Effectiveness and Safety of Procalcitonin-Guided Antibiotic Therapy in Lower Respiratory Tract Infections in “Real Life”:  An International, Multicenter Poststudy Survey (ProREAL) FREE

The efficacy, feasibility, and safety of procalcitonin (PCT)-guided antibiotic stewardship in lower respiratory tract infections (LRTIs) and sepsis has been documented in several randomized controlled trials (RCTs).^1- 12 Procalcitonin-guided antibiotic stewardship reduced initial antibiotic prescription rates by 40% to 50% in patients with LRTI presenting to the emergency department⁵ and by 70% to 80% in ambulatory patients presenting to their general physician¹³ and reduced total antibiotic exposure in community-acquired pneumonia (CAP) by 40% to 50%.⁴ The main effect was by discouraging antibiotic initiation in bronchitis and acute exacerbation of chronic obstructive pulmonary disease and by shortening antibiotic courses in CAP without increased rates of adverse outcomes.^1- 7

Most evidence regarding PCT-guided antibiotic stewardship derives from RCTs, with little data outside of controlled study conditions. Results from RCTs may not unconditionally be generalized because of exclusion criteria or nonenrollment¹⁴ and are frequently not adequately implemented in daily practice.

In this context, we previously performed a single-center poststudy surveillance¹⁵ to investigate the real-life effectiveness of PCT-guided antibiotic stewardship after completion of the ProHOSP RCT.⁷ Median duration of antibiotic treatment was significantly shorter in this survey than in the standard of care ProHOSP control group (6 vs 7 days).¹⁵ Compliance with the prespecified algorithm was excellent (90%), without differences in adverse outcomes. However, physicians had previous experience with the algorithm, which potentially increased adherence.

To assess whether these results also apply to different health care settings, we investigated the effects of PCT guidance on inpatients and outpatients in hospitals and general physician offices in 3 countries with diverse antibiotic-prescribing cultures.

This prospective, observational, international, multicenter, quality control survey (ProREAL) monitored initiation and duration of antibiotic therapy, adherence to the published PCT algorithm (Figure 1), and outcome of patients with community-acquired LRTI in centers in Switzerland (n = 10), France (n = 3), and the United States (n = 1) between September 12, 2009, and February 28, 2011. Three of the Swiss hospitals had previously participated in the ProHOSP study^2,7 and were considered algorithm experienced; all others were considered algorithm naive.

Measurement of PCT levels was recommended in all patients using highly sensitive immunoassays (Kryptor [BRAHMS AG] or miniVidas [bioMérieux]). Both assays provide similar PCT results.^16- 17

Diagnostic workup and treatment were left to the discretion of the treating physicians. Consecutive patients with LRTI who were seen at an emergency department or a physician's office were registered by the physician on duty on a password-secured website that displayed the PCT algorithm (Figure 1).^3,7 In the US center, study nurses screened hospital admission records and enrolled patients within 24 hours. There were no exclusion criteria. Physicians and study personnel were instructed in initial face-to-face 1-hour seminars. Throughout the study, weekly e-mails were sent to local coordinators that contained current enrollment status, encountered problems, and suggested solutions. If necessary, local coordinators or individual physicians were contacted by e-mail or telephone.

Bronchitis, LRTI, acute exacerbation of chronic obstructive pulmonary disease, and CAP were defined according to guidelines.⁷ We defined compliance as antibiotic treatment that was initiated and discontinued in accord with the PCT cutoff ranges or, if the PCT levels suggested no antibiotic therapy, with the predefined overruling criteria. Overruling of the algorithm was possible if any of the following criteria were met: admission to the intensive care unit, life-threatening comorbidity, severe immunosuppression, chronic or non–respiratory tract infection requiring antibiotics, complications (abscess and empyema), difficult-to-treat organisms, or high clinical severity scores (Figure 1). Conversely, noncompliance was defined if antibiotic therapy was initiated or not discontinued despite low PCT levels in the absence of these predefined overruling criteria, that is, if the algorithm was overruled based only on clinical judgment.

On presentation and during hospitalization, baseline characteristics, comorbidities, clinical severity scores (CURB-65 [confusion, serum urea nitrogen, respiratory rate, blood pressure, and age ≥65 years]¹⁸ and the Pneumonia Severity Index [PSI]¹⁹), course of PCT levels and antibiotic therapy, length of hospital stay, and complications were prospectively collected and entered by treating physicians or study personnel (US site) on the website. Adverse medical outcomes, including all-cause and LRTI-related hospital mortality, intensive care unit admission, disease-related complications (ie, empyema, acute respiratory distress syndrome, shock, and requirement for a ventilator or vasopressor support), and recurrence, were obtained from hospital records and were documented on the website. Thirty days after enrollment, telephone interviews were performed by study personnel to identify adverse outcomes, including death, recurrence, rehospitalization, requirement for additional or repeated antibiotic therapy, and antibiotic-related adverse effects. Any adverse outcomes and repeated antibiotic courses were confirmed with hospital medical records or with general physicians. Informed consent was obtained from all the patients at the US site but was waived by Swiss and French ethics committees.

The primary end point was the total duration of antibiotic treatment within 30 days. To calculate antibiotic use duration, we divided the number of total doses by the number of daily doses. In case of combination therapy, antibiotic use duration was determined by the duration of the antibiotic that was given longest; doses of different but simultaneously given antibiotics were not summed. Secondary end points were duration of antibiotic therapy at the index presentation, adherence to the PCT algorithm, and adverse medical outcomes in the index hospitalization.

Discrete variables are expressed as number (percentage) and continuous variables as median (interquartile range), unless stated otherwise. Frequency comparison was performed using the χ² test and 2-group comparisons by the t test. All the regression models were adjusted for potential confounders, including LRTI diagnosis, country, inpatient or outpatient treatment, sex, PSI score, previous experience with the algorithm, multilobar involvement in CAP, and different comorbidities (diabetes mellitus, malignant disease, cerebrovascular disease, congestive heart failure, renal and hepatic disease, chronic lung disease, and peripheral arterial disease). A multivariable generalized linear model was calculated to assess the effect of algorithm compliance and other independent predictors of antibiotic therapy duration. Logistic regression models were performed to investigate the risk of algorithm compliance, with adverse outcomes defined as in-hospital complications (including acute respiratory distress syndrome, empyema, requirement for vasopressors or mechanical ventilation, admission to the intensive care unit, and death), any 30-day complication (any of the in-hospital complications, death within 30 days, recurrence of LRTI, or rehospitalization), or antibiotic adverse effects (including Clostridium difficile –associated diarrhea, nausea, vomiting, rash, allergic reactions, abdominal pain, and candidiasis). For each adverse outcome, 2 multivariable analyses were performed to assess the safety of initially withholding antibiotic therapy based on low PCT levels (≤0.25 ng/L) and to assess the safety of cessation of antibiotic use after a decrease in the PCT concentration (≤0.25 ng/L or a ≥80% decrease from maximum). Statistical analyses were performed using commercially available software programs (SAS, version 9.2 [SAS Institute, Inc] and Epi Info 2002 [Centers for Disease Control and Prevention]). All the testing was 2-tailed, and P < .05 was considered statistically significant.

Of 1810 patients enrolled, 1759 had complete data sets from the index visit (Swiss: 1361; US: 295; and French: 103). Of 1520 patients (86.4%) with a final diagnosis of LRTI, 1425 (93.8%) had sufficient follow-up information at day 30. Patients with CAP who were overruled correctly according to the predefined criteria in the algorithm were sicker than those treated strictly according to PCT values (PSI class: 2.8 vs 2.2; CURB-65 score: 1.5 vs 1.2; P = .002), who, in turn, were sicker than those who were not correctly overruled, that is, “not compliant” with the algorithm (PSI class: 2.2 vs 1.7; CURB-65 score: 1.2 vs 0.8; P < .001). Baseline characteristics were similar to those of patients in the former ProHOSP study⁷ in the 3 centers participating in both studies except for a higher severity of patients with CAP in the ProHOSP group (Table 1).

Of 1520 patients with LRTIs, 1208 (79.5%) received at least 1 antibiotic dose. The overall mean duration of antibiotic therapy was 6.9 days (inpatients, 7.3 days; outpatients, 3.5 days; P < .001).

After multivariate adjustment, the predicted mean duration of antibiotic therapy in LRTI was shorter in algorithm-experienced compared with algorithm-naive centers (6.0 vs 6.7 days; absolute difference in days [95% CI], −0.71 [−1.25 to −0.17]; P < .01) and if the PCT algorithm was followed compared with if not (5.9 vs 7.4 days; absolute difference in days [95% CI], −1.51 [−2.04 to −0.98]; P < .001). Other risk factors for longer antibiotic use duration were CAP (vs bronchitis), renal insufficiency, treatment in France (vs Switzerland), in-hospital (vs ambulatory) treatment, and higher PSI classes. More detailed results of the multivariable analysis are displayed in Table 2.

Antibiotic therapy was longer when blood cultures yielded positive results (adjusted regression coefficient, 1.06; 95% CI, 0.13 to 1.99; P = .03) and in patients with positive sputum culture results (adjusted regression coefficient, 2.03; 95% CI, 1.53 to 3.08; P < .001).

Comparing results with those of historic controls, antibiotic therapy duration during ProREAL (6.2 days; 95% CI, 5.8 to 6.7 days) was longer compared with that of the ProHOSP PCT group (5.0 days; 95% CI, 4.4 to 5.6 days; P = .001) but shorter than in the ProHOSP control group (7.9 days; 95% CI, 7.3 to 8.4 days; P < .001) for the 3 centers that participated in both studies (Figure 2).

During the entire index presentation (ie, clinic visit, emergency department visit, or entire hospitalization), overall algorithm compliance was 68.2%. Antibiotic therapy followed PCT cutoff ranges on initial presentation in 72.4% of patients and predefined overruling criteria in 8.6% (Figure 1), resulting in overall algorithm compliance of 81.0%. The most important overruling reasons were high clinical severity and respiratory or hemodynamic instability. Overruling due to clinical judgment without prespecified reason occurred in 19.0% of patients.

Algorithm compliance was higher in algorithm-experienced than algorithm-naive centers (P < .001), higher in outpatients than in inpatients (P < .001), higher in Switzerland and France than in the US center (P < .001), and highest in bronchitis and influenza (Figure 3).

In-hospital mortality was 5.2%; 10.5% of patients were admitted to the intensive care unit. Overall, the in-hospital complication rate was 19.3%, 30-day mortality was 7.6%, and the recurrence rate was 6.8%. After controlling for comorbidities and severity of disease, withholding antibiotics at initial presentation in patients with low PCT values (≤0.25 ng/L) was not associated with in-hospital complications (adjusted odds ratio [OR], 0.63; 95% CI, 0.30 to 1.31; P = .22) or with any complications within 30 days (adjusted OR, 0.83; 95% CI, 0.44 to 1.55; P = .56) but with a reduction in the risk of antibiotic-related adverse events (adjusted OR, 0.23; 95% CI, 0.06 to 0.91; P = .04) in multivariable models (Table 3).

Early cessation of antibiotic therapy according to PCT levels (if the PCT level decreased to ≤0.25 ng/L or decreased ≥80% from its maximum) was not associated with risk of in-hospital complications (adjusted OR, 1.10; 95% CI, 0.61 to 1.97; P = .76) or any complication within 30 days (adjusted OR, 0.61; 95% CI, 0.36 to 1.04; P = .07) in multivariable logistic models (Table 4).

In this first international multicenter poststudy survey of PCT-guided antibiotic therapy for LRTI, the duration of antibiotic therapy was significantly shortened if the algorithm was followed. The mean overall reduction was approximately 20% (from 7.4 to 5.9 days), which is in the range of that (25%) of a randomized trial using a single PCT value at initial presentation.²⁰ A single PCT measurement in the primary care setting was recently shown to safely reduce the antibiotic treatment rate by 41.6%.¹³

Multivariable analyses controlling for severity of illness and other confounders confirmed that neither withholding antibiotics on hospital admission in patients with low PCT levels nor discontinuation of antibiotic therapy in patients with appropriate decreases in PCT levels was associated with an increased risk of mortality or of developing other adverse events during hospitalization or over 30 days. The risk of antibiotic-associated adverse events was significantly lower if antibiotics were withheld according to the algorithm. This confirms the safety of PCT-guided antibiotic stewardship outside study conditions and is in accordance with all RCTs and meta-analyses, where there was no difference in outcomes despite strict implementation of the algorithm.^21- 22

In a single-center pilot survey,¹⁵ we had detected similar results for the Kantonsspital Aarau, Aarau, Switzerland, with extensive previous PCT algorithm experience, where we reported 7 and 6 days of antibiotic therapy in patients with CAP and LRTIs, respectively, who were treated according to the PCT algorithm. In the single-center and the present multicenter surveys, we recruited patients with severe comorbidities and immunodeficiency who were formerly excluded from the ProHOSP study. This inclusion at least partially explains why, in the present study, the mean (unadjusted) antibiotic therapy duration in the 3 algorithm-experienced centers was 1.2 days longer than that in the ProHOSP intervention group but 1.7 days shorter than that in the ProHOSP control group. In experienced centers, antibiotic treatment duration was similar to that in the previous single-center pilot study and significantly shorter than that in algorithm-naive centers. These results indicate that PCT safely allows individualized and shortened treatment duration also in real-life conditions in different health care settings.

The finding that CAP and in-hospital treatment were predictors of longer antibiotic treatment is plausible since those patients have a higher likelihood of a bacterial etiology than do patients with bronchitis, who more frequently undergo outpatient treatment. Likewise, its association with increasing severity of illness might explain why renal insufficiency was an independent predictor of longer antibiotic treatment. Renal insufficiency does not prolong the half-life of PCT.²³ Patients with an identified cause had a longer duration of antibiotic therapy than if no causative agent was identified. The PCT levels have been associated with the presence of bacteremia^24- 26 and bacterial load,²⁶ supporting our observation and the validity of our algorithm.

The second important focus of this survey was to evaluate algorithm adherence outside of stringent study conditions in countries with considerably different antibiotic-prescribing cultures and PCT experience. The observed level of compliance is remarkable, especially considering that initial introductory seminars were short, that only 3 of the 14 participating sites were algorithm-experienced, and that 1 of the 3 countries had limited experience with PCT (the United States). In the previous survey,¹⁵ we had achieved overall algorithm compliance of 90% outside of study conditions in a hospital well used to the algorithm. Aujesky et al²⁷ found that 37.4% of patients with low PSI scores were hospitalized mainly for comorbid illnesses, reflecting the difficulty of implementing guidelines into daily clinical practice and limited confidence in clinical scores. Although the PSI algorithm is more complex than the PCT algorithm, resulting in a lower likelihood of being implemented, we assessed PCT compliance not only at presentation but throughout the entire index presentation, that is, hospitalization in inpatients.

Implementation of algorithms is more challenging with increasing severity of illness,⁹ when physicians might rely on self-perceived levels of experience. This might explain the observation of decreasing compliance from bronchitis to acute exacerbation of chronic obstructive pulmonary disease and CAP. Algorithm compliance was higher in experienced than in naive centers and in Europe than in the United States, probably reflecting a learning curve of the laboratory and the physicians. Therefore, ongoing reinforcement is necessary to increase compliance and the rational use of antibiotics, minimizing antibiotic selection pressure. This also demonstrates the importance of preexisting regional and cultural differences in antibiotic-prescribing habits between Europe and North America,^28- 29 which influence the success of such interventions. Antibiotic-prescribing patterns reflect multilayer decision making, including sociocultural, health care system, legal, and health belief factors. It, thus, is not surprising that compliance was highest in Switzerland, with the lowest antibiotic consumption and resistance rate.^30- 31

The strengths of this study are its size as the largest single study of PCT-guided antibiotic therapy and its international web-based design. The inclusion of consecutive patients of all different levels of care and experience without exclusion criteria and the relatively small implementation efforts support its generalizability. Although the study design did not permit a direct control group, comparison with historic controls from an RCT performed at the same locations supports the effectiveness of the algorithm.

A potential limitation is the discrepancy between the number of participating centers in Switzerland (n = 10), France (n = 3), and the United States (n = 1). It is, therefore, difficult to make a general conclusion for all countries based on the findings of this survey. Improved PCT assay availability, preferably as a bedside point-of-care test, and reduced assay costs are prerequisites for widespread implementation.

In conclusion, this real-life effectiveness study complements the excellent efficacy and safety record of PCT-guided antibiotic stewardship from many previous RCTs and extends those beyond the stringent study setting. We demonstrate that good compliance with the PCT algorithm is possible in real-life conditions but has to be reinforced to achieve optimal benefit. Regional and cultural differences in preexisting antibiotic-prescribing habits pose a challenge to its successful implementation that requires particular attention.