Duration of antibiotic therapy in critically ill patients: a randomized controlled trial of a C-reactive protein-based protocol versus an evidence-based best practice strategy without biomarkers

Background: The rational use of antibiotics is one of the main strategies to limit the development of bacterial resistance . We therefore sought to evaluate the effectiveness of a C reactive protein-based protocol in reducing antibiotic treatment time in critically ill patients. Methods: A randomized, open-label, controlled clinical trial conducted in two intensive care units of a university hospital in Brazil. Critically ill infected adult patients were randomly allocated to: i) intervention to receive antibiotics guided by daily monitoring of CRP levels, and ii) control to receive antibiotics according to the best practices for rational use of antibiotics. Results : 130 patients were included in the CRP (n=64) and control (n=66) groups. In the intention to treat analysis, the median duration of antibiotic therapy for the index infectious episode was 7.0 (5.0-8.8) days in the CRP and 7.0 (7.0-11.3) days in the control (p = 0.011) groups. A significant difference in the treatment time between the two groups was identified in the curve of cumulative suspension of antibiotics, with less exposure in the CRP group (p = 0.007). In the per protocol analysis, involving 59 patients in each group, the median duration of antibiotic treatment was 6.0 (5.0-8.0) days for the CRP and 7.0 (7.0- 10.0) days for the control (p = 0.011) groups. Conclusions: Daily monitoring of CRP levels may aid in the reduction of antibiotic treatment time of critically ill patients, even in a scenario of judicious use of antimicrobials. Trial Registry : Registered 2016,


Background
Increasing concerns about antimicrobial abuse and development of bacterial resistance have fueled the search for the objective and rational use of antibiotics. There is growing evidence supporting the use of shorter antibiotic courses to treat various types of infection, with clinical outcomes similar to those obtained with longer treatments 1,2,3,4,5,6 . Individualization of antibiotic treatment time has been gaining importance 7,8,9,10 . This measure prevents unnecessary exposure to antibiotics while reducing the risk of therapeutic failure in those with a late response.
Circulating inflammatory biomarkers have been used as a guide to support treatment individualization. One useful marker is procalcitonin (PCT), whose benefit in reducing antibiotic treatment time was demonstrated in several studies 11,12,13,14,15,16 , including a potential reduction in the mortality of critically ill patients 14,15,16 . Nevertheless, the high cost of PCT testing reduces its availability in some settings 10,17 . In this context, C-reactive protein (CRP) may be an reasonable lowcost alternative 10,16,18 . A recent meta-analysis demonstrated that the PCT-guided algorithms only showed a survival benefit when used in combination with CRP, along with other specificities 16 .
Nevertheless, few studies testing a CRP-guided strategy have been conducted in adult critical ill patients. A recent single center clinical trial involving patients with sepsis suggested that CRP may be a useful marker to guide antibiotic treatment time, when compared to a PCT strategy 19 .
The objective of this study was to test the impact of a decision flowchart based on CRP serum levels on the duration of antibiotic therapy in critically ill infected patients, compared to a control group treated according to the best available evidence for rational antibiotic treatment in this population. Participants and randomization All adult patients (age ≥ 18 years) admitted to the ICUs were considered for potential inclusion according to: clinical suspicion or microbiological confirmation of infection and the prospect of an ICU stay longer than 24 hours. The diagnosis of sepsis or septic shock was considered according to current definitions 21 . The exclusion criteria were: patients using antibiotics for more than 48 hours at the time of screening; severe immunosuppression; patients under full and exclusive palliative care; death expectancy for the next 24 hours; diagnose of infections known to require prolonged antibiotic therapy; and patients that underwent major surgery in the previous five days. For details of definitions and justifications of inclusion and exclusion criteria see Additional file 1.

Study design
Eligible patients were randomly assigned to one of two groups: a CRP-guided therapy or a control group. Randomization was performed individually at an allocation rate of 1: 1, using a computergenerated random number table. Due to the nature of the intervention, the investigators and the assistant physicians were aware of the group in which the patients had been included. Preparation and conduction of the study followed the recommendations of the CONSORT Statement 22 .

Intervention
The main purpose of this study was to evaluate if a strategy of CRP-guided therapy adds value to an already rational clinical practice of antibiotic use, applied to all study participants. Accordingly, in the control group, decisions about treatment time were taken according to the best evidence established in the literature for duration of antibiotic treatment 1,2,3,4,5,6 . The study protocol provided that antibiotic therapy should be discontinued in the control group when the patient reached clinical criteria for suspension, which did not include CRP levels and was based on clinical improvement, microbiological results and stipulated time according to infectious focus. This stipulated duration is objectively described in Additional file 2. Apart from the values used in the diagnostic workup for infection, CRP levels were not provided to the clinical assistants and research teams during the antibiotic therapy period in this group of patients.
In the intervention group, the duration of antibiotic therapy was defined through a clinical protocol based on the daily serum levels of CRP (Fig. 1). The researchers recommended discontinuation of antibiotic treatment when the criteria in the study protocol were obtained. For patients with persistently elevated CRP serum levels but with clinical improvement and absence of signs of active infection, the duration of antibiotic therapy was the same as that suggested for the control group. For both groups, the study protocol determined that patients with bacteremia received at least five full days of adequate antibiotic coverage. In all cases included, a strategy of antibiotic de-escalation was recommended whenever possible. Any decision on antimicrobial suspension was ultimately the responsibility of clinical assistants. We considered "non-adherence to the protocol" any case in where the treating team chose not to follow the investigators' recommendations, either stopping antibiotics earlier or later than recommended by the protocol. The study protocol established longer antibiotic treatments independently of the levels of the biomarker (in the intervention group) or of the predetermined time (in the control group) in cases of unfavorable clinical evolution, maintenance of uncontrolled focus, and clinical or microbiological findings that require extended antibiotic therapies.
These cases were monitored daily by the research team, who recommended discontinuation of antibiotics as soon as possible.
Outcomes and data collection Patients were followed up by researchers from the time of inclusion until hospital discharge, death, or up to 90th day, whichever occurred first. Each patient was included only once.
The primary outcome analyzed was the duration of antibiotic therapy of the index infection episode.
The sample size calculation was done with 5% alpha error and 80% power, and estimated the need for 53 patients per group, plus a 15% correction due to a non-normal distribution of data. Detailed calculation is described in Additional file 3. Secondary outcomes were: mortality for any cause on Primary and secondary outcomes were primarily analyzed according to the intent to treat (ITT). Both groups were compared using chi-square or Student's T-test / Mann-Whitney U-test as indicated. For additional analysis of the primary outcomes, a cumulative curve of antibiotic discontinuation was compared between both groups (time-to-event analysis) using the Wilcoxon test 27 . When relevant, the results are presented by odds ratio and 95% confidence intervals. Two-tailed tests and a significance level of 0.05 were used for all analyzes.

Baseline demographics and clinical characteristics
We evaluated 437 patients for eligibility and a total of 135 patients were randomized. After randomization, five patients were excluded, and 130 patients were included in the intention-to-treat analysis ( Figure 2).

Figure 2
Demographics and clinical characteristics at inclusion were similar between patients of the two groups ( Table 1). The established empiric antimicrobial therapy was adequate in 90% of the cases 28 .
The values of CRP upon admission and during the first 10 days of follow-up did not reveal significant differences between the groups (Additional file 5). Table 1 Outcome results: Intention-to-treat analysis There was no between-group significant difference regarding the total days of antibiotic exposure during follow-up and antibiotic-free days. Similar results were found for ICU and hospital length of stay, mechanical ventilation time, intensive care and 28 th day mortality, sepsis related death, recurrence of infection, new nosocomial infections, and infection with multidrug-resistant bacteria (Table 2). Table 2 In a time-to-event data analysis, with the target event defined as 'antibiotic suspension' and censoring patients who did not experience this outcome (due to death or hospital discharge in use of antibiotic therapy), a significant lower antibiotic exposure was observed in the CRP group (p = 0.007) ( Figure 3).

Figure 3
The workflow of interventions in each subgroup, completed with outcome related to each of the subpopulations is depicted in Additional file 8.

Outcome results: Per-protocol and subgroup analysis
The rate of adherence to the protocol was 91% (92.2% CRP vs. 89.4% control; p = 0.764). Analysis per protocol revealed a reduction of one day in the median duration of antibiotic therapy in the CRP group in comparison to the control (6 (5-8) days vs 7 (7-10) days; p = 0.011) (Additional file 9).
Similar results were observed in post hoc analysis restricted to patients with SAPS 3 less than or equal to 59 (n = 66) at admission, patients with community-acquired infections (n = 56); patients with lower-respiratory tract infections (n = 58); and those who had adequate initial empirical antimicrobial therapy (n = 117) (Additional file 9).

Discussion
In this randomized clinical trial, we investigated the usefulness of a CRP-based protocol to reduce the duration of antibiotic therapy in critically ill patients undergoing an evidence-based judicious use of antibiotics strategy. We found lower antibiotic exposure in the intervention patients in comparison to controls, who were treated according to the best practice in antibiotic therapy 1,2,3,5,8 . Despite a similar median time of therapy, there was a narrower distribution of this parameter in the CRP-arm patients. Moreover, in the CRP group, more patients had their antimicrobial therapy suspended up to the fifth day of follow-up, with a significant lower antibiotic exposure in the time-to-event analysis.
Finally, the analysis per protocol revealed a reduction of one day in the median duration of antibiotic therapy in the intervention group.
Given the benefits offered by the rational use of antibiotics, including the reduction of multiresistant bacteria 5,29,30 , treatment costs 31 , frequency of adverse effects 32 , and less interference with microbiome, objective criteria to define the ideal treatment length is warranted. CRP is a low-cost and affordable biomarker 10 , routinely used in intensive care, that has been shown relate to prognosis in studies involving different populations with serious infectious conditions 17,20,33,34 .
Previous studies using biomarkers, notably PCT, included control groups in which the therapeutic strategy was freely determined by the assistant team. This strategy may have led to excessively long treatment duration of the control groups 10 , which varied from 10 11 to 15 days 12 . A meta-analysis involving data from more than 4000 patients on PCT-guided antibiotic therapy of acute respiratory infections, revealed that the PCT-guided group was treated for seven days in comparison with the control, which received 10 days treatment, or 14 days for patients in intensive care settings 35 . In our study we used the best standard of care in the control group, not the usual care as described above.
More recently, two studies have tested the usefulness of biomarker-guided antibiotic therapy compared to controls using shorter therapies. In a single center study, Oliveira et al found that a PCTbased protocol was not superior to a protocol based on serum CRP levels for reducing the use of antibiotics in sepsis. It is worth highlighting the fact that in this study the researchers originally stipulated a maximum of seven days for the duration of the therapy, independently of the levels of biomarkers 19 . In a larger Dutch study, de Jong et al. showed the usefulness of PCT to reduce the duration of antibiotic therapy in critically ill patients, with five days as the median treatment time compared to seven days for controls 14 . It is noteworthy that the population included in the present work was significantly more severely ill than the patients included in the Dutch study 14 (septic shock 32.3% vs. 18.5%, respectively), which reinforces the value of our results.
Although our study showed no difference in median duration of antibiotics in the ITT analysis, there were more patients which received shorter durations of antibiotics in the CRP-arm. Also, there was less exposure in the CRP group in the cumulative curve of antibiotic suspension. Further, one-day reduction in median duration of antibiotic treatment was found in the per-protocol analysis and in different post-hoc analyzes of subgroups. Specifically, in patients of lower severity and complexity (eg, community-acquired infections and SAPS-3 < 50%) the difference found may be justified by the easier application of the decision flowchart. In patients with respiratory tract infection, there is an already known better CRP performance in patients with pneumonia 36 . Patients who had appropriate initial empirical antibiotic therapy may have presented better results by the lower interference of inadequate initial antibiotics in treatment time 28;37 . These preliminary findings reinforce the potential role of a CRP-guided protocol in reducing antibiotic exposure in hospitalized infected patients.
Interestingly, in a recent published meta-analysis, authors found that the use of PCT-algorithms to guide antibiotic therapy was associated with increased survival especially when combined with a CRPguided strategy 16 .
The rate of adherence to the protocol reported herein was higher than that reported in previous clinical trials 13;14 . Patients were included when in intensive care and followed up until hospital discharge or death. Therefore, interventions were also applied in other hospital units. This strategy allowed the high rates of adhesion to the protocol, and proved feasible from the logistic point of view.
Other relevant investigated outcomes, such as antibiotic-free days and total time of antibiotic therapy during follow-up revealed similar between groups. Also, no statistically significant differences were found in safety and survival outcomes. These findings suggest that CRP-guided therapy is safe, although further studies, with a sample size powered for these analysis, should be conducted to confirm this hypothesis.
Our study has several limitations that should be mentioned. This was a single center study, restricted to two intensive care units of a high complexity hospital. Therefore, the findings lack external validity and cannot be extrapolated to other populations. In addition, the inherent open design may have biased the results, favoring the alternative hypothesis. Third, there was a high rate of non-inclusion among the patients evaluated for potential eligibility. Although this scenario has been observed in several similar studies 11;14 , this fact limits the population to which the protocol can be applied, especially immunosuppressed population. It remains unclear how these patients, including those with immunosuppressive dose corticosteroid therapy, respond to CRP-guided antibiotic therapy. Also, although infections with non-fermenting Gram-negative bacteria was not an exclusion criterion, patients with these kind of infection were poorly represented in this study. Fourth, there was an apparent trend towards higher mortality in the CRP-guided therapy group, with no statistically significant difference. However, sepsis-associated mortality was quite similar between the two groups, as well as recurrent infection rates. Also, mortality rate of patients who had early suspension of antibiotics, according to study protocols, was lower than the overall mortality rate, and similar between intervention and control groups (Additional file 8).

Conclusions
A protocol based on daily monitoring of CRP levels may support a tailored time of antibiotic therapy in critically ill patients, reducing the exposure of these patients to antimicrobials, even in a scenario of judicious use of these drugs. Although subtle, the time reduction observed in the group of patients undergoing CRP-guided therapy is potentially impacting. CRP-guided strategy is feasible and apparently safe. Further studies are needed to confirm these findings, to properly assess safety outcomes and to evaluate the real impact of this strategy in clinical practice. cript. CO, SS and MVA gave routine support for the daily inclusions and patients follow-up. RC and RB gave assistance in the inclusion process, with collection of clinical data and biological material. LM and EC gave great assistance with statistical analysis. All authors read and approved the final manuscript.

Availability of data and materials
The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Acknowledgments:
We thank the ICUs staff for the routine support. We also thank the students, Andrea Jerusa, Arthur