Study setting

This prospective investigation was conducted at Haukeland University Hospital's Emergency Department (ED) in Bergen, Norway, from September 25, 2020, to April 19, 2023. Haukeland University Hospital serves as a local healthcare provider for approximately 430,000 individuals, while also serving as a referral center for a broader population of 1,000,000. During the study period, the yearly number of admissions from the ED ranged from 38,000 to 45,000[12].

National guidelines

The Norwegian Directorate of Health publishes the National Professional Guidelines for the use of antibiotics in hospitals [13]. For community-acquired pneumonia, treatment recommendations are based on CRB-65 score severity and the presence of penicillin allergy, as outlined in Table 1.

Table 1 National guidelines for empirical antibiotic treatment and therapy duration in suspected CAP and study adherence definitionsStudy design and patient cohort

Patients from two study cohorts were included for the purpose of this study: a randomized controlled trial (RCT) investigating the impact of rapid syndromic testing using the FAP plus (ClinicalTrials.gov Identifier: NCT04660084), with enrolment occurring from September 25, 2020, to June 21, 2022; and a subsequent prospective study, with recruitment spanning from August 22, 2022, to April 19, 2023. Both studies were approved by the Regional Committee for Medical and Health Research Ethics, Norway (registration no. 31935).

Identical inclusion and exclusion criteria were applied to both cohorts to minimize selection bias and ensure they were representative of the same patient population, enabling cohesive analysis across the groups.

Targets for improvement

We focused on two critical aspects of guideline adherence in antibiotic stewardship: (1) the choice of empirical antibiotic treatment, and (2) the duration of antibiotic therapy [8]. The guidelines recommend specific empirical treatments for patients with suspected CAP admitted to the hospital and suggest a treatment duration of five days, including both IV and oral antibiotics, for mild to moderate CAP (defined as CRB-65 score 0–2) and seven days for more severe CAP (defined as CRB-65 score > 2). From the original study cohort, we formed two sub-cohorts for analysis—one for each antimicrobial stewardship target.

Data collection

Eligible patients were included shortly after presenting to the ED. Baseline data were collected by study nurses or investigating physicians and documented in the electronic case report form Viedoc (Viedoc Technologies, Uppsala, Sweden). Clinical data, including patient demographics, comorbidities, initial antibiotic choice, therapy duration, and discharge prescriptions, were extracted from electronic medical records.

Inclusion and exclusion criteriaGeneral inclusion and exclusion criteria for the cohort

The inclusion and exclusion criteria were the same for both study cohorts, detailed previously [14, 15]. Briefly, adults aged 18 and over, presenting to the ED with suspected CAP and meeting at least two predefined clinical indicators, were considered. These indicators included symptoms like new or intensified cough, expectoration, dyspnoea, haemoptysis, pleuritic chest pain, fever (≥ 38.0 °C), or evidence of pneumonia through radiological imaging or abnormalities detected during chest auscultation or percussion. Exclusion criteria encompassed recent hospitalization (within the last 14 days before ED presentation), cystic fibrosis, palliative care status, or unwillingness or inability to provide a lower respiratory tract (LRT) sample. Additional inclusion and exclusion criteria for the analysis of the two different antimicrobial stewardship targets are outlined below. Written informed consent was obtained from all patients, or their legal guardians/close relatives when applicable.

Criteria for the sub-cohorts analyzed in the current study

The empirical antibiotic treatment sub-cohort included all patients enrolled in the two original study cohorts who received antibiotic treatment. Patients were excluded from this sub-cohort if they had initiated antibiotic therapy more than 48 h after admission or if their initial antibiotic choice was guided by microbiology test results. These exclusions were necessary because such patients were unlikely to have community-acquired infections, and their therapy could not be classified as purely empirical.

For the antibiotic therapy duration sub-cohort, we included all patients from the two original study cohorts who received antibiotic treatment. Patients with discharge diagnoses other than CAP or community-acquired COPD exacerbations were excluded from the analysis of treatment durations, as the guideline recommendations were not applicable to these groups. Both CAP and COPD exacerbations were considered within the scope of the study, provided the patients met the inclusion criteria. Additionally, patients who died before they could have received antibiotic therapy beyond the recommended duration were excluded. To focus on patients intended to receive a full course of antibiotics, we also excluded patients treated with antibiotics for < 24 h. This short time frame was chosen as studies have shown that antibiotic courses as short as 72 h can be safe in CAP [17]. We also excluded patients treated with antibiotics for more than 20 days, as this typically indicates complicated pulmonary infections. Furthermore, we excluded patients who were assumed to be clinically unstable at day five or day seven of their antibiotic treatment. Detailed clinical stability assessments at these time points were not available from the trial and could not be easily extracted from the electronic medical records.Therefore, we used discharge status as a proxy for stability. For patients discharged to a nursing home or other healthcare institutions, we manually reviewed their electronic medical charts to assess clinical stability. Patients were considered clinically stable if they were afebrile (body temperature < 37.6 °C), had a respiratory rate ≤ 24/minute, peripheral oxygen saturation > 90% while breathing ambient air, and were able to eat [16, 17].

1.

Choice of empiric antibiotic therapy within 48 h of admission.

The assessment focused solely on the choice of empiric antimicrobial agents, excluding any dosing considerations. Guideline adherence was evaluated solely for the initial therapy administered.

The initial therapy was defined as the first antibacterial agent(s) administered during admission. If a second or third antibiotic was administered within three hours of the first, it was considered combination therapy. To establish the most appropriate cutoff between combination therapy and an antibiotic switch, we evaluated time intervals of 1, 2, 3, 4, 5, and 6 h. For each interval, we manually reviewed whether the second and third antibiotics were part of a combination regimen or indicated an early change in therapy. Based on this assessment, a three-hour cutoff was determined to be the most appropriate. The initial therapy was then classified as adherent or non-adherent using a rule-based approach, as detailed in Table 1.

2.

Duration of antibiotic therapy.

The Norwegian guidelines recommend a fixed total antibiotic treatment duration of 5 days for clinically stable patients with CAP or COPD exacerbations, based on the initial severity of infection. Specifically, for patients with a CRB-65 score ≤ 2, indicating mild to moderate disease, the recommended treatment duration is 5 days. For patients with more severe infections (CRB-65 score > 2), the recommended treatment duration is 7 days [13].

Importantly, these recommendations are based on the baseline severity of illness and do not explicitly adjust for clinical stability at specific points during treatment. However, the guidelines do reference the NICE guidelines, which recommend extending therapy if patients have not achieved clinical stability by day three [18]. These aspects were not directly addressed in the fixed duration recommendations from the Norwegian guidelines but are pertinent when evaluating adherence.

Therapy duration included both in-hospital (IV and oral) and planned post-discharge treatment. Guideline-adherent therapy was defined as ≤ 6 days (144 h) for mild to moderate disease and ≤ 8 days (192 h) for severe disease, with a 24-h margin to ensure a practical and clinically relevant definition of adherence. In September 2020, one month before the start of study inclusion, the guideline recommendations for the duration of antibiotic treatment were changed from 5 to 7 days to a fixed five days for stable CAP. Since guideline implementation might be delayed in clinical practice, we also analysed the time from the guideline change to admission as part of our study to assess its impact on adherence.

Statistics

For descriptive analysis, categorical variables are presented as counts and percentages of available data. Continuous variables are shown as medians with interquartile ranges (IQR) based on available data. We compared categorical variables using Pearson’s Chi-squared test and continuous variables using the Wilcoxon rank-sum test.

To handle missing data, we used the missRanger package in R, which employs chained random forests with predictive mean matching. This method leverages the non-linear relationships between variables to accurately predict missing values, thereby enhancing our analyses' robustness by mitigating bias from incomplete data [19].

To assess the effect of three continuous variables (maximum C-reactive Protein (CRP) value, length of stay, and time since the guideline change) on the probability of antibiotic therapy durations longer than recommended, we employed a logistic regression model that accounted for non-linearity using natural splines. We visualized the effect of time since the guideline change using the sjPlot package in R [20].

In our multivariate regression analyses, all variables were converted to binary forms. Established cutoffs from the literature were used where available. For numeric variables lacking established cutoffs, we explored the relationship between each variable and the probability of the outcome, assessing for both linear and non-linear patterns. If no relationship was detected, the cutoff was set at the median value. When a relationship was identified, the cutoff was determined based on the observed pattern.

We utilized multivariate regression using Poisson regression model with generalized estimating equations (GEE) to estimate the risk of antibiotic treatment deviating from guidelines. This approach allows for the direct estimation of risk ratios, which are more intuitive and interpretable than odds ratios (OR) [21, 22]. The results are presented as risk ratios (RR) with 95% confidence intervals (CI) and associated p-values. In addition, Population Attributable Fractions (PAF) were calculated for variables with a significant effect on guideline adherence.

The multivariate models included age, sex, and variables likely to influence treatment. For empirical antibiotic therapy the model included age, gender, clinical frailty scale, admission from a nursing home, hospital admission within the last month, COPD, chronic kidney disease (CKD), immunodeficiency, ongoing antibiotics at admission, antibiotics use in the last month, antibiotic allergy, SOFA score, and CRB-65 score. For duration of therapy the model included age, sex, immunodeficiency, COPD, use of systemic steroids, length of stay, Clinical Frailty Scale, Charlson Comorbidity Index, SOFA-score, ventilatory support, ICU admission, maximum CRP value during admission, and microbial detections.

For both univariate and multivariate analyses, we set a two-sided significance level of 0.05.

We conducted all analyses using R statistical software, version 4.4.0 (http://www.r-project.org) [23].