Determinants of Deescalation Failure in Critically Ill Patients with Sepsis: A Prospective Cohort Study

Introduction. Deescalation refers to either discontinuation or a step-down of antimicrobials. Despite strong recommendations in the Surviving Sepsis Guidelines (2012) to deescalate, actual practices can vary. Our objective was to identify variables that are associated with deescalation failure. Methods. In this prospective study of patients with sepsis/septic shock, patients were categorized into 4 groups based on antibiotic administration: no change in antibiotics, deescalation, escalation (where antibiotics were changed to those with a broader spectrum of antimicrobial coverage), or mixed changes (where both escalation to a broader spectrum of coverage and discontinuation of antibiotics were carried out). Results. 395 patients were studied; mean APACHE II score was 24 ± 7.8. Antimicrobial deescalation occurred in 189 (48%) patients; no changes were made in 156 (39%) patients. On multivariate regression analysis, failure to deescalate was significantly predicted by hematologic malignancy OR 3.3 (95% CI 1.4–7.4) p < 0.004, fungal sepsis OR 2.7 (95% CI 1.2–5.8) p = 0.011, multidrug resistance OR 2.9 (95% CI 1.4–6.0) p = 0.003, baseline serum procalcitonin OR 1.01 (95% CI 1.003–1.016) p = 0.002, and SAPS II scores OR 1.01 (95% CI 1.004–1.02) p = 0.006. Conclusions. Current deescalation practices reflect physician reluctance when dealing with complicated, sicker patients or with drug-resistance or fungal sepsis. Integrating an antibiotic stewardship program may increase physician confidence and provide support towards increasing deescalation rates.


Introduction
Early administration of broad-spectrum, empiric antimicrobial therapy reduces mortality and improves outcomes in patients with severe sepsis and septic shock. However, broadspectrum therapy favors the emergence of drug-resistance and adds excessively to the costs of care. Deescalation refers to a strategy whereby clinicians either discontinue or change to a narrower spectrum antimicrobial drug and is usually carried out after culture results become available. The objective of this study was to identify variables associated with deescalation failure.

Methods
This study is reported following the STROBE statement checklist for observational studies [1].  of Helsinki and its later amendments. No individual patient data is presented.

Study Design and Setting.
In this prospective, cohort study we reviewed consecutive adult (>14 years) patients admitted to the intensive care unit (ICU) with a diagnosis of sepsis or septic shock. The period of study was from 1st January 2013 to 1st January 2014. Patients who were not for resuscitation (DNR) or were expected to die within 48 hours were excluded.

Operational Definitions.
Antibiotic therapy was considered appropriate based on in vitro sensitivity on culture. On day 7 after ICU admission, we categorized patients into four groups based on antibiotic administration: no change in antibiotics, deescalation (defined as stopping or changing to a narrower spectrum antibiotic), escalation (where antibiotics were changed to those with a broader spectrum of antimicrobial coverage), or mixed changes (where both escalation to a broader spectrum of coverage and discontinuation of antibiotics were carried out).

Statistical
Analysis. Continuous data was tested for normality; measures of central tendency were reported as means ± standard deviations (SD) and compared using Student'stest for normally distributed variables and reported as medians (interquartile range, IQR) and compared using the Mann-Whitney test for skewed data. Categorical variables were compared using the 2 test or the Fisher Exact test for < 5. Logistic regression analysis was performed to determine the predictive ability of variables for antibiotic deescalation. Univariate and multivariate techniques were used, and, for multivariate regression, a backward mode with a threshold 0.10 was used for elimination. Multivariate associations were reported as odds ratios, Exp(B) with 95% confidence intervals. A two-sided value of < 0.05 was considered as statistically significant. All analyses were carried out using IBM SPSS version 22.0.
Please refer to Table 2 for frequencies of all isolates. Empiric therapy was appropriate in 57% cases. The median ICU length of stay was 6 days (IQR 4-43) with a 28day survival rate of 71% (281 patients).
Antimicrobial deescalation was carried out in 189 (48%) patients; in 156 (39%) patients no changes in the antimicrobial regimen were made; 42 (11%) patients had their antimicrobial coverage escalated and in 8 (2%) patients mixed changes were made.
Please refer to Table 3 for differences in patient characteristics by final antibiotic grouping.
In a comparison of patients that were deescalated compared to patients "not deescalated" (combination of groups with escalation, no changes, or both escalation and deescalation, i.e., mixed changes), rates of malignancy, multidrug resistant (MDR) organisms, fungal sepsis, chronic organ failure (renal, liver), baseline APACHE II, SAPS II, and serum procalcitonin were significantly different. Deescalation rates were not significantly different between patients with positive cultures and those with negative cultures or single versus multiple positive culture sites or when the patient continued to be vasopressor-dependent. Deescalation was associated with a significantly lower ICU mortality compared to patients not deescalated, 27  = 0.011, and MDR isolates OR 2.9 (95% CI 1.4-6.0) = 0.003 remained significant predictors for no deescalation.
Please refer to Table 4 showing multivariate regression analysis to indicate variables associated with no deescalation of antimicrobials.

Discussion
In this prospective study of critically ill, septic patients antimicrobial deescalation was carried out in less than half of all patients, with higher baseline procalcitonin levels, greater organ dysfunction scores, comorbid hematologic malignancy, isolation of drug-resistant bacteria, and fungal organisms identified as independent predictors of failure to deescalate. The morbidity and costs of continued broad-spectrum antimicrobials and the safety of deescalation are now well established in the medical literature. A deescalation strategy has not been shown to be harmful in patients with varied immune statuses or systemic or limited infections or in fungal septicemia [7,11,12,[16][17][18][19][20][21][22][23][24][25] and in fact may even exert a protective effect as reported by Lee et al. [11] and Garnacho-Montero et al. [9]. Please see Table 5 for a summary of recent studies on antibiotic deescalation.
Despite reports of benefit, deescalation remains variably practiced with rates from 10 to 60% [26]. Please refer to Table 5. In our cohort of multidisciplinary critically ill patients with sepsis and shock, antimicrobial deescalation was carried out in 48% patients, which is comparable to that reported by other investigators.
The real question then is when are physicians less likely to deescalate? Recent studies have shown that antibiotic deescalation becomes less likely with severe, complicated infections and drug-resistance and when initial antibiotic therapy is inadequate. Please refer to Table 5. In our study, deescalation failure was predicted by the isolation of drugresistant bacteria and fungal organisms, greater severity of illness as demonstrated by higher initial organ dysfunction scores, underlying hematologic malignancy, and a procalcitonin level that may suggest a greater bacterial load.
What then appears to be a common theme here is that physicians are uncomfortable deescalating antimicrobials when faced with sicker patients with a higher possibility of complications. Antimicrobial stewardship is a strategy that employs availability of either an infectious disease specialist and/or a clinical pharmacist to assist in decision-making at the bedside. Stewardship programs have been shown to successfully reduce resistance patterns, reduce antibiotic usage, and reduce costs without increasing adverse outcomes [27][28][29][30][31][32][33][34]. Our study confirms the results of others that there is a real need for and potential benefits of implementing antimicrobial stewardship programs across all areas where broad-spectrum antimicrobials are utilized. Whether they are specialty driven or pharmacy-led should be tailored to the resources available to individual centers.
This study's strengths are its large numbers of patients and the generalizability of our results to other ICUs. We have a varied case-mix from our surgical and medical ICUs with causative organisms that are similar to those isolated from most ICUs. The limitation of our results is that this is a singlecenter study.
Assisting the stewardship model is a recent publication from de Jong and colleagues [35] where, in a controlled trial in 15 Netherland hospitals, ICU admissions were randomized to usual care versus antibiotic deescalation once procalcitonin levels decreased by 80% or more of its peak value or to 0.5 g/L or lower. Mortality was significantly lower in the procalcitonin-guided group, between-group absolute difference 5.4% (95% CI 1.2-9.5, = 0.0122). Therefore, procalcitonin absolute levels and patterns may assist bedside decision-making incorporated into an antibiotic stewardship program.

Conclusions
Current deescalation practices reflect physician reluctance when dealing with complicated, sicker patients or with drug-resistance or fungal sepsis. Integrating an antibiotic 6 Critical Care Research and Practice stewardship program may increase physician confidence and provide support towards increasing deescalation rates.

Competing Interests
There are no competing interests to declare.