Online Appendix: Ecologic-level Analysis
Figure . Time-Series Comparing 30-Day Mortality and Administration Rates for GST and Select GST-Based Antibiotic Categories.
Methods
An ecological time-series analysis was conducted in which admission-level variables were aggregated by month of study (72 months). Time-dependent changes in variables were analyzed with regression.1 Changes in monthly means (e.g., age, heart rate) were estimated with linear models. Changes in event rates (e.g. GST administration, 30-day mortality) were estimated with negative binomial models, where regression coefficients (βi) estimated changes in the natural logarithm of event rate per admission. For convenience, observed rates pre- and post-guidelines were tabled as percentages of admissions. For antibiotic administration and clinical outcome rates, segmented models were used to test for changes associated with guideline publication. Each segmented regression analysis started with a full model consisting of a monthly time trend before guideline publication (β1), a change in level starting the month immediately after publication (β2), and a change in time trend after publication (β3).1,2 Backwards selection was used to develop a full, subset, or no model at a 0.10 significance level. If the change in time trend was significant, the full model was reported.
To determine if monthly changes in GST utilization were associated with 30-day mortality, a best model of mortality that included the administration rate of GST as a predictor was selected stepwise by Akaike’s Information Criterion (base model).3 Model selection included all non-antibiotic covariates significant in univariate regression at the 0.20 level. As GST utilization was expected to be associated with a reduction in mortality, the GST significance test was one-sided.
To evaluate the associations between components of GST and the 30-day mortality rate, GST was omitted from the base model and antibiotic component rates were added individually. Associations between the 30-day mortality andantibiotic administration rates were expressed as mortality rate ratios (MRR). MRR was the estimated ratio of the mortality rate with x+10% absolute increase in the administration of an antibiotic regimen to mortality rate with x% administration. All statistical analyses were performed using R version 2.10.0.
Results
The cohort was comprised of 499 and 824 admissions occurring before and after guideline publication, respectively (Table 1). Predominant healthcare exposures included recent prior hospitalization (69.2%), nursing home residence (38.4%), and recent infusion therapy (23.5%); and 27.5% were directly admitted to the ICU. Notable time trends included increasing rates of patients admitted with a primary diagnosis of aspiration, septicemia, or respiratory failure; decreasing rates of nursing home residence and recent infusion therapy; an increasing mean Charlson score; increasing rates of metastatic cancer, prior surgery, tiotropium therapy; and collection of bacterial cultures.
Table 1. Characteristics of HCAP admissions before and after publication of the Guidelines including monthly time trends.
Characteristic / Level* / Monthly Time Trend†Before / After / Change / / P
Demographics
Age (y) / 72.7 / 71.5 / -1.7 / — / —
Gender (% Male) / 98.3 / 96.8 / -1.6 / — / —
Primary Diagnosis
Pneumonia (%) / 73.5 / 63.7 / -13.4 / -0.0042 / 0.01
Aspiration with pneumonia (%) / 22.7 / 28.2 / 24.3 / 0.0055 / 0.04
Septicemia with pneumonia (%) / 2.3 / 4.7 / 107.1 / 0.0197 / <0.01
Respiratory failure with pneumonia (%) / 1.9 / 3.4 / 83.5 / 0.0212 / 0.01
HCAP Criteria
Nursing Home residence (%) / 42.7 / 35.9 / -15.8 / -0.0037 / 0.09
Hospitalization of 2 days within 90 days (%) / 68.1 / 69.8 / 2.4 / — / —
IV therapy within 30-days (%) / 27.1 / 21.5 / -20.7 / -0.0088 / <0.01
Wound care in last 30-days (%) / 2.1 / 3.9 / 87.9 / — / —
Chronic dialysis within 30-days / 2.7 / 2.7 / 0.7 / — / —
Other Guideline MDR Criteria
Antibiotics within 90 days (%) / 60.8 / 65.0 / 6.9 / — / —
Recent immunosuppression (%) / 18.3 / 21.0 / 14.5 / — / —
Cultures Collected Upon Admission
MRSA Surveillance nares culture (%) / 0.2 / 24.7 / 11743 / 0.1321 / <0.01
Blood cultures (%) / 76.3 / 84.9 / 11.4 / 0.0026 / 0.08
Sputum culture (%) / 41.0 / 47.6 / 15.9 / 0.0045 / 0.02
Bronchoscopy culture (%) / 1.3 / 1.8 / 42.4 / — / —
Acute Measures of Illness
Heart rate (BPM) / 90.4 / 90.0 / -0.5 / — / —
Temperature (F) / 98.6 / 98.6 / 0.0 / — / —
Systolic Blood Pressure (mm Hg) / 124.5 / 124.9 / 0.3 / — / —
Respiratory rate (RPM) / 21.5 / 21.1 / -2.1 / -0.0174 / 0.03
Sodium (meq/dl) / 135.0 / 135.9 / 0.7 / 0.0152 / 0.09
Blood Urea Nitrogen (mg/dl) / 28.6 / 28.4 / -0.7 / — / —
White Blood Cell count (cells/mm3) / 14.3 / 13.3 / -7.0 / -0.0316 / 0.05
Hematocrit (%) / 35.5 / 35.6 / 0.3 / — / —
Admitted to Intensive Care Unit (%) / 26.5 / 28.1 / 6.3 / — / —
Mechanical Ventilation (%) / 6.3 / 8.2 / 31.0 / — / —
Chronic Measures of Illness
Charlson Comorbidity Score / 3.9 / 4.5 / 15.8 / 0.0156 / <0.01
Metastatic Cancer (%) / 9.4 / 13.8 / 46.8 / 0.0084 / 0.05
Stroke (%) / 5.8 / 3.9 / -32.9 / — / —
Diabetes (%) / 32.9 / 33.1 / 0.5 / — / —
Congestive Heart Failure (%) / 30.6 / 27.6 / -9.7 / — / —
Chronic Obstructive Lung Disease (%) / 53.3 / 62.6 / 17.4 / — / —
> 1 AECOPD within last 365 days (%) / 9.2 / 11.0 / 20.3 / — / —
Any Prior Surgery (%) / 19.0 / 30.4 / 60.2 / 0.0163 / <0.01
Proton Pump Inhibitor therapy (%) / 40.6 / 39.0 / -3.9 / — / —
Tiotroprium therapy (%) / 0.4 / 9.4 / 2149 / 0.0418 / <0.01
Note:* Level is mean value for a characteristic measured on a continuous scale and is event rate per 100 admissions (%) for a nominal characteristic. Before and After are levels of a characteristic before and after publication of the HCAP guidelines. Change (%) is relative change.† Time trend () is the estimated change per month in mean for a characteristic measured on a continuous scale and natural logarithm of event rate per admission for a nominal characteristic.
The percentage of admissions that received GST increased throughout the study ( = 0.0259, P < 0.01)(Table 2). This was primarily due to increased administration of vancomycin ( = 0.0252, P < 0.01), and secondarily, increased administration of piperacillin/tazobactam ( = 0.0094, P < 0.01). Category B antibiotic utilization was decreasing prior to guideline publication but lacked trend after guideline publication (post-guideline slope = 0.0086, P = 0.28). The change coincided with the VA preferred fluoroquinolone formulary change from levofloxacin to moxifloxacin in 2004. Anti-pseudomonal double coverage was uncommon; however consistent with Guideline recommendations, administration of double-coverage in conjunction with anti-MRSA therapy increased throughout the study.
Table 2. Monthly antibiotic utilization expressed as a percentage of admissions and segmented regression analysis before and after publication of the Guidelines.
Antibiotic Regimen / Administration Rate (%) / Slope BeforeGuidelines / Change in Level / Change in Slope
Before / After / Change / / P / / P / / P
Guideline-similar Therapy (Categories [A or B] + C) / 11.9 / 35.6 / 199.7 / 0.0259* / <0.01 / — / — / — / —
Anti-pseudomonal β-lactams
(Category A) / 44.0 / 51.0 / 16.6 / 0.0061* / <0.01 / — / — / — / —
Piperacillin/Tazobactam / 31.7 / 41.3 / 30.4 / 0.0094* / <0.01 / — / — / — / —
Ceftazidime/Cefepime / 10.4 / 6.8 / -35.1 / — / — / -0.4321 / 0.03 / — / —
Imipenem/Meropenem / 4.2 / 5.0 / 19.6 / — / — / — / --- / — / —
Aztreonam / 0.2 / 0.7 / 241.6 / 0.0445* / 0.05 / — / --- / — / —
Other anti-pseudomonal antibiotics(Category B) / 21.5 / 10.4 / -51.4 / -0.0482 / <0.01 / -0.2897 / 0.35 / 0.0568 / <0.01
Ciprofloxacin/Levofloxacin / 19.0 / 10.0 / -47.4 / -0.0515 / <0.01 / -0.1820 / 0.58 / 0.0610 / <0.01
Gentamicin/Tobramycin / 2.5 / 0.5 / -81.0 / — / — / -1.6618 / <0.01 / — / —
Anti-MRSA antibiotics
(Category C) / 15.0 / 41.8 / 178.4 / 0.0252* / <0.01 / — / — / — / —
Vancomycin / 14.6 / 40.9 / 180.6 / 0.0252* / <0.01 / — / — / — / —
Linezolid / 0.4 / 0.9 / 127.8 / — / — / — / — / — / —
Any anti-pseudomonal coverage(Category A or B ) / 56.0 / 54.1 / -3.4 / 0.0064* / 0.05 / -0.2619 / 0.06 / — / —
Anti-pseudomonal double coverage(Category A+B) / 6.9 / 1.2 / -82.7 / — / — / -1.7616 / <0.01 / — / —
Anti-pseudomonal β-lactam + anti-MRSA antibiotic
(Category A+C) / 8.1 / 27.8 / 241.6 / 0.0293* / <0.01 / — / — / — / —
Other anti-pseudomonal antibiotic + Anti-MRSA antibiotic(Category B+C) / 1.5 / 1.4 / -2.4 / -0.1756 / 0.02 / 2.4221 / 0.14 / 0.1932 / 0.01
Guideline recommended antibiotics(Category A+B+C) / 2.3 / 6.2 / 169.2 / 0.0214* / <0.01 / — / — / — / —
Other Non-HCAP Antibiotics / 40.6 / 39.6 / -2.5 / -0.0049 / 0.59 / 0.2575 / 0.12 / -0.0204 / 0.04
Note: Before (%) and After (%) are rates of antibiotic regimen administration to HCAP admissions before and after publication of the HCAP guidelines. Change (%) is relative change. For each antibiotic regimen, Slope Before Guidelines (), Change in Level (), and Change in Slope () are estimated regression coefficients from a negative binomial generalized linear model of administration rate. These coefficients describe changes in the natural logarithm of administration rate per admission over time. If the interaction was significant, then the full model was reported. * Estimated slope was for the entire study period, i.e, Slope Before Guidelines was unchanged after guideline publication.
Time-dependent changes in clinical outcome were evident. The 30-day mortality rate decreased throughout the study ( = -0.0058, P = 0.07) from 19.2% prior to guideline release to 16.5% post publication.
Eleven aggregated non-antibiotic covariates were associated with the 30-day mortality rate at the 0.20 significance level: concurrent acute respiratory failure; heart rate, systolic blood pressure, blood urea nitrogen (BUN) upon admission; > 1 acute exacerbation of chronic obstructive pulmonary disease or prior surgery within the past year; use of tiotroprium or proton pump inhibitor medications; WBC <2,500 cells/mm3 30-days prior to admission;blood and nasal MRSA surveillance culture rates upon admission. The base modelof the 30-day mortality rate included five predictors (Table 3). GST administration rate was associated with decreased 30-day mortality rate with and without adjustment for non-antibiotic covariates (unadjusted = -0.0089, P = 0.02; adjusted = -0.0092, P = 0.02).
Table 3: Regression model of 30-day mortality rate for HCAP admissions.
Predictors / / SE() / 95% CI / P-value(Intercept) / -1.7918 / 0.0689 / (-1.9268, -1.6568) / <0.01
Monthly % of admissions with acute respiratory failure / 0.0461 / 0.0188 / (0.0092, 0.0830) / 0.01
Monthly % of admissions with BUN > 30 mg/dl / 0.0103 / 0.0064 / (-0.0022, 0.0228) / 0.10
Monthly average HR upon admission / 0.0272 / 0.0159 / (-0.0040, 0.0584) / 0.09
Monthly % admissions with WBC <2,500 cells/mm.3 / 0.0426 / 0.0167 / (0.0099, 0.0753) / 0.01
Monthly % guideline-similar therapy (GST) / -0.0092 / 0.0046 / ( -0.0017) / 0.02
Note: Monthly counts of 30-day mortality were modeled with a negative binomial generalized linear model. is estimated change per month in the natural logarithm of 30-day mortality rate per admission for a one unit change in predictor value (e.g, +1% GST). Predictors were centered so that exponentiation of the intercept is an estimate of mortality rate conditional on predictors at their study mean.BUN – blood urea nitrogen, HR – heart rate, WBC – white blood cell count.
The MRR for GST was 0.912, or equivalently, an 8.8% relative reduction in the 30-day mortality rate for a 10% absolute increase in monthly utilization of GST (Table 4). Utilization of anti-pseudomonal or anti-MRSA therapy alone was not associated with the 30-day mortality rate. However, the combination of an anti-pseudomonal β-lactam and anti-MRSA therapy was associated with a reduction in the 30-day mortality rate (MRR = 0.903; P = 0.02). Utilization of non-GST antibiotics was not associated with a reduction in the 30-day mortality rate.
1
Table 4. Association between administration rate for antibiotic regimens and the 30-day mortality rate including estimated 30-day mortality rate ratios (MRRs) for a hypothesized 10% absolute increase in antibiotic administration rate.
Antibiotic Regimen / / P-value / MRR / MRR UCL / RelativeChange (%) / Relative Change (%) UCL
Guideline-similar Therapy (GST) / -0.0092 / 0.02 / 0.912 / 0.983 / -8.8 / -1.7
Anti-pseudomonal β-lactam (Cat. A) / -0.0042 / 0.22 / 0.959 / 1.049 / -4.1 / 4.9
Anti-pseudomonal FQ or AG (Cat. B) / 0.0056 / 0.84 / 1.058 / 1.157 / 5.8 / 15.7
Anti-MRSA antibiotics (Cat. C) / -0.0061 / 0.07 / 0.940 / 1.006 / -6.0 / 0.6
Anti-pseudomonal β-lactam + Anti-MRSA antibiotic (Cat. A+C) / -0.0102 / 0.02 / 0.903 / 0.980 / -9.7 / -2.0
Guideline recommended antibiotics (Cat. A+B+C) / -0.0110 / 0.12 / 0.819 / 1.075 / -18.1 / 7.5
Other non-GST antibiotics (Not Cat. A, B, or C) / -0.0006 / 0.46 / 0.994 / 1.087 / -0.6 / 8.7
Note: Each row summarizes a separate model of the association between the 30-day mortality rate and the administration rate of an antibiotic regimen while adjusting for the fivenon-antibiotic covariates in the base model (Table 3). – estimated regression coefficient from a negative binomial generalized linear model of mortality rate per admission, P – Pr(β0). MRR was the ratio of mortality rate with x+Δ% administration of an antibiotic regimen to mortality rate with x% administration. The estimated MRR for a Δ=10% absolute increase in administration of an antibiotic regimen was . UCL – 95% upper confidence limit, Relative Change – relative percent change in mortality rate.
Supplement Bibliography
1.Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time series studies in medication use research. J Clin Pharm Ther. 2002;27(4):299-309.
2.Madaras-Kelly KJ, Remington RE, Lewis PG, Stevens DL. Evaluation of an intervention designed to decrease the rate of nosocomial methicillin-resistant Staphylococcus aureus infection by encouraging decreased fluoroquinolone use. Infect Control Hosp Epidemiol. 2006;27(2):155-169.
3. Akaike H. A new look at the statistical model identification. IEEE Trans Automat Contr. 1974;19(6):716–723.
1