Frequency (percentage) were recorded for categorical variables. Inorder to explore the independent nature of the variables, ?2 were used.
To estimatethe predictive model that will allow us to differentiate between Candidaspecies and proven candidal infection, the crude odds ratio (OR) for each riskfactor associated with proven candidal infection was estimated. Statisticalsignificance was accepted at the 5% level. Statistically significant variablesin the univariate analysis were included in the model, and through a stepwiseelimination process, the so-called “Modified Candida score” was obtained usinglogistic regression. The discriminatory power of this score was evaluated bythe area under the receiver operating characteristics (ROC) curve and the 95%confidence interval (CI). Then, a cut-off specificity in the validation set wasselected.
The results were analyzed using SPSS 14.0 software (SPSS, Chicago,IL, USA) and SAS version 9.2 statistical software (SAS Institute, Inc). ResultsOf the 750 patients, 85 (113%) patients werediagnosed with invasive candidiasis (IC). There were 64 (75.
3%) men and 21(24.7%) women in this group, with a mean age of 58.5 (±16.9) years.
The frequencies of the patients based on the risk factors,were recorded in the Table 1. As the table1 shows, rates of patients with risk factors are higher than indiagnosed IC patients except of diabetes mellitus, hemodialysis and mechanical ventilation. There was no statisticallysignificant difference concerning the corticosteroids consumption,immunosuppressive therapy, chronic renal failure, diabetes mellitus,hemodialysis and mechanical ventilation between two groups.In the logit model adjusted for possible confoundingvariables, surgery on ICU admission, total parenteral nutrition, ICU stay>7 days, broad antibiotic therapy, pancreatitis, central venous catheter, and severesepsis were independently associated with a greater risk for proven candidalinfection (Table 2). Through a stepwise elimination process, the Candidascore was obtained (Table 3). The discriminatory power of this score,assessed by the area under the ROC curve and its main cut-off values, is shownin Figure 1.
Area under curve (AUC) was estimated 0.86 and sensitivityand specificity in cutoff point 4 are suitable, 0.83 and 0.80 respectively.
Table4 shows the result of the comparison of sensitivity and specificity basedon performance of Candida score and Modified Candida score systemsaccording to test introduced by Newcombe (16).There is no significant difference between sensitivity systems (p=0.234) but in contrast, specificity modified candida score wassignificant higher than specificity candida score (p=0.025).