Blood Cultures Versus Respiratory Cultures: 2 Different Views of Pneumonia

Abstract

Background.

Choice of empiric therapy for pneumonia depends on risk for antimicrobial resistance. Models to predict resistance are derived from blood and respiratory culture results. We compared these results to understand if organisms and resistance patterns differed by site. We also compared characteristics and outcomes of patients with positive cultures by site.

Methods.

We studied adult patients discharged from 177 US hospitals from July 2010 through June 2015, with principal diagnoses of pneumonia, or principal diagnoses of respiratory failure, acute respiratory distress syndrome, respiratory arrest, or sepsis with a secondary diagnosis of pneumonia, and who had blood or respiratory cultures performed. Demographics, treatment, microbiologic results, and outcomes were examined.

Results.

Among 138 561 hospitalizations of patients with pneumonia who had blood or respiratory cultures obtained at admission, 12 888 (9.3%) yielded positive cultures: 6438 respiratory cultures, 5992 blood cultures, and 458 both respiratory and blood cultures. Forty-two percent had isolates resistant to first-line therapy for community-acquired pneumonia. Isolates from respiratory samples were more often resistant than were isolates from blood (54.2% vs 26.6%; P < .001). Patients with both culture sites positive had higher case-fatality, longer lengths of stay, and higher costs than patients who had only blood or respiratory cultures positive. Among respiratory cultures, the most common pathogens were Staphylococcus aureus (34%) and Pseudomonas aeruginosa (17%), whereas blood cultures most commonly grew Streptococcus pneumoniae (33%), followed by S. aureus (22%).

Conclusions.

Patients with positive respiratory tract cultures are clinically different from those with positive blood cultures, and resistance patterns differ by source. Models of antibiotic resistance should account for culture source.

Bacterial pneumonia is an important cause of morbidity, mortality, and hospitalizations in the United States [1, 2]. Pneumonia is difficult to treat with certainty because, even in cohort studies with periodically scheduled cultures, a microbial etiology is found in only half of cases [3]. Thus, much of antibiotic prescribing for pneumonia is based on regional or syndromic microbial trends, guideline recommendations, or prescribing habits rather than on an individual patient’s infecting organism. The 3 main sources of data that inform our understanding of the etiology of bacterial pneumonia are respiratory tract cultures, blood cultures, and antigenic testing of serum or urine. Each form of testing has limitations, including high rates of contamination from oral flora among respiratory cultures (low specificity), low rates of positivity among blood cultures (low sensitivity), and limited targets for antigenic testing; thus, no single form of testing can be used to extrapolate the causes of all bacterial pneumonia.

In 2005, the Infectious Diseases Society of America (IDSA) pneumonia guidelines [4] recognized a new entity dubbed healthcare-associated pneumonia (HCAP). Patients with HCAP were believed to have higher rates of resistant organisms, and guidelines recommended that empiric antibiotic therapy for HCAP parallel that of hospital-acquired pneumonia (HAP). A subsequent meta-analysis [5] of 24 HCAP studies challenged the HCAP concept as having a low positive predictive value for determining which patients were infected with multidrug-resistant organisms (MDROs), and HCAP has subsequently been dropped from the guidelines [6]. Several risk assessment models, including the Drug Resistance in Pneumonia Score, have attempted to identify pneumonia patients at high risk of harboring MDROs [7, 8]. However, it is not known to what extent respiratory cultures identify true pathogens. If the resistance patterns in respiratory cultures differ substantially from those of blood cultures, models incorporating respiratory cultures may overestimate the likelihood of MDROs. Moreover, patients with positive respiratory cultures may not be representative of patients more broadly. That is, even if respiratory cultures represent true infection, but models to predict resistance incorporate only the minority of pneumonia patients who have positive respiratory cultures, these models may still overestimate the prevalence of resistance among patients who do not have a positive respiratory culture. We explored the differences in the organisms, resistance patterns, likelihood of susceptibility to guideline-recommended antibiotics for community-acquired pneumonia (CAP), and patient characteristics between patients with positive respiratory cultures and those with positive blood cultures in order to inform empiric prescribing recommendations.

MATERIALS AND METHODS

Setting and Patients

The Premier Healthcare Database (Premier Inc, Charlotte, North Carolina), an inpatient data service developed for measuring quality and healthcare utilization and which is frequently used for healthcare research [9], was used to identify patients. Premier member hospitals are located in all regions of the United States, include both community-based and academic medical centers, and range broadly in size. Multiple data elements are included such as sociodemographic information, International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes, hospital and physician information, treatments received, source of admission, and discharge status. A subset of 177 Premier hospitals also provides microbiology data, including all cultures, organisms, and antibiotic sensitivity testing through the SafetySurveillor infection tracking tool. Patients selected for this study were ≥18 years old and admitted to any of the 177 hospitals that submitted microbiology data between July 2010 and June 2015. Because the dataset excluded patient identifiers, the Cleveland Clinic Institutional Review Board determined that this study did not constitute human subjects research.

For each hospitalization, we extracted patient age, sex, principal and secondary diagnoses, demographic information, comorbidities, treatments, and microbiologic data including final culture results and susceptibility profiles. We included hospitalizations of patients of age ≥18 years with a principal diagnosis of pneumonia, or a secondary diagnosis of pneumonia paired with a principal diagnosis of respiratory failure, acute respiratory distress syndrome, respiratory arrest, or sepsis, and who had blood or respiratory cultures collected on admission. Respiratory cultures included any samples from the respiratory tract, including those labeled as sputum cultures, bronchial cultures, and others simply designated “respiratory cultures.” We did not include cultures of the nares. Cultures for which an organism and antibiotic sensitivities were reported were considered to be positive. We did not include probable contaminants (eg, coagulase-negative staphylococci) or organisms that are not known to cause pneumonia (eg, enterococci). Patients with either CAP or HCAP were included, and patients with HAP or ventilator-associated pneumonia (VAP) were excluded [4]. The methodology for determining guideline-concordant antibiotics for CAP and HCAP is outlined in the Appendix. Markers of severity of illness included intensive care unit (ICU)–level care, mechanical ventilation, or vasopressor use on admission, which have been shown previously to have comparable discriminatory ability to clinical data [10]. Hospitalizations were excluded if there was evidence of an alternate site of infection or inability to verify clinical status or outcomes, as described in Figure 1. Outcomes included antibiotic resistance, as reported by each hospital’s laboratory, and the following clinical endpoints: inpatient case-fatality, late deterioration (as evidenced by admission to ICU, or initiation of mechanical ventilation or vasopressor medications, on hospital day 2 or later), length of stay, and inpatient cost. We included only 1 randomly selected admission for patients with multiple hospitalizations. Comorbidities were identified from ICD-9-CM secondary diagnosis codes and Diagnosis Related Groups using Healthcare Cost and Utilization Project Comorbidity Software, version 3.1, based on the work of Elixhauser [11]. Combined comorbidity scores were also calculated using the method described by Gagne et al [12].

Figure 1.

Flowchart indicating derivation of the study sample. Abbreviations: CLABSI, central line–associated bloodstream infection; CT, computed tomography; LOS, length of stay; POA, present on admission.

Statistical Analysis

Continuous baseline characteristics were summarized by quartiles, and categorical variables by frequencies and proportions. We compared baseline characteristics between patients with positive respiratory culture only, positive blood culture only, and both positive cultures for the same organism, excluding 113 patients with positive respiratory and blood cultures for different organisms. For continuous or ordinal characteristics, we used the Kruskal-Wallis rank analysis of variance test, and for categorical variables we used Pearson χ² test without continuity correction, or Fisher’s exact test when expected values were <5.

We examined the associations of antibiotic resistance with each putative exposure site (CAP vs HCAP), culture site, and Gram stain result. To properly account for potential multiple positive culture sites and/or cultured organisms within a hospitalization, in these analyses we treated hospitalizations as clusters and used the second-order, null variance–based Rao-Scott analogue to the Pearson χ² statistic for hypothesis testing [13].

Generalized linear mixed models (GLMMs) were used to assess associations between culture results and outcomes, specifically, logistic regression for dichotomous variables (in-hospital case-fatality, transfer to ICU, and introduction of mechanical ventilation and vasopressor on or after hospital day 2), and gamma GLMMs with log link function, as is common practice, for the continuous and highly right-skewed variables hospital length of stay and cost [14]. Models included simple random hospital effects to account for clustering of patients within the hospitals contributing to the Premier database, and additive covariates for age, sex, insurance status, race/ethnicity, and comorbidities and markers of initial illness severity (Table 1). GLMMs were fit by maximizing the residual subject-specific pseudo-likelihood [14]. Results of logistic models are reported as odds ratios (ORs), and results of gamma models as ratios of means, each with corresponding 95% Wald confidence intervals (CIs), with hypothesis tests based on Wald statistics. All analyses were performed using SAS version 9.4 software (SAS Institute, Cary, North Carolina).

Table 1. Characteristics of Patients With Pneumonia Who Had Positive Blood Cultures Only, Positive Respiratory Cultures Only, and Both Types of Cultures Positive With the Same Organism.

	Blood	Respiratory	Both
Factor	(n = 5992)	(n = 6438)	(n = 458)	P Valuea
Age, median, y (Q1, Q3)	70.0 (57.0, 82.0)	70.0 (58.0, 80.0)	65.0 (54.0, 75.0)	< .001b
Sex, %				< .001
Female	47.8	43.4	41.5
Male	52.2	56.6	58.5
Race, %
White	74.6	76.5	72.3
Black	14	10.6	15.1
Hispanic	0.75	1.1	0.66
Others	10.5	11.9	11.6
Unknown	0.03	0.06	0.44
Marital status, %				.007
Married	41.8	41	38.9
Single	51.9	50.9	55.2
Other	6.2	7.8	5.9
Unknown	0.18	0.25	0
Admission source, %				< .001
Emergency room	89.8	86.5	89.3
SNF/ICF	6.6	8.9	8.1
Clinic	3.4	4.5	2.6
Others	0.15	0.11	0
CAP/HCAP, No. (%)				< .001c
CAP	3856 (64.4)	3979 (61.8)	246 (53.7)
HCAP	2136 (35.6)	2459 (38.2)	212 (46.3)
Insurance payor, %				< .001
Medicare	66.6	71.5	59.6
Medicaid	9.5	11	16.8
Managed care	12.3	8.8	12
Commercial indemnity	3.7	2.8	2.8
Others	7.8	5.9	8.7
Principal diagnosis, %				< .001
Pneumonia	20.7	43.2	12.4
Aspiration pneumonia	2.5	8.3	1.3
Sepsis	75	37.8	82.3
Respiratory failure	1.8	10.8	3.9
Comorbidity score, median (Q1, Q3)	3.0 (2.0, 6.0)	3.0 (1.00, 5.0)	4.0 (2.0, 6.0)	< .001b
Comorbidities, %
Hypertension	63.3	62.3	57.9	.048
Fluid and electrolyte D/O	62.5	55.5	76.6	< .001
Chronic pulmonary disease	38.8	59.4	50	< .001
Anemia	38.1	36.5	38.4	.17
Diabetes	33.6	33.4	34.7	.86
CHF	27.6	27.7	26.4	.83
CKD	20.3	16.1	16.6	< .001
Neurological D/O	13.3	18.9	13.1	< .001
UTI	18.7	14.3	12.9	< .001
Obesity	12.8	12.4	13.5	.69
Weight loss	18.5	18.2	21.8	.16
Valvular disease	12.2	8	10	< .001
Coagulopathy	18.4	11.1	24	< .001
PVD	9.1	9	6.6	.17
Paralysis	4.5	7.8	3.9	< .001
Alcohol abuse	5.9	5.5	8.7	.015
Drug abuse	5.1	4.9	9	< .001
Liver disease	5.8	3.7	9.8	< .001
Lymphoma	2.9	1.6	3.1	< .001
Bleeding peptic ulcer	0.03	0.02	0	.65c
Markers of illness severity, %
ICU	35.8	45.8	70.5	< .001
IMV	8.5	28	43.7	< .001
NIV	10.3	12.3	16.6	< .001
Vasopressor use	13.3	17.7	39.7	< .001
Blood lactate	58.8	54.3	72.5	< .001
Arterial and venous blood gas	43.4	57.3	73.6	< .001

Abbreviations: CAP, community-acquired pneumonia; CHF, congestive heart failure; CKD, chronic kidney disease; D/O, disorder; HCAP, healthcare-associated pneumonia; ICF, intermediate care facility; ICU, intensive care unit; IMV, invasive mechanical ventilation; NIV, noninvasive mechanical ventilation; PVD, peripheral vascular disease; Q1, quartile 1; Q3, quartile 3; SNF, skilled nursing facility; UTI, urinary tract infection.

^aPearson χ² test, without continuity correction, unless otherwise footnoted. Comparison is across all 3 groups.

^bKruskal-Wallis test.

^cFisher exact test.

To determine whether use of guideline-recommended empiric antibiotics for CAP would have been adequate to treat patients who ultimately were found to have positive cultures, we compared the susceptibilities of the cultured organisms to the IDSA guideline for empiric antibiotics for treatment of patients with CAP (ie, respiratory quinolone or ceftriaxone plus azithromycin) [15]. For this portion of the analysis, 209 patients were excluded because they had organisms for which there are no Clinical and Laboratory Standards Institute (CLSI) breakpoints to determine antibiotic susceptibilities [16].

RESULTS

Of 171 935 patients with pneumonia, 138 561 had blood or respiratory cultures collected and met inclusion criteria. Among these, 94 350 (68%) had CAP and 44 211 (32%) had HCAP. Ninety-nine percent of patients had blood cultures obtained, whereas only 18% of patients had respiratory cultures collected. Overall, positive cultures were infrequent (8.6% of patients with CAP and 11% of patients with HCAP), and blood cultures were much less likely than respiratory cultures to grow an organism (4.7% vs 27.9%; P < .001). Among those with respiratory cultures, HCAP patients were more likely to have positive cultures than were CAP patients (33.0% vs 25.4%; P < .001).

Clinical Characteristics

Table 1 compares demographic characteristics, principal diagnoses, underlying comorbidities, and markers of severity of illness for which subsequent analyses of outcomes are adjusted, among patients with positive respiratory cultures or blood cultures or both. Patients who had only positive respiratory cultures were more likely than patients with only positive blood cultures to have received a principal diagnosis of pneumonia, aspiration pneumonia, or respiratory failure, to have been transferred from a skilled nursing facility, to have chronic pulmonary disease and neurologic disease, and on admission to be treated in an ICU, and to require invasive or noninvasive mechanical ventilation and blood gas monitoring. Patients who had only positive blood cultures were more likely than patients with only positive respiratory cultures to have a principal diagnosis of sepsis and to have coagulopathy, heart disease, and fluid and electrolyte imbalance. Of the 3 groups, patients who had both types of cultures positive were the most likely to be treated in an ICU and to require invasive and noninvasive mechanical ventilation, vasopressors, and monitoring of blood lactate and blood gasses on admission.

Microbiology and Antibiotics

Table 2 shows the bacterial etiology, which varied by culture site. Among patients with positive respiratory cultures, the most common pathogens were Staphylococcus aureus (34%) and Pseudomonas aeruginosa (17%), whereas patients with positive blood cultures most commonly grew Streptococcus pneumoniae (33%), followed by S. aureus (22%).

Table 2. Bacterial Etiology Among Hospitalized Patients With Culture-positive Pneumonia.

	Source of Positive Culture, No. (%)
Bacteria	Respiratory	Blood	Both
Staphylococcus aureus	2631 (33.58)	1404 (22.24)	207 (44.52)
Methicillin sensitive	1488 (18.99)	881 (13.96)	120 (25.81)
Methicillin resistant	1118 (14.27)	502 (7.95)	85 (18.28)
Streptococcus pneumoniae	1127 (14.38)	2102 (33.3)	151 (32.47)
Pseudomonas aeruginosa	1334 (17.02)	218 (3.45)	36 (7.74)
Escherichia coli	501 (6.39)	747 (11.83)	17 (3.66)
Klebsiella pneumoniae	599 (7.64)	290 (4.59)	20 (4.3)
Haemophilus influenzae	317 (4.05)	124 (1.96)	8 (1.72)
Proteus mirabilis	194 (2.48)	100 (1.58)	1 (0.22)
Streptococcus agalactiae	31 (0.4)	170 (2.69)	1 (0.22)
Serratia marcescens	153 (1.95)	19 (0.3)	5 (1.08)
Stenotrophomonas (Xanthomonas) maltophilia	140 (1.79)	11 (0.17)	0 (0)

Figure 2 shows the prevalence of resistance to IDSA guideline–recommended CAP therapy among organisms overall and by culture site. Overall, 41.8% of patients with positive cultures grew an organism that was resistant to CAP therapy, with gram-negative organisms more likely to be resistant than gram-positive organisms (51.8% vs 35.4%). Resistant organisms were more common in respiratory than in blood samples (59.1% vs 37.3% of gram-negative organisms and 48.7% vs 24.4% of gram-positive organisms).

Figure 2.

Rates of bacterial resistance to Infectious Diseases Society of America guideline–recommended antibiotics for community-acquired pneumonia, stratified by culture source and Gram stain result. Standard errors of all depicted proportions are ≤1.1%. Abbreviations: CAP, community-acquired pneumonia.

Table 3 shows the empiric antibiotics given to patients at the time of admission. Compared to patients who had only positive blood cultures or only positive respiratory cultures, patients with both more often received anti–methicillin-resistant Staphylococcus aureus (MRSA) antibiotics (67.5% vs 42.4% and 47.6%; P < .001 for blood only and respiratory only), fully HCAP-guideline concordant antibiotics (27.3% vs 11.8% and 15.7%), and ≥4 empiric antibiotics (33% vs 17.5% and 21%).

Table 3. Empiric Antibiotics Administered at the Time of Admission to Hospitalized Patients With Pneumonia With Only Positive Blood Cultures, Only Positive Respiratory Cultures, and Positive Cultures of the Same Organism From Both Sites.

	Blood	Respiratory	Both
Factor	(n = 5992)	(n = 6438)	(n = 458)	P Valuea
Third-generation cephalosporin, %	44.5	37.3	39.1	< .001
Respiratory quinolone (periadmission), %	42.7	39	46.3	< .001
Anti-MRSA agents, %	42.4	47.6	67.5	< .001
Antipseudomonal quinolone, %	40	38.2	46.1	.002
Macrolide (periadmission), %	36.3	35.3	36.7	.5
Piperacillin-tazobactam, %	26.7	31	35.8	< .001
Antipseudomonal cephalosporin, %	10	13.1	15.5	< .001
Antipseudomonal carbapenem, %	4.2	6.1	10	< .001c
Aminoglycosides, %	3.1	3.5	3.1	.46
No. of antibiotics received, %				< .001
1	18.7	15.9	8.1
2	38	37.9	26
3	25.9	25.2	33
≥4	17.5	21	33
Guideline antibiotics, %				< .001
CAP	51.4	43.1	32.8
Fully HCAP	11.8	15.7	27.3
Partial HCAP	21.7	24.9	30.8
Other antibiotics	15.1	16.3	9.2

Abbreviations: CAP, community-acquired pneumonia; HCAP, healthcare-associated pneumonia; MRSA, methicillin-resistant Staphylococcus aureus.

^aPearson χ² test, without continuity correction.

Table 4 shows the unadjusted outcomes by culture source and Table 5 shows the same outcomes adjusted for demographics, comorbidities, and severity of illness indicators. After adjustment, patients with both cultures positive had statistically significantly higher case fatality than those with only positive blood (OR, 1.47 [95% CI, 1.12–1.94]) or respiratory culture (OR, 1.59 [95% CI, 1.22–2.08]), and 24% higher costs along with 18% longer stays (all P < .001) relative to those with only positive respiratory cultures. Although upon admission, patients with positive respiratory cultures were more often admitted to an ICU and received mechanical ventilation and broad-spectrum empiric antimicrobials than patients with positive blood cultures, after adjustment for demographics, baseline comorbidities, and illness severity, patients with only positive respiratory cultures had nonsignificantly lower case-fatality and significantly fewer late introductions of vasopressors, shorter lengths of stay, and lower costs. Case-fatality by specific organism appears in the Supplementary Table.

Table 4. Observed Patient Outcomes by Source(s) of Positive Culture(s), Among Hospitalized Patients With Pneumonia With Only Positive Blood Cultures, Only Positive Respiratory Cultures, and Positive Cultures of the Same Organism From Both Sites.

	Blood	Respiratory	Both
Factor	(n = 5992)	(n = 6438)	(n = 458)
In-hospital mortality, %	12	11.4	25.3
30-day readmission, %	4.7	5.6	4.8
Late ICU (day 2 or later)a, %	8.6	5.2	4.1
Late IMV (day 2 or later)a, %	10.2	6.6	12
Late vasopressor (day 2 or later)a, %	10.5	7.1	11.6
Cost, median (Q1, Q3)	12 057 (7055, 22 151)	12 342 (6882, 23 789)	21 711 (11 623, 41 389)
Length of stay, median (Q1, Q3)	7 (4, 11)	6.5 (4, 10)	9 (5, 16)

Abbreviations: ICU, intensive care unit; IMV, invasive mechanical ventilation; Q1, quartile 1; Q3, quartile 3.

^aAmong patients not previously in ICU, on IMV, or vasopressors, respectively.

Table 5. Associations of Outcomes of Hospitalized Patients With Culture-positive Pneumonia With Source(s) of Positive Cultures, Adjusted for Demographics, Baseline Comorbidities, and Illness Severity.

Outcome	Respiratory vs Blood	Both vs Blood	Both vs Respiratory
Case fatality	0.93 (.81–1.06)	1.47 (1.12–1.94)	1.59 (1.22–2.08)
30-day readmission	1.16 (.97–1.39)	1.12 (.71–1.78)	0.97 (.61–1.53)
Late ICUa	0.98 (.81–1.17)	0.95 (.53–1.70)	0.97 (.54–1.75)
Late IMVa	1.52 (1.28–1.81)	1.64 (1.11–2.42)	1.08 (.72–1.60)
Late vasopressora	0.78 (.67–.92)	1.02 (.70–1.49)	1.31 (.90–1.90)
Mean multiplier
Length of stay	0.90 (.87–.93)	1.06 (.96–1.17)	1.18 (1.07–1.29)
Costs	0.87 (.82–.92)	1.08 (.97–1.20)	1.24 (1.14–1.35)

Data are presented as odds ratio or multiplier (95% confidence interval). Abbreviations: ICU, intensive care unit; IMV, invasive mechanical ventilation.

^aAmong patients not previously in ICU, on IMV, or vasopressor, respectively.

DISCUSSION

In this large multicenter US cohort of adults with pneumonia treated in real-world settings, we found that patients with positive respiratory cultures differed from those with positive blood cultures or both in their clinical characteristics, presentations, and outcomes. Patients with positive respiratory cultures were more likely than those with positive blood cultures to have been transferred from a skilled nursing facility and to have chronic respiratory or neurologic disease, experience aspiration or respiratory failure, and need mechanical ventilation. In contrast, patients with positive blood cultures were more apt than those with positive respiratory cultures to have sepsis, but less likely to have been admitted initially to ICU and to receive mechanical ventilation or vasopressors. Patients who had both cultures positive had markers of more severe acute and chronic illness, and ultimately higher case-fatality. The microbiology and resistance patterns of isolated organisms also differed among these populations. Patients with positive respiratory cultures were more frequently infected with organisms resistant to CAP therapy, including MRSA and P. aeruginosa, whereas patients with positive blood cultures were most often infected with S. pneumoniae and methicillin-sensitive Staphylococcus aureus. Overall, patients with positive respiratory cultures were twice as likely to harbor organisms resistant to CAP therapy.

The results of this study underscore the difficulty in determining the etiology of pneumonia, which depends upon what cultures we believe to be the true representation of the disease. This debate has been active in the medical literature for more than a hundred years [17]. Our beliefs about the etiology and antibiotic susceptibility of bacterial pneumonia in large part depend upon the population studied. If conclusions are drawn only from the tiny fraction of patients who have both culture sites positive, then bacterial pneumonia is caused overwhelmingly by S. aureus and S. pneumoniae, and the outcomes are poor. If the population is limited to patients with positive blood cultures, then S. pneumoniae is the dominant organism and resistance is uncommon. Examining respiratory cultures, which are taken from a minority of patients who tend to be debilitated and suffering from chronic lung disease, paints a picture of pneumonia in which highly resistant organisms play a major role.

Our results provide important perspective on prior efforts to determine pneumonia microbiology, as done by Kollef and colleagues in their large analysis of the Cardinal Health-Atlas Research Database [18]. In the first large study to describe HCAP as distinct from CAP, HAP, and VAP, they found that more than three-quarters of HCAP was caused by either gram-negative organisms or MRSA. The fact that they enrolled subjects based only on positive respiratory cultures may have biased these results, which have been challenged in later studies that relied on a wider range of culture sources [19, 20].

More recent studies have shown that the risk factors for HCAP that were outlined in the 2005 American Thoracic Society/IDSA guideline [4] have poor predictive value for patients who harbor MDROs [21], and at least 12 studies have attempted to create prediction models to identify patients at high risk for MDROs [22]. These studies have generally not discriminated between blood and respiratory sources, and assume that risk factors for resistant infection seen in this combination of sources apply to all patients, although validation studies are generally limited to patients with positive cultures (ie, the same population from which the models were derived). Performance of the models based on the area under the receiver operating characteristic curve ranged from 0.64 to 0.90. However, the models are not meant to be applied to patients who are already known to have positive cultures. They are meant to be applied to patients who are admitted to the hospital with a working diagnosis of pneumonia and require empiric antibiotic therapy.

Future prediction models should incorporate our findings in 2 ways. First, we need to understand the importance of positive blood cultures vs respiratory cultures, which are primarily sputum. The latter may be particularly subject to colonization and may not represent true infection. This point could be settled by high-quality studies of the impact of adequate vs inadequate antibiotic therapy based on the resistance patterns observed in respiratory samples. If patients with inadequate therapy have similar outcomes to those with adequate therapy, then these pathogens are not likely to be important and should not be included in the models. Second, models should include patients with negative cultures with the assumption that their pathogens are susceptible, although some patients with negative cultures may not have bacterial infection at all. Because the models will be applied to such patients, they should be included in the derivation and validation as well.

There are several limitations to our findings. Cultures obtained from the patients in this cohort were real-world specimens and were not limited to those obtained in a controlled setting such as bronchoalveolar lavage, which could mean that some positive cultures were due to colonizing organisms rather than true pathogens. While the very large majority of patients had blood cultures, only a minority of patients had respiratory cultures and there was no standard protocol for determining who would get them. This may have introduced a selection bias into our sample. However, the use of clinical specimens is representative of the use and quality of cultures obtained at both community and academic hospitals outside of the research setting and thus is a more accurate reflection of the data that are managed and interpreted daily by practicing clinicians. Our comorbidities were based on ICD-9-CM codes, which are subject to differential coding at different hospitals. Some patients may have received outpatient antibiotics that could have decreased culture yields, although we somewhat mitigated this problem by excluding patients who were transferred from other hospitals. Finally, it is not possible to determine the antibiotic susceptibilities for any organism for which no CLSI breakpoints exist [16]. Consequently, we excluded these organisms from our analysis of whether the current guideline-recommended empiric therapy for CAP would have been effective treatment, and from determinations of resistance to antipseudomonal agents.

In summary, our study has shown that patients with positive respiratory tract cultures are clinically different than those with positive blood cultures, and that estimates of the prevalence of resistance to CAP therapy change depending on the source of the cultures. This fact adds to the difficulty in determining which cultures most accurately reflect the etiology of bacterial pneumonia, and thus which cultures should drive empiric antibiotic selection and antimicrobial stewardship activities. Future prediction models should take these findings into account when determining both the appropriate outcome measure for the model and the appropriate patient population for both derivation and validation sets.

APPENDIX:

Methodology Used to Determine Community-acquired Pneumonia Treatments and Healthcare-associated Pneumonia Treatments.

1.	Antibiotic Class	Antibiotic or Definition
	Aminoglycosides	Amikacin, gentamicin, tobramycin
	Anti-MRSA agents	Vancomycin, linezolid, quinupristin-dalfopristin, ceftaroline, telavancin
	Antipseudomonal carbepenem	Imipenem, meropenem, doripenem
	Nonpseudomonal carbepenem	Ertapenem
	Third-generation cephalosporin	Ceftriaxone, cefotaxime, ceftizoxime
	Antipseudomonal cephalosporin	Cefepime, ceftazidime, cefoperazone
	Antipseudomonal β-lactam/β-lactamase inhibitor	Piperacillin-tazobactam
	Nonpseudomonal β-lactam/β-lactamase inhibitor	Ampicillin-sulbactam
	β-lactam	Ampicillin
	Respiratory quinolone	Levofloxacin, gemifloxacin, moxifloxacin
	Antipseudomonal quinolone	Levofloxacin, ciprofloxacin
	Macrolide	Azithromycin, clarithromycin, erythromycin (erythro base, erythro est, erythro ethyl, erythro stear), dirithromycin
	Doxycycline	Doxycycline
2.	CAP therapy definition (class)	A respiratory quinolone alone OR a β-lactam or third-generation cephalosporin or nonpseudomonal carbepenem AND a macrolide or doxycycline
3.	HCAP therapy definition (class)	• Fully concordant: anti-MRSA antibiotic AND antipseudomonal cephalosporin or antipseudomonal carbepenem or antipseudomonal β-lactam or aztreonam AND antipseudomonal quinolone or aminoglycoside • Partially concordant: anti-MRSA antibiotic AND antipseudomonal cephalosporin or antipseudomonal carbepenem or antipseudomonal β-lactam or aztreonam OR antipseudomonal quinolone or aminoglycoside

Abbreviations: CAP, community-acquired pneumonia; HCAP, healthcare-associated pneumonia; MRSA, methicillin-resistant Staphylococcus aureus.