ABSTRACT
To assess interobserver variability of venography for screening for postoperative deep venous thrombosis (DVT), we used 185 bilateral ascending contrast venograms (366 lower limbs) which were performed using the long-leg film technique, in the frame of a trial of the efficacy of two low-molecular-weight-heparin fractions (initial evaluation). These venograms were submitted in a multicenter setting to three further readers who performed a serial scoring as DVT, no DVT or non-evaluable. DVTs were diagnosed in 78, 55 and 59/366 limbs (initial evaluation: 58). The proximal locations of DVTs were 16, 9 and 16 (initial evaluation: 15) and the non evaluable limbs 3, 5 and 18 (initial evaluation: 0). Apparent pairwise agreement between the three readers ranged from 87 to 90% (true coefficient of agreement Kappa 0.63-0.70). It ranged from 89 to 93% (Kappa 0.63-0.74) for the comparison between the readers and the initial evaluation. This considerable degree of disagreement among experienced readers should be taken into account in calculating sample sizes in prospective thromboprophylactic studies. Indeed, one of the two prophylactic regimens proved to be significantly (p = 0.012, p = 0.031, p = 0.049) or non-significantly (p = 0.073) superior to the other one depending upon the reading of venograms.
