Statistics from Altmetric.com
We must express serious concerns about the internal validity and conclusions of the recent paper by Egermayer et al 1 in which the authors suggest that normal results of d-dimer, arterial blood gas tensions, and respiratory rate measurements can be used to rule out pulmonary embolism. With respect to blood gases, two earlier well designed studies reported that the Pao 2 and Paco 2, alone or in combination, did not exclude pulmonary embolism.2 3 If a low Paco 2 is taken as a reasonable surrogate for tachypnoea, these studies directly contradict Egermayer’s findings. We attribute this discrepancy to a serious flaw in study design.
In any valid evaluation of the accuracy of a diagnostic test, comparison must be made with an appropriate reference standard.4 Being able to conclude that any test can exclude pulmonary embolism, as the authors have done, mandates that the selected reference standard accurately and objectively rules out pulmonary embolism in all patients who truly do not have it and confirms the diagnosis of pulmonary embolism in all those who truly do. What was the reference standard in this case? The authors apply strict criteria for establishing an objective diagnosis of pulmonary embolism but make no attempt to rule it out with any degree of objectivity. It is clear that their composite reference standard does not divide patients into those with and without pulmonary embolism, but into those who meet the authors’ criteria for “objective pulmonary embolism” and everyone else. In only 214 of the 507 patients with suspected pulmonary embolism (154 with normal lung scans, 36 with high probability scans, and 24 with pulmonary angiograms) was a diagnosis made; in the remaining 58% of patients pulmonary embolism was neither proven nor excluded. Furthermore, among the 27 patients who died within 10 days of evaluation, only five had a necroscopic examination. The cause of death and, specifically, the possibility of fatal pulmonary embolism in the remaining 22 is unknown.
Among the 317 patients with non-diagnostic scans there were 135 indeterminate scan results and 182 low probability scans. In the PIOPED study5 pulmonary embolism was present in 29% and 12% of such patients, respectively. If we assume that the percentages here are similar, 61 of these patients would be expected to have pulmonary embolism. Yet from the results and conclusions published it is apparent that this entire group was designated as having had pulmonary embolism excluded!
The results as presented are highly misleading and cannot justify the conclusions. It is impossible to determine whether a test excludes pulmonary embolism when the reference standard to which it is being compared does not itself exclude pulmonary embolism. Unfortunately, by neglecting this fundamental aspect of study design the authors have invalidated their findings. All that has been shown is that patients with normal d-dimer levels, blood gas tensions, or respiratory rates are unlikely to have the combination of a high probability lung scan together with a high clinical suspicion of pulmonary embolism. Patients with “negative” test results still have a significant probability of pulmonary embolism and it is a dangerous mistake to think otherwise.
authors’ reply Thank you for the opportunity to respond to the concerns raised by Stanbrook and Geerts regarding our recent study.1-1 Two retrospective studies have been cited to justify the conclusion that a Pao 2 of >80 mm Hg (10.6 kPa) does not exclude pulmonary embolism.1-2 1-3 Neither was “well designed” from the point of view of answering the questions we were addressing since neither gave any information concerning the proportion of patients referred for pulmonary angiography who actually had their blood gas tensions measured. In our institution this figure would be approximately 50% (excluding those who were enrolled in our prospective trial). We did not conclude that normoxaemia excludes pulmonary embolism since in our study four of 34 patients (12%) with objectively diagnosed pulmonary embolism had a Pao 2 of >80 mm Hg. Allowing for the different methodology, this is similar to the findings of the two studies quoted earlier. It would indeed be surprising if subsegmental pulmonary embolisms caused significant hypoxaemia. What we do conclude from our study and from previous studies is that the finding of normoxaemia significantly reduces the likelihood of clinically important pulmonary embolism.
Contrary to what is stated, hypocarbia is not a reasonable surrogate for tachypnoea since in our study over 30% of patients with a respiratory rate of ⩾20 breaths/min had an arterial Paco 2 of >35 mm Hg (authors’ unpublished data). Stanbrook and Geerts may be confusing tachypnoea with hyperventilation, which implies a high minute volume. We have not drawn any conclusions about the usefulness of normocarbia or of the arterial–alveolar gradient for excluding pulmonary embolism. As far as we are aware, our study is the first to evaluate prospectively the hypothesis that a normal respiratory rate excludes pulmonary embolism. We have found that this simple but useful observation is also often overlooked by doctors who wrongly assume that pulmonary angiography or a lung scan will provide a definitive diagnosis in all cases.1-4
It is suggested that a reference standard with 100% sensitivity and specificity is required to evaluate diagnostic tests properly. Such a standard does not exist in the area of venous thromboembolism. For example, pulmonary angiography for pulmonary embolism falls far short of this standard due to technical limitations and interobserver disagreement in nearly 40% of cases involving smaller emboli.1-5
We were interested in assessing the usefulness of various observations for predicting the absence of “objectively diagnosed” pulmonary embolism (according to our predetermined criteria).1-1 The most important calculation for this purpose is the proportion of correct exclusions—that is, the predictive value—which is defined as true positives/true positives + false positives. This calculation does not require the accurate identification of true negatives. We were already aware from previous studies that the d-dimer test was likely to have a very poor specificity for diagnosing pulmonary embolism and did not consider it worthwhile to demonstrate this further.
Drs Stanbrook and Geerts are concerned about the possibility of unrecognised pulmonary embolism among the 22 patients who died and did not have a necropsy. They also rightly point out that inadequate investigation of many of the patients with intermediate probability ventilation perfusion lung scans undoubtedly led to cases of pulmonary embolism remaining undiagnosed. However, it would be unrealistic to assume that more aggressive investigations would detect most cases of major pulmonary embolism since these are often asymptomatic.1-6 Nevertheless, subsequent analysis of outcomes over two years in untreated patients with pulmonary embolism showed an excellent prognosis even without treatment.1-7They overlook the more obvious problem of false positive diagnoses of pulmonary embolism in two of the five patients who did have a necropsy. Anticoagulant treatment was the direct cause of death in one of these. A previous study conducted at a different New Zealand hospital showed a similar false positive rate for diagnosis of pulmonary embolism of nearly 50% among patients who underwent a necroscopic examination following a perfusion lung scan.1-4 In addition to the 40 cases of objectively diagnosed pulmonary embolism in our study, there were a further 68 patients who received a diagnosis of pulmonary embolism without adequate supporting evidence or, in many cases, despite evidence to the contrary such as a normal lung scan or normal pulmonary angiogram. Of this group of 68 patients 19 (28%) had a negative d-dimer test, the result of which was not known to the physicians responsible for the care of the patient. It is possible that greater utilisation of tests which help to exclude pulmonary embolism could reduce the dangers of misdiagnosis and inappropriate treatment.
Excluding venous thromboembolism is a concept that many clinicians find difficult. The impulse is to continue searching until some evidence of thrombosis is found to justify the use of anticoagulant therapy. It is, of course, impossible to prove that a patient does not have venous thromboembolism. The best that can be aimed for is to reach a point where it is considered no longer profitable to continue the search. There is really only one possible methodology to achieve this purpose: one must first carefully define what one is looking for and then prospectively search a large series of cases to see whether the entity exists. We identified 93 consecutive patients with a negative SimpliRED test and Pao 2 of >80 mm Hg and did not find any with objectively diagnosed pulmonary embolism. We concluded that this combination of findings excluded objectively diagnosed pulmonary embolism with a very high level of confidence.
Whether or not it is “dangerous” to withhold anticoagulant therapy in patients with negative test results remains to be determined. Further prospective studies with analysis of clinical outcomes are being planned to investigate this question.1-8