Antibiotics account for one-third of hospital’s pharmacy budget, with between 25% and 50% hospitalized patients receiving antibiotics
. In hospitals, nearly all physicians prescribe antibiotics, and their prescriptions are influenced by characteristics of patient population, physician’s prescribing habits and local resistance patterns
. Inappropriate use of antibiotics results in a variety of adverse outcomes: narrow coverage increases the risk of therapeutic failure whereas broad coverage increases the risk of superinfection
The aim of the present study was to investigate adherence by treating physicians to written ID recommendations inserted in clinical records and the potential influence that this adherence had on clinical outcome in hospitalized intravenous antibiotic-treated patients in a tertiary university teaching hospital in Spain.
Patients and methods
A prospective, randomized, controlled study was carried out from January 2008 to December 2008 in the Hospital Universitario Marques de Valdecilla, Santander, Spain, a tertiary universitary hospital with 874 beds. The study protocol was approved by the Ethics Committee of the Autonomous Community of Cantabria (IFIMAV, Spain), the Fondo de Investigaciones Sanitarias [Registered number: FIS PI06/90094], and the Instituto de Formación e Investigación Marqués de Valdecilla (IFIMAV) [Registered number: API 06/03]. The patient’s informed consent was waived because the study was directed to treating physicians, not to patients, since the primary objective of the study was to assess adherence by treating physicians to ID recommendations using a new method described below. Potential influence of adherence by treating physicians on patient's outcomes was only secondarily assessed. The possibility of ID consultations is part of normal practice in the hospital, and as in routine practice, treating physicians participating in the study could ask for ID oral consultation at any time.
Patients admitted in General Surgery, Pneumology and Endocrinology departments that were receiving intravenous antimicrobial therapy prescribed by their responsible physicians for 3 days were identified by the Pharmacy department. The ID team was daily contacted and patients were randomised by groups (stratified randomization by clinical units) to intervention or non-intervention using the EPIDAT 3.1 programme (Dirección Xeral de Saúde Pública, Xunta de Galicia & Organización Panamericana de la Salud. Santiago de Compostela, Coruña, Spain, 2003). Interventions consisted of insertion in medical records of written treatment recommendations by an ID physician based on International
[11–13] and Spanish
[14–16] treatment guidelines adapted to local data on antimicrobial susceptibility after examination of patients, clinical records and microbiological data (if any) for each patient in the intervention group. Written recommendations (in specifically designed study forms) were inserted in the patient’s records in order to make it available to treating physicians, with a statement indicating availability for oral consultation if desired. Consultation was also available as part of daily practice for physicians treating patients in the non-intervention group (without insertion of written recommendations), but in the case of consultation the patient was withdrawn from the study. Paediatric patients, chronic dialysis patients, those receiving oral antibiotic therapy and those in which a member of the ID department had prescribed the initial antimicrobial regimen, were excluded.
Demographic and clinical data (including comorbidities), reason for antibiotic administration, regimen of antibiotic treatment, imaging and microbiological data, and microbiological and clinical outcomes were recorded both for patients in the intervention and non-intervention groups. The Charlson comorbidity index unadjusted by age
 (age and sex were considered as separate variables), and the McCabe score
 were calculated. Regimens of antibiotic treatment were classified as empirical, microbiologically-based or prolonged prophylaxis (antibiotic prophylaxis administered for >24 hours). Length of hospitalization (time from admission to discharge/death) was calculated. Patients with definite infection were those who met clinical criteria for infection with or without microbiological documentation. Infection was considered as nosocomial when signs/symptoms initiated 48-72h after admission and as healthcare-associated infection if infection occurred in patients who came from nursing homes or chronic care centres, had domiciliary hospitalization, had attended the hospital for oncological chemotherapy or had been hospitalized within the previous 30 days.
The appropriateness of empirical treatments prescribed by treating physicians (physicians of the primary service) was assessed and treatments were classified as adequate (agree with antibiotic choice, dosage, dosing interval and duration of treatment) or inadequate. Treatments were considered inadequate when one or several criteria following the Erbay’s modification
 of Kunin’s criteria
 were met. When IDs disagreed with the need for an antibiotic, treatments were classified as “unnecessary” by assigning a new category and thus eliminating this criterion from the Erbay’s category “inadequate”. In the intervention group, the degree of adherence to intervention was assessed at day +7 and +10 (from initiation of intravenous antibiotic treatment) and classified as complete (administration within 24h after consultation of the recommended antibiotic therapy with deviations lower than 20% for the dose and lower than ±30% for treatment duration), partial (administration of the recommended antibiotic therapy with deviations greater than 20% for the dose and/or greater than ±30% for duration and/or initiation between 24h and 48h after consultation) or non-adherence (non prescription of the recommended regimen or initiation after 48h from consultation). When the recommended duration of treatment was longer than 10 days, adherence was assessed at discharge.
Patients were followed during hospitalisation and at discharge they were clinically evaluated as “clinical success” (cure or improvement) or “clinical failure” (absence of improvement, worsening or death during hospitalization). Cure was considered when patients were discharged from the hospital with resolution of acute signs and symptoms of infection and improvement when patients were discharged from the hospital with oral antibiotic treatment. Patients with an initial positive culture and at least one follow-up culture were considered microbiologically evaluable and were evaluated as “microbiological success” (eradication) or “microbiological failure” (persistence or superinfection) at discharge.
Comparisons between proportions were performed by the χ2 test and the Fisher’s exact test, when necessary. For quantitative variables, since data did not showed normality in the Kolmogorov-Smirnoff test, the Kruskal-Wallis and Mann–Whitney tests, when necessary, were used. Bivariate analyses were performed comparing all variables between the group of patients with treatment showing adherence (complete adherence or complete plus partial adherence) versus those showing non-adherence, as well as between the group of patients showing clinical success versus those showing clinical failure. In order to avoid false associations in multiple comparisons in the bivariate analyses, the Bonferroni correction was applied and the p<0.001 was considered statistically significant to minimize type 1 error.Two different logistic regression models (step–wise procedure) were performed: one using “adherence” as dependent variable and another using “clinical failure”. In both multivariate analyses, independent variables were those showing differences (p≤0.1) in the previous bivariate analyses performed. In the multivariate analysis using “clinical failure” as dependent variable, adherence (both considered as complete adherence and as complete + partial adherence) was one of the independent variables introduced. Interactions and linear dependence between independent variables were previously controlled. Statistical analyses were performed using SPSS v 18 programme (SPSS Inc, Chicago IL). The models showing the highest R2 were considered.