INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Pneumonia is the leading infectious cause of death and the fourth overall cause of mortality in the elderly (1). The presence of underlying comorbid diseases, an impaired mucociliary clearance, and a waning immunity contribute to the increased incidence of pneumonia in that age group.

The diagnosis of a specific microbial etiology for pneumonia in the elderly remains imprecise. Although many studies have reported series of potential pathogens in the acute and the long-term settings (2), solid data to base treatment upon has been lacking. Nearly all studies utilizing expectorated sputum for diagnosis have failed to identify a clear etiologic organism in 30 to 50% of the cases. In addition, several studies have shown that the interpretation of these cultures is fraught with problems because of poor quality specimens, upper airway contamination, and oropharyngeal colonization (5, 6).

The knowledge of the predominant microbial patterns in the elderly with severe pneumonia represents the basis for empirical antimicrobial treatment. Because of the substantial mortality from pneumonia among the elderly, it is essential that initial antimicrobial therapy have activity against the causative organisms. The ATS guidelines have issued a series of recommendations for initial treatment based on the patient's age, severity of pneumonia, and coexistent comorbid illnesses (7). However, these guidelines have not been validated prospectively in the elderly group. Furthermore, no study has addressed comprehensively the impact of functional status on the microbial etiology of severe pneumonia in the geriatric population.

We therefore conducted an interventional prospective study in patients 75 yr of age and older to determine the prevalence of respiratory pathogens, and the impact of comorbidity and functional status on the microbial etiology of severe pneumonia in the very elderly.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient Population

A prospective cohort study was carried out between June 1996 and September 1999 at two university-affiliated hospitals after obtaining approval from the Institutional Review Board of the University at Buffalo. An informed consent was obtained from the next of kin to authorize the study.

All patients  75 yr of age admitted to the intensive care unit (ICU) with the diagnosis of pneumonia requiring mechanical ventilation were included in the study if they met all the following criteria: (1) the presence of symptoms of lower respiratory tract infections; (2) a new radiographic infiltrate compatible with pneumonia; (3) the presence of any two of the following clinical parameters: temperature > 38° C, leukocytosis (> 11.0 × 10⁹/L) or neutropenia (< 3.5 × 10⁹/L), purulent sputum, or change in character of respiratory secretions, and increasing arterial-alveolar gradient; and (4) acquisition of the infection outside the confine of a hospital. Patients were excluded if they were transferred to the ICU after 24 h of initial admission to the ward, hospitalized within 30 d of admission, or for whom consent could not be obtained. Patients with pneumonia as a result of witnessed aspiration or as an expected terminal event from metastatic cancer were also excluded.

Data Collection at Study Entry

Data collected on ICU admission included age, sex, residence prior to admission, comorbid illnesses, Activity of Daily Living (ADL) score, clinical symptoms, and chest radiographic pattern and distribution. The most abnormal readings of vital signs, laboratory data, arterial blood gases, and mechanical ventilation requirements were recorded within the first 24 h. In addition, Glasgow Coma Scale and 24-h urinary output were abstracted from ICU flowsheets. Antimicrobial therapy received prior to hospital admission and during hospitalization was retrieved. Antibiotic treatment was considered inadequate if the microbial susceptibility profile showed a resistant pattern to the empiric drug(s) used on admission. At the clinical end points of hospital discharge or death, the length of stay, days of intubation, septic shock as defined by Bone and colleagues (8), pleural empyema, renal failure defined as a rise in serum creatinine > 2 mg/dl from baseline or requiring hemodialysis, and code status with regard to "do not resuscitate" (DNR) orders were additionally recorded.

The ADL score was abstracted from a standardized patient-review instrument included in all patients' charts (9). Patients were assigned an ADL score in each of the six major areas of activity: eating, toileting, feeding, bathing, mobility, and continence, ranging from 1 if they were fully independent, 2 if they were partially independent, and 3 if they were completely dependent. The ADL score was calculated by adding the points assigned for each activity, and it ranged from 6 to 18. Three categories were arbitrarily created: ADL I that corresponded to ADL scores from 6 to 8, ADL II with scores from 9 to 13, and ADL III with scores from 14 to 18. The severity of illness was assessed by the Acute Physiology and Chronic Health Evaluation II (APACHE II) (10).

Comorbidities were classified as follows: a cardiac comorbidity is considered present if a treatment is being provided for coronary artery disease, congestive heart failure, arrhythmias, or valvular heart disease; pulmonary comorbidity: treatment for chronic obstructive lung disease, or interstitial lung disease; renal comorbidity: preexisting renal disease with documented abnormal creatinine level prior to hospitalization; hepatic: preexistent chronic viral hepatitis or liver cirrhosis; endocrine: treatment for diabetes mellitus; central nervous system: presence of symptomatic acute or chronic vascular or nonvascular encephalopathy; neoplastic disease: presence of active malignancy (solid tumor or hematologic malignancy) at the time of presentation; immunosuppression: the use of steroids at a dose  20 mg/d for more than 2 mo, HIV infection with CD4 count < 200 cells/mm³, neutropenia with absolute neutrophil count < 1,000 cells/mm³, or the use of cytotoxic drugs.

Microbiologic Evaluation

Specimen collection comprising blood for blood cultures, and serology, pleural fluid, urine samples, and tracheobronchial secretions. Sera were collected on admission and 4 to 6 wk thereafter and tested for evidence of complement-fixing antibody to influenza A and B, parainfluenza viruses 1 to 3, and adenovirus. The presence of circulating immunoglobulin G (IgG) antibodies to Chlamydia pneumoniae and Mycoplasma pneumoniae was detected with the indirect microimmunofluorescence test and the indirect fluorescence antibody system, respectively. Legionella pneumophila (serogroups 1 to 6) was diagnosed using indirect microimmunofluorescence antibody technique IgG. Urine samples were assayed for the Legionella antigen by the enzyme-linked immunosorbent assay. Specimens of lower respiratory tract secretions were obtained within the first 24 h of ICU admission via (1) distal blindly placed catheter for protected bronchoalveolar lavage (PBAL), (2) flexible fiberoptic bronchoscopy (FOB)-guided protected bronchoalveolar lavage (FOB-PBAL), (3) FOB-guided protected specimen brush (FOB-PSB), and (4) postmortem transthoracic needle aspiration (TNA) biopsy. Undiluted and serially diluted respiratory secretions samples were plated onto 5% sheep's blood, chocolate agar, and MacConkey plates. All cultures were incubated at 37° C under aerobic and anaerobic conditions and in CO₂-enriched atmosphere. After a 3-d incubation period, plates were assessed for bacterial and fungal growth. Gram and Ziehl-Neelsen stains and direct immunofluorescence for L. pneumophila were performed using undiluted samples. All bacterial species isolated by quantitative culture were identified by standard microbiologic technique. Results of quantitative cultures were expressed as colony-forming units per milliliter (CFU/ml).

Diagnostic Criteria

The etiology of pneumonia was determined by one of the following criteria: (1) blood cultures or pleural fluid yielding a bacterial or fungal pathogen in the absence of an apparent extrapulmonary focus; (2) seroconversion (fourfold increase in IgG titers for C. pneumonia, M. pneumoniae, L. pneumophila, and respiratory viruses such as influenza viruses A and B, parainfluenza viruses 1 to 3, and adenovirus); (3) positive urinary antigen for L. pneumophila type 1; (4) bacterial growth of  10³ colony-forming units (CFU)/ml of a pathogenic microorganism from a PSB specimen or TNA; growth of  10⁴ CFU/ml of a pathogenic microorganism from PBAL; or  10⁵ CFU/ml in BAL. Isolation of fungi from respiratory samples was considered diagnostic only in the presence of positive concomitant blood culture yielding the same organism.

Statistics

Data were analyzed using the NCSS 2000 statistical software (NCSS Statistical Analysis System, Kaysville, UT). Results are expressed as means ± SD. Univariate analysis was carried out using the chi-square test and Fishers exact test for categorical data and the t test for independent samples for continuous variables. The association between the most frequent pathogens and comorbid illness was done by using chi-square tests with odds ratios (OR) and 95% confidence intervals (CI). Multiple linear regression was used to evaluate differences in length of stay variables (days receiving ventilation and days in hospital) after controlling for sex, APACHE II, and ADL scores among those admitted from the community compared with nursing home residents. These covariates were selected because they may act as confounders of the relation between severity of illness and death and severity of illness and length of stay. Multivariate analysis was applied by using a stepwise logistic regression in order to assess which criteria present on admission were independently associated with mortality. Pairwise correlations between predictor variables and the variance inflation factor were computed to assess for multicollinearities according to the method described by Slinker and Glantz (11). Variables with association among each other were not included in the analysis. As for influential observations, no patients with outlier values in any variable were detected. All reported p values are two-tailed. The level of significance was set at 5%.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient Characteristics

A total of 136 patients were eligible for enrollment. Eleven refused to participate in the study, and 15 had no health proxy or a relative to sign the consent form. Overall, 104 patients (mean age, 82.3 ± 5.5 yr) were admitted to the ICU for respiratory failure. Fifty-seven patients (55%) were residing in the community and 47 (45%) were admitted from nursing homes. There were no characteristic clinical or chest roentgengraphic findings among those patients admitted from the community and those from nursing homes (Table 1). Seventy-eight percent of the patients (n = 81) had at least one comorbid illness. Congestive heart failure was present in 22 patients (21%). Thirty-two (31%) had prior documentation of chronic obstructive pulmonary disease (COPD), and 17 (16%) had evidence of neurologic deficit from a cerebrovascular accident.

The length of time receiving mechanical ventilation was similar between those with community-acquired pneumonia (CAP) and nursing home-acquired pneumonia (NHAP) (Table 2). After we used multiple linear regression to control for sex, APACHE II score, and ADL score, the difference in time receiving ventilation between the two groups remained nonsignificant.

Twenty-three patients (23%) had received antimicrobial agents prior to their ICU admission. Eight patients were receiving oral quinolone, three were receiving an oral macrolide, one was receiving an oral penicillin, and 11 had an injectable form of cephalosporin.

Diagnostic Yield of Applied Techniques

One hundred six invasive bronchial specimens (IBS) were recovered (Table 3). Sixty-one (58%) were obtained by PBAL, 19 (18%) by FOB-PBAL, 12 (11%) by FOB-PSB, and 14 (13%) by TNA. Twenty-nine organisms (48%) were isolated from PBAL, six (32%) from FOB-BAL, four (33%) from PSB, and five (36%) from TNA.

Ninety-five percent of the patients admitted from home and 83% of the nursing home patients had blood cultures on admission. Fifteen pathogens were identified for a yield of 14%. A diagnostic thoracocentesis was performed on seven patients, among whom empyema was diagnosed in 4 and they required chest tube drainage. Serology had a yield of 12% and was performed more significantly on patients admitted from the community compared with those from nursing homes (p < 0.02). A similar trend was noted also for those who had less comorbidities and in the subgroup of patients with central nervous system (CNS) disorders.

Staphylococcus aureus was identified predominantly from invasive bronchial sampling in 15 (83%) of the 18 isolates, whereas a pneumococcal etiology was as likely to be determined from blood cultures (50%) as from bronchial aspirates (42%). Legionellosis was diagnosed by urinary antigen in three of five patients (60%), and by direct fluorescent antibody in one of the respiratory specimens. All five cases were confirmed by serology. Mycobacterium tuberculosis was recovered from one patient only who had been living in a shelter for 5 yr.

The diagnostic yield was not significantly different in those admitted from the community when compared with those who were residing in nursing homes (53% versus 47%, p = 0.86). However, a positive etiology was more likely to be identified in those who had not received preadmission antimicrobial treatment (56 versus 14%, p < 0.01).

Complication rates attributed to the invasive diagnostic techniques were minimal and consisted of transient worsening hypoxemia (11%), postbronchoscopy fever (5%), and transient cardiac arrhythmia (2%). No pneumothoraces were noted.

Microbial Etiology

A microbial etiology could be determined in 55 (53%) of the 104 patients (Table 4). Overall, 70 pathogens were isolated, of which 44 were detected by invasive bronchial sampling and 26 by other techniques. S. pneumoniae represented 19% of the total isolates (13 out of 70) and was the predominant pathogen isolated from patients with CAP (14%). Legionella sp. was the second most frequent pathogen (9%) in CAP followed by S. aureus (7%) and H. influenzae (7%). Three of the five patients with Legionnaire's disease had underlying comorbidities. Two were steroid-dependent COPD, and one had myelodysplastic syndrome. Furthermore, six of the eight patients with S. aureus and H. influenzae CAP had either an underlying renal disease or COPD.

S. aureus was recovered primarily from patients admitted from nursing homes (29% in NHAP compared with 7% in CAP). The majority of the isolates were methicillin-sensitive except for three (21%) of the 14 who were methicillin-resistant. The prevalence of gram-negative enteric bacilli (GNEB) was similar in both groups (14% in the CAP group versus 15% in the NHAP group) with K. pneumoniae (n = 5) and Escherichia coli (n = 5) being the predominant isolates (10 out of 14). Pseudomonas aeruginosa was recovered from three patients, one from the CAP group who had bronchiectasis acquired from a previous history of pulmonary tuberculosis, and two from the NHAP group, one with underlying diabetes mellitus and the other with dense hemiplegia secondary to a cerebrovascular accident, respectively.

Mixed infections were present in 10 (18%) of the 55 patients in whom a microbial etiology was determined (four from the community and six from the nursing homes). Of these, four had at least three comorbidities or more, and three had two. COPD was the most reported of the comorbidities and was noted in five of the 10 cases.

Bacteremia was seen more frequently in patients admitted with NHAP (10 out of 15 patients with positive blood cultures) although the difference did not reach statistical significance (OR, 2.8; 95% CI: 0.9 to 8.8; p = 0.1). The pathogens isolated included S. pneumoniae, S. aureus, E. coli, Serratia sp., P. aeruginosa, and Enterococcus faecalis (Table 3). Four (67%) of the six cases of S. pneumoniae bacteremia were recovered from patients with CAP, whereas five (83%) of the six cases of gram-negative bacteremia were observed in those admitted from nursing homes.

Mortality was highest in cases of pneumonia caused P. aeruginosa (three out of the three cases, 100%), and L. pneumophila (three out of five cases, 60%). It was 44% for S. aureus (eight out of 18 cases), and 33% for S. pneumonia (four out of 12 cases).

Impact of Comorbidity and Functional Status on Microbial Etiology

There was no discernible association between microbial etiology and patients with cardiac comorbidity or those with underlying congestive heart failure. Patients with COPD, however, were more likely to have pneumonia caused by S. pneumoniae. Diabetes mellitus was associated with GNEB pneumonia, whereas the presence of CVA predisposed to infection with S. aureus. Pneumonia caused by Legionella sp., on the other hand, was significantly associated with immunosuppression from either prolonged steroid use or hematologic malignancy (Table 5).

Pneumonia caused by S. aureus tended to be more frequently present as the functional status worsened (9% in ADL I versus 56% in ADL III, p = 0.0034). A similar trend was observed for infections with GNEB and P. aeruginosa (17% in ADL I versus 39% in ADL III, p = 0.16). Conversely, pneumonia with S. pneumonia was less likely to occur, although not significantly, as the ADL scores increases (22% in ADL I versus 11% in ADL III, p = 0.4) (Figure 1).

View larger version (37K): [in this window] [in a new window]   Figure 1.   Impact of functional status on microbial etiology. Activity of Daily Living (ADL) I corresponds to a score of 6 to 8; ADL II, 9 to 13; ADL III, 14 to 18. Asterisk refers to a statistical significance among the ADL groups.

Antimicrobial Therapy

One hundred one patients (97%) received antimicrobial therapy on admission to the ICU. Two of the 3 patients who did not receive antimicrobial therapy within the first day of admission did not survive hospitalization. The most frequent drugs prescribed were second- and third-generation cephalosporins (47%), beta-lactam/beta-lactam inhibitors (41%), macrolide (33%), and quinolones (16%). Forty-seven percent received monotherapy, 51% received a combination of two antibiotics, and 3% had a combination of three drugs. There was no difference in mortality among those who received monotherapy versus those who were given a combination of antimicrobial agents (57% versus 52%, respectively, p = 0.7). However, mortality was significantly higher in those who received inadequate antimicrobial therapy (39% versus 4%, respectively, p = 0.007).

Prognostic Factors and In-Hospital Mortality

The actual mortality for the entire cohort group was 54.8% (95% CI, 43.4 to 66.0%) and was not significantly different between those admitted from the community and those who were residing in nursing homes (53% versus 57%, respectively; p = 0.8). Among those patients who died 41 of 57 (72%) compared with only seven of the 47 survivors (15%) received DNR orders (p < 0.001). Although the ADL scores were significantly lower for those admitted with CAP in comparison with those with NHAP (p < 0.001) (Figure 2), the difference in the mean ADL scores among the survivors and the nonsurvivors were not statistically significant (p = 0.3). Twenty-one patients (20%) returned to the community, and 26 (25%) were discharged to nursing homes. The 30-d, 3-mo, and 1 yr mortality rates were 51, 59.6, and 73.1%, respectively.

View larger version (52K): [in this window] [in a new window]   Figure 2.   Impact of functional status on mortality. There was no difference in the ADL scores between those who survived hospital discharge and those who died from the community or from the nursing homes.

Risk factors for in-hospital mortality, and the level of significance are given in Table 6. Only four variables were found to be independent predictors of hospital mortality by multivariate analysis. These were multilobar involvement, septic shock, 24-h urine output, and inadequate antimicrobial therapy (Table 7).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The primary findings of the current study were: (1) the microbial etiology of severe pneumonia based on invasive diagnostic techniques is not uniformly consistent with the oropharyngeal pathogens isolated from sputum cultures, (2) the respiratory pathogens responsible for severe pneumonia differs with the location of residence prior to admission, (3) the presence of specific comorbidities had distinct impact on etiology, (4) survival is not related to functional status in acute ICU settings, and (5) inadequate antimicrobial therapy is associated with poor outcome.

Studies examining the etiology of CAP in the elderly have reached different conclusions about the responsible organisms because of the heterogeneity of the population studied and the lack of utilization of standardized techniques (3, 4, 6). Almost invariably, S. pneumonia remains the most common isolate causing 20 to 30% of cases of CAP. Nontypable H. influenzae, L. pneumophila, C. pneumoniae, and gram-negative bacilli are responsible for the rest of 20 to 40% of infectious agents. In our study, S. pneumoniae was the most frequent etiology in those admitted from the community. Atypical pathogens were present in 10% of our population, with legionellosis accounting for the majority of the cases. We have found a lower prevalence of C. pneumonia infection compared with that in previous studies (12). The discrepancy is likely to be attributed to the variation in geographic settings and to the limited number of serologic tests performed in our study population. Similarly, H. influenzae was less common than previously reported (2, 15). This finding agrees with recent studies on severe community pneumonia in the elderly where H. influenzae accounted from 0 to 11% (14, 16, 17). Prior use of antibiotics would partially explain this finding as in vitro studies have indicated that subinhibitory concentrations of antibiotics can reduce colonization of human epithelial cells with H. influenzae (18). We have also found a relatively high percentage of gram-negative bacilli among the pathogens responsible for CAP. Three studies had reported the isolation of GNEB in excess of 6% as the causative agent in CAP (3, 19, 20). Only one had a comparable population to our study (3). The underlying mechanism in most cases of gram-negative bacillary pneumonia is thought to be secondary to aspiration of endogenous flora. Because older, chronically ill patients are susceptible to a higher rate of oropharyngeal colonization with GNEB (21), it is plausible that the increased predisposition to infection with GNEB is related to the multitude of comorbidities encountered in our elderly population.

In the current study, we found that the distribution of microbial pathogens in NHAP is significantly different from those admitted from the community. S. aureus surpassed S. pneumoniae as the leading pathogen. Six percent of the S. aureus isolates were identified as methicillin-resistant, in accordance with Muder and colleagues (22) who reported that as much as one third of invasive staphylococcus infection in long-term patients may be due to methicillin resistant strains (22). We were not able to document a significant role of atypical pathogens in the nursing home patients with severe pneumonia. Similarly, Drinka and coworkers (4) were unable to detect a rise in the serologic titers for Legionella, Mycoplasma, and Chlamydia in a group of nursing home residents followed for a period of 1 yr. It is speculated that the waning immunity of those with advanced age and debilitating comorbid illnesses might have blunted a rise in antibody titers to the presence of these microbes.

Pneumococcal pneumonia was significantly associated with COPD, whereas pneumonia with GNEB was significantly related to the presence of diabetes mellitus. Whereas COPD may predispose to pneumococcal pneumonia as consequence of altered pulmonary clearance and tracheobronchial colonization (23), the predisposition of elderly diabetic patients to the development of gram-negative aerobic pneumonia could be explained by a combination of factors. First, diabetics have an increased rate of upper airway colonization with these organisms (24) because of a greater ability of these pathogens to adhere to the upper respiratory epithelium. Second, the higher susceptibility of diabetics to laryngeal trauma during intubation could predispose to gram-negative colonization (25). Third, the in vitro defects described in circulating immune cells of diabetics may contribute to the perpetuating respiratory infection caused by ineffective recruitment of circulating neutrophils to the site of infection in the lungs (26).

The association between immunosuppression and Legionnaire's disease has been attributed to the depressed cell-mediated immunity. Our findings have been consistent with this widely held observation, as 60% of the elderly population admitted with Legionella pneumonia had been receiving prolonged corticosteroid therapy.

It is advanced that patient's functional status is the most important determinant of pneumonia outcome in this age group. Several studies have reported a significant association between dependency in activities of daily living and short-term mortality (27). However, our study failed to show a significant difference in ADL scores between those who died and those who survived. The reason for the discrepancy has been the lack of distinction between ICU and floor admission in previous studies. Aggressive intervention might have diminished the impact of functional status for the short-term mortality. Moreover, the studies did not distinguish between those who had a "do not intubate order" and those who received treatment in monitored environment.

The overall mortality in our study was 54.8%, a figure higher than reported in other series (3, 30). These differences in mortality are largely attributed to the older age group of our study population, and to the difference in number of patients needing mechanical ventilation, both factors strongly related to mortality (31). Using logistic regression analysis, our group showed that the presence of septic shock, multilobar involvement, urine output in the first 24 h, and inadequate antibiotic coverage are highly correlated with poor prognosis. These results fully agree with an increasing number of studies focusing on prognostic indicators of mortality in the elderly group. Venkatesan and colleagues (32) have shown that hypotension, worsening hypoxemia, urinary incontinence of recent onset, and apyrexia were indicators of fatal outcome. Other studies have found that septic shock, bacteremia, radiographic spread of pneumonia, inadequate antibiotic coverage, renal failure, and SAPS > 12 (33, 34) were associated with increased mortality.

Of these factors, the most amenable to medical intervention is the administration of appropriate antibiotics. Mehr and coworkers (29) found that after adjusting for activities of daily living, inadequate initial antimicrobial therapy was the single factor related to poor outcome. This finding has also been observed in cases of ventilator-associated pneumonia, and in Legionella pneumonia (35). The recommendations regarding initial antibiotic treatment from the ATS study state that in case of severe CAP, a regimen of two antibiotics (a macrolide plus a third-generation cephalosporin with antipseudomonal activity or a fluoroquinolone) is highly advisable (7). This strategy has been recommended because of lack of specificity of clinical (17) and radiographic features of suspected organisms, and the need to provide adequate antimicrobial therapy pending the results of microbial investigation and susceptibility patterns. Although these guidelines have not been validated in the elderly, we found that only 52% of the patients had received a combination therapy. Because 18% of those patients with identifiable pathogen had either Legionella or Chlamydia infection, we support the use of an empirical regimen of a fluoroquinolone, or the combination of an intravenous macrolide plus a broad-spectrum beta-lactam antibiotic for those admitted from the community. As for those residing in a long-term care facility, empiric regimen should include a fluoroquinolone plus a third-generation cephalosporin with antipseudomonal activity, with or without vancomycin depending on the local antimicrobial susceptibility for S. aureus.

Several potential limitations should be considered in interpreting the results of this study. The size of the population is relatively small; however, to our knowledge, the current study has included the largest number of very elderly patients where invasive diagnostic techniques have been performed to assess the microbial etiology of severe pneumonia. Another potential limitation is the possibility of a selection bias in the population under investigation. The population we studied represents a highly selected group of patients older than 75 yr of age who received mechanical ventilation. This bias would have been introduced by physicians who elected to refer patients to the ICU for aggressive medical treatment, whereas others restricted further intensive care from patients in whom they thought would not benefit from such an approach. In addition, there were many confounding variables we were unable to control such as the treating physicians, hospital discharge, and input from the patients' families.

Although the design of the study was made to exclude patients with witnessed aspiration, it has been estimated that the overall incidence of anaerobic lung infection caused by microaspiration could be as high as 21 to 33%, and that anaerobic organisms such as Bacteroides, Fusobacterium, and Peptostreptococcus sp. may be secondary only to S. pneumoniae as a cause of pneumonia (36). In our study, we were not able to recover anaerobic bacteria from the lower airways. Further investigations are needed to investigate the role of these organisms in elderly patients with severe pneumonia.

In conclusion, we have shown the utility and safety of diagnostic bronchial sampling in elderly patients with suspected bacterial pneumonia. This approach should be strongly considered to differentiate active pneumonia from either upper airway colonization or noninfectious pneumonitis, thus circumventing the difficulties inherent in the interpretation of routine culture of sputum or endotracheal aspirates. Our findings emphasize the need for an initial broad-based antimicrobial therapy based on local microbial and susceptibility patterns.