Kathleen Kerr, BA

Study Population

A retrospective cohort study of all adults (age 18 years) discharged from the inpatient general medicine service of Moffitt‐Long Hospital at the University of California, San Francisco, was conducted from January 1999 through July 2001. Eligibility criteria included a principal discharge diagnosis of CAP and a chest radiograph demonstrating an infiltrate within 48 hours of admission. Exclusion criteria included infection with the human immunodeficiency virus, history of organ transplantation or use of immunosuppressive therapy (including prednisone > 15 mg/day), cystic fibrosis, postobstructive pneumonia, active tuberculosis, recent hospitalization (within 10 days), or admission for comfort care. The study protocol and procedures were reviewed and approved by the UCSF Committee for Human Research.

Data Collection

Medical record review by trained research assistants blinded to the research question was used to gather demographic data, comorbid illnesses, physical examination findings on initial presentation, and laboratory or radiographic results on initial presentation. The pneumonia severity index (PSI) score was calculated for each patient using the above data.14 In addition, data were collected on antibiotic allergies, antibiotics used within the 30 days prior to admission, results of sputum or blood cultures, and admission location (intensive care unit [ICU] versus medical floor).

Data from TSI (Transition Systems Inc., Boston, MA), the hospital administrative database, were used to identify the initial empiric antibiotic regimen. All antibiotics prescribed within the first 48 hours of hospitalization were considered initial empiric therapy with few exceptions. Initial empiric therapy was classified as 1) ceftriaxone plus doxycycline (including patients treated with these agents alone in the first 48 hours, as well as patients treated with both agents in the first 24 hours who were switched to alternative therapy [broader coverage] on the second day), or 2) other appropriate therapy (treatment consistent with current national guideline recommendations including at least a beta‐lactam plus a macrolide or a beta‐lactam plus a fluoroquinolone, or fluoroquinolone monotherapy). Patients receiving therapy inconsistent with current national guideline recommendations were excluded.

Outcomes

TSI data were used to identify length of stay, death during the index hospitalization, and return to the emergency department or readmission within 30 days of discharge. The National Death Index was used to identify all deaths that occurred after hospital discharge. The 30‐day mortality data included deaths occurring during the index hospitalization and in the 30 days after the index hospitalization discharge.

Statistical Analysis

For the purposes of this analysis we compared patients treated with ceftriaxone plus doxycycline to patients treated with other appropriate therapy. To examine demographic and clinical differences between the two groups, statistical tests of comparison were performed using chi‐square tests for the dichotomous variables and t tests for the numeric variables, all of which were normally distributed (after log transformation in the case of length of stay).

To adjust for clinical variables that might contribute to differences in outcomes between the two groups, we used backward stepwise logistic regression analysis to construct a propensity score15 for the likelihood of ceftriaxone plus doxycycline use. The propensity score reflected the conditional probability of exposure to ceftriaxone plus doxycycline and allowed for stratification and, subsequently, comparisons by quintiles of propensity score. Propensity scores often have distinct advantages over direct adjustment for a large number of confounding variables and allow direct comparisons between groups with a similar propensity for receiving ceftriaxone plus doxycycline.15 Unlike random assignment of treatment, however, the propensity score cannot balance unmeasured variables that may affect treatment assignment. Thus, the possibility of bias remains. The variables used to build the score included age, presence of comorbid illness, admission from a nursing home or long‐term care facility, antibiotic allergy, prior antibiotic use, PSI score, PSI risk class, diagnosis of aspiration, admission to the ICU, and positive blood cultures. The propensity score was then stratified and used as an adjustment variable in comparisons between groups for in‐hospital mortality, 30‐day mortality, and 30‐day readmission rates. As expected, length of stay was highly skewed and was therefore log‐transformed and compared between groups with adjustment for the propensity score.

To further address issues related to potential selection bias, a separate analysis was performed on a subset of the original cohort that excluded patients for whom ceftriaxone plus doxycycline would not generally be recommended as first‐line therapy. For this analysis, patients admitted from a nursing home or long‐term care facility, patients admitted to the ICU, and patients with a principal diagnosis of aspiration pneumonia were excluded. A propensity score was rederived for this subset, which was used to adjust for differences in outcomes. All statistical procedures were performed using STATA (Ver. 7.0, Stata Corporation, College Station TX).

Patient Characteristics

A total of 341 patients were eligible for analysis. Of this group, 216 were treated with ceftriaxone plus doxycycline and 125 received other appropriate therapy. Both groups of patients were similar in age. Patients treated with ceftriaxone plus doxycycline had a lower median PSI score and fewer comorbid illnesses than did patients treated with other appropriate therapy (Table 1). Blood cultures were positive in 30 (8.8%) of the 341 patients included in the analysis, with S. pneumoniae the most commonly isolated organism (n = 17, 5.0%). Of S. pneumoniae isolates, 4 (24%) were resistant to penicillin (MIC 1 g/mL), and 2 (12%) were resistant to tetracycline (MIC 8 g/mL).

Common antibiotic choices in patients receiving other appropriate therapy included a beta‐lactam/beta‐lactamase inhibitor plus doxycycline or a macrolide (n = 36, 29%), fluoroquinolone monotherapy (n = 16, 13%), and a variety of other antibiotic combinations with activity against S. pneumoniae and atypical bacteria (n = 52, 42%).

Clinical Outcomes

Analyses of unadjusted outcomes showed that patients treated with ceftriaxone plus doxycycline had significantly lower inpatient (2% vs. 14%, P < .001) and 30‐day (6% vs. 20%, P < .001) mortality compared to patients treated with other regimens (Table 2). Multivariable logistic regression analysis identified three variables (diagnosis of congestive heart failure, admission to the ICU, and the presence of comorbid illness) associated with initial antibiotic selection, which were used to build a propensity score. After adjustment for the propensity score, use of ceftriaxone plus doxycycline remained significantly associated with lower inpatient mortality (OR = 0.26, 95% CI: 0.080.81) and 30‐day mortality (OR = 0.37, 95% CI: 0.170.81). Differences in length of stay and 30‐day readmission rates between the treatment groups were not significant (Table 2).

Subset Analysis

To address issues related to selection bias, we performed an analysis of a subset of the patients after excluding those admitted from a nursing home, diagnosed with aspiration, or admitted to the ICU, for whom ceftriaxone plus doxycycline would not be considered recommended (or first‐line) therapy. The two resulting groups were similar, except there were fewer patients with comorbid illness in the ceftriaxone plus doxycycline group (34% vs. 50%, P = .015). The propensity score was rederived for this subset and used for adjustment. Unadjusted and adjusted outcomes are shown in Table 3. Use of ceftriaxone plus doxycycline in this subset also was associated with reduced odds of inpatient mortality (OR = 0.17, 95% CI: 0.040.77). The odds of 30‐day mortality also were reduced but not significantly, as the confidence interval included 1.0 (OR = 0.43, 95% CI: 0.141.31). There were no differences between groups in length of stay or in 30‐day readmission rate.

Study Population

A retrospective cohort study of all adults (age 18 years) discharged from the inpatient general medicine service of Moffitt‐Long Hospital at the University of California, San Francisco, was conducted from January 1999 through July 2001. Eligibility criteria included a principal discharge diagnosis of CAP and a chest radiograph demonstrating an infiltrate within 48 hours of admission. Exclusion criteria included infection with the human immunodeficiency virus, history of organ transplantation or use of immunosuppressive therapy (including prednisone > 15 mg/day), cystic fibrosis, postobstructive pneumonia, active tuberculosis, recent hospitalization (within 10 days), or admission for comfort care. The study protocol and procedures were reviewed and approved by the UCSF Committee for Human Research.

Data Collection

Medical record review by trained research assistants blinded to the research question was used to gather demographic data, comorbid illnesses, physical examination findings on initial presentation, and laboratory or radiographic results on initial presentation. The pneumonia severity index (PSI) score was calculated for each patient using the above data.14 In addition, data were collected on antibiotic allergies, antibiotics used within the 30 days prior to admission, results of sputum or blood cultures, and admission location (intensive care unit [ICU] versus medical floor).

Data from TSI (Transition Systems Inc., Boston, MA), the hospital administrative database, were used to identify the initial empiric antibiotic regimen. All antibiotics prescribed within the first 48 hours of hospitalization were considered initial empiric therapy with few exceptions. Initial empiric therapy was classified as 1) ceftriaxone plus doxycycline (including patients treated with these agents alone in the first 48 hours, as well as patients treated with both agents in the first 24 hours who were switched to alternative therapy [broader coverage] on the second day), or 2) other appropriate therapy (treatment consistent with current national guideline recommendations including at least a beta‐lactam plus a macrolide or a beta‐lactam plus a fluoroquinolone, or fluoroquinolone monotherapy). Patients receiving therapy inconsistent with current national guideline recommendations were excluded.

Outcomes

TSI data were used to identify length of stay, death during the index hospitalization, and return to the emergency department or readmission within 30 days of discharge. The National Death Index was used to identify all deaths that occurred after hospital discharge. The 30‐day mortality data included deaths occurring during the index hospitalization and in the 30 days after the index hospitalization discharge.

Statistical Analysis

For the purposes of this analysis we compared patients treated with ceftriaxone plus doxycycline to patients treated with other appropriate therapy. To examine demographic and clinical differences between the two groups, statistical tests of comparison were performed using chi‐square tests for the dichotomous variables and t tests for the numeric variables, all of which were normally distributed (after log transformation in the case of length of stay).

To adjust for clinical variables that might contribute to differences in outcomes between the two groups, we used backward stepwise logistic regression analysis to construct a propensity score15 for the likelihood of ceftriaxone plus doxycycline use. The propensity score reflected the conditional probability of exposure to ceftriaxone plus doxycycline and allowed for stratification and, subsequently, comparisons by quintiles of propensity score. Propensity scores often have distinct advantages over direct adjustment for a large number of confounding variables and allow direct comparisons between groups with a similar propensity for receiving ceftriaxone plus doxycycline.15 Unlike random assignment of treatment, however, the propensity score cannot balance unmeasured variables that may affect treatment assignment. Thus, the possibility of bias remains. The variables used to build the score included age, presence of comorbid illness, admission from a nursing home or long‐term care facility, antibiotic allergy, prior antibiotic use, PSI score, PSI risk class, diagnosis of aspiration, admission to the ICU, and positive blood cultures. The propensity score was then stratified and used as an adjustment variable in comparisons between groups for in‐hospital mortality, 30‐day mortality, and 30‐day readmission rates. As expected, length of stay was highly skewed and was therefore log‐transformed and compared between groups with adjustment for the propensity score.

To further address issues related to potential selection bias, a separate analysis was performed on a subset of the original cohort that excluded patients for whom ceftriaxone plus doxycycline would not generally be recommended as first‐line therapy. For this analysis, patients admitted from a nursing home or long‐term care facility, patients admitted to the ICU, and patients with a principal diagnosis of aspiration pneumonia were excluded. A propensity score was rederived for this subset, which was used to adjust for differences in outcomes. All statistical procedures were performed using STATA (Ver. 7.0, Stata Corporation, College Station TX).

Patient Characteristics

A total of 341 patients were eligible for analysis. Of this group, 216 were treated with ceftriaxone plus doxycycline and 125 received other appropriate therapy. Both groups of patients were similar in age. Patients treated with ceftriaxone plus doxycycline had a lower median PSI score and fewer comorbid illnesses than did patients treated with other appropriate therapy (Table 1). Blood cultures were positive in 30 (8.8%) of the 341 patients included in the analysis, with S. pneumoniae the most commonly isolated organism (n = 17, 5.0%). Of S. pneumoniae isolates, 4 (24%) were resistant to penicillin (MIC 1 g/mL), and 2 (12%) were resistant to tetracycline (MIC 8 g/mL).

Common antibiotic choices in patients receiving other appropriate therapy included a beta‐lactam/beta‐lactamase inhibitor plus doxycycline or a macrolide (n = 36, 29%), fluoroquinolone monotherapy (n = 16, 13%), and a variety of other antibiotic combinations with activity against S. pneumoniae and atypical bacteria (n = 52, 42%).

Clinical Outcomes

Analyses of unadjusted outcomes showed that patients treated with ceftriaxone plus doxycycline had significantly lower inpatient (2% vs. 14%, P < .001) and 30‐day (6% vs. 20%, P < .001) mortality compared to patients treated with other regimens (Table 2). Multivariable logistic regression analysis identified three variables (diagnosis of congestive heart failure, admission to the ICU, and the presence of comorbid illness) associated with initial antibiotic selection, which were used to build a propensity score. After adjustment for the propensity score, use of ceftriaxone plus doxycycline remained significantly associated with lower inpatient mortality (OR = 0.26, 95% CI: 0.080.81) and 30‐day mortality (OR = 0.37, 95% CI: 0.170.81). Differences in length of stay and 30‐day readmission rates between the treatment groups were not significant (Table 2).

Subset Analysis

To address issues related to selection bias, we performed an analysis of a subset of the patients after excluding those admitted from a nursing home, diagnosed with aspiration, or admitted to the ICU, for whom ceftriaxone plus doxycycline would not be considered recommended (or first‐line) therapy. The two resulting groups were similar, except there were fewer patients with comorbid illness in the ceftriaxone plus doxycycline group (34% vs. 50%, P = .015). The propensity score was rederived for this subset and used for adjustment. Unadjusted and adjusted outcomes are shown in Table 3. Use of ceftriaxone plus doxycycline in this subset also was associated with reduced odds of inpatient mortality (OR = 0.17, 95% CI: 0.040.77). The odds of 30‐day mortality also were reduced but not significantly, as the confidence interval included 1.0 (OR = 0.43, 95% CI: 0.141.31). There were no differences between groups in length of stay or in 30‐day readmission rate.

Study Population

A retrospective cohort study of all adults (age 18 years) discharged from the inpatient general medicine service of Moffitt‐Long Hospital at the University of California, San Francisco, was conducted from January 1999 through July 2001. Eligibility criteria included a principal discharge diagnosis of CAP and a chest radiograph demonstrating an infiltrate within 48 hours of admission. Exclusion criteria included infection with the human immunodeficiency virus, history of organ transplantation or use of immunosuppressive therapy (including prednisone > 15 mg/day), cystic fibrosis, postobstructive pneumonia, active tuberculosis, recent hospitalization (within 10 days), or admission for comfort care. The study protocol and procedures were reviewed and approved by the UCSF Committee for Human Research.

Data Collection

Medical record review by trained research assistants blinded to the research question was used to gather demographic data, comorbid illnesses, physical examination findings on initial presentation, and laboratory or radiographic results on initial presentation. The pneumonia severity index (PSI) score was calculated for each patient using the above data.14 In addition, data were collected on antibiotic allergies, antibiotics used within the 30 days prior to admission, results of sputum or blood cultures, and admission location (intensive care unit [ICU] versus medical floor).

Data from TSI (Transition Systems Inc., Boston, MA), the hospital administrative database, were used to identify the initial empiric antibiotic regimen. All antibiotics prescribed within the first 48 hours of hospitalization were considered initial empiric therapy with few exceptions. Initial empiric therapy was classified as 1) ceftriaxone plus doxycycline (including patients treated with these agents alone in the first 48 hours, as well as patients treated with both agents in the first 24 hours who were switched to alternative therapy [broader coverage] on the second day), or 2) other appropriate therapy (treatment consistent with current national guideline recommendations including at least a beta‐lactam plus a macrolide or a beta‐lactam plus a fluoroquinolone, or fluoroquinolone monotherapy). Patients receiving therapy inconsistent with current national guideline recommendations were excluded.

Outcomes

TSI data were used to identify length of stay, death during the index hospitalization, and return to the emergency department or readmission within 30 days of discharge. The National Death Index was used to identify all deaths that occurred after hospital discharge. The 30‐day mortality data included deaths occurring during the index hospitalization and in the 30 days after the index hospitalization discharge.

Statistical Analysis

For the purposes of this analysis we compared patients treated with ceftriaxone plus doxycycline to patients treated with other appropriate therapy. To examine demographic and clinical differences between the two groups, statistical tests of comparison were performed using chi‐square tests for the dichotomous variables and t tests for the numeric variables, all of which were normally distributed (after log transformation in the case of length of stay).

To adjust for clinical variables that might contribute to differences in outcomes between the two groups, we used backward stepwise logistic regression analysis to construct a propensity score15 for the likelihood of ceftriaxone plus doxycycline use. The propensity score reflected the conditional probability of exposure to ceftriaxone plus doxycycline and allowed for stratification and, subsequently, comparisons by quintiles of propensity score. Propensity scores often have distinct advantages over direct adjustment for a large number of confounding variables and allow direct comparisons between groups with a similar propensity for receiving ceftriaxone plus doxycycline.15 Unlike random assignment of treatment, however, the propensity score cannot balance unmeasured variables that may affect treatment assignment. Thus, the possibility of bias remains. The variables used to build the score included age, presence of comorbid illness, admission from a nursing home or long‐term care facility, antibiotic allergy, prior antibiotic use, PSI score, PSI risk class, diagnosis of aspiration, admission to the ICU, and positive blood cultures. The propensity score was then stratified and used as an adjustment variable in comparisons between groups for in‐hospital mortality, 30‐day mortality, and 30‐day readmission rates. As expected, length of stay was highly skewed and was therefore log‐transformed and compared between groups with adjustment for the propensity score.

To further address issues related to potential selection bias, a separate analysis was performed on a subset of the original cohort that excluded patients for whom ceftriaxone plus doxycycline would not generally be recommended as first‐line therapy. For this analysis, patients admitted from a nursing home or long‐term care facility, patients admitted to the ICU, and patients with a principal diagnosis of aspiration pneumonia were excluded. A propensity score was rederived for this subset, which was used to adjust for differences in outcomes. All statistical procedures were performed using STATA (Ver. 7.0, Stata Corporation, College Station TX).

Patient Characteristics

A total of 341 patients were eligible for analysis. Of this group, 216 were treated with ceftriaxone plus doxycycline and 125 received other appropriate therapy. Both groups of patients were similar in age. Patients treated with ceftriaxone plus doxycycline had a lower median PSI score and fewer comorbid illnesses than did patients treated with other appropriate therapy (Table 1). Blood cultures were positive in 30 (8.8%) of the 341 patients included in the analysis, with S. pneumoniae the most commonly isolated organism (n = 17, 5.0%). Of S. pneumoniae isolates, 4 (24%) were resistant to penicillin (MIC 1 g/mL), and 2 (12%) were resistant to tetracycline (MIC 8 g/mL).

Common antibiotic choices in patients receiving other appropriate therapy included a beta‐lactam/beta‐lactamase inhibitor plus doxycycline or a macrolide (n = 36, 29%), fluoroquinolone monotherapy (n = 16, 13%), and a variety of other antibiotic combinations with activity against S. pneumoniae and atypical bacteria (n = 52, 42%).

Clinical Outcomes

Analyses of unadjusted outcomes showed that patients treated with ceftriaxone plus doxycycline had significantly lower inpatient (2% vs. 14%, P < .001) and 30‐day (6% vs. 20%, P < .001) mortality compared to patients treated with other regimens (Table 2). Multivariable logistic regression analysis identified three variables (diagnosis of congestive heart failure, admission to the ICU, and the presence of comorbid illness) associated with initial antibiotic selection, which were used to build a propensity score. After adjustment for the propensity score, use of ceftriaxone plus doxycycline remained significantly associated with lower inpatient mortality (OR = 0.26, 95% CI: 0.080.81) and 30‐day mortality (OR = 0.37, 95% CI: 0.170.81). Differences in length of stay and 30‐day readmission rates between the treatment groups were not significant (Table 2).

Subset Analysis

To address issues related to selection bias, we performed an analysis of a subset of the patients after excluding those admitted from a nursing home, diagnosed with aspiration, or admitted to the ICU, for whom ceftriaxone plus doxycycline would not be considered recommended (or first‐line) therapy. The two resulting groups were similar, except there were fewer patients with comorbid illness in the ceftriaxone plus doxycycline group (34% vs. 50%, P = .015). The propensity score was rederived for this subset and used for adjustment. Unadjusted and adjusted outcomes are shown in Table 3. Use of ceftriaxone plus doxycycline in this subset also was associated with reduced odds of inpatient mortality (OR = 0.17, 95% CI: 0.040.77). The odds of 30‐day mortality also were reduced but not significantly, as the confidence interval included 1.0 (OR = 0.43, 95% CI: 0.141.31). There were no differences between groups in length of stay or in 30‐day readmission rate.