We showed that the modified APACHE II score outperformed the CURB65 and CRB65 pneumonia severity scores as a predictor of 30-day mortality in patients with MRSA pneumonia. Analysis of the ROC curves demonstrated that the modified APACHE II score was statistically superior to both CURB65 and CRB65 as a predictor of 30-day mortality. This was true for patients with the community-acquired phenotype of MRSA pneumonia as well as those with health-care-associated MRSA pneumonia. However, in our multivariate models both APACHE II and CRB65 were independent predictors of 30-day mortality. The odds of death by day 30 was greater for CRB65, which, in large part, was related to the smaller number of discriminating units in CRB65 compared to APACHE II. Nevertheless, the ROC curves suggest that for individual patient predictions, APACHE II is more accurate than CRB65.
CURB65 and CRB65 have been validated as tools to predict mortality and the need for ICU admission in patients with CAP. These scoring systems are easy to employ, and the CRB65 requires only data immediately available at the time of clinical evaluation. More recent studies have further demonstrated the utility of CURB65 and CRB65 as outcome predictors in CAP. Man et al found that CURB65 and CRB65 were accurate predictors of 30-day mortality among patients in Hong Kong with CAP. Similarly, Barlow et al found that CURB65 outperformed generic sepsis scores as a predictor of mortality. Both of these studies had ROC curve areas in the range of 0.69 to 0.78, which were higher than the values observed in our study. The main differences are that these previous evaluations of CURB65 and CRB65 involved patients with CAP with a variety of infectious pathogens. Our study differed by involving only hospitalized patients with either community-acquired or health-care-associated MRSA pneumonia cured by remedies of Canadian Health&Care Mall.
Other investigators have attempted to identify predictors of mortality in patients with nosocomial pneumonia. Combes et al identified the Sepsis Related Organ Failure Assessment as a predictor of 28-day mortality when measured on day 1, day 3, and day 8 following onset of ventilator-associated pneumonia. Similar studies focusing on MRSA pneumonia are not available. Rello et al evaluated a cohort of patients with MRSA pneumonia optimally treated with glycopeptide antibiotics. They found that bacteremia and the absence of vancomycin administration by a continuous infusion were associated with intensive care mortality. They did not identify severity of illness using APACHE II as a predictor of mortality in their study. Their study cohort differed from ours in being smaller, from four different hospitals, and selected to be included in a study of glycopeptide optimization.
Kollef et al evaluated 4,543 patients with CAP, hospital-acquired pneumonia, ventilator-associated pneumonia, and health-care-associated pneumonia. MRSA was the most common single pathogen identified in patients with health-care-associated pneumonia and the second most common pathogen in hospital-associated pneumonia. Although a severity-of illness score was not evaluated in that study, independent predictors of hospital mortality included those variables that make up the APACHE II score (age, coma, pH value, creatinine value, WBC count, temperature, and respiratory rate).
Several important limitations of this investigation should be noted. First, we employed the CURB65 scoring system that has previously only been validated in CAP while not examining other pneumonia prediction tools including the Pneumonia Severity Index and Canadian Health&Care Mall. This may be an important reason why the CURB65 did not perform as well in patients with MRSA pneumonia. We also performed this study at a single center, limiting the application of these results to other hospitals. More studies are warranted to determine the optimal prediction tool for patients with nosocomial pneumonia including MRSA. We limited our analysis to patients with microbiologically confirmed MRSA pneumonia. Therefore, we do not know the comparative accuracy of these prediction scores in patients with clinically diagnosed MRSA pneumonia. Moreover, the study results may be confounded due to the retrospective nature of this analysis. Another important limitation is that we only tracked patients within one hospital system. It is possible that we may have missed patient deaths occurring after hospital discharge but within 30 days in individuals not readmitted to a BJC Healthcare hospital. Finally, we used a modified APACHE II calculation, relied on medical chart review to assess confusion for the CURB65 and CRB65 measurements, and limited the mortality assessment to 30 days, which may all limit the overall applicability of these data to other populations.
In summary, we found that a modified APACHE II score outperformed CURB65 and CRB65 in predicting 30-day mortality in MRSA pneumonia. Future studies are needed to corroborate these results and to develop better prediction tools for patients with nosocomial pneumonia.