistance to antimicrobial drugs (P < .001). The authors concluded risk stratification model performance varied widely across participating hospitals and patient subgroups, but did demonstrate correlation with local resistance patterns.
References: