Background
Community-acquired pneumonia (CAP) is an important reason for hospitalization worldwide [
1‐
3]. It has been estimated that the total costs associated with CAP amount to approximately 11 billion euros annually in Europe, with approx. 5 billion euros accounting for in-hospital CAP costs [
1]. In the Netherlands there are an estimated 25,000-36,000 hospital admissions for CAP each year, [
4] with an estimated total costs of about 100 to 178 million euro annually [
5,
6]. The intramural costs are mainly determined by the length of hospitalization and site of care (medical ward or intensive care unit, ICU) [
5,
6].
In choosing the optimal antibiotic treatment strategy for CAP, effectiveness, cost-effectiveness and ecological effects of antibiotics should be taken into account. Optimally, this would consist of a strategy associated with the best patient outcome at the lowest price and with least selective pressure for antibiotic resistance. The three treatment strategies most widely used are beta-lactam monotherapy, beta-lactam/macrolide combination therapy, and fluoroquinolone monotherapy. From an ecological perspective beta-lactam monotherapy is preferred over beta-lactam/ macrolide combination therapy, and fluoroquinolone monotherapy, since the latter two drug classes have been associated with resistance development during treatment [
7,
8].
In a cluster-randomized cross-over trial of patients hospitalized with CAP to non-ICU wards, a strategy of beta-lactam monotherapy was non-inferior to beta-lactam/macrolide combination therapy, and fluoroquinolone monotherapy in terms of all-cause day-90 mortality (CAP-START study) [
9]. The quinolone monotherapy strategy was associated with a shorter length of intravenous treatment, but this was not reflected in a statistically significant shorter length of stay. In the current study, we set out to conduct a cost-minimization analysis of these different antibiotic strategies and a cost-effectiveness analysis from a third payer and a social perspective.
Discussion
In these analyses, we have demonstrated that the differences in costs associated with either of three preferred empirical antibiotic treatment strategies (i.e., beta-lactam monotherapy, beta-lactam/macrolide combination therapy, or fluoroquinolone monotherapy) for patients hospitalized for community-acquired pneumonia did not reach statistical significance. Together with non-inferiority of the beta-lactam monotherapy strategy for day-90 mortality [
9] and the perceived preference of beta-lactam monotherapy from an ecological perspective, the current analysis supports the use of beta-lactam monotherapy as preferred empirical treatment for these patients.
This is the first comparison of costs and cost-effectiveness for different preferred antibiotic treatment strategies in patients hospitalized with CAP. Our study has several strengths. First, because this was a pragmatic trial, where patients were included during strategy periods regardless of the actual antibiotics used, the intention-to-treat analysis of our study is well generalizable to daily clinical practice. All patients that received antibiotic treatment for a working diagnosis of CAP and who were hospitalized to a non-ICU medical ward, were eligible. Second, the cluster-randomized design allowed the immediate start of the allocated antibiotic treatment because individual randomization was not needed. This minimizes effects of other antibiotics prescribed in the Emergency Departments before study randomization. Third, because of the cross-over design, all hospitals applied all three strategies, thus minimizing confounding bias. As a result, baseline characteristics of the three strategies were very comparable. Fourth, we have collected comprehensive data on antibiotic treatment and medical procedures that allowed us to estimate hospitalization costs per patient. Using 2,000 bootstrapping samples and five imputations per sample, we were able to provide robust estimates and confidence intervals for the different cost categories. Our estimated costs per CAP admission are in line with previously published data from the Netherlands [
5,
6]. Fifth, different economic viewpoints were pursued in the current analysis. The (reduced) third payer perspective and the societal perspective taking the friction approach all gave the same direction and magnitude of effect. The large confidence intervals observed in the societal perspective with human capital approach was due to the low number of fatal cases under 65 years of age and due to working status being unknown for unreturned 28
th day questionnaires. This led to unstable imputation of working status, since these variables also interact i.e. the proportion of returned questionnaires was lower for patients that had died at day 90, thus increasing confidence intervals.
Our approach had certain limitations. We had limited data on medication use other than antibiotics. Although it seems unlikely that one of the antibiotic treatment strategies would be associated with other patterns of non-antibiotic medication use, if so, we may have slightly underestimated the costs. 28th day questionnaires, used for DNHC and INHC estimation, were returned by approximately 40% of the participants. We used multiple imputation to deal with missing data because response to the 28th day questionnaire was obviously dependent on clinical outcome and was related to baseline characteristics (e.g. dependency in activities of daily living or hospitalizations in the previous year). This may have increased uncertainty for the third payer and societal perspectives, and it certainly did for the societal perspective with human capital approach, as explained previously.
The number of days on intravenous antibiotic treatment was significantly lower during the fluoroquinolone monotherapy strategy (hazard ratio for time to switch to oral treatment 1.29, 95% CI 1.15–1.46) [
9]. This was fully explained by the larger proportion of patients starting with oral treatment from the day of admission, despite the similar baseline characteristics between the different strategies, and can, therefore, not be attributed to a faster clinical response. The known high bioavailability of oral fluoroquinolones [
16] may have stimulated physicians to directly start with oral antibiotics and this may have contributed to the more favourable point estimate of difference in costs seen in the fluoroquinolone monotherapy period. Whether the same proportion of patients could start with oral beta-lactam monotherapy without compromising patient outcome remains to be elucidated.
In an open-label randomized controlled trial from Switzerland, beta-lactam monotherapy was not non-inferior to beta-lactam/macrolide combination therapy in establishing clinical stability after seven days of antibiotic treatment [
17]. This study was not designed to determine non-inferiority for day-30 or day-90 mortality, and there were no statistically significant or clinically relevant differences in outcome between both study arms. Time to clinical stability was not determined in our study, however, length of stay was significantly longer for the beta-lactam/macrolide combination strategy, and consequently also the costs per patient were higher, although not statistically significant. This seemingly opposite finding might in part be explained by the maximized adherence to the allocated antibiotic, i.e. the strict criteria for switching antibiotic treatment, which could only have disadvantaged the beta-lactam monotherapy arm in the Swiss study. The current analysis shows that any benefit of beta-lactam/macrolide combination treatment on time to clinical stability, if present, does not lead to cost reduction.
Generalizability of the estimated costs may depend on several factors, the most important of which are the duration of hospitalization, ICU admission, the length of intravenous and oral antibiotics, and post discharge patterns of healthcare use. Although the actual reported costs are obviously specific for the Netherlands, the relative differences in costs for medication are comparable internationally [
18,
19]. As the generalizability of clinical outcome may depend on the proportion of CAP caused by pathogens not covered by beta-lactam monotherapy, as discussed previously [
9], we think that the cost-efficacy will be similar in most regions with comparable etiology.