Summary
We evaluated selection bias in a prospective study of 1162 consecutive patients with suspected pulmonary embolism. Of these, 983 were eligible, and 627 could actually be included. During two months extensive data were collected on all non-included patients. Finally, our patient characteristics were compared with those of the PIOPED study (1990) and the study of Hull et al. (1994). Compared with included patients, the non-included patients had more often non-diagnostic V/Q scans (50% vs. 36%, p <0.01) and were more often already hospitalized (31% vs. 22%, P = 0.04). The subgroup of patients not included due to refusal or inability to give informed consent (IC) was older (mean age 61 vs. 53 years, P <0.01), more often suffered from malignancies (26% vs. 11%, P <0.01) and frequently had non-diagnostic V/Q scans (57%) as compared to included patients. In our study, 54% of all patients screened was eventually included versus 27% in the PIOPED study. In the PIOPED study patients who had contra-indications for pulmonary angiography were excluded, while in the study of Hull et al. those with inadequate cardiorespiratory reserve were excluded. In studies on new diagnostic technologies, patient selection bias does occur. The potential for such a selection bias should be taken into account when diagnostic strategies are devised to improve their generalizability and acceptability.
Keywords
Pulmonary embolism - diagnosis - selection bias - perfusion scintigraphy - spiral CT
