Two-way analysis for detecting factors affecting ventilator-associated pneumonia.

ABSTRACT

The "clinically required ventilation period" for assessing ventilator-associated pneumonia (VAP) has not been studied because this period could not be clinically predicted. We addressed this problem using both rate analysis and failure-time analysis. A total of 325 patients who had received mechanical ventilatory support in the intensive care unit of a university hospital were reviewed. The total ventilation period and the ventilation period before VAP were compared using logistic regression and the Cox proportional hazard model for univariate and multivariate analyses. The Frechet distribution model was also used. Fifty patients were excluded for having pneumonia before intubation or for being admitted to a department in which no VAP occurred; 12 patients had VAP. Discrepancies in both methods caused by time-dependent bias were observed in patients emergently admitted (odds ratio, 1.435; hazard ratio, 0.3928). This reduced hazard ratio remained with the multivariate Frechet distribution model. Longer operation time significantly increased the VAP rate in the logistic model only. Low body mass index increased the rate of VAP in both models, especially in female patients (hazard ratio, 0.1707; 95% confidence interval, 0.02105-0.6728). The results of rate analysis and failure-time analysis were similar for most factors but differed somewhat for several factors, such as emergency admission. Unknown factors might be obscured by this type of difference, and this two-way method might be able to reveal artificial effects.