Rationale and objectives: To determine the diagnostic quality, performance characteristics and interreader reliability of computed tomography pulmonary angiography (CTPA) and venography (CTV) in intensive care unit (ICU) patients with suspected venous thromboembolism (VTE).
Materials and methods: A total of 100 consecutive ICU patient CT examinations performed for clinically suspected VTE on a four-row CT scanner were reviewed. Three readers rated the diagnostic quality of each CTPA and CTV examination as excellent, acceptable, or nondiagnostic. Readers scored the overall determination for pulmonary embolism (PE) and deep venous thrombosis (DVT) using a 5-point scale, and scored the determination for PE by anatomic level. Receiver operator characteristic (ROC) analysis was performed for each reader and the original clinical report, using consensus interpretation as the reference standard. Interobserver variability for PE and DVT was determined using kappa analysis, and was stratified by examination quality.
Results: A total of 25% of CTPA examinations were nondiagnostic, most commonly because of motion artifact and poor contrast opacification. A total of 24% of CTV examinations were nondiagnostic, most commonly because of poor contrast opacification and metallic hardware. Using receiver operating characteristic analysis, the areas under the curve (Az) for PE diagnosis were 0.875, 0.923, 0.888, and 0.674 for the three readers and clinical reading, respectively, and for DVT diagnosis were 0.842, 0.859, 0.952 and 0.669. Interobserver agreement for detection of PE was moderate at the supralobar level (kappa = 0.55), very good at the lobar level (kappa = 0.69), and moderate for segmental (kappa = 0.54) and subsegmental arteries (kappa = 0.44). Overall reader agreement was good for excellent/good quality CTPA examinations (kappa = 0.52-0.56), and poor when examination quality was poor (kappa = 0.06).
Conclusions: CTPA and CTV are sufficiently accurate and reliable techniques for evaluating VTE in ICU patients, particularly in light of patient complexity.