Prediction of Staphylococcal Etiology among Patients with Septicemia with or without Endocarditis by Multivariate Statistical Methods

Abstract

Multivariate statistical methods, multiple regression (RA) and automatic interaction detector analysis (AID) were used to study the possibility of an early prediction of Staphylococcal etiology in 249 of 851 patients with verified septicemia or endocarditis. The variables included pertinent symptoms and signs and laboratory data available soon after admission. 10 of the 70 variables initially studied showed simple, or in various combinations, a statistically significant partial correlation to Staphylococcal etiology in the AID. The highest predictive value with a high probability for Staphylococcal etiology was recorded for combinations of the variables: i.v. narcotic addiction and septic pulmonary embolism; non-addiction, wound infection, and hospitalization within 4 weeks; non-addiction, absence of skin infection, presence of foreign body, and age < 60 yr. Staphylococcal etiology was contradicted by the absence of i.v. narcotic addiction, skin infection, foreign body, septic skin manifestation, surgical procedure within 4 weeks, joint symptom and a C-reactive protein ≤10 mm. Thus, a prediction of etiology may be valuable in choosing therapy before definite confirmation by positive blood cultures or when blood cultures remain sterile.