Initial Antibiotic Selection and Patient Outcomes: Observations from the National Pneumonia Project

SUPPLEMENT ARTICLE Initial Antibiotic Selection and Patient Outcomes: Observations from the National Pneumonia Project Dale W. Bratzler, Allen Ma, and Wato Nsa Oklahoma Foundation for Medical Quality, Oklahoma City Background. Guidelines for empirical treatment of hospitalized patients with pneumonia provide specific recommendations for antibiotic selection that are primarily based on findings from observational studies. Methods. We conducted a retrospective study of 27,330 community-dwelling, immunocompetent Medicare patients (age, 165 years) with pneumonia who were hospitalized in 1998 1999 and 2000 2001. Associations between initial antimicrobial regimens and risk-adjusted mortality were assessed, accounting for differences in patient characteristics, comorbidities, illness severity, geographic location, and processes of care. Treatment with nonpseudomonal third-generation cephalosporin monotherapy constituted the reference group for comparisons. Results. For patients not in the intensive care unit, initial treatment with fluoroquinolone monotherapy was associated with reduced in-hospital mortality, 14-day mortality, and 30-day mortality rates (adjusted odds ratio [AOR] for 30-day mortality, 0.7; 95% confidence interval [CI], 0.6 0.9; P p.001 ). The combination of a cephalosporin plus a macrolide was associated with reduced 14-day and 30-day mortality rates (AOR for 30-day mortality, 0.7; 95% CI, 0.6 0.9; P!.001). For intensive care unit patients, the combination of a cephalosporin and a macrolide was associated with reduced in-hospital mortality (AOR, 0.6; 95% CI, 0.3 0.9; P p.018). Conclusions. Initial antimicrobial treatment with the combination of a second- or third-generation cephalosporin and a macrolide or initial treatment with a fluoroquinolone was associated with a reduced 30-day mortality rate, compared with treatment with third-generation cephalosporin monotherapy, among non intensive care unit patients. Although our results are consistent with other observational studies, controversy continues to exist about the use of nonexperimental cohort studies to demonstrate associations between processes of care, such as antibiotic selection, and patient outcomes. Initial treatment of community-acquired pneumonia (CAP) is usually empirical because the microbial etiology cannot be predicted on the basis of clinical presentation and because accurate and reliable rapid diagnostic methods to identify specific pathogens are not currently available for most cases. Guidelines published by the Infectious Diseases Society of America and the American Thoracic Society provide consensus recommendations for initial antimicrobial treatment of CAP [1]. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does the mention of trade names, commercial products, or organizations imply endorsement by the US Government. Reprints or correspondence: Dr. Dale W. Bratzler, Oklahoma Foundation for Medical Quality, 14000 Quail Springs Pkwy., Ste. 400, Oklahoma City, OK 73134 (dbratzler@okqio.sdps.org). Clinical Infectious Diseases 2008; 47:S193 201 2008 by the Infectious Diseases Society of America. All rights reserved. 1058-4838/2008/4711S3-0016$15.00 DOI: 10.1086/591404 Although many antimicrobial agents are approved for the treatment of CAP, guidelines for empirical treatment for hospitalized patients provide specific recommendations for antibiotic selection. For instance, recommendations for patients who are not admitted to the intensive care unit (ICU) include the combination of a b-lactam plus a macrolide or monotherapy with a fluoroquinolone [1]. These recommendations are primarily based on the findings from observational studies that have demonstrated a reduction in mortality among patients given treatment with a b-lactam plus a macrolide or monotherapy with a fluoroquinolone, compared with patients given treatment with a cephalosporin alone [2 5]. In addition, a number of observational studies have suggested improved patient outcomes for patients who receive guideline-concordant antimicrobial therapy [6 11]. In 1994, the Centers for Medicare & Medicaid Services began data collection from a large national cohort of inpatients with pneumonia, data that Gleason et al. [2] evaluated for the relationship between initial anti- Initial Antibiotic Selection and Patient Outcomes CID 2008:47 (Suppl 3) S193

biotic selection and patient 30-day mortality rates. The Centers for Medicare & Medicaid Services subsequently implemented the National Pneumonia Project in 1999 with the goal of improved quality of care for patients with pneumonia. Routine measurement of performance and the feedback of process-measure data to hospitals has occurred since 1999. Current measures of pneumonia quality that have been implemented are limited to processes of care, including measures that reflect care at the time of hospital admission (oxygenation assessment, blood cultures, antibiotic timing, and antibiotic selection) [12]. As part of this project, the Centers for Medicare & Medicaid Services collected medical record data from 176,000 pneumonia hospitalizations that occurred during 1998 1999 and 2000 2001. Using these data, we have studied the relationship between the initial antibiotic regimen and patient mortality and present the results here. In addition, we discuss some of the limitations of using nonexperimental cohort studies to evaluate associations between care processes and patient outcomes. METHODS The National Pneumonia Project used Medicare fee-for-service hospital claims to identify pneumonia hospitalizations with a principal diagnosis of pneumonia (International Classification of Disease, Ninth Edition, Clinical Modification [ICD-9-CM] codes 480.0 483.8, 485 486, or 487.0) or with a principal diagnosis of septicemia or acute respiratory failure (ICD-9-CM codes 038.XX or 518.81) and a secondary diagnosis of pneumonia [13]. As many as 850 cases were selected in each state and territory for review during each of 2 sampled time periods. Records were sampled during July March during 1998 1999 and 2000 2001. Consent and institutional review board approval were not required because federal law provides the Centers for Medicare & Medicaid Services with access to medical records for administration of the Medicare program. Data collection. Hospitals sent photocopies of medical records to 1 of 2 clinical data abstraction centers. Abstractors used computerized tools with explicit predefined entry criteria to record data elements that included patient demographic characteristics and clinical values, antibiotic selection, performance of cultures and culture results, and timing of antibiotic administration. Data quality was monitored continuously through interabstractor reliability testing. We used Medicare enrollment data to detect mortality and Medicare part A claims to detect readmission. Readmission and mortality rates were calculated from the date of hospital admission. Study population. We applied multiple exclusion criteria (table 1) to produce an analytic database representing patients who had a radiographically supported working diagnosis of pneumonia at the time of admission, were admitted for curative therapy, were aged 65 years, and were immunocompetent. We limited analysis to CAP by excluding patients who had been in the hospital during the 14 days before admission and by Table 1. Exclusion criteria for the analysis of empirical antibiotic selection and patient outcomes. Measure in 1998 1999 in 2000 2001 No. of cases abstracted a 39,242 (100) 37,123 (100) General exclusions No working diagnosis of pneumonia at time of hospital arrival 4864 (12.4) 4114 (11.1) Transferred from another acute care facility 563 (1.4) 591 (1.6) Received comfort care only 1104 (2.8) 1505 (4.1) Age!65 years 3369 (8.6) 3478 (9.4) Resident of Virgin Islands or Puerto Rico 414 (1.1) 492 (1.3) Chest radiograph findings not consistent with pneumonia 3567 (9.1) 3077 (8.3) Analysis-specific exclusions Readmitted within 14 days after previous discharge 1577 (4.0) 1532 (4.1) HIV/AIDS, leukemia, lymphoma, or immunosuppression 1665 (4.2) 1556 (4.2) Receipt of chemotherapy or immunosuppressive therapy within previous 3 months 2853 (7.3) 2677 (7.2) Died or was discharged on the same date as hospital admission 118 (0.3) 111 (0.3) No antimicrobials given during the hospital stay 88 (0.2) 61 (0.2) All antimicrobials were given after the initial 36 h of hospital stay 200 (0.5) 162 (0.4) Unable to determine whether antimicrobials were given during the initial 36 h of hospital stay 153 (0.4) 108 (0.3) Unable to determine whether antimicrobials were given within 4 h after hospital arrival 48 (0.1) 33 (0.1) Unable to determine whether a blood culture specimen was obtained within 24 h after hospital arrival 284 (0.7) 166 (0.4) Multiple admissions for pneumonia during the study period 161 (0.4) 134 (0.4) Resided in a nursing home or extended care facility before hospital arrival 4063 (10.4) 3747 (10.1) No. of records analyzed 14,151 (36.1) 13,579 (36.6) NOTE. Data are no. (%) of patients, unless otherwise indicated. a Medicare inpatients with pneumonia were randomly selected from every state and territory as part of the National Pneumonia Project. All medical records were independently abstracted by clinical data abstraction centers contracted by the Centers for Medicare & Medicaid Services. S194 CID 2008:47 (Suppl 3) Bratzler et al.

Table 2. Demographic characteristics of patients. Characteristic Age group, years All patients combined ( n p 27,730) in 1998 1999 ( n p 14,151) in 2000 2001 ( n p 13,579) 65 74 8616 (31.1) 4507 (31.8) 4109 (30.3).004 75 84 11,903 (42.9) 6031 (42.6) 5872 (43.2).294 85 7211 (26.0) 3613 (25.5) 3598 (26.5).067 Sex Female 14,306 (51.6) 7246 (51.2) 7060 (52.0).190 Race/ethnicity White 24,478 (88.3) 12,517 (88.5) 11,961 (88.1).340 African American 1626 (5.9) 825 (5.8) 801 (5.9).807 Other 1626 (5.9) 809 (5.7) 817 (6.0.288 Census region Midwest 6435 (23.2) 3242 (22.9) 3193 (23.5).234 Northeast 4901 (17.7) 2563 (18.1) 2338 (17.2).051 South 8654 (31.2) 4318 (30.5) 4336 (31.9).011 West 7740 (27.9) 4028 (28.5) 3712 (27.3).036 Mortality In hospital 1309 (4.7) 742 (5.2) 567 (4.2)!.001 14 Day 1479 (5.3) 809 (5.7) 670 (4.9).004 30 Day 2303 (8.3) 1230 (8.7) 1073 (7.9).017 Readmission within 30 days 4122 (14.9) 1914 (13.5) 2208 (16.3)!.001 NOTE. Data are no. (%) of patients, unless otherwise indicated. a P values reflect pairwise comparisons of the 1998 1999 and 2000 2001 patient cohorts by use of x 2 tests. P a excluding patients who resided in a long-term-care facility. Only the first of a patient s multiple hospitalizations was included in the analysis. Cases with death or discharge on the day of admission were excluded. Geographic regions were those used for the US Census (i.e., West, Midwest, South, and Northeast). We calculated the pneumonia severity index (PSI) score for each patient [14]. Patients from the 2 cohorts were then classified into groups by initial antimicrobial regimen. Only antimicrobials administered within the first 36 h after hospital arrival were considered in the classification of patients. All antimicrobials administered within the first 36 h were first grouped into 7 mutually exclusive categories, including macrolides, fluoroquinolones, b-lactam/ b-lactamase inhibitors, aminoglycosides, second- and thirdgeneration cephalosporins, and other. Fluoroquinolones and macrolides were evaluated if they were administered orally or parenterally. All other antimicrobials had to be administered parenterally to be included in subsequent analyses with orally administered antibiotics assigned to the other regimen. Patients were then assigned to 1 of 9 initial antimicrobial regimens, composed of the 6 of the 7 previously mentioned categories (excluding second-generation cephalosporin monotherapy) and 3 combinations: macrolide plus cephalosporin (second or third generation), fluoroquinolone plus cephalosporin (second or third generation), and macrolide plus b- lactam/b-lactamase inhibitor. Any other antimicrobials or combination of antimicrobials not classified into one of these groups was put into the other regimen group. The definitions for the monotherapy and dual-therapy regimens were strict, meaning that the patient could not receive any additional antimicrobials within 36 h after hospital arrival. Data analysis. At univariate analysis, differences across selected subgroups were assessed using x 2 tests and ORs for categorical variables. Multivariate logistic regression was used to produce adjusted ORs (AORs) that describe the association between antibiotic selection and clinical outcomes while controlling for potential confounding. We analyzed the 2 cohorts of patients separately and as a combined population. Because the results were similar for the 2 cohorts, we present the analyses of the combined cohort in this article. The multivariate model adjusted for age, sex, neoplastic disease, cardiovascular disease, altered mental status, respiratory rate 30 breaths/min, systolic blood pressure!90 mm Hg, temperature!35 C or 40 C, pulse 125 beats/min, arterial ph!7.35, blood urea nitrogen level 111 mmol/l, sodium level!130 meq/l, hematocrit!30%, partial pressure of oxygen!60 mm Hg, and presence of pleural effusion. In addition, admission to the intensive care unit and first antibiotic dose within 4 h after hospital arrival were included in the multivariate model. The association between antibiotic selection and patient mortality was also assessed by Initial Antibiotic Selection and Patient Outcomes CID 2008:47 (Suppl 3) S195

Table 3. Clinical characteristics of patients at the time of hospital admission. Characteristic All patients combined ( n p 27,730) in 1998 1999 ( n p 14,151) in 2000 2001 ( n p 13,579) Neoplastic disease 913 (3.3) 486 (3.4) 427 (3.1).176 Chronic liver disease 325 (1.2) 150 (1.1) 175 (1.3).077 Congestive heart failure 8447 (30.5) 4058 (28.7) 4389 (32.3)!.001 Cerebrovascular disease 4343 (15.7) 2656 (18.8) 1687 (12.4)!.001 Chronic renal disease 1819 (6.6) 1222 (8.6) 597 (4.4)!.001 Altered mental status 4928 (17.8) 2579 (18.2) 2349 (17.3).044 Respiratory rate 30 breaths/min 4766 (17.2) 2564 (18.1) 2202 (16.2)!.001 Systolic blood pressure!90 mm Hg 661 (2.4) 303 (2.1) 358 (2.6).007 Temperature!35 C or 40 C 548 (2.0) 299 (2.1) 249 (1.8).095 Pulse 125 beats/min 2416 (8.7) 1218 (8.6) 1198 (8.8).525 Arterial ph!7.35 1491 (5.4) 785 (5.5) 706 (5.2).199 Blood urea nitrogen level 111 mmol/l 6644 (24.0) 3205 (22.6) 3439 (25.3)!.001 Serum sodium level!130 meq/l 1737 (6.3) 870 (6.1) 867 (6.4).416 Glucose level 114 mmol/l 1637 (5.9) 870 (6.1) 767 (5.6).078 Hematocrit!30% 1896 (6.8) 929 (6.6) 967 (7.1).067 Arterial PO 2!60 mm Hg or SaO 2!90% 8108 (29.2) 4164 (29.4) 3944 (29.0).486 Pleural effusion 7236 (26.1) 3653 (25.8) 3583 (26.4).278 Admission to ICU within 24 h after hospital arrival 2950 (10.6) 1612 (11.4) 1338 (9.9)!.001 PSI risk class II 2481 (8.9) 1280 (9.0) 1201 (8.8).558 III 7630 (27.5) 3872 (27.4) 3758 (27.7).560 IV 13,389 (48.3) 6679 (47.2) 6710 (49.4)!.001 V 4230 (15.3) 2320 (16.4) 1910 (14.1)!.001 NOTE. Data are no. (%) of patients, unless otherwise indicated. ICU, intensive care unit; PO 2, partial pressure of oxygen; SaO 2, arterial saturation of oxygen. a P values reflect pairwise comparisons of the 1998 1999 and 2000 2001 patient cohorts by use of x 2 tests. P a calculating mortality rate, stratified on the basis of the initial PSI score. To evaluate the impact of clustering of care within hospitals on patient outcomes, we first examined the number cases included in each hospital and examined the mean, median, and range of number of cases per facility. We used the generalized linear mixed model (SAS PROC GLIMMIX) for outcomes, including hospital as a random cluster effect. We compared the model results from GLIMMIX to those from multiple logistic regression models. Because of the limited number of cases sampled at the individual hospital level, the results of this analysis were similar to those found in multivariate analysis and are not presented here. All analyses were completed using SAS statistical software (version 9.1.3; SAS Institute). P values are 2 sided. Statistical significance is defined by a 95% CI that excludes 1.0 or by P!.05. RESULTS From the initial sample of 39,242 cases abstracted in 1998 1999 and 37,123 cases abstracted in 2000 2001, there were 14,151 (36.1%) cases that met our inclusion criteria in the 1998 1999 cohort and 13,579 (36.6%) cases remaining for analysis in 2000 2001 (table 1). The number of cases analyzed by hospital ranged from 1 to 122 cases in each of the 2 cohorts (median, 3 cases per hospital for both cohorts). Patient demographic characteristics are summarized in table 2. Although there were some statistically significant differences between 1998 1999 and 2000 2001 in the patient population, the absolute differences were small. In-hospital, 14-day, and 30-day mortality rates all decreased from 1998 1999 to 2000 2001; however, readmission rates were increased. Patient clinical characteristics are summarized in table 3. There were significant but relatively small absolute differences in the proportion of patients admitted with congestive heart failure, cerebrovascular disease, chronic renal disease, altered mental status, tachypnea, hypotension, and elevation of blood urea nitrogen level. The proportion of patients admitted to the ICU decreased in 2000 2001, as did the proportion of patients in PSI risk classification V. Unadjusted patient mortality for non-icu and ICU patients stratified by initial antimicrobial regimen is summarized in ta- S196 CID 2008:47 (Suppl 3) Bratzler et al.

Table 4. Unadjusted mortality rates for community-dwelling Medicare patients, according to initial antibiotic treatment. Initial antimicrobial regimen In-hospital mortality, % 14-Day mortality, % 30-Day mortality, % Non-ICU patients ICU patients a Non-ICU patients ICU patients a,b Non-ICU patients ICU patients a Third-generation cephalosporin 4.0 (178/4463) 12.1 (51/423) 5.3 (237/4463) 11.3 (48/423) 8.4 (376/4463) 15.4 (65/423) Macrolide monotherapy 1.4 (10/717) 0.0 (0/43) 2.1 (15/717) 0.0 (0/43) 3.9 (28/717) 0.0 (0/43) Second-generation cephalosporin 3.5 (36/1037) 15.1 (13/86) 4.7 (49/1037) 11.6 (10/86) 7.9 (82/1037) 17.4 (15/86) Quinolone monotherapy 2.9 (146/5045) 11.5 (48/418) 3.6 (180/5045) 10.8 (45/418) 6.3 (318/5045) 15.6 (65/418) Any aminoglycoside 7.4 (38/513) 28.5 (53/186) 7.6 (39/513) 26.9 (50/186) 10.1 (52/513) 35.5 (66/186) Cephalosporin plus macrolide 2.9 (173/5963) 6.7 (45/673) 3.5 (211/5963) 7.4 (50/673) 5.7 (338/5963) 10.5 (71/673) Cephalosporin plus quinolone 4.0 (60/1484) 15.9 (33/207) 5.1 (76/1484) 15.0 (31/207) 7.7 (114/1484) 20.3 (42/207) b-lactam or b-lactamase inhibitor plus macrolide 2.3 (4/171) 15.0 (3/20) 3.5 (6/171) 10.0 (2/20) 4.7 (8/171) 15.0 (3/20) Other 4.6 (248/5387) 19.0 (170/894) 5.2 (279/5387) 16.9 (151/894) 8.4 (455/5387) 22.9 (205/894) NOTE. Data are for the combined 1998 1999 and 2000 2001 cohorts ( n p 27,730). The numerator and denominator for calculation of mortality rates are given in parentheses. Mortality rates were calculated from the date of hospital admission. ICU, intensive care unit. a Admitted to the ICU within 24 h after hospital arrival. b The 14-day mortality rate is lower than the reported in-hospital mortality rate because the date of death for 20% of ICU-admitted patients with CAP was 114 days after hospital admission for patients who were still hospitalized. ble 4. The in-hospital and 30-day mortality rates for patients not admitted to the ICU within 24 h after arrival were 3.6% and 7.1%, respectively. For those patients admitted to the ICU, the in-hospital and 30-day mortality rates were 14.1% and 18.0%, respectively. AORs for in-hospital, 14-day, and 30-day mortality, stratified by ICU admission and by initial antimicrobial treatment, are summarized in table 5. For non-icu patients, initial treatment with fluoroquinolone monotherapy was associated with reduced in-hospital, 14-day, and 30-day mortality rates (AOR for 30-day mortality, 0.7; 95% CI, 0.6 0.9; P p.001). The combination of a cephalosporin plus a macrolide was associated with a reduced 14-day and 30-day mortality (AOR for 30-day mortality, 0.7; 95% CI, 0.6 0.9; P!.001). For ICU patients, the use of an aminoglycoside was associated with the elevated risk of death during the hospital stay, at 14 days, and at 30 days Table 5. Adjusted ORs (AORs) for mortality among community-dwelling Medicare patients, according to initial antibiotic treatment. Patient group, initial antimicrobial regimen Non-ICU patients ( n p 24,780) In-hospital mortality 14-Day mortality 30-Day mortality AOR (95% CI) P AOR (95% CI) P AOR (95% CI) P Third-generation cephalosporin Reference Reference Reference Macrolide monotherapy 0.6 (0.3 1.1).085 0.5 (0.3 0.9).029 0.5 (0.3 0.9).009 Second-generation cephalosporin 1.0 (0.6 1.4).824 1.0 (0.7 1.4).805 1.0 (0.8 1.4).776 Quinolone monotherapy 0.8 (0.6 1.0).034 0.7 (0.5 0.8).001 0.7 (0.6 0.9).001 Any aminoglycoside 2.0 (1.3 3.0).001 1.4 (0.9 2.1).095 1.2 (0.8 1.7).302 Cephalosporin plus macrolide 0.8 (0.6 1.1).120 0.7 (0.6 0.9).003 0.7 (0.6 0.9)!.001 Cephalosporin plus quinolone 1.0 (0.7 1.5).825 1.0 (0.7 1.3).922 0.9 (0.7 1.2).500 b-lactam or b-lactamase inhibitor plus macrolide 0.8 (0.3 2.2).640 0.9 (0.4 2.1).751 0.7 (0.3 1.5).346 Other 1.2 (1.0 1.5).077 1.0 (0.8 1.2).904 1.0 (0.9 1.2).773 ICU patients ( n p 2950) Third-generation cephalosporin Reference Reference Reference Macrolide monotherapy 0.0 (0.0 1.0).977 0.0 (0.0 1.0).977 0.0 (0.0 1.0).974 Second-generation cephalosporin 1.6 (0.8 3.2).211 1.3 (0.6 2.7).559 1.4 (0.7 2.8).295 Quinolone monotherapy 0.9 (0.5 1.5).657 0.9 (0.6 1.6).841 0.9 (0.5 1.4).525 Any aminoglycoside 2.2 (1.3 3.6).002 2.0 (1.2 3.3).009 2.2 (1.4 3.5).001 Cephalosporin plus macrolide 0.6 (0.3 0.9).018 0.7 (0.4 1.1).081 0.7 (0.5 1.1).096 Cephalosporin plus quinolone 1.2 (0.7 2.2).472 1.2 (0.7 2.2).504 1.2 (0.7 2.0).535 b-lactam or b-lactamase inhibitor plus macrolide 1.6 (0.4 6.1).485 0.9 (0.2 4.2).890 1.0 (0.3 3.9).95 Other 1.7 (1.1 2.5).009 1.5 (1.0 2.2).042 1.5 (1.1 2.2).020 NOTE. Data are for the combined 1998 1999 and 2000 2001 cohorts ( n p 27,730 ). Results were adjusted for age, sex, neoplastic disease, cardiovascular disease, altered mental status, respiratory rate 30 breaths/min, systolic blood pressure!90 mm Hg, temperature!35 C or 40 C, pulse 125 beats/min, arterial ph!7.35, blood urea nitrogen level 111 mmol/l, sodium level!130 meq/l, hematocrit!30%, partial pressure of oxygen!60 mm Hg, geographic region, first dose of antibiotics administered within 4 h after hospital arrival, and presence of pleural effusion. ICU, intensive care unit. Initial Antibiotic Selection and Patient Outcomes CID 2008:47 (Suppl 3) S197

Table 6. Adjusted ORs (AORs) for 30-day mortality, according to initial antibiotic treatment and discharge time frame. Initial antimicrobial regimen July September discharge October December discharge January March discharge AOR (95% CI) P AOR (95% CI) P AOR (95% CI) P Third-generation cephalosporin Reference Reference Reference Macrolide monotherapy 1.0 (0.5 2.0).989 0.3 (0.1 0.6).001 0.4 (0.1 1.2).107 Second-generation cephalosporin 1.7 (1.1 2.6).019 0.8 (0.5 1.1).152 1.4 (0.7 2.8).302 Quinolone monotherapy 0.8 (0.6 1.2).347 0.7 (0.6 0.9).003 0.7 (0.5 1.0).063 Any aminoglycoside 2.1 (1.3 3.4).001 1.3 (0.9 1.8).236 1.2 (0.6 2.4).594 Cephalosporin plus macrolide 0.9 (0.7 1.3).646 0.6 (0.5 0.8)!.001 0.6 (0.4 0.9).022 Cephalosporin plus quinolone 1.0 (0.6 1.7).989 0.9 (0.6 1.1).291 1.1 (0.7 1.8).761 b-lactam or b-lactamase inhibitor plus macrolide 1.8 (0.7 4.9).236 0.7 (0.3 1.8).491 0.0 (0.0 1.0).969 Other 1.5 (1.1 2.0).011 1.0 (0.8 1.2).662 1.1 (0.8 1.6).556 NOTE. Data are for the combined 1998 1999 and 2000 2001 cohorts ( n p 27,730 ). Results were adjusted for age, sex, neoplastic disease, cardiovascular disease, altered mental status, respiratory rate 30 breaths/min, systolic blood pressure!90 mm Hg, temperature!35 Cor 40 C, pulse 125 beats/min, arterial ph!7.35, blood urea nitrogen level 111 mmol/l, sodium level!130 meq/l, hematocrit!30%, partial pressure of oxygen!60 mm Hg, presence of pleural effusion, admission to the intensive care unit within 24 h after hospital arrival, geographic region, and administration of antibiotic treatment within 4 h after hospital arrival. after admission (AOR at 30 days, 2.2; 95% CI, 1.4 3.5; P p.001). The combination of a cephalosporin plus a macrolide was associated with a reduced risk of in-hospital mortality (AOR, 0.6; 95% CI, 0.3 0.9; P p.018 ) but not associated with significant reductions in 14-day or 30-day mortality rate. AORs for 30-day mortality, stratified by discharge time frame and by initial antimicrobial treatment, are summarized in table 6. Significant reductions in risk of death were noted for macrolide monotherapy (AOR, 0.3; 95% CI, 0.1 0.6; P p.001), fluoroquinolone monotherapy (AOR, 0.7; 95% CI, 0.6 0.9; P p.003), and the combination of a cephalosporin plus macrolide (AOR, 0.6; 95% CI, 0.5 0.8; P!.001) for patients discharged in October December. Patients given treatment with the combination of a cephalosporin plus a macrolide and discharged in January March also had reduced mortality (AOR, 0.6; 95% CI, 0.4 0.9; P p.022). We also evaluated the relationship between initial antimicrobial selection and patient outcomes stratified by region of the country (data not shown). At 30 days, significant reductions in mortality were demonstrated with the combination of a cephalosporin plus a macrolide in the Northeast (AOR, 0.6; 95% CI, 0.4 1.0; P p.05), South (AOR, 0.7; 95% CI, 0.5 0.9; P p.01), and West (AOR, 0.7; 95% CI, 0.5 0.9; P p.01). Fluoroquinolone monotherapy was associated with significant reductions in mortality in the South (AOR, 0.7; 95% CI, 0.5 0.9; P!.01) and West (AOR, 0.7; 95% CI, 0.5 1.0; P p.03). With the exception of increased mortality among those patients receiving an aminoglycoside, no other statistically significant associations were noted between anti- Table 7. Adjusted ORs (AORs) for 30-day mortality, according to initial antibiotic treatment and pneumonia severity index (PSI) risk class. Initial antimicrobial regimen PSI risk classes II and III PSI risk classes IV and V AOR (95% CI) P AOR (95% CI) P Third-generation cephalosporin Reference Reference Macrolide monotherapy 0.7 (0.3 1.8).427 0.5 (0.3 0.8).003 Second-generation cephalosporin 1.0 (0.4 2.1).944 1.1 (0.8 1.5).477 Quinolone monotherapy 0.9 (0.6 1.5).750 0.7 (0.6 0.9).001 Any aminoglycoside 1.7 (0.7 3.9).203 1.4 (1.1 1.9).011 Cephalosporin plus macrolide 0.9 (0.6 1.4).609 0.7 (0.6 0.8)!.001 Cephalosporin plus quinolone 1.4 (0.8 2.5).260 0.9 (0.7 1.2).411 b-lactam or b-lactamase inhibitor plus macrolide 0.7 (0.1 5.1).712 0.8 (0.4 1.5).420 Other 1.0 (0.6 1.5).875 1.1 (1.0 1.3).156 NOTE. Data are for the combined 1998 1999 and 2000 2001 cohorts ( n p 27,730). ORs are adjusted for ad- mission to the intensive care unit within the first 24 h after hospital arrival, antibiotic treatment administered within 4 h after hospital arrival, and geographic region. S198 CID 2008:47 (Suppl 3) Bratzler et al.

Table 8. Trends in initial antibiotic treatment among community-dwelling Medicare patients admitted to the hospital with pneumonia. Patient group, initial antimicrobial regimen 1998 1999 2000 2001 2002 2003 2004 2005 a 2006 a Non-ICU patients No. of patients 12,539 12,241 11,995 11,128 11,968 357,891 254,465 Third-generation cephalosporin 22.3 13.7 10.6 8.0 6.4 5.6 3.8 Macrolide monotherapy 3.2 2.6 1.6 1.1 1.2 1.1 0.8 Second-generation cephalosporin 6.2 2.2 0.9 0.4 0.2 0.1 0.0 Quinolone monotherapy 12.5 28.4 27.2 25.5 22.8 24.3 25.3 Any aminoglycoside 2.5 1.6 1.7 1.5 1.4 1.2 0.9 Cephalosporin plus macrolide 25.7 22.4 22.4 27.2 30.2 31.3 32.8 Cephalosporin plus quinolone 5.1 6.9 6.6 6.0 6.0 6.1 6.1 b-lactam or b-lactamase inhibitor plus macrolide 0.7 0.6 0.3 0.5 0.5 0.6 0.5 Other b 21.9 21.5 28.7 29.7 31.0 29.6 29.9 ICU patients No. of patients 1612 1338 1207 1126 1363 39,835 28,587 Third-generation cephalosporin 22.3 10.8 6.0 4.9 4.1 3.8 2.4 Macrolide monotherapy 3.2 1.1 0.9 0.4 0.6 0.6 0.3 Second-generation cephalosporin 6.2 1.2 0.7 0.0 0.1 0.0 0.0 Quinolone monotherapy 12.5 20.3 17.5 14.7 14.2 14.1 13.6 Any aminoglycoside 2.5 4.1 5.7 4.0 3.2 4.0 2.8 Cephalosporin plus macrolide 25.7 20.9 19.9 22.5 22.2 22.7 22.6 Cephalosporin plus quinolone 5.1 8.8 7.5 6.9 6.9 7.4 8.1 b-lactam or b-lactamase inhibitor plus macrolide 0.7 0.7 1.1 1.7 0.7 1.1 0.8 Other b 21.9 32.0 40.7 45.0 48.0 46.3 49.4 NOTE. Data are % of patients, unless otherwise indicated. ICU, intensive care unit. a Antimicrobial use for 2005 and 2006 is based on hospital self-abstracted data submitted to the Quality Improvement Organization National Clinical Warehouse as a part of the Reporting Hospital Quality Data for Annual Payment Update initiative (pay-for-reporting) sponsored by the Centers for Medicare & Medicaid Services. All hospitals are subject to random validation audits of submitted data. b The marked increase in the number of patients receiving the other regimen is primarily because of patients to whom 3 antibiotics were administered within 36 h after hospital arrival. biotic selection and 30-day mortality rate for patients discharged from a hospital in the Midwest region during the time frames studied. AORs for 30-day mortality, stratified by PSI score and by initial antimicrobial treatment, are summarized in table 7. Treatment with macrolide monotherapy (AOR, 0.5; 95% CI, 0.3 0.8; P p.003), fluoroquinolone monotherapy (AOR, 0.7; 95% CI, 0.6 0.9; P p.001), and the combination of a cephalosporin plus a macrolide (AOR, 0.7; 95% CI, 0.6 0.8; P!.001) resulted in lower odds of mortality for patients in PSI risk classes IV and V. No significant associations between antibiotic selection and 30-day mortality were noted for patients in PSI risk classes II and III. National trends in initial antibiotic treatment for Medicare patients admitted to the hospital with pneumonia, stratified by care in the ICU or in a non-icu ward, are summarized in table 8. Consistent with guideline recommendations and national performance measures that are publicly reported for hospitals, use of second- and third-generation cephalosporin monotherapy has decreased dramatically, whereas the use of fluoroquinolone monotherapy and combination therapy with a cephalosporin plus a macrolide have increased. Patients receiving 3 antimicrobials within 36 h after hospitalization have increased substantially particularly among the population of ICU patients. DISCUSSION Our findings are similar to those of other observational studies demonstrating an association between antibiotic selection and patient outcomes [2 5, 15 18]. The magnitude of benefit of fluoroquinolone monotherapy or a cephalosporin plus a macrolide, compared with third-generation cephalosporin monotherapy, was similar to that demonstrated by Gleason et al. [2]. We found greater benefit of fluoroquinolone monotherapy and of a cephalosporin plus a macrolide for patients with moresevere pneumonia (PSI risk classes IV and V). However, for the 10.6% of our sample who were admitted to the ICU, we found no association between antibiotic selection and patient mortality. Initial Antibiotic Selection and Patient Outcomes CID 2008:47 (Suppl 3) S199

Although our results are consistent with those of a number of other studies highlighting the association between antimicrobial selection and patient outcomes, the benefit of giving patients treatment with respiratory fluoroquinolones or the combination of a cephalosporin plus a macrolide has been challenged [19 24]. Multiple issues have been raised about using a nonexperimental cohort study to identify associations between a process of care (such as antibiotic selection) and patient outcomes. Perhaps the greatest risk is the potential for residual confounding of patient outcomes due to unmeasured variables [19, 23, 24]. The finding in our study and similar findings by Gleason et al. [2] that macrolide monotherapy was associated with a reduced 30-day risk-adjusted mortality and that the use of any aminoglycoside was associated with elevated risk-adjusted 30-day mortality in patients with pneumonia in PSI risk classes IV and V highlight this concern. As noted by others, the use of macrolide monotherapy for patients with pneumonia may actually be a marker of less-severe disease, as determined by the clinician at the bedside. Similarly, the use of an aminoglycoside may reflect a clinical decision at the bedside to use the agent in a patient with severe pneumonia. It is possible that unmeasured clinical variables prevent adequate risk adjustment of patient outcomes and result in the findings demonstrated in our study and the studies of others. Another controversy raised about the evaluation of antibiotic selection and patient outcomes is the time frame after admission that should be used to assess outcomes such as mortality [25]. Are deaths at 30 days after a pneumonia hospitalization attributable to decisions made about antibiotic selection at the time of hospital admission? Finally, our study and others have been limited to patients in the Medicare-eligible age group because of the ability to follow up with patients for outcomes such as readmission and death on the basis of administrative data. The results of this study may not be generalizable to other populations. Despite the limitations, observational studies do present some advantages for evaluating the association between processes of care and outcomes. The most obvious advantage is the very large sample of patients available for evaluation in studies from efforts such as the National Pneumonia Project. In addition to being relatively cost-effective, observational studies allow investigators to test hypotheses that might not be considered ethical in a prospective study (e.g., evaluating the association between delayed antibiotic delivery and patient outcomes). Because the treatment effect of choosing one antibiotic combination over another may result in relatively small absolute differences in mortality, the number of patients who would be required in a prospective study sample to evaluate the outcome of interest may be prohibitive. We estimate, for example, that, on the basis of a 30-day mortality rate of 8.4% for a community-dwelling Medicare patient who was admitted to a non-icu hospital bed, it would take 12700 patients per treatment arm to demonstrate a 2% absolute reduction in mortality of an investigational antibiotic, compared with third-generation cephalosporin monotherapy. In addition, observational studies may allow for evaluation of a population of patients from the community setting that is more representative than the population often enrolled in prospective, randomized trials on the basis of strict study enrollment criteria [26]. How might the results of our study and other observational studies help to inform the design of clinical trials of antibiotic therapy for pneumonia? Observational studies provide useful information about the likely clinical response to treatment and the likelihood of outcomes, such as death and readmission, for large populations of patients with pneumonia. This will clearly inform sample-size considerations for prospective clinical trials. The results of observational studies may also inform the magnitude of the absolute differences in outcomes between various treatment regimens, which would also inform the design of clinical trials, including sample-size considerations. Finally, observational studies may identify unforeseen issues that could have an impact on the evaluation of a clinical trial. For instance, the results of our study suggest that outcomes of pneumonia, stratified by antibiotic selection, may vary with both the region of the country in which the patient is hospitalized and the time of the year of hospitalization (perhaps linked to coexistent viral infections or differences in the prevalence in infections caused by atypical organisms). CONCLUSION In summary, our study of community-dwelling patients with pneumonia who were admitted to the hospital demonstrated that initial antimicrobial treatment with the combination of a second- or third-generation cephalosporin plus a macrolide or initial treatment with a fluoroquinolone was associated with a reduced 30-day mortality rate, compared with treatment with third-generation cephalosporin monotherapy in the non-icu population. Although our results are consistent with those of other observational studies of antibiotic selection and patient outcomes in pneumonia, controversy continues to exist about the use of nonexperimental cohort studies to demonstrate associations between processes of care, such as antibiotic selection, and patient outcomes. This study continues to identify the need for randomized trials to definitively answer questions about antibiotic selection and patient mortality. The results of these large-scale observational studies may provide useful information that will inform decisions in the development and design of future randomized trials of antimicrobial treatment for patients with pneumonia. S200 CID 2008:47 (Suppl 3) Bratzler et al.

Acknowledgments Financial support. The analyses on which this publication is based were performed in part under contract numbers 500-02-OK-03 funded by the Centers for Medicare & Medicaid Services, an agency of the US Department of Health and Human Services (4-598-OK-0208). Supplement sponsorship. This article was published as part of a supplement entitled Workshop on Issues in the Design and Conduct of Clinical Trials of Antibacterial Drugs for the Treatment of Community-Acquired Pneumonia, sponsored by the US Food and Drug Administration and the Infectious Diseases Society of America. Potential conflicts of interest. All authors: no conflicts. References 1. Mandell LA, Wunderink RG, Anzueto A, et al. Infectious Diseases Society of America/American Thoracic Society consensus guidelines on the management of community-acquired pneumonia in adults. Clin Infect Dis 2007; 44:S27 72. 2. Gleason PP, Meehan TP, Fine JM, Galusha DH, Fine MJ. Associations between initial antimicrobial therapy and medical outcomes for hospitalized elderly patients with pneumonia. Arch Intern Med 1999; 159: 2562 72. 3. Houck PM, MacLehose RG, Niederman MS, Lowery JK. Empiric antibiotic therapy and mortality among Medicare pneumonia inpatients in 10 western states: 1993, 1995, and 1997. Chest 2001; 119:1420 6. 4. Dudas V, Hopefl A, Jacobs R, Guglielmo BJ. Antimicrobial selection for hospitalized patients with presumed community-acquired pneumonia: a survey of nonteaching US community hospitals. Ann Pharmacother 2000; 34:446 52. 5. Brown RB, Iannini P, Gross P, Kunkel M. Impact of initial antibiotic choice on clinical outcomes in community-acquired pneumonia: analysis of a hospital claims-made database. Chest 2003; 123:1503 11. 6. Battleman DS, Callahan M, Thaler HT. Rapid antibiotic delivery and appropriate antibiotic selection reduce length of hospital stay of patients with community-acquired pneumonia: link between quality of care and resource utilization. Arch Intern Med 2002; 162:682 8. 7. Marras TK, Jamieson L, Chan CK. Inpatient care of community-acquired pneumonia: the effect of antimicrobial guidelines on clinical outcomes and drug costs in Canadian teaching hospitals. Can Respir J 2004; 11:131 7. 8. Mortensen EM, Restrepo M, Anzueto A, Pugh J. Effects of guidelineconcordant antimicrobial therapy on mortality among patients with community-acquired pneumonia. Am J Med 2004; 117:726 31. 9. Dean NC, Bateman KA, Donnelly SM, et al. Improved clinical outcomes with utilization of a community-acquired pneumonia guideline. Chest 2006; 130:794 9. 10. Frei CR, Restrepo MI, Mortensen EM, Burgess DS. Impact of guidelineconcordant empiric antibiotic therapy in community-acquired pneumonia. Am J Med 2006; 119:865 71. 11. Mortensen EM, Restrepo MI, Anzueto A, Pugh JA. Antibiotic therapy and 48-hour mortality for patients with pneumonia. Am J Med 2006; 119:859 64. 12. Centers for Medicare & Medicaid Services. Pneumonia. Available at: http://www.qualitynet.org/dcs/contentserver?cidp1089815967023 &pagenamepmedqic%2fcontent%2fparentshelltemplate&parent NamepTopic&siteVersionpnull&cpMQParents. Accessed 7 October 2008. 13. Public Health Service. International classification of diseases, ninth revision: clinical modification. 4th ed. Washington, DC: US Department of Health and Human Services, 1991:report PHS91-1260. 14. Fine MJ, Auble TE, Yealy DM, et al. A prediction rule to identify lowrisk patients with community-acquired pneumonia. N Engl J Med 1997; 336:243 50. 15. Stahl JE, Barza M, DesJardin J, Martin R, Eckman MH. Effect of macrolides as part of initial empiric therapy on length of stay in patients hospitalized with community-acquired pneumonia. Arch Intern Med 1999; 159:2576 80. 16. Baddour LM, Yu VL, Klugman KP, et al. Combination antibiotic therapy lowers mortality among severely ill patients with pneumococcal bacteremia. Am J Respir Crit Care Med 2004; 170:440 4. 17. Lodise TP, Kwa A, Cosier L, Gupta R, Smith RP. Comparison of b- lactam and macrolide combination therapy versus fluoroquinolone monotherapy in hospitalized Veterans Affairs patients with community-acquired pneumonia. Antimicrob Agents Chemother 2007;51: 3977 82. 18. Metersky ML, Ma A, Houck PM, Bratzler DW. Antibiotics for bacteremic pneumonia: improved outcomes with macrolides but not fluoroquinolones. Chest 2007; 131:466 73. 19. Dowell SF. The best treatment for pneumonia: new clues, but no definitive answers. Arch Intern Med 1999; 159:2511 2. 20. Burgess DS, Lewis JS 2nd. Effect of macrolides as part of initial empiric therapy on medical outcomes for hospitalized patients with community-acquired pneumonia. Clin Ther 2000; 22:872 8. 21. Shefet D, Robenshtok E, Paul M, et al. Empirical atypical coverage for inpatients with community-acquired pneumonia. Arch Intern Med 2005; 165:1992 2000. 22. Mills GD, Oehley MR, Arrol B. Effectiveness of b lactam antibiotics compared with antibiotics active against atypical pathogens in nonsevere community acquired pneumonia: meta-analysis. BMJ 2005; 330: 456 62. 23. Harbarth S, Barbino J, Pugin J, Romand JA, Pittet D. Lack of effect of combination antibiotic therapy on mortality in patients with pneumococcal sepsis. Eur J Clin Microbiol Infect Dis 2005; 24:688 90. 24. Oosterheert JJ, Bonten MJ, Hak E, Schneider MM, Hoepelman IM. How good is the evidence for the recommended empirical antimicrobial treatment of patients hospitalized because of community-acquired pneumonia? A systematic review. J Antimicrob Chemother 2003;52: 555 63. 25. Mortensen EM, Coley CM, Singer DE, et al. Causes of death for patients with community-acquired pneumonia: results from the Pneumonia Patient Outcomes Research Team cohort study. Arch Intern Med 2002; 162:1059 64. 26. Weiss NS, Koepsell TD, Psaty BM. Generalizability of the results of randomized trials. Arch Intern Med 2008; 168:133 5. Initial Antibiotic Selection and Patient Outcomes CID 2008:47 (Suppl 3) S201