Study design and setting

A retrospective cohort study was conducted using data from four secondary care hospitals in the Netherlands (Tergooi Medical Centre (Hilversum), Haga hospital (the Hague), Hospital Group Twente (Almelo and Hengelo), and Diakonessenhuis (Utrecht)). This study was approved by the Institutional Research Board of Tergooi MC (study number 2024.54) and local approval was obtained at each of the participating hospitals. No consent procedure was required, as this was a retrospective study with pseudonymized data. Data was not collected from patients who, according to their medical records, had objected to the use of their data for research purposes.

Study population

The study population consisted of patients who were hospitalized for at least 12 hours with a community-onset infection between January 1, 2021, and December 31, 2023, and who met all of the following inclusion criteria: age ≥18 years; an International Classification of Diseases and Related Health Problems-10 (ICD-10) diagnosis related to a RTI, UTI or SSTI (Supplement S1) during hospitalization; a discharge letter confirming a diagnosis of RTI, UTI or SSTI; and administration and prescription of the first systemic antibiotic dose within 48 hours of hospitalization. Patients were excluded if they met any of the following exclusion criteria: tested positive for, or were diagnosed with, COVID-19 within 48 hours before or after hospitalization (see Supplement S2 for definition), due to the risk of antibiotic overprescribing ^{Reference Langford, So and Raybardhan14} ; were transferred to or from another hospital, as treatment duration (length of therapy (LOT)) could not be reliably determined; had a concurrent infectious ICD-10 diagnosis during hospitalization (Supplement S1); clinical antibiotic LOT was below 0 days, which indicates a prescription error; antibiotic LOT exceeded 21 days, to exclude prophylactic antibiotic therapy; or died during hospitalization before completing the recommended duration of treatment, as they could not have received prolonged antibiotic therapy. Patients with empyema, for which only a minimal treatment duration is recommended, ¹⁵ or with legionella infection, for which one local guideline recommends a total treatment duration of two to three weeks, ¹⁶ making it impossible to distinguish between prolonged and chronic therapy, were also excluded from the RTI analysis.

Patients with a UTI were classified into a single category in the following order: catheter-associated UTI, UTI with systemic symptoms (eg, fever, delirium, flank pain), cystitis, or UTI—Other if none of the aforementioned criteria were documented in the patient’s electronic health record (EHR) (see Supplement S2 for definitions). ^{Reference Terpstra, Geerlings and van Nieuwkoop9,Reference Klinkhamer, Geerlings and Knottnerus10} Guideline adherence and prolonged treatment for UTI—Other were based on the cystitis treatment guidelines. Due to uncertainty about this classification, in an unplanned sensitivity analysis, they were instead based on treatment guidelines for UTI with system symptoms. Patients with a RTI were similarly classified in the following order: hospital-acquired pneumonia (HAP), aspiration pneumonia, severe community-acquired pneumonia (CAP-s) (defined as Pneumonia Severity Index (PSI) class 5 or CURB-65/AMBU-65 score of 3–5), mild-to-moderate community-acquired pneumonia (CAP-m) (PSI class 1–4 or CURB-65/AMBU-65 score 0–2), or a RTI not fulfilling any of these classifications (RTI—Other) (Supplement S2). ^{Reference van Daalen, Boersma and van de Garde11} Guideline adherence and prolonged treatment for RTI—Other were assessed based on the CAP-m treatment guidelines. Patients with a SSTI were classified as having either a panaritium/paronychia or cellulitis/erysipelas (Supplement S2). ^{Reference Bons, Bouma and Draijer12,13}

Study outcomes

The primary outcome was prolonged antibiotic treatment. This was calculated separately for UTI, RTI, and SSTI and based on nationally available guidelines (Supplement S3, Table 2). ^{Reference Terpstra, Geerlings and van Nieuwkoop9–13} To calculate the LOT, the duration was measured from the initiation of antibiotic treatment at hospitalization until the end of a consecutive period of antibiotic prescriptions, including discharge antibiotics prescribed in the hospitals’ electronic prescribing system. When stop dates were missing from in-hospital antibiotics, the discharge date was used. Preoperative antibiotic prophylaxis prescriptions were excluded, defined as cefazolin prescriptions for ≤24 hours or prescribed by anesthesiologists or surgeons around surgical procedures. Consecutive antibiotic treatment was defined as antibiotic prescriptions with a gap of ≤24 hours between the stop and start date of a new prescription. Guideline adherence was defined as treatment durations within 0.5 days below the minimum to 0.5 days above the upper recommended range. To determine prolonged treatment, the upper limit of the guideline-recommended duration was used, with an additional margin of 0.5 days added to correct for administration time (Table 2).

Table 1. Baseline characteristics

Table 2. Length of therapy, guideline adherence, and prolonged treatment per infection

n, number of patients; IQR, interquartile range; UTI, urinary tract infection; UTI-CAD, catheter-associated urinary tract infection; RTI, respiratory tract infection; CAP-m, community-acquired pneumonia mild-to-moderate; CAP-s, community-acquired pneumonia severe; HAP, hospital-acquired pneumonia; PSI, pneumonia severity index; SSTI, skin and soft tissue infection.

a. In accordance with Dutch guidelines ^{Reference Terpstra, Geerlings and van Nieuwkoop9–13} ; b. Guideline adherence was defined as treatment durations within 0.5 days below the minimum to 0.5 days above the upper recommended range; c. Prolonged treatment was defined as the upper limit of the guideline-recommended duration, with an additional 0.5 days margin to correct for administration time; d. Determined only in patients with prolonged treatment. Excess days were calculated as the difference between the length of therapy and the guideline-recommended upper limit plus 0.5 days; e. Other urinary tract infection; f. Other pneumonia or CAP without an PSI, AMBU-65 or CURB-65 score; g. 7 days for women with diabetes mellitus, women receiving immunosuppressive medication, pregnant women not treated with amoxicillin clavulanic acid, or male patients. 5 days for pregnant women receiving amoxicillin clavulanic acid or healthy nonpregnant women; h. 7 days in case of doxycycline use.

To address the secondary objective, identification of risk factors associated with prolonged antibiotic treatment, we used the same outcome of prolonged treatment. The following risk factors were included: age, sex, body mass index, medical specialty (surgical vs nonsurgical), hospital of admission, intravenous-to-oral antibiotic switch (oral-only, intravenous-only, or switch), antibiotic use at admission, being discharged before guideline-recommended treatment duration, positive blood cultures, positive sputum cultures, positive urinary cultures, positive wound cultures, infectious diagnosis category, use of systemic immunosuppressive medication, initial C-reactive protein (CRP) level, change in CRP between the final and prior maximum measurement (increased/remained the same level, decreased, or not measured), initial leukocyte count, change in leukocyte count between the final and prior maximum measurement (increased/remained the same level, decreased, or not measured), presence of fever during the first 24 hours of hospitalization, absence of fever during the last 24 hours measured within the guideline-recommended treatment duration, highest Early Warning Score (EWS) during the first 24 hours of hospitalization, and highest EWS during the last 24 hours measured (see Supplement S2 for definitions).

Data collection

All four hospitals have ChipSoft HiX as their EHR and prescribing system. Pseudonymized data was extracted from the EHR using CTCue (version v4.15.2). CTcue is a text-mining software application designed to construct rule-based queries and perform contextual analysis for patient selection and data extraction from both structured and unstructured EHR data. ^{Reference van Laar, Kapiteijn, Gombert-Handoko, Guchelaar and Zwaveling17} The following parameters were extracted uniformly across all hospitals, using predefined queries developed by one researcher (SvdE): baseline characteristics, including treatment restriction defined as the presence of any documented care limitation such as “do not resuscitate” or “no admission to the intensive care unit,” hospital data, antibiotics prescribed during hospitalization and at discharge, laboratory results, vital signs, microbiology findings, and (co-)infections diagnoses.

Statistical analysis

Continuous baseline characteristics were reported using mean and standard deviation or median and interquartile range, as appropriate. Categorical baseline characteristics were reported as frequencies and percentages. Differences in characteristics between patients receiving prolonged treatment and those not receiving prolonged treatment were analyzed using the chi-square test, Fisher’s exact test, Student’s t-test, or Mann–Whitney U-test, as appropriate. Multiple imputation of missing predictors was performed, with the number of imputation sets determined by the covariate with the highest percentage of missingness. Accordingly, twenty imputation sets were used in the analyses. Backward stepwise multivariable logistic regression was used to identify risk factors for prolonged treatment for RTI, UTI, and SSTI separately. Variable selection was based on Wald tests, and variables with a P value ≥ .2 were removed from the model. Model performance was evaluated using the C-statistic and calibration slope (“shrinkage factor”). Internal validity was assessed through 50 bootstrap replications to obtain the shrinkage factor and the bias-corrected estimates of the C-statistic. Based on the bias-corrected C-statistic, the model was classified as poor (<0.6), fair (0.6–0.7), moderate (0.7–0.8), good (0.8–0.9), or strong (0.9–1.0). Effect estimates were reported as odds ratios (ORs) with 95% confidence intervals (CIs), including both unadjusted and adjusted ORs (aOR). The unadjusted ORs were corrected for optimism using the shrinkage factor obtained through bootstrapping, resulting in the aORs.