Methicillin-resistant Staphylococcus aureus (MRSA) is a common cause of nosocomial pneumonia (NP) and healthcare-associated pneumonia (HCAP) [1,2,3,4]. Although incidence is higher in ventilator-associated pneumonia (VAP) [1], that observed in non-ventilated patients is not negligible in areas with a high prevalence of MRSA [3,4,5,6]. Mortality in this patient population is high (40–60%) and, therefore, MRSA should be considered when designing empirical therapy regimens. Despite the remarkable efforts made in recent years to improve prognosis in these patients [2,3,4, 7], the clinical settings that warrant additional coverage against MRSA are poorly defined outside the context of VAP [4]. Therefore, the identification of predictors for MRSA would be helpful in selecting patients that might benefit from including anti-MRSA drugs in empirical therapy. In this context, this study aimed to characterize if the odds of MRSA etiology in certain non-ventilated hospitalized patients with NP/HCAP could be defined by clinical predictors that can be valuable for empirical treatment.

Materials and methods

Study subjects and design

This is a retrospective, observational, multicenter, case–control study (SENECA study, Pfizer Protocol NRA5950067), performed in 15 university hospitals in Spain. The case patient population included non-ventilated adult patients hospitalized for > 48 h with NP or HCAP caused by MRSA over a period of 5 years (from January 2008 to December 2013). Patients admitted to the intensive care unit (ICU) were excluded. Table 1 shows the clinical criteria used to diagnose pneumonia. We considered that pneumonia was nosocomial when it debuted at least 48 h after hospital admission. HCAP was defined as community-onset pneumonia in patients with any of the following: intermediate care facility residents (e.g., nursing home or rehabilitation centers), hospitalization for > 48 h in the 90 days prior to the current admission, or reception of home intravenous therapy or dialysis in the previous month [8]. MRSA was defined as an oxacillin minimum inhibitory concentration (MIC) ≥ 4 μg/mL, according to the guidelines of the Clinical and Laboratory Standards Institute (CLSI) [9].

Table 1 Clinical criteria for pneumonia diagnosis

The cases were identified from the records of the local microbiology services databases. Patients from whom MRSA was isolated from valid sputum (> 25 neutrophils and < 10 epithelial cells per field), quantitative bronchial aspirate (BAS)/bronchoalveolar lavage (BAL), brush catheter, pleural fluid, or blood cultures were eligible. Medical records were reviewed in search of patients meeting the diagnostic criteria for pneumonia.

Two controls were included for each case, matched by hospital and type of pneumonia (NP or HCAP). Patients with NP or HCAP and with microbiological isolation of bacteria other than MRSA from respiratory samples, pleural fluid, or blood cultures were eligible as controls; those dated immediately before or after the respective case in the same hospital were selected. Controls of NP cases were not matched for healthcare exposure before present admission.


The ethics committee of the Hospital Universitario Reina Sofía approved the use and analysis of clinical data needed to conduct the study and waived the need for informed consent due to the retrospective design of the study (code NRA 5950067, reference 1893). The database used for the analysis was anonymized. The study was classified by the Spanish Agency of Medicaments, Ministry of Health, as an “observational study” (No-EPA).

Clinical variables and definitions

Clinical variables and definitions are shown in Table 2.

Table 2 Univariate and multivariate analysis of risk factors for methicillin-resistant Staphylococcus aureus (MRSA) pneumonia in non-ventilated hospitalized patients

Calculation of sample size

To estimate an odds ratio (OR) ≥ 1.6 with a 95% significance level, a two-sided test power of 80%, and assuming a 60% exposure rate in the control group, we estimated that 761 patients should be included (254 cases and 507 controls). Estimating a replacement rate of 10%, a sample size of 837 patients (279 cases and 558 controls) was calculated and aimed.

Statistical analysis

Quantitative variables were expressed as mean ± standard deviation (SD). We calculated univariate ORs and the 95% confidence intervals (CIs) for all the potential risk factors of MRSA pneumonia using univariate logistic regression. For all analyses, patients without respiratory sample cultures in the previous 12 months were considered to have no MRSA isolates. All variables with p < 0.25 in the univariate conditional logistic regression analysis were included in a multivariate conditional logistic regression model [10] and selected using a stepwise backward process. The area under the receiver operating characteristic (AUROC) curve for the final multivariable model was determined; the sensitivity, specificity, and predictive values of the model with respect to observed data were calculated. The goodness-of-fit of the model was determined by the Hosmer–Lemeshow test.

All analyses were conducted with SPSS software. p-Values ​​ < 0.05 were considered to be statistically significant. All reported p-values ​​are two-tailed.


Description of cases and controls

A total of 140 cases and 280 controls were included in the study. The 140 cases of MRSA pneumonia in non-ventilated patients were identified in the selected centers: 77 cases of NP (55%) and 63 cases of HCAP (45%); 12 cases of NP (15.6%) and 18 cases of HCAP (28.6%) were residents of a long-term care facility. The median (range) of the proportion of methicillin resistance among S. aureus isolates in the hospitals in which the participating patients were admitted was 32% (16–47). A full description of the clinical and demographic variables of cases and controls is shown in Table 2. The etiologies of the 280 controls are shown in Table 3.

Table 3 Etiology of pneumonia in the 280 control patients

Risk factors for MRSA pneumonia

The variables that were significantly associated with the risk of MRSA pneumonia in the univariate analysis were age, dependency, MRSA respiratory infection/colonization within the previous year, hospitalization in the previous 90 days, and the use of proton pump inhibitors in the previous month. For the analysis, patients without cultures in the 12 months prior to diagnosis were considered to have no MRSA infection/colonization (25/140 [17.9%] vs. 4/280 [1.4%]). The variables associated with the risk of MRSA pneumonia in the conditional multivariate logistic regression model were: (i) MRSA infection/colonization in the previous year (OR 14.81, 95% CI 4.13–53.13, p < 0.001); (ii) hospitalization in the previous 90 days (OR 2.41, 95% CI 1.21–4.81, p = 0.012); and (iii) older age (OR per year, 1.02, 95% CI 1.001–1.05, p = 0.040) (Table 2). The AUROC curve for the multivariable model was 0.72 (95% CI, 0.66–0.78). The multivariate model predicted MRSA pneumonia for the observed data with a sensitivity of 74.5% (95% CI 65.3–83.6), a specificity of 63.3% (95% CI 56.0–70.6), a positive predictive value of 52.5% (95% CI 43.9–61.2), and a negative predictive value of 82.0% (95% CI 75.3–88.8) (Table 2).


We have studied whether some clinical variables can define the risk of MRSA NP/HCAP in order to characterize the situations in which coverage for MRSA in the empirical treatment of NP or HCAP is needed. Our major findings are: (1) MRSA respiratory infection/colonization within the previous year, hospitalization in the previous 90 days, and older age are associated with the risk of MRSA pneumonia; (2) These risk factors are applicable to both HCAP and NP in settings with a high prevalence of MRSA; and (3) The identified clinical variables define a low-risk population in whom coverage for MRSA could be avoided.

Other studies have reported risk factors for community-acquired pneumonia due to MRSA [8] or other multidrug-resistant bacteria [11,12,13,14]. The primary risk factors in the “Aliberti Score” [11, 12] included hospitalization in the previous 3 months and residence in a nursing home. Other specific MRSA risk factors are related to dependence on the health system (advanced age, parenteral antibiotic therapy, or neurological impairment) [15,16,17], which may be surrogate risk factors for respiratory colonization.

Our study confirms the above risk factors for patients in the community, such as advanced age and hospitalization in the previous 3 months. However, a further contribution of our study is that MRSA colonization/infection in the previous year is a risk factor for MRSA HCAP and NP. The value of the isolation of MRSA to guide the empirical treatment of NP/HCAP has been discussed. Although the 2005 Infectious Diseases Society of America (IDSA)/American Thoracic Society (ATS) guidelines recognize the value of MRSA colonization to guide empirical therapy [4], no prospective study has evaluated the use of MRSA screening to guide the empirical treatment, so there is not enough evidence to consider this as a definitive risk factor for MRSA pneumonia. The 2016 IDSA/ATS guidelines insist on the negative predictive value of the MRSA screening only when the prevalence of MRSA is low [18]. Our results should be understood and interpreted within the framework of the high MRSA prevalence (32%) observed in the centers included in the study. In this context, our multivariate model has a high negative predictive value (82%) and permits identifying patients at low risk in whom colonization studies at admission or empirical coverage for MRSA could be avoided. It is difficult to make a recommendation in the remaining patients, since the model has a positive predictive value of 52.5%.

This study has some limitations which have to be pointed out, mainly its retrospective design and a much lower recruitment rate than previously estimated. This was due to the fact that the actual incidence of NP/HCAP in the centers included in the study was lower than expected. Furthermore, to avoid inclusion of a non-diagnosed MRSA pneumonia, the controls used in the study were patients with pneumonia of known etiology, and, therefore, the results cannot be applied to patients in whom the etiology of pneumonia is not known. Finally, our results would be mainly applicable to centers with a similar epidemiology of MRSA.

In summary, our study establishes a set of clinical predictors of MRSA NP/HCAP that can be used to identify patients at low risk for MRSA pneumonia. Given the scarce available information to take decisions with regard to coverage against MRSA in HCAP and NP outside ICUs, our findings may be clinically valuable in helping to optimize the diagnosis and empirical treatment of MRSA pneumonia.