Materials and Methods

Study design. We performed secondary data analysis of the Improving Medicine through Pathway Assessment of Critical Therapy of Hospital-Acquired Pneumonia (IMPACT-HAP) database. The IMPACT-HAP is a retrospective, multicenter, observational study of patients with hospital-acquired pneumonia, health care-associated pneumonia, and VAP who were treated in 4 academic institutions: University of Louisville, Louisville, Kentucky; Ohio State University Medical Center, Columbus; Henry Ford Health System, Detroit, Michigan; and Jackson Memorial Hospital, Miami, Florida. Data were collected from February 2006 through August 2007. Only patients whose cases met the Centers for Disease Control and Prevention's clinical criteria for VAP were analyzed in this study [7]. In each participating health care center, we reviewed nonconsecutive medical records for patients with VAP. Each investigator completed a case report form that was transferred via the Internet to the IMPACT-HAP study center at the University of Louisville. A sample data collection form is available at the study Web site ( http://www.impact-hap.net). Validation of data quality was performed at the study center before the case was entered into the database. The study was approved by the institutional review board at each participating center.

Study variables. We collected data on a total of 60 variables related to the patients' demographic characteristics, comorbidities, and physical examination, laboratory, and chest radiograph findings. Immunocompromise was defined as follows: use of steroids (>10 mg of prednisolone or equivalent for >5 days) or other immunomodulators, active malignancy (i.e., any cancer, except basal or squamous cell cancer of the skin, that was active at the time of diagnosis of VAP or diagnosed ⩽1 year before the current episode of VAP), AIDS, or active chemotherapy or radiotherapy (⩽30 days before the VAP diagnosis). Hypotension was defined as a systolic blood pressure ⩽90 mm Hg or mean arterial blood pressure ⩽70 mm Hg. Thrombocytopenia was defined as a platelet count <100,000/mm³. Multilobar lung involvement was defined as pulmonary parenchymal involvement in >1 lobe noted by chest radiography on the day of VAP diagnosis. The outcome variable for this study was all-cause mortality during the 14-day study period.

Deriving the prediction rule. The prediction rule was derived in 2 steps. First, we performed univariate analysis of all predictor variables, with patient mortality as the independent variable. Second, from the group of variables with P values <.1, the investigators selected a group of 5 variables; the 5 variables needed to be easily obtainable and needed to represent a combination of the patient's history, physical examination findings, laboratory values, and manifestations noted by chest radiography.

Comparison of APACHE II score with the new prediction rule. The APACHE II score was calculated on the day of diagnosis of VAP for each patient. Sensitivity, specificity, positive predictive value, negative predictive value, and the areas under the receiver operating characteristic (ROC) curves (AUCs) for prediction of mortality for the APACHE II score and the new prediction tool were compared. An AUC of 1 was defined as a perfect performance, whereas an AUC of 0.5 was interpreted as an equal probability that the test correctly identified outcome. ROC curves were generated to compare the new prediction rule with the APACHE II score.

Statistical analysis. The relationships between predicted values of the APACHE II score and a new prediction rule with regard to mortality in VAP patients were investigated using 2 propensity-adjusted logistic regression models, 1 for each prediction score [8].

Categorical variables were reported as counts and percentages or examined as predictors using the odds ratio. Variables were tested using the χ²test or, if applicable, Fisher's exact test. Univariate analysis was used to compare variables for survivors versus persons who died. Comparisons were unpaired, and all tests of statistical significance were 2-tailed. Continuous variables were compared using Student's t test for normally distributed variables and the Wilcoxon rank sum test for nonnormally distributed variables. Propensity-adjusted logistic regression analyses were performed for the APACHE II score and the new prediction rule, with mortality used as the outcome [8]. Results were reported as adjusted odds ratios with 95% confidence intervals (CIs). The goodness-of-fit of the final model was examined by the Hosmer-Lemeshow test. All data analyses were performed using SPSS software for Windows, version 16.0 (SPSS), and P values ⩽.05 were considered to be statistically significant.

Results

Study population. A total of 178 patients with VAP were included in the study. The mean age of subjects (± standard deviation) was 55.8±17.8 years (range, 16–93 years), and 116 patients (65%) were male. Table 1presents demographic characteristics, severity of disease, and clinical, laboratory, and radiologic findings for the study population. VAP was microbiologically confirmed in 142 (80%) of 178 patients. The isolated organisms included methicillin-resistant Staphylococcus aureus (67 cases), methicillin-susceptible S. aureus (23 cases), Klebsiella pneumoniae (11 cases), Pseudomonas aeruginosa (20 cases), Enterobacter species (18 cases), Stenotrophomonas maltophilia (5 cases), Serratia marcecsens (3 cases), Acinetobacter baumannii (11 cases), Citrobacter freundii (3 cases), and other organisms (12 cases). Two different pathogens were isolated in 31 patients. We were not able to identify any significant delays in the administration of antibiotics in our population.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 1

Mortality rate for Acute Physiology and Chronic Health Evaluation II (APACHE II) and IBMP-10 scores for ventilator-associated pneumonia. Categorization for APACHE II scores: 0, score of <15; 1, score of 15–19; 2, score of 20–24; 3, score of 25–29; and 4, score of ⩾30. Categorization for IBMP-10 scores: 0, score of 0; 1, score of 1; 2, score of 2; 3, score of 3; and 4, score of 4.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 2

Receiver operating curves for Acute Physiology and Chronic Health Evaluation II (APACHE II) and IBMP-10 scoring systems. The area under the receiver operating curves are as follows: APACHE II, 0.743 (95% CI, 0.628–0.857; P<.001); IBMP-10, 0.808 (95% CI, 0.721–0.895; P<.001).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 3

Propensity-adjusted risk (%) of mortality for ventilator-associated pneumonia, by Acute Physiology and Chronic Health Evaluation II (APACHE II) and IBMP-10 scores. APACHE II: adjusted odds ratio, 1.36; 95% confidence interval (CI), 1.2–1.5; P<.001; P=.919, by Hosmer and Lemeshow goodness-of-fit test. IBMP-10: adjusted odds ratio, 2.4; 95% CI, 1.7–4.1; P<.001; P=.241, by Hosmer and Lemeshow goodness-of-fit test.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1

Characteristics of patients with ventilator-associated pneumonia (VAP), stratified by mortality.

Deriving the prediction rule: the IBMP-10. The following 5 variables were selected to derive the predictive rule: (1) the presence of immunodeficiency; (2) blood pressure <90 mm Hg (systolic) or <60 mm Hg (diastolic); (3) multilobar infiltrates noted on a chest radiograph; (4) platelet count, <100,000 platelets/mm³; and (5) duration of hospitalization before the onset of VAP of >10days. The IBMP-10 score was derived by assigning 1 point to each of the 5 variables. The mortality rate for the study population with an IBMP-10 score ranging from 0 to 5 points and the mortality for APACHE II score are depicted in figure 1.

Comparison of APACHE II score with the IBMP-10. The sensitivity, specificity, and positive and negative predictive values for mortality at different cutoff points for each scoring system are depicted in table 2. Both prediction rules had high negative predictive values but relatively low positive predictive values at most of the cutoff points examined. ROCs for the APACHE II and IBMP-10 scoring systems are shown in figure 2. The AUC for IBMP-10 (0.808; 95% CI, 0.721–0.895; P<.001) was statistically significantly greater than that for the APACHE II score (0.743; 95% CI, 0.628–0.857; P<.001). The relationship between the predicted mortality rates for the APACHE II score and the IBMP-10 score for each variable of interest, as determined with a propensity-adjusted logistic regression model, is depicted in figure 3.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 2

Test characteristics of IBMP-10 and Acute Physiology and Chronic Health Evaluation II (APACHE II) scores with different prediction scores for mortality in patients with ventilator-associated pneumonia.

Discussion

This study indicates that the new IBMP-10 score, which is based on patients' immunosuppression status (I), low blood pressure (B), multilobar lung involvement (M), low platelet count (P), and a duration of hospitalization of >10 days before development of VAP (10), may be used to predict mortality in patients with VAP. In our study population, the IBMP-10 score had higher sensitivity, specificity, and AUC to predict mortality, compared with the APACHE II scoring system.

The APACHE II score has been modified in the attempt to improve the accuracy of predictions of mortality for intensive care unit patients. As a result, the APACHE III and APACHE IV scores were developed to be more accurate mortality predictors, which have become increasingly important in the field of clinical research [9]. However, because the mean time required for data abstraction to calculate the APACHE scores is 30 min, these tools are not viable for use in clinical practice. The IBMP-10 score is a simple-to-calculate alternative developed specifically for predicting mortality in patients with VAP that can be easily incorporated into clinical practice.

Immunosuppression, hypotension, multilobar lung infiltrates, and thrombocytopenia have been previously demonstrated as risk factors for mortality in patients with community-acquired pneumonia and in patients with VAP [10-12]. Studies of the effect of length of stay in the hospital, length of stay in the intensive care unit, or prior days on a ventilator have indicated that patients who developed late-onset VAP are at an increased risk for poor outcomes [1, 13, 14].

For the IBMP-10 scoring system, we included the factor of >10 days of hospital stay on the basis of the significant differences in mortality observed when duration of hospitalization extended beyond 10 days for the patients in our database.

One important weakness of our data is that we evaluated the predictive value of the IBMP-10 score with the same cohort of patients that we used to derive the score. Because of this important limitation, we believe that the IBMP-10 score should be interpreted only as ground work in the field of outcome prediction for patients with VAP. It will be necessary to validate the IBMP-10 with use of larger databases before the score can be used in clinical practice.

In conclusion, a 5-point score, the IBMP-10, has a good discriminatory power to predict mortality in patients with VAP. If future studies validate the IBMP-10 score, physicians may be have a simple tool to evaluate disease severity and to predict outcomes in patients with VAP.

IMPACT-HAP Investigators

Marty Allen, Raul Nakamatsu, Forest Arnold, and Timothy L. Wiemken (University of Louisville, Kentucky); Marcus Zervos and Nadia Haque (Henry Ford Health System); Dan Kett and Ennie Cano (University of Miami, Florida); Julie Mangino, Carol Myers, Lindsay Pell, and David Taylor (Ohio State University, Columbus); and Kimbal Ford and Ernesto Scerpella (Pfizer).