This article is more than 5 years old. Information may no longer be current.
Decision tree method addresses antimicrobial resistance in gram-negative bacteria
Click Here to Manage Email Alerts
A decision tree method can help clinicians adjust treatments to address possible antimicrobial resistance, according to researchers.
Decision trees — an approach to predicting whether patients develop resistance to antibiotics and tailored specifically to each institution — can aid clinicians in prescribing the drug or drug combination needed to treat patients with bloodstream infections caused by gram-negative bacteria (GNB), they wrote in Clinical Infectious Diseases.
“This methodology can be relatively easily applied to derive local decision trees,” researcher M. Cristina Vazquez Guillamet, MD, of the University of New Mexico division of pulmonary, critical care and sleep medicine, and colleagues wrote. “These findings provide a framework for how empiric antibiotics can be tailored according to decision tree patient clusters.”
The researchers analyzed the records of 1,618 patients with sepsis or septic shock treated at Barnes-Jewish Hospital in St. Louis between 2008 and 2015. All patients had bloodstream infections caused by gram-negative bacteria.
Using isolates from the patients, the researchers measured the rates of antimicrobial resistance to piperacillin-tazobactam, cefepime and meropenem — representing the “most common empiric antibiotic choices for the treatment of suspected GNB infection” at their institution, the researchers said.
They found that 462 isolates (28.6%) were resistant to piperacillin-tazobactam, 352 (21.8%) were resistant to cefepime and 138 (8.5%) were resistant to meropenem. In addition, 153 isolates (9.5%) were resistant to both piperacillin-tazobactam and cefepime, and 106 (6.6%) were resistant to all three drugs.
Three factors —nursing home residence, transfer from another hospital and antibiotic treatment within 30 days before admission — were associated with increased risk of resistance to all three antibiotics.
Resistance to piperacillin-tazobactam was more than two and a half times more likely after admission from a nursing home (OR = 2.55; 95% CI, 1.70-3.82), and the increase in risk was also significant for the latter two factors (OR = 1.77; 95% CI, 1.30-2.42 and OR = 1.62; 95% CI, 1.24-2.10, respectively).
The increased risk for resistance to cefepime was similar across the three factors (OR = 1.78; 95% CI, 1.18-2.68; OR = 1.64; 95% CI, 1.22-2.21; and OR = 1.54; 95% CI, 1.20-1.99, respectively). The increased risk was greatest for meropenem (OR= 2.62; 95% CI, 1.38-4.96; OR = 2.55; 95% CI, 1.56-4.15; and OR = 2.13; 95% CI, 1.26-3.6, respectively).
Pseudomonas spp. and Acinetobacter spp. were most heavily linked to resistance to meropenem (OR = 6.65; 95% CI, 4.37-10.10, for either).
To analyze decision trees, the researchers devised algorithms factoring in variables including central venous catheters, mechanical ventilation and surgery, among others. In all, they identified two patient clusters with low risk for resistance to meropenem and four with high risk for resistance to all three antibiotics. Examples included a cluster of 233 patients who had central venous catheters and were admitted from a nursing home. With both of those variables present, these patients had a 47.2% probability of having an isolate with resistance to piperacillin-tazobactam.
Another cluster consisting of 173 patients with mechanical ventilation, septic shock and a central venous catheter had a 37.6% probability of resistance to cefepime, the researchers said.
The decision trees can provide guidance on which drugs or drug combinations, and in what quantities, should be used when antimicrobial resistance is a possibility.
“Combining user-friendly clinical decision trees and multivariable logistic regression models may offer the best opportunities for hospitals to derive local models to help with antimicrobial prescription,” they wrote. – by Joe Green
Disclosure: The researchers report no relevant financial disclosures.
Perspective
Back to Top
As Guillamet and colleagues note, increasing rates of antibiotic resistance put clinicians in a difficult position when selecting empiric therapy, particularly for critically ill patients, in whom there is little margin for error.
It can be tempting to use the carbapenems in these situations because they offer assurance of coverage for multidrug-resistant organisms. However, as discussed in the paper, this is unnecessary in most patients and may lead to additional increases in resistance.
The modeling strategy employed in this study to identify patients at highest risk for resistance to piperacillin-tazobactam, cefepime and meropenem therefore has important clinical implications. The authors use regression and decision tree analysis to create divisions of patients into low-, intermediate- and high-risk categories, with subsequent proposed algorithms for antibiotic selection, which they appropriately note may require validation in individual institutions.
Different or additional antibiotics might be selected for comparison depending on an institutional antibiogram, and the variables that are significant in regression analysis and creation of risk clusters may differ depending on patient population, hospital size and location. Implementation of this type of stratification within a health care institution would also require a consensus regarding what constitutes an acceptable resistance risk level.
For example, institutions might have different levels of comfort with what resistance rate should be designated low as opposed to intermediate risk when constructing their clinical decision trees, so this would require multidisciplinary collaborative discussion.
Finally, the authors mention that they were able to consider only prior antibiotic use data within their institution, which may underestimate risk in some patients. Improving methods for electronic record communication between health care institutions, including long–term care facilities, about recent antibiotic prescriptions might aid in more accurate modeling of risk.
Click Here to Manage Email Alerts