The diagnosis of pulmonary embolism is often missed, so it is important to request the appropriate diagnostic tests. Antonio Vaz Carneiro takes us through this systematic review, which compares diagnostic tests for pulmonary embolism

This month's paper is Roy PM, Colombet E, Durieux P, Chattelier G, Sors H, Meyer G. Systematic review and meta-analysis of strategies for the diagnosis of suspected pulmonary embolism. BMJ 2005;331:259-63. You can read it by going to studentbmj.com and clicking on the link.

Why do the study?
Pulmonary embolism is a serious complication of deep vein thrombosis with a 30% mortality rate, mostly from recurrent embolism. Emboli normally arise in the deep venous system of the lower extremities and in right heart cavities. The venous bed in the kidney, pelvic, and upper extremities can also be a source of emboli.

Clinical presentation is variable, and it is believed that the diagnosis is missed in about 50% of patients. Treatment reduces long term mortality to less than 8%, and it is therefore crucial to achieve a correct diagnosis.

Several diagnostic tests are available, some of which have been assessed and shown variable discriminatory power. Given the high incidence of pulmonary embolism, the judicious use of diagnostic tests is essential. Doctors should be aware of the diagnostic properties of the available tests in clinical practice.

The authors carried out this systematic review to assess the likelihood ratios for several tests for suspected pulmonary embolism, including ventilation perfusion lung scan, spiral computed tomography, ultrasonography of leg veins, d-dimer concentration in the blood, magnetic resonance angiography, and echocardiography. They also estimated the range of pretest probabilities over which each test is capable of confirming or excluding pulmonary embolism (pretest probability is the probability of the target condition being present before the results of a diagnostic test are available; it is the same as the prevalence of the disease).

This systematic review is a good example of secondary research (as opposed to primary research — that is, with patients). Secondary research uses exactly the same methods as a primary study — introduction, methods, results, and conclusions — but the subjects are the primary studies themselves (for example, randomised clinical trials, diagnostic or prognostic studies).

What is the study design?
This is a systematic review with meta-analysis of diagnostic studies confirming or excluding pulmonary embolism, using likelihood ratios (see below). A systematic review is a critical assessment and evaluation of research that attempts to address a focused clinical question using methods designed to reduce the likelihood of bias (that is, systematic error). It differs from a classical review article (also called narrative), normally done by one or more authors, in several ways:

1.  It answers a focused question, not a general one — for example, “Long term ACE inhibitor therapy in patients with heart failure or left ventricular dysfunction” and not “Treatment of heart failure”
2. Their sources are pre-specified and the evidence is selected by an explicit, criteria based, and thorough strategy — for example, the authors state, “We searched Medline from 1980-2005, EMBASE from 1990-2005, reviewed bibliography from review articles” using the keywords “heart failure, systolic dysfunction, ACE inhibitors”
3. Each article is critically appraised using questions such as, “Were the criteria for inclusion (or exclusion) of the studies appropriate?” or, “Were all the clinical relevant outcomes considered?”
4. If possible, its results are combined through further statistical analysis to produce a quantitative summary or overview, called meta-analysis. This allows a generalisation of the results — that is, the bigger the final sample of patients in the systematic review (by combining samples from individual studies) with consistent results, the more one believes, for example, in the efficacy of the intervention, diagnostic test, or prognostic estimation.

In the present study, the inclusion criteria for the selected studies (from several well known sources) were the use of pulmonary angiography as a gold standard, a prospective design, consecutive recruitment of patients, and independent evaluation of diagnostic tests in each patient. Each study was analysed and then graded according to the reference method and the patients' characteristics, and likelihood ratios were calculated. This is a classic method of doing a systematic review.

The likelihood ratio of a diagnostic test expresses the relative likelihood that a given test would be expected in a patient with (as opposed to one without) the specific disorder. It can either be positive for strategies confirming the diagnosis of pulmonary embolism, or negative for strategies that exclude it.

What did the researchers find?
From a total of 1012 selected articles, only 48 were included for final analysis, representing a total of 11 004 patients (with a 30% (n=3329) prevalence of pulmonary embolism in this sample). Table 1 shows the positive likelihood ratios for diagnostic tests. Positive likelihood ratios above 5 are considered strong diagnostic evidence and above 10 convincing diagnostic evidence; the same reasoning applies to negative likelihood ratios below 0.2 (strong evidence) and 0.1 (convincing evidence).

Table 1 Positive likelihood ratios for diagnostic tests (95%CI)

Spiral computed tomography	24.1 (12.4 to 46.7)
High probability ventilation perfusion lung scan	18.3 (10.3 to 32.5)
Ultrasonography of leg veins	16.2 (5.6 to 46.7)

Table 2 Negative likelihood ratios for diagnostic tests (95% CI)

Negative result on spiral computed tomography with negative result on ultrasonography	0.04 (0.03 to 0.06)
Normal or near normal appearance on lung scan	0.05 (0.03 to 0.10)
A d-dimer concentration <500 µg/l	0.08 (0.04 to 0.18)

In patients with a moderate (±35%) or high (±70%) pretest probability (the probability of the target condition being present before the results of a diagnostic test are available; also called prevalence of disease), these findings were associated with a post-test probability (the probability of the target condition being present after the results of a diagnostic test are available) of pulmonary embolism greater than 85% — a positive diagnosis by the author's definition. In other words, the higher the prevalence, the better the diagnostic test confirms the disease.

Table 2 shows negative likelihood ratios for diagnostic tests. In patients with a low (±10%) or moderate pretest probability, these findings were associated with a post-test probability of pulmonary embolism below 5% — a negative diagnosis. In other words, the lower the prevalence, the better the diagnostic test excludes the disease.

It is important to remember that most people with pulmonary embolism do not have the classic signs and symptoms at presentation. Based on the PIOPED study (JAMA 1990;263:2753-9), the best source for diagnosis of pulmonary embolism, and in patients with pre-existing cardiopulmonary disease, the most common symptoms are dyspnoea, pleuritic pain, cough, haemoptysis, and occasionally fever. On the other hand, the most common signs found in these patients are tachypnoea, tachycardia, rales, and typical changes in cardiac auscultation (a fourth heart sound and accentuated pulmonary component of S2). These signs and symptoms, however, have a low sensitivity and specificity — that is, they can be absent in patients with pulmonary embolism and present in patients without pulmonary embolism. That is why ancillary diagnostic tests are needed, and hence, the importance of knowing the discriminatory characteristics.

From a practical point of view, the results of this study allow us to define a diagnostic strategy, based on the pretest probability of pulmonary embolism:

If a patient has a low (±10%) clinical probability of having pulmonary embolism, then a test (d-dimer, lung scan, spiral computed tomography, or magnetic resonance angiography) with normal results confidently excludes pulmonary embolism. In this case, only pulmonary angiography could make the diagnosis

DEPT. OF NUCLEAR MEDICINE, CHARING CROSS HOSPITAL/SPL
Nothing beats the lung scan in terms of artistic potential

If a patient has an intermediate (±35%) clinical probability of having pulmonary embolism — for example, he or she has some risk factors such as stroke, history of deep venous thrombosis, malignancy, immobilisation, surgery within the past three months — then the diagnosis can be excluded if the d-dimer test is <500 µg/L, the lung scan is normal (or near normal), and spiral computed tomography and leg vein ultrasonography are also normal. In this risk group, diagnostic confirmation can be achieved through positive leg vein ultrasonography, a high probability ventilation perfusion lung scan (a scan with perfusion defects without ventilation abnormalities), or a positive result with spiral computed tomography or magnetic resonance angiography

If the patient has a high (±70%) clinical probability of having pulmonary embolism, only a pulmonary angiography can confidently exclude the diagnosis. All other tests (echocardiography, leg ultrasonography, lung scan, spiral computed tomography and magnetic resonance angiography), if abnormal, will confirm it.

Diagnostic algorithms — which look like decision trees, branching from a starting point into several hypotheses — are available to help clinicians define the prevalence of the disease (low, moderate, and high). For example, one of them — known as the Wells criteria (Ann Int Med 2001;135:98) includes several predictive factors that provide a final score: clinical symptoms of deep vein thrombosis (3 points), other diagnosis less likely than pulmonary embolism (3 points), tachycardia >100 beats/minute (1.5 points), immobilisation or surgical intervention in the previous four weeks (1.5 points), history of previous deep venous thrombosis or pulmonary embolism (1.5 points), presence of haemoptysis (1 point) and/or malignancy (1 point). The final score is obtained by adding all the individual scores of each predictive factor present in the patient: a score higher than 6 is very suggestive of pulmonary embolism, and a score less than 2 tends to exclude it.

Keep out of trouble to avoid getting on the “downward spiral”

In summary, an accurate pretest probability (prevalence) will allow the sequential ordering of confirmatory or exclusion tests for the diagnosis of pulmonary embolism.

What are the problems with the study?
A systematic review is only as good as the quality of the individual studies that form the basis of it. The better the individual papers, the higher the quality of the systematic review.

As stated by the authors, in this report only a few of the included studies were cross sectional studies in unselected patients, with independent and blinded assessment of test and reference methods. The best way to describe diagnostic accuracy or percentage of correct diagnoses is one in which every patient is subject to both tests, the new and the gold standard. In these studies, both tests (new and gold standard) are compared in every patient, with a subsequent building of a 2x2 table to derive parameters such as sensitivity, specificity, positive and negative predictive values. These parameters define the diagnostic discriminative power of a test. The gold standard of diagnosis of pulmonary embolism — pulmonary angiography — is an invasive test and is not normally used in all patients.

Some studies use clinical follow-up to detect the status of untreated patients with negative test results. This method may suffer from absence of blindness (both the patients and the researchers know the results of the tests and can therefore be influenced in their evaluation of the results) allowing detection bias (of false positive results—when the patient is wrongly classified as having pulmonary embolism, and false negative results—when a patient is wrongly classified as not having pulmonary embolism).

If the studies are heterogeneous, then the systematic review is difficult to conduct because the generalisation of its results may be compromised. This is the case in the studies of quantitative d-dimer ELISA (probably the best diagnostic technique for detecting fibrin degradation products through d-dimer—a consequence of the normal physiological mechanisms involved in reopening a clotted artery), which was used in several studies with high heterogeneity.

Was it a good study?
A search of the literature for diagnostic tests will disclose many articles dealing with several aspects of the tests. However, they rarely include assessment of outcomes for patient management (for example, differentiating between stable and unstable angina, they use only angina). The authors of this review hoped to overcome this limitation by trying to define the accuracy of diagnostic tests for pulmonary embolism and the appropriate clinical setting for their use.

There are challenges in conducting systematic reviews of diagnostic technologies with differing magnitude and importance. The first is the identification of all (published and unpublished) studies. This is important because of the need to have a clear and unbiased view of the true diagnostic properties of the tests and its generalisation to different clinical settings. The authors searched several biomedical databases from 1990 to 2003, manually retrieved several studies from reference lists of the published papers, and looked into personal files.

The second challenge concerns the need to assess the methodological quality of the individual papers. This is important in terms of assuring the validity of the results and is based on the design of the individual studies. The authors went to great lengths to ensure that they would not miss any important study, thus assuring the usability of the results for other patients.

The third problem with systematic reviews is a correct synthesis of test accuracy (see above). In this paper, the authors carefully state that they calculated the positive likelihood ratio for confirmation diagnostic strategies and the negative likelihood ratio for exclusion diagnostic strategies, as well as 95% confidence intervals. They also carried homogeneity tests to evaluate the consistency of findings across the studies with the classic statistics (Cochran's Q and I2), to be sure that the studies could be combined statistically and give a meaningful synthesis.

Finally, it is always desirable to report the assessment of the impact of diagnostic technology, known as life years saved by the diagnostic test — that is, the impact of diagnosis of the disease in society. It is also useful to give an economic analysis, in which the consequences are expressed in natural units — units that describe events or characteristics that are important to patients, such as cost per life saved, deep vein thrombosis prevented. The authors of this paper failed to do so.

Overall, this systematic review is a welcome addition to our knowledge of the diagnosis of pulmonary embolism.

Antonio Vaz Carneiro, clinical professor of medicineFaculty of Medicine, University of Lisbon,Portugal
Email: avc@fm.ul.pt


studentBMJ 2005;13:397- 440 November ISSN 0966-6494