Background

Bacterial infections (BIs) are widespread in Paediatric and Neonatal Intensive Care Units (ICUs) and 30–61% of neonates and children hospitalised in ICUs receive antibiotic treatments [1, 2]. For suspected sepsis or septic shock, the Surviving Sepsis Campaign (SSC) recommends the early administration of empiric broad-spectrum antibiotic therapy with one or more antimicrobials to cover all likely pathogens, appropriate routine microbiologic cultures (including blood) before starting antibiotic therapy if doing so causes no substantial delay in starting antimicrobials as well as early de-escalation of antimicrobials based on culture results, susceptibility results and clinical improvement [3, 4]. The choice of antimicrobials, number of antibiotic doses in 24 h, daily dose and duration are determined by local epidemiology, patient characteristics (age, patient history, allergies, multidrug-resistant [MDR] status), BI characteristics (infection site(s), community or hospital-acquired infection, microbiological results) as well as clinical and biological evolution in line with published guidelines. Their recommendations are numerous: French (SPILF, GPIP, SFAR, SRLF, HAS) [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21], European (ESCMID) [22, 23] and American (IDSA) [24,25,26,27,28] guidelines. Long courses of antibiotic therapy and broad-spectrum antimicrobials increase the length of hospitalisation and are associated with changes in the microbiome [29], emergence of multidrug resistant organisms [30, 31] and antibiotic-associated adverse events (toxicity, overdose, allergy) [32]. On the other hand, too short antibiotic therapy can expose the patient to a risk of BI recurrence. In 2017 and 2021, the Société de Pathologie Infectieuse de Langue Française (SPILF or French Language Society for Infectious Diseases) and Groupe de Pathologies Infectieuses en Pédiatrie (GPIP or French Group for Paediatric Infectious Diseases) published French recommendations for the shortest treatment durations for BIs [5, 6]. To our knowledge, compliance with these recommendations has never been evaluated.

In this context, we assessed compliance with recommendations for antibiotic prescriptions, and factors associated with non-compliance for children hospitalised in ICU and receiving systemic antibiotics for an episode of suspected or proven BI.

Methods

Study design, setting and participants

We conducted an observational, prospective, multicentre study in eight French Paediatric and Neonatal ICUs (Appendix 1) during six separate time-periods between June 2020 and May 2021. After an inclusion period testing the feasibility of data collection at the Coordinator Centre (Toulouse) from June to August 2020, we then arbitrarily chose a priori five weeks each spread two months apart for multicentre inclusion over the course of one year. All participating ICUs had the possibility of an audit with an infectious disease specialist over the telephone and most had an antimicrobial stewardship programme (ASP) organised in once weekly multidisciplinary (intensivists, microbiologists and paediatric infectious disease specialists) staff meetings with prospective audit and feedback. Intensivists were the only prescribers. This study was supported by the Groupe Francophone de Réanimation et d’Urgences Pédiatriques (GFRUP or French-Speaking Group for Paediatric Intensive and Emergency Care). All methods were performed in accordance with relevant guidelines and regulations. In accordance with French Ethics and Regulatory Law (Public Health Code), this trial is covered by reference methodology MR-004 from the French Data Protection Commission (CNIL). It was approved by Toulouse University Hospital and is registered on its Study Data Register under number RnIPH2019-79 and on the ClinicalTrials.gov website under number NCT04642560. The date of first trial registration on ClinicalTrials.gov was 24/11/2020.

During the study periods, all consecutive neonates and children hospitalised in ICUs and receiving systemic (intravascular, intramuscular or oral) antibiotic treatment for a suspected or proven community-acquired or nosocomial BI were assessed for eligibility. Antibiotics had to be initiated in ICUs during the study periods or no more than 24 h prior to ICU admission occurring over the study periods. We called this episode “first suspected or proven BI episode” to distinguish it from a possible BI recurrence at a later stage. Exclusion criteria were: newborns < 72 h old; neonates < 37 weeks post-menstrual age; age ≥ 18 years; children under surgical antimicrobial prophylaxis; and children previously included in an ongoing interventional study. Informed verbal consent was obtained from the parents or legal guardians of the patient prior to study enrolment.

Data collection

Data on patient characteristics and the first suspected or proven BI episode (characteristics, organ dysfunction scores, antibiotic therapy and concomitant therapeutics other than antibiotics) were prospectively collected on a daily basis by medical study site investigators so long as the patient was receiving antibiotics in hospital for the first suspected or proven BI episode (ICU and general paediatric ward if antibiotic therapy was not completed in the ICU) (Appendix 2). D0 was the day of antibiotic therapy initiation. Additional data on the length of ICU and hospital stay, mortality, outpatient antibiotic therapy (if this was the case) and recurrence of BI occurring within 28 days following D0 were registered upon hospital discharge. Antibiotics used for any other infections during the 28 days following D0 and surgical antimicrobial prophylaxis were not taken into account.

The primary endpoint was the number of first episodes for which antibiotics were prescribed inappropriately on the basis of current recommendations (non-compliance) involving one or more of the following parameters: length of antibiotic therapy, duration of each antimicrobial treatment, choice of antimicrobials, number of antibiotic doses in 24 h, daily dose of antibiotic therapy and reassessment of antibiotic therapy at 72 h. Secondary endpoints were: number of first episodes with non-compliance for any of the parameters; length of antibiotic therapy for the first suspected or proven BI episode; duration of broad-spectrum antibiotic therapy for the first suspected or proven BI episode; and recurrence of BI within 28 days following D0.

Recurrence of infection was defined as the isolation of one or more of the initial causative bacteria from the same or another site at 48 h or more after cessation of antibiotics, combined with clinical signs or symptoms of infection or the need to prescribe a new antimicrobial therapy covering this pathogen [33]. Only recurrences during the same hospital stay as for the first BI episode and occurring within 28 days following D0 were taken into consideration.

Finally, we used two separate definitions to identify broad-spectrum antibiotics: the standard definition [34] and the 2019 AWaRe (Access, Watch, Reserve) classification (Watch and Reserve groups) [35, 36].

Analysis of compliance with antibiotic recommendations

For each first suspected or proven BI episode, the same paediatric infectious disease expert committee (CB and EG) analysed compliance with antibiotic recommendations for length of antibiotic therapy, duration of each antimicrobial treatment, choice of antimicrobials, number of antibiotic doses in 24 h, daily dose of antibiotic therapy and reassessment of antibiotic therapy at 72 h (Appendix 2 and 3). The two experts relied principally on the SPILF and GPIP French guidelines [5, 6] for the duration of antibiotic therapy and each antimicrobial treatment and on a combination of French, European and American guidelines [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28] for the choice of antimicrobials, number of antibiotic doses in 24 h and daily dose of antibiotic therapy. Non-compliance (inappropriate antibiotic prescription based on the guidelines) was only determined if both experts agreed. To detect a high background rate of resistant pathogens in ICUs which would justify the use of initial broad-spectrum empiric therapy during the study period, we asked participating centres for their local microbiological data from the two years prior to the study period (2018 and 2019).

Statistical analysis

The results of descriptive statistics were presented as absolute frequencies (%) for qualitative variables and as medians (IQR) for continuous variables.

For all first suspected (or proven) BI episodes and also for only first confirmed (documented or not) BI episodes, we performed univariate analyses to assess factors that might be associated with non-compliance for each parameter and for all parameters combined. For independent qualitative variables, we used the χ² test or Fisher’s exact test while independent quantitative variables employed a two-sample t-test or a Wilcoxon-Mann-Whitney rank-based test. To convert quantitative variables into classes when necessary, we determined the thresholds using Youden’s index.

To establish the independent predictive factors for non-compliance, we then carried out multivariate analyses by stepwise logistic regression after selecting the independent qualitative variables associated with the dependent variable with p < 0.20. The association between variables was significant if p < 0.05. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for the significant variables. When one or more centres demonstrated significance during the multivariate analysis, ORs were adjusted for each centre.

Additionally, to evaluate if the use of an ASP during a suspected (or proven) BI episode is a factor modifying the effect of independent variables on non-compliance, we conducted ASP-stratified univariate logistic regression analyses of all factors potentially associated with non-compliance with recommendations for all parameters and for each parameter (length of antibiotic therapy, duration of each antimicrobial treatment, choice of antimicrobials, daily dose of antibiotic therapy and reassessment of antibiotic therapy at 72 h). The ASP sub-group included all episodes that received auditing with an antimicrobial stewardship team for antibiotic duration and/or choice of antimicrobials. Stratified ORs and 95% CIs were calculated for all independent variables in both the ASP and non-ASP sub-groups and were compared with ORs and 95% CIs of all suspected (or proven) BI episodes.

All statistical analyses were performed using SAS software (version 9.4, Cary, NC, USA).

Results

Local microbiological data of participating intensive care units

Local ICU microbiological data from the two years prior to the study period (years 2018–2019) were available at 5 of the 8 centres (Appendix 4). Isolation frequencies for major pathogens was quite similar between the centres. Resistance rates to major antibiotics (intermediate or resistance categories) remained at a low background rate overall except for Enterobacterales resistance to third-generation cephalosporins in some centres.

Population characteristics

During the study periods, 868 children were hospitalised in the eight participating ICUs. One hundred thirty-nine first suspected (or proven) BI episodes occurring in 134 children were included. All children meeting the entire inclusion criteria and with no non-inclusion criteria were included. No one was missed and none of the children had legal guardians who refused consent. The median age was 0.8 (IQR 0.1–6.3) years and 20.1% of episodes affected neonates. Patient characteristics, infection characteristics and antibiotic therapy for first suspected (or proven) BI episodes, and outcomes are set out in Table 1. The most frequent initially suspected BI sites were respiratory (56.1%), catheter-related bacteraemia (20.1%) and intra-abdominal (9.4%). Final diagnosis was no BI for 26.6% of the episodes, presumed (i.e., not documented) BI for 40.3% and documented BI for 35.3%. Three episodes combined both presumed and documented BI. The median length of antibiotic therapy for the first suspected (or proven) BI episode was 7.1 (IQR 4.0-10.5) days. Recurrence occurred for only two episodes (1.4%).

Table 1 Patient characteristics, infection characteristics and antibiotic therapy for first suspected (or proven) bacterial infection episodes, and outcome: univariate analysis of factors associated with non-compliance with antibiotic recommendations for all parameters
Full size table

Diagnosis of BI (documented or not) was confirmed for 102 of the 139 first suspected BI episodes (Table 2). The 2005 IPSC infection severity was as follows: 25.5% infection without sepsis; 52.0% sepsis (excluding severe sepsis and septic shock); 11.8% severe sepsis (excluding septic shock); and 10.8% septic shock. Among the 76 episodes of sepsis, severe sepsis or septic shock, respiratory dysfunction was the most common acute organ dysfunction related to the BI and found in 60.5% of the episodes. The most frequent BI sites ultimately identified were respiratory (57.8%), mainly community-acquired pneumonia (13.7%) and aspiration pneumonia (23.5%), catheter-related bacteraemia (18.6%) and intra-abdominal (10.8%).

Table 2 Characteristics of first episodes of confirmed (documented or not) bacterial infection: univariate analysis of factors associated with non-compliance with antibiotic recommendations for all parameters
Full size table

Documented BIs concerned 49 first confirmed BI episodes. Microbiological data are presented in Table 3. The most prevalent causative bacteria encountered were Staphylococcus sp (34.7%), Enterococcus sp (16.3%) and Klebsiella sp (16.3%). Coagulase-negative Staphylococci (22.4%) were mostly methicillin-resistant (90.9% of the 11 episodes where coagulase-negative Staphylococci were isolated) while Staphylococci aureus (14.3%) were predominantly methicillin-sensitive (85.7% of the 7 episodes where Staphylococcus aureus was isolated). Extended-spectrum β-lactamase (ESBL)-producing Enterobacteriaceae were involved in only one of the 17 episodes in which Enterobacteriaceae were identified and no carbapenem-resistant pathogens were isolated.

Table 3 Microbiological data for first episodes of documented bacterial infection (n = 49)
Full size table

Non-compliance with antibiotic recommendations

As a result of death occurring during antibiotic treatment for 6 episodes (antibiotic therapy not completed), compliance for all parameters combined could only be analysed for 133 episodes. Non-compliance with recommendations for all parameters occurred in 51.1% of cases. Results of non-compliance with recommendations for each parameter are detailed in Table 4. The main reasons for non-compliance were inappropriate choice of antimicrobials (27.3%), duration of each antimicrobial treatment (26.3%) and length of antibiotic therapy (18.0%). In most cases, this duration was longer than that recommended by infection guidelines.

Table 4 Non-compliance with antibiotic recommendations for first suspected (or proven) bacterial infection episodes: type and source of non-compliance
Full size table

Factors associated with Non-compliance with antibiotic recommendations

Results of univariate analyses to determine the significant factors associated with non-compliance with recommendations for all parameters are presented in Table 1 for all first suspected (or proven) BI episodes and in Table 2 for only first confirmed (documented or not) BI episodes. In multivariate analyses (Tables 5 and 6), independent risk factors of non-compliance for all parameters were the use of 2 or more antimicrobials per episode (OR 4.06, 95% CI 1.69–9.74, p = 0.0017) and the duration of broad-spectrum antibiotic therapy based on the standard definition of 4 days or more (OR 2.59, 95% CI 1.16–5.78, p = 0.0199) while independent protective factors of non-compliance (i.e., factors increasing compliance) for all parameters were respiratory failure as the reason for ICU admission (OR 0.36, 95% CI 0.14–0.90, p = 0.0281), using department protocol for antibiotic duration (OR 0.42, 95% CI 0.19–0.92, p = 0.0313) and aspiration pneumonia as BI site ultimately identified (OR 0.37, 95% CI 0.14–0.99, p = 0.0486).

Table 5 Significant independent factors for non-compliance with antibiotic recommendations for first suspected (or proven) bacterial infection episodes: results of multivariate analyses
Full size table
Table 6 Significant independent factors for non-compliance with antibiotic recommendations for first episodes of confirmed (documented or not) bacterial infection: results of multivariate analyses
Full size table

Similarly, we performed univariate and multivariate analyses to establish the independent predictive factors of non-compliance with recommendations for each parameter. Results of multivariate analyses are set out in Tables 5 and 6. In addition to those mentioned above, other independent risk factors for non-compliance were: (1) neurologic compromise as the reason for ICU admission; (2) suspected catheter-related bacteraemia; (3) suspected or confirmed BI site classified as “other” (for which antibiotic therapy is often poorly defined by guidelines); (4) late-onset ventilator-associated pneumonia (VAP) or non-ventilator hospital-acquired pneumonia (HAP); (5) presence of ≥ 1 risk factor for ESBL Enterobacteriaceae; and (6) duration of broad-spectrum antibiotic therapy ≥ 3 days according to the AWaRe classification (Watch and Reserve antibiotics). For children with neurologic compromise at time of ICU admission, the length of antibiotic therapy was prolonged for all non-compliant episodes and concerned aspiration pneumonia in half of all cases, catheter-related bacteraemia and one episode of BI that was finally ruled out. For late-onset VAP, non-compliance regarding the choice of antimicrobials involved empiric therapy for all non-compliant episodes. Otherwise, the independent factors for compliance were respiratory failure at the onset of suspected BI and suspected respiratory BI site. Notably, patient severity during the first suspected (or proven) BI episode, evaluated by PELOD-2 and pSOFA scores, did not represent an independent factor for non-compliance with recommendations. Serious infection (sepsis, severe sepsis and septic shock), particularly with one or more organ dysfunctions, constituted an independent factor for compliance in terms of length and daily dose of antibiotic therapy while the presence of ≥ 2 organ dysfunctions related to infection was an independent risk factor for non-compliance regarding the duration of each antimicrobial treatment. For sepsis with ≥ 2 organ dysfunctions, the duration of one or more antimicrobials was prolonged for all non-compliant episodes. This extended duration concerned antimicrobials used for VAP and non-ventilator HAP for 3 episodes and catheter-related bacteraemia, primary bacteraemia, pneumonia with parapneumonic pleural effusion, BI site classified as “other” and a final diagnosis of no BI for one episode each.

Finally, the use of an ASP during a suspected (or proven) BI episode did not modify the effect of independent variables on non-compliance with recommendations for all parameters (Table 7) and for each compliance parameter (results not shown). Indeed, all 95% CIs for stratified ORs overlapped in the two sub-groups although some independent variables were significant only in the ASP sub-group.

Table 7 Antimicrobial stewardship programme-stratified univariate analysis: significant factors associated with non-compliance with antibiotic recommendations for all parameters and comparison with all first suspected (or proven) bacterial infection episodes
Full size table

Discussion

In this prospective observational multicentre study, we assessed compliance with recommendations for antibiotic prescriptions made by intensivists in a large population of children hospitalised in 8 French ICUs, most of which have an ASP (once weekly infection multidisciplinary staff meeting with audit and feedback). We also analysed the factors associated with non-compliance. Half of the prescriptions complied with guidelines. In the cases where recommendations were not followed, the main reasons for non-compliance were inappropriate choice of antimicrobial(s), inappropriate duration of one or more antimicrobials and inappropriate length of antibiotic therapy (most frequently prolonged duration for both). In multivariate analyses, we identified situations where intensivists were more attentive to recommendations: patients with respiratory failure, when a respiratory site was initially suspected, when aspiration pneumonia was the ultimately identified site and when antibiotic protocols were available in their ICU. Conversely, we highlighted contexts where risk of non-compliance with guidelines is likely to exist: prescribing at least two antibiotics, duration of broad-spectrum antibiotic therapy ≥ 3–4 days, neurologic compromise at time of ICU admission, suspicion of catheter-related bacteraemia, suspecting or confirming a BI site for which antibiotic therapy is often poorly defined by guidelines (classified as “other”), late-onset VAP, non-ventilator HAP and presence of ≥ 1 risk factor for ESBL Enterobacteriaceae. In sepsis patients, the presence of ≥ 2 organ dysfunctions related to infection represented an independent factor of non-compliance for duration of each antimicrobial treatment while one or more organ dysfunctions were an independent factor of compliance for length of antibiotic therapy.

A few studies have looked at antibiotic prescribing in paediatric ICUs. A prospective multicentre paediatric study of ventilator-associated lung disease showed 70% compliance before implementation of a local protocol and 76% afterwards [37]. Another study described inappropriate antibiotic prescribing in paediatric ICUs ranging from 16.7 to 61.9% depending on the evaluator and the period [38]. In ICU adults with sepsis, studies have demonstrated a comparable compliance rate between 47% and 58% [39, 40].

Lindberg et al. reported that non-compliance was independently associated with an increased risk of 30-day mortality corresponding to 1.86 (95% CI 1.34 to 2.58, p < 0.001) for partial compliance and 2.18 (95% CI 1.34 to 3.40, p < 0.001) for complete non-compliance [40]. In our study, we did not find a statistically significant difference in all-cause mortality between the compliant and non-compliant episodes (7.7% versus 5.9% respectively). However, mortality in our population was 10.8%, much lower than in critically ill adults and our study was not designed to evaluate the impact of non-compliance on mortality. Similarly, non-compliance was not associated with recurrence of bacterial infection in our work although the recurrence rate was only 1.4%.

Good compliance with recommendations for respiratory patients is consistent with clinical practice. Indeed, respiratory infections were mainly community-acquired in our study. Antibiotic therapy for these infections is well-defined by current guidelines for children [5, 6, 9] which allows doctors to better comply with recommendations.

The advantage of antibiotic protocols has already been identified. Protocols, based on current guidelines and updated regularly, are available 24/7 and enable harmonisation of prescriptions made by intensivists that can change from day to day. In adults, the implementation of computerised local antibiotic therapy protocols has been associated with reduced antibiotic exposure and mortality in adult ICUs [41].

To steer decisions about antibiotic therapy in ICUs, a European expert panel has created an “antibiotic care bundle” (ABC-Bundle) with evidence-based recommendations for antibiotic prescribing [42]. The six steps are: (1) provide rationale for antibiotic start; (2) perform appropriate microbiological sampling; (3) prescribe empiric antibiotic therapy according to guidelines (day 1); (4) review diagnosis; (5) assess de-escalation based on microbiological results (days 2–5); and (6) consider discontinuation of treatment in patients with negative culture results and clinical improvement (days 3–5). To mitigate the risk of antibiotic resistance emergence, recent European recommendations have defined antimicrobial de-escalation (ADE) as discontinuing one or more antimicrobials in the empiric combination therapy or replacing broad-spectrum antimicrobials with narrower spectrum agents [43]. In our study, ADE was evaluated by the duration of each antimicrobial treatment, reassessment of antibiotic therapy at 72 h and choice of antimicrobials (narrow antimicrobial therapy once pathogen identification and susceptibility testing results are available). Based on the DIANA study [44], we decided that ADE should take place within the first 3 days of initiation of empiric therapy to be appropriate. In the DIANA study in which 152 adult ICUs in 28 countries participated, ADE within the first 3 days of empiric therapy occurred in only 16% of patients while combination therapy was prescribed in half of the patients and infections were documented in 56% of the study population. Our results concerning inappropriate choice of antimicrobials and duration of one or more antimicrobials as the most frequent reasons for non-compliance and the number of antimicrobials used per episode and length of broad-spectrum antibiotic therapy to be representative of independent risk factors for non-compliance, show that ADE is essential in clinical practice. Recent recommendations on sepsis management in children have also emphasised that dual therapy is no longer recommended for children who are not immunocompromised and present no risk of carrying multidrug-resistant bacteria [4].

ADE is more frequently used in patients with a favourable course [45]. To analyse compliance with antibiotic recommendations, our paediatric infectious disease experts took into account patient improvement. Despite a favourable trajectory, ADE was often missing in patients with the most serious infections (sepsis with ≥ 2 organ dysfunctions). We hypothesise that intensivists probably delayed or did not achieve de-escalation due to the initial severity of the infection. However, de-escalation is safe for patients with favourable evolution. De Bus and colleagues reported no significant difference in 28-day mortality and infection relapse for ADE patients versus non-ADE patients in adult ICUs [44]. The 2021 SSC adult guidelines also suggested daily assessment for ADE and early ADE based on adequate clinical improvement for adults with sepsis or septic shock [3].

In patients with neurologic compromise at time of ICU admission, non-compliance in terms of length of antibiotic therapy was due to aspiration pneumonia in 50% of the patients while aspiration pneumonia represented a factor exhibiting good compliance for the entire study population. We assume that intensivists considered these patients to be at higher risk. A recent study, including 27,455 hospitalised children with neurologic impairment and pneumonia, showed that there were more systemic complications (acute respiratory failure, sepsis or ECMO) in neurologically impaired children with aspiration pneumonia [46].

Early administration of appropriate empiric therapy in patients with sepsis is crucial to reduce mortality [3]. In children with sepsis-associated organ dysfunction and septic shock, the 2020 SSC paediatric guidelines recommended empiric broad-spectrum therapy with one or more antimicrobials to cover all likely pathogens, taking into account local epidemiology, patient characteristics (age, patient history, allergies, MDR status) and suspected BI sites [4]. We found that late-onset VAP was an independent factor for non-compliance regarding the choice of antimicrobials and empiric therapy was inappropriate for all non-compliant episodes. Furthermore, Mangino and colleagues reported a low rate of appropriate empiric therapy (44%) for HAP in adult ICUs despite the implementation of multimodal educational activities to teach ICU staff about the guidelines. The presence of ≥ 1 risk factor for ESBL Enterobacteriaceae was another aetiology for inappropriate empiric therapy in our study. However, local microbiological data from participating ICUs showed a low rate of ESBL pathogens and ESBL Enterobacteriaceae was isolated in only 2.0% of BI episodes while children presented one or more risk factors for ESBL Enterobacteriaceae in 29.5% of episodes. Indications of probabilistic antibiotic therapy covering ESBL Enterobacteriaceae are mainly based on adult recommendations [10, 20] and should perhaps be adapted to paediatrics.

In French Paediatric and Neonatal ICUs, antibiotic prescriptions are issued by intensivists. They have access to a possible audit with an infectious disease specialist over the telephone (at their discretion) on weekdays and sometimes during weekends as well as an ASP (multidisciplinary staff meeting with intensivists, microbiologists and paediatric infectious disease specialists) occurring once a week at most. Previous studies have reported that ASPs decreased inappropriate prescriptions, antibiotic consumption and drug resistance [47, 48]. We could not demonstrate better compliance with the use of an ASP. In stratified analyses, the use of an ASP occurring once a week at most did not modify the effect of independent variables on non-compliance with recommendations. However, a once weekly ASP does not allow for recommendations to be given to all patients on antibiotics or for the daily “correction” of non-compliance concerning the duration of antimicrobials or re-evaluation at 72 h. Moreover, we only identified the presence or absence of non-compliance without quantifying the number of days for each instance of non-compliance on a daily basis (except for duration of antibiotics).

This study has several limitations. Firstly, we arbitrarily chose several separate weeks to include patients and not just one study period. This is explained by the very time-consuming nature of data collection and by the lack of funding for this study. These separate periods enabled each participating centre to organise their data collection. As with point prevalence studies, such data collection can potentially be a source of selection bias and could limit the generalisation of our results to the entire paediatric ICU population. However, consecutive patients were included during the study periods and since these weeks were spread over one year, we were able to take into account seasonal variations specific to paediatrics. Furthermore, at the time of the study, three centres did not have an ASP but all centres had access to audits with an infectious disease specialist over the telephone on weekdays and sometimes during weekends. For centres implementing an ASP, there was a once weekly multidisciplinary staff meeting, with advice given for antibiotic prescription. Finally, local ICU microbiology data from the two years prior to the study period were missing for three participating centres, which may have impacted analysis of compliance regarding the choice of antimicrobials by the paediatric infectious disease expert. However, the low MDR rate for the episodes included and in the local epidemiology of the other centres suggest a minimal influence.

Conclusions

In French Paediatric and Neonatal ICUs, most of which hold a once weekly ASP, half of antibiotic prescriptions remain non-compliant with guidelines. For respiratory illnesses with clear treatment guidelines or for which antibiotic protocols already exist at a given centre, antibiotic use by intensivists tended to be more compliant. This highlights the importance of developing consensus/guidelines about treating specific illnesses and antibiotic protocols based on current guidelines that are updated regularly for use by clinicians. The benefit of using several antimicrobials and broad-spectrum antibiotic therapy should be reassessed daily by intensivists. Based on microbiological results and patient evolution, intensivists must perform an early de-escalation and stop antibiotics that are no longer indicated. A daily ASP could improve compliance with guidelines in these non-compliance prone situations.