ベイズ解析を用いた診断確率定量法による MRSA 下気道感染症診断の試み

Abstract

Using quantitative Bayesian analysis as a clinical epidemiological approach, we developed a diagnosis for lower respiratory tract infection (LRTI) due to Methicillin-resistant Staphylococcus aureus (MRSA). We retrospectively reviewed the charts of 181 subjects--a derivation cohort-with MRSA retrieved from lower respiratory specimens June 2006 to March 2008. Dividing them into infection or colonization (no infection) groups, we compared them for the presence or absence of clinical parameters, including fever > 38 degrees C, MRSA >106 CFU (colony-forming units)/mL, phagocytosis on Gram staining, serum albumin < 3.0 g/dL, and peripheral WBC count > 15,000/mL. We them determined positive and negative likelihood ratios (LR +, LR -) for these parameters to quantify MRSA-LRTI diagnostic probability based on combined likelihood ratios (Bayesian analysis). We then determined Bayesian MRSA-LRTI diagnostic probabilities (BDPs) in 40 subjects with respiratory MRSA--a validation cohort-from May 2008 to October 2008 clinically judged with either infection (n = 14) or colonization (n = 26) by infection control personnel (ICP) blinded to the test (parameter LR+ and LR -). BDPs (mean +/- SD) quantified by combining the four parameters-fever, MRSA CFU, phagocytosis, and serum albumin-were 62.3 +/- 25.4% for 14 judged with infection, and 40.2% +/- 20.4% for 26 patients judged with colonization (p = 0.005). Using a diagnostic probability of 51% as the cut off, we compared positive and negative predictive Bayesian diagnoses ICP judgment, i.c., 77% vs. 85%. The Bayesian approach proved useful in quantitatively diagnosing infectious disease such as MRSA-LRTI that lack established diagnostic, and may aid physicians in deciding the need for specific antimicrobial therapy.