Predictors for length of hospital stay in patients with community-acquired Pneumonia: Results from a Swiss Multicenter study

Herein, we used the clinical data from all patients with CAP enrolled in the multicenter ProHOSP study [12]. The design of this study has been reported in detail elsewhere [12]. In brief, from October 2006 to March 2008, a total of 1359 consecutive patients with presumed lower respiratory tract infection (LRTI) from six different hospitals located in the northern part of Switzerland were included. In total, 925 patients had the definite diagnosis of CAP. Patients were randomly assigned to an intervention group, where guidance of antibiotic therapy was based on procalcitonin cut-off ranges, and to a standard group in which guidance of antibiotic therapy was based on enforced guideline recommendations without knowledge of procalcitonin. The baseline characteristics for both groups were similar and both were treated according to current guidelines.

Patients had to be over 18 years old with a definite diagnosis of CAP to be eligible for this analysis. Inclusion criteria included the presence of at least 1 respiratory symptom (cough, sputum production, dyspnea, tachypnea, pleuritic pain), plus at least 1 finding during auscultation (rales, crepitation), or 1 sign of infection (core Body temperature >38.0°C, shivering, or Leukocyte count >10000/μL or < 4000/μL) independent of antibiotic pretreatment. CAP was defined as a new infiltrate on chest radiograph. Exclusion criteria included patients with active intravenous drug use, severe immunosuppression other than corticosteroid use, life-threatening medical comorbidities leading to possible imminent death, patients with hospital acquired pneumonia and patients with chronic infection necessitating antibiotic treatment. We only included patients who survived their hospital stay for this analysis, as early death would induce a bias for LOS prediction. Patients were examined on admission to the emergency department by a resident supervised by a board-certified specialist in internal medicine. The standardized baseline assessment included medical history, clinical examination, lab tests and chest X-ray. For all patients with CAP, the PSI and the CURB65 was calculated on admission as described elsewhere [6, 7]. The study protocol was approved by all local ethical committees, and written informed consent was obtained from all participants.

The primary endpoint of this analysis is LOS defined as the time from hospital admission to hospital discharge. We recorded baseline data including demographic characteristics (age, gender), comorbid diseases, and living situation (at home, nursing home residents or living with continuous nursing support at home), insurance status, clinical presentation on admission and initial results from blood analysis. All patients were followed until hospital discharge and we further assessed adverse events during the hospital stay including need for ICU admission and development of CAP specific complications (empyema).

Variables are presented as medians and interquartile range and a two group comparison was made with Wilcoxon-MWU tests. In a first step, we assessed the association of different baseline characteristics with LOS in univariate Cox regression models with time to hospital discharge of patients as the endpoint of interest. For the time to event analysis, patients were censored at the time of hospital discharge. Further, we calculated multivariate Cox regression analysis adjusted for all covariates to assess which parameters had an independent association with longer LOS. Of note, hazard ratios (HRs) lower than 1 correspond to an association of the factor with longer LOS, while high HRs correspond to earlier discharge. The proportional hazards assumption of Cox regression models was evaluated by graphical display and analysis of the scaled Schoenfeld residuals as recommended [13]. In a 50:50 split sample analysis, we used all independent predictors from the multivariate model to calculate two weighted clinical decision rules assigning points to each predictor based on the magnitude of association (i.e. HR 0.8-0.9 = 1 point, 0.7-0.8 2 points, 0.6-0.5 = 3 points, <0.5 = 4 points); the prediction of based on the first 50 % of CAP patients was then used to predict LOS in the second half of patients to assess calibration. We developed two models: a first one considered baseline criteria only, and a second model also considered factors during follow up. To illustrate the association of this rule and LOS, we displayed data in Kaplan Meier survival curves stratified by the number of predictors.

In a sensitivity analysis we also investigated whether randomization of the initial study had an effect on this analysis. We found no association of the intervention on LOS (HR 1.05, 95%CI 0.92-1.20). Also, when including an interaction term into the regression model, we found no evidence of effect modification. Therefore treatment assignment was not further considered in this analysis.

All testing was two-tailed and P values less than 0.05 were considered to indicate statistical significance. All calculations were performed using STATA 9.2 (Stata Corp, College Station, Texas).

From a total of 925 patients, 875 (94.6 %) patients survived their hospital stay and were included in the further analysis. Baseline characteristics of the study cohort are presented in Table 1. The median age of the patients at the time of study enrolment was 72 years and 42 % were women. Patients had important comorbidities including chronic pulmonary disease in 29 %, congestive heart failure in 16 % and diabetes in 17 %. A total of 49 % of patients were classified in high risks PSI classes IV or V.

A total of 5 % of patients were not hospitalized. The mean LOS of the overall cohort was 9.8 days (95%CI 9.3-10.4). We assessed LOS in Cox regression models where low HRs correspond to an association of the factor with longer LOS. Older age, respiratory rate >20 pm, confusion, residing in a nursing home, or need for regular nursing assistance at home, as well as different pre-existing comorbid conditions including chronic pulmonary disease, congestive heart failure and diabetes showed a significantly decreased chance to be discharged from the hospital in the univariate Cox regression analysis (Table 2). Furthermore, markers of high disease severity such as high procalcitonin levels >0.5ug/L, high CRP levels >150 mg/dL and PSI class were associated with longer time until hospital discharge. Similarly, positive blood cultures and multilobar CAP, as well as development of empyema or admission to the ICU during the hospital stay were also significantly associated with higher LOS.

Using a 50:50 split sample analysis approach, we calculated two multivariate prediction models: one for factors being present on hospital admission only, and one with all factors on admission and during hospital stay (Table 3). In the admission only model, older age, respiratory rate >20 pm, nursing home residence, chronic pulmonary disease, diabetes, multilobar CAP and the pneumonia severity index class were independently associated with longer LOS. We calculated a weighted prediction score assigning points based on the magnitude of association. This was also confirmed in Kaplan Meier curves for time to hospital discharge, where more points were associated with significantly longer time until hospital discharge (p logrank <0.001) (Figure 1). This prediction rule also showed a good calibration when used in the validation cohort (Figure 2).

An overall model including not only baseline factors, but also follow up information, namely blood culture results, need for ICU admission and development of complications (empyema) was also calculated. Both, need for ICU admission and development of complications (empyema), as well as initial albumin level were strong LOS predictors while multilobar CAP did not reach statistical significance in this second model. A weighted score based on these factors showed again a high discrimination of LOS and a good calibration when used in the validation cohort.