Performance of algorithms and pre-test probability scores is often overlooked in the diagnosis of pulmonary embolismBMJ 2013; 346 doi: http://dx.doi.org/10.1136/bmj.f1557 (Published 19 March 2013) Cite this as: BMJ 2013;346:f1557
- Michael Newnham, respiratory registrar1,
- Helen Stone, respiratory registrar1,
- Ruth Summerfield, radiology consultant1,
- Naveed Mustfa, respiratory consultant1
The mantra that pre-test probability (PTP) scoring should be the first stage in diagnosing pulmonary embolism is supported by a wealth of evidence.1 2 3 Unfortunately, the performance of diagnostic algorithms and PTP scores in clinical settings is often overlooked; when applied incorrectly the predictive potential is negated.
Our retrospective study of 323 computed tomography pulmonary angiograms (CTPAs) at a large teaching hospital found that half (51%) of the Wells PTP scores were inaccurate, significantly more so in the pulmonary embolism negative cohort (P=0.001).4 The original submitted PTP scores did not differ between groups (3.7 v 3.8; P=0.631), but when retrospectively rescored with the correct clinical information, the positive cohort had a significantly higher PTP score (3.1 v 2.0; P=0.001). Although we agree that PTP alone cannot diagnose pulmonary embolism, the positive predictive value improved from 32% to 52% when used accurately.
The finding that clinicians have poor adherence with PTP scoring to meet CTPA requesting criteria is not surprising. However, it is worrying that the predictive potential of PTP is abolished by inaccurate usage. Training and stringent rechecks might improve accuracy, but probably only negligibly. A more radical solution is to dispense with PTP scoring and use initial clinical signs and symptoms together with D-dimer testing, which often acts as the final arbiter of further investigation. The inconvenient truth is that we find it hard to accept there will always be deficiencies in the system. Instead of striving for perfection we should accept and incorporate clinicians’ fallibilities to improve our real world diagnostic algorithms.
Cite this as: BMJ 2013;346:f1557
Competing interests: None declared.