A rational approach to the diagnostic workup of acute pulmonary embolism

This essay only deals with certain controversial, or problematic,  issues that clinicians may not fully consider during the diagnostic workup of a suspected PE patient, and the author presumes that interested readers already know the basic principles pertaining to PE diagnostic algorithms. An excellent background article on PE diagnostic testing by Joe Lex is freely available online at eMedHome.com -- 25 Years of Naked Emperor: Why Can't We Diagnose Pulmonary Embolism Yet

The problem of how best to diagnose a PE is one of the most difficult diagnostic problems that emergency physicians (EP) face, because no single diagnostic test can provide a definitive answer. An EP often has to use a number of diagnostic tests in sequence in order to diagnose, or exclude, a PE, and he has to decide which sequence of tests is optimum for his patients. Different PE experts offer different diagnostic algorithms, and the average EP may be perplexed by the large number of alternative choices. In this essay, I will be logically analysing the different choices, and I will offer some guidance as to why one choice may be better than another choice -- based on my knowledge of the EBM literature and my logical reasoning. I will start off this essay by dissecting a standard diagnostic algorithm, and I will end this essay by presenting my own recommended diagnostic algorithm.

Why do we need a diagnostic algorithm to diagnose PE?

It is common knowledge that PE is suspected in patients who present with a variety of symptoms or signs suggestive of PE (eg. chest pain, sudden dyspnea, worsening of chronic dyspnea, unexplained tachycardia, unexplained hypoxia), and that the clinical predicament lies in the fact that there is no combination of clinical symptoms/signs that will enable an EP to diagnose PE, or exclude PE, with any degree of certainty. Why is there a need for diagnostic certainty? The need for diagnostic certainty arises because undetected, and untreated PE, can be associated with significant morbidity and mortality, and an EP therefore feels impelled to make, or exclude, the diagnosis of PE at the time of initial patient presentation. Many EPs think that a diagnosis of PE is made when a specific diagnostic test is positive for PE, and that diagnostic certainty arises from the diagnostic test rather than from the mental process of clinical judgement. It is my belief that many incorrect diagnoses are made because clinicians think that diagnostic truth lies in the diagnostic test, rather than being derived from a clinician's interpretation of the diagnostic test's result in the context of the pre-test likelihood of the disease-in-question. Consider a practical example of that type of error. In the diagnostic workup of a PE, a V/Q scan is often the first diagnostic test ordered, and many EPs would regard a high probability V/Q scan as being diagnostic of PE. In fact, a high probability scan is only diagnostic of PE if the clinical suspicion of PE is moderate-high, but it may not be diagnostic of PE if the clinical suspicion of PE is low. In his review article on the diagnosis of PE [1] Wells states --- "using pulmonary angiography as the gold standard, that two studies have demonstrated that between 45% to 66% of highly positive V/Q scans are falsely positive when a skilled clinician deems that the patient's pre-test probability of PE is low". This is a critical point -- EPs need to understand that diagnostic certainty is not solely derived from a positive diagnostic test result, but is in fact derived from the effect that a positive test result has on a clinician's thinking when he mentally weighs the evidence of a positive test result in conjunction with his prior estimation of the patient's pretest probability of PE. If after weighing the evidence in his mind, a clinician moves from a point of suspecting PE to a point of being certain that PE truly exists, then he has simply had a change in his mental attitude with respect to the diagnosis of PE. Different clinicians when faced with the same data (clinical symptoms/signs + positive diagnostic test results) can end up harboring different degrees of diagnostic certainty, and this mental process is often purely subjective and individual clinician-specific. It is obvious that the medical profession needs to standardize the evaluation process and minimize the subjectivity element, so that different clinicians can come to identical conclusions when faced with the same combination of facts. Evidence-based medicine (EBM) provides clinicians with the necessary tools and skills to enable this process to occur in a consistently objective manner -- via the process of Bayesian analysis and thinking.

The process of Bayesian analysis is essentially an objective mathematical process whereby an estimated pre-test probability of a particular disease (strength of a clinician's suspicion of disease) is altered by the result of a diagnostic test to create a post-test probability of that particular disease. It can be expressed simplistically by the following formula with respect to the diagnosis of PE: Pretest probability of PE (or prior probability of PE) x (LR+ of a PE-diagnostic test) => Post-test probability of  PE (or posterior probability of PE). The LR+ stands for the likelihood ratio of a positive test, and the LR+ figure is derived from the test's sensitivity and specificity. A test with high sensitivity/specificity will result in large LR+ numbers, which in turn will have the greatest power to change the pre-test probability of PE in the direction of confirming the presence of PE. A LR- figure is also derived from a test's sensitivity/specificity and a LR- allows a clinician to alter his pre-test suspicion of PE towards a point of diagnostic certainty that PE is excluded. Although the mathematical Bayesian process is fixed, there are apparently no consensus definitions of "PE-confirmed" and "PE-excluded" that are uniformly accepted. I have noted that some PA experts regard a post-test probability figure of >80% as a sufficiently high endpoint (test-treat threshold) to warrant treatment of PE. However, it is also important to understand that the test-treat threshold figure should not be regarded as a fixed goalpost, and individual clinicians may decide to drive the posterior probability of PE to an even higher threshold level (eg. >90% post-test probability) if the risks of anticoagulation therapy are considerable (eg. patient has a bleeding diathesis or recent major surgery). From the perspective of EPs, most EPs are particularly concerned about not missing a PE, and they therefore aim for a really low post-test probability figure of <1-2% (no treatment-test threshold) as the desired endpoint that excludes PE. Although a no-treatment-test threshold of <2% is a commonly used endpoint, some PE experts also regard a no treatment-test threshold of <5% as being an acceptable endpoint (see Kearon's review article [2]).

(* It is critical that a reader understand the basic concept of Bayesian analysis if they want to follow the arguments expressed in this essay, because my arguments are mainly based on Bayesian calculations)

Consider the usual diagnostic algorithmic approach to PE, that has been recommend by many PE experts over the past two decades, in very simple terms. The diagnostic algorithmic process usually starts with an initial screening test such as a V/Q scan. A positive (high probability) V/Q scan test result confirms the diagnosis of PE, while a negative (normal) V/Q scan result excludes PE. Patients with non-diagnostic V/Q scans (indeterminate or low/intermediate probability) require further diagnostic testing. A leg ultrasound study is usually the next diagnostic test recommended and a positive test result confirms the presence of venous-thromboembolic disease (VTE). A negative leg ultrasound is regarded as non-diagnostic, and a third diagnostic test is recommended -- usually a pulmonary angiogram (PA). A negative PA excludes PE while a positive PA confirms PE. The only major variation on this theme is the preliminary use of D-dimer testing in low probability patients, and a negative D-dimer test result excludes PE while a positive test result is non-diagnostic. Patients with a positive D-dimer test result proceed to have a screening V/Q scan, which is also recommended as the initial diagnostic test in moderate-high probability patients. The entire diagnostic process seems clear-cut, but it is fraught with many complexities, that I will now discuss.

Before I dissect the entire diagnostic algorithmic process in piecemeal fashion, consider what happens to PE patients during the diagnostic process. Many clinicians may be surprised to discover how many PEs are potentially missed if definitive diagnostic testing using a "gold standard" diagnostic test (PA) is not routinely performed during the workup of suspected PE patients.

A hypothetical diagnostic algorithm that predicts what happens to PE patients during the diagnostic process, and what ultimately happens to patients with undetected PE after hospital discharge

Consider the hypothetical diagnostic algorithm (see below) that was presented by Clive Kearon in his recent review article [2]. Note that Kearon starts off with an outpatient population of 1,000 hypothetical patients who have a prevalence of PE of 20% (200 true-positive PE patients out of 1,000 suspected PE patients). He then demonstrates what will happen to those true-positive PE patients as they are worked up by his proposed diagnostic algorithm.

Note that Kearon makes certain assumptions that are likely to be reasonably consistent with the known facts. Even if one disagrees somewhat with his numbers, I think that they are sufficiently accurate so as to enable one to get a good perspective of the problematic situation of PEs that are undetected by initial diagnostic tests.

Consider the following facts relating to his hypothetical algorithm:-

note that he first divides the suspected PE-patients into three groups based on their estimated pre-test probability of PE (low-moderate-high) and he estimates that 55% (550 out of 1,000) of suspected PE patients will have a low probability of PE, 35% (350 out of 1,000) a moderate probability of PE, and 10% (100 out of 1,000) a high probability of PE [a reasonably accurate estimation for most EDs that have a prevalence of PE of 20% in their population of suspected PE patients]

he then estimates what percentage of those suspected PE patient subgroups will have a PE -- 10% (55/550) of the low probability group, 24% (85/350) of the moderate probability group, and 60% (60/100) of the high probability subgroup [a reasonable assumption based on "average" figures from many previously published studies]

he then demonstrates what will happen to those PE patients as they are subjected to a sequence of diagnostic tests [based on reasonable assumptions regarding the known sensitivity/specificity of the different diagnostic tests -- D-dimer test, lung V/Q scan, and leg ultrasound study)

pay particular attention to the number of patients who finally end up in the subgroup -- non-diagnostic V/Q scan + negative leg ultrasound  (67 PE patients/373 suspected PE patients) -- which is at point 4 in the algorithm

Note that at the time of completion of the PE workup in an ED setting (using a common protocol involving D-dimer testing for low probability patients => a V/Q scan for low probability patients who have a positive D-dimer test, and for all moderate and high probability patients => a leg ultrasound study for patients with a non-diagnostic V/Q scan) that 373 patients (37% of the original sample) will have a non-diagnostic V/Q scan + negative leg ultrasound study (point 4 in the algorithm), and that the prevalence of PE in that subgroup is approximately 18% (67/373). That means that 33% of the original sample of PE patients (67/200) do not have their PE diagnosed after a preliminary PE-workup (which is completed at the time of their initial presentation and prior to hospital discharge) and that the prevalence of PE in that subgroup of patients (18%) is near-identical to the overall prevalence of PE in the original sample of suspected PE patients (20%). Point 4 is a common diagnostic endpoint for many clinicians, who do not perform further diagnostic testing (PA) to definitively exclude PE -- if the suspected PE patient is deemed to have a  low-moderate probability of PE + good cardiopulmonary reserves + no signs of a severe PE.

What happens to those 33% of PE patients if further definitive diagnostic testing (PA) is not performed and the patients are discharged from the ED without further testing?  According to Kearon's algorithm, serial ultrasound testing performed in an outpatient setting over the next 2 weeks will pick up an additional 9 PEs, which means that 58 PEs will remain undetected (58/200 = 29% of the original sample). What happens to that large number of undetected PE patients. According to Kearon, clinical outcome studies have demonstrated that there is only a <1% incidence of VTE during 3 months' follow-up. Therefore, the inescapable conclusion must be that spontaneous PE resolution occurs in the majority of undetected PE patients, who fall into this particular category. Does the EBM evidence support Kearon's figures? It is my belief that the EBM literature evidence is not entirely clear about this critical point, and that divisive sentiment prevails. There are two major schools of thought with respect to this critical dilemma. The conservative school does not trust the results of clinical outcome studies, and they believe that all undetected PE patients are at high risk of a subsequent fatal PE (or the development of chronic VTE disease). They therefore believe that mandatory PA testing is needed to definitively exclude PE in all the patients in that large group of undetected PE patients (33% of the original sample). Their main criticism of clinical outcome studies is that PE fatalities may have been missed, because autopsies were not routinely performed, and that the definition of a PE-death is an arbitrary definition, that is derived from an adjudication committee. My own review of a number of clinical outcome studies suggests that this adjudication issue is indeed a weak point of those studies, because the diagnosis of a PE-death is performed by an adjudication process, which has no "gold standard" criteria of adjudication that are uniformly regarded as "airtight". However, my own personal sentiment still favors the position of the other school of thought -- that believes that it is safe to ignore the fact that the estimated posterior probability of PE in that subgroup of patients is >2% (above the not test/treat threshold) because follow-up clinical outcome studies show a <2% risk of subsequent VTE (after recommended serial ultrasound studies are routinely performed to pick up a small percentage of PEs). Each clinician has to think carefully about this critical issue, because his final decision-choice should be based on informed clinical judgement, which means that each clinician should be willing to quote the EBM literature evidence that defends his own position.

(* The same dilemma exists with respect to CT scan negative patients if a CT scan is performed as the initial diagnostic test. However, I will deal with that subject at greater length later in this essay)

Consider another controversial issue that is common to many PE diagnostic algorithms.

What is the value of D-dimer testing in low probability patients?

What is the main value of dividing suspected PE patients into pretest probability subgroups -- low, moderate, high probability -- prior to diagnostic testing. The main value of this subdivision lies in the fact, that if one can accuratedly define a low probability subgroup of patients who have a low prevalence of PE (arbitrarily defined as a <20% prevalence of PE), then one could theoretically use a D-dimer test to "screen-out" a certain percentage of low probability patients who have a negative test, as having "PE-excluded". Note in Kearon's algorithm, that he recommends using a D-dimer test of moderate sensitivity (eg. SimpliRed  D-dimer test with a sensitivity of 85%), and he demonstrates that a negative test is found in roughly 64% of the low probability patients (354/550), who therefore do not need to undergo further diagnostic testing -- potentially a great advantage in terms of convenience and cost-effectiveness. Kearon obviously accepts the fact that 15% of PEs (8 out of 55) in his low probability patients are missed by this diagnostic approach, and he calculates that the D-dimer test has a false-negative rate of approximately 2% (8/354) in those low probability patients. He apparently believes that if a D-dimer test has a negative predictive value of >98% in low probability patients that he has essentially excluded PE, and he supports his belief by indicating that clinical outcome studies show a <2% incidence of VTE during 3 months of follow-up in those D-dimer negative patients. There are many clinicians who are sympathetic to this approach, and who believe that it is safe to use a D-dimer test to exclude PE in low probability patients. In fact, ACEP has a clinical policy statement on this issue [3] that states that level B evidence demonstrates that a negative quantitative D-dimer assay (turbidimetric or ELISA) excludes PE in low probability patients.

Let's examine this D-dimer issue more thoroughly. Kearon's hypothetical algorithm presumes that the prevalence of PE in low probability patients is 10%. However, there is no guarantee in "real life" clinical practice that the prevalence of PE in patients who are labelled "low probability" will really be that low. Two main approaches are used to determine the clinical probability of PE in clinical practice -- an empiric clinical estimation approach and an approach based on clinical prediction rules. The two most frequently used strategies involving clinical prediction rules are the Wicki model and the Wells model. When I considered the percentage of low probability patients who actually have a PE in studies that used empirical estimation strategies or clinical prediction rules, I noted that the results varied from 3.4% to over 20% with an average of 10-15%. However, it is important to note that the specific results only pertain to the patients in each particular study, and there is no guarantee that patients in "real life" clinical practice, who are labelled low probability by one prediction model, would be labelled low probability by another prediction model. In fact, when the Wells model was compared to the Wicki model [4] with respect to a fixed set of patients, there was a 30% chance that patients labelled "low probability" by one model would be labelled "moderate probability" by the other model (and vica versa). What are the implications of this critical fact?

Consider the following table showing the negative predictive value (NPV) of two D-dimer tests of high and moderate sensitivity -- the highly sensitive rapid ELISA test (sensitivity 97%, specificity 42%) and the moderatedly sensitive SimpliRed test (sensitivity 85%, specificity 75%) -- for different pre-test probabilities of PE (prevalence of PE = pre-test probability of PE). The figures were obtained by using an online Bayesian calculator -- which is freely available at http://www.intmed.mcw.edu/clincalc/bayes.html.

Note that the SimpliRed D-dimer test only has a NPV of  >98% (<2% posterior probability of PE) if the prevalence of PE is <10%. Note that the SimpliRed test cannot drive the posterior probability of PE to <2% in low probability patients who have a pre-test probability of PE of 10-20% (a <2% posterior probability of PE is the endpoint that is acceptable to the majority of PE experts who are proponents of D-dimer testing). More importantly, note that there is no safety margin when using the SimpliRed test (or any other D-dimer test that has a sensitivity of 85%), and that the SimpliRed test would be relatively useless in excluding PE if the actual prevalence of PE was between 20-40% (a situation that could occur in ~30% of cases when using both the Wells and Wicki clinical prediction rule to determine the pre-test probability of PE if their results are discordant). Note that it is much safer to use the rapid ELISA test because one can be relatively certain that the post-test probability of PE will never be >5% -- even if the prevalence of PE is as high as 40%. Although a post-test probability of PE of 5% is higher than the standard desired endpoint (post-test probability of PE <2%), it may be a relatively safe endpoint for low probability patients in "real life" clinical practice. The only advantage of the SimpliRed test is that it has a higher specificity than the rapid ELISA test, and a D-dimer test with higher specificity is associated with a lower proportion of false-positive tests relative to true negative tests. That situation is advantageous because it means that less suspected PE patients have to undergo further diagnostic testing (CT scan or V/Q scan). To make the latter statements more readily understandable, consider the following practical example.

Presume that a SimpliRed test  (sensitivity 85%, specificity 75%) and rapid ELISA test (sensitivity 97%, specificity 42%) are performed on 1,000 low probability PE patients who have a PE prevalence of 10%. Then consider the number of patients with true-positive, false-positive, true-negative and false-negative test results.

Note that the SimpliRed test has the advantage of only producing 225 false-positive tests for every 1,000 patients tested, while the rapid ELISA test produces 522 false-positive tests for every 1,000 patients tested. The disadvantage of the rapid ELISA test is that >60% of patients have a positive test, and all those patients have to undergo further diagnostic testing, while only 30% of SimpliRed tests are positive. However, note that the SimpliRed test has 5x the number of false-negative tests as the rapid ELISA test. What is better -- a D-dimer test with a very low rate of false-negative results (safe test) + a high rate of false-positive results (inefficient test), or a D-dimer test with a higher rate of false-negative results (less safe test) + a lower rate of false-positive results (more efficient test)? That is an important issue that all clinicians, who are supportive of D-dimer testing, must carefully consider.

(* As an aside -- It is important not to use rapid ELISA D-dimer testing too liberally as a screening test, because if patients who are very unlikely to have PE are unnecessarily tested, then a fair number of non-PE patients will have positive D-dimer tests, and be subjected to further PE diagnostic testing (V/Q scan, sCT scan, leg ultrasound study). A small proportion of those non-PE patients will have positive PE diagnostic studies, and virtually all of the positive studies will be false-positive studies. If a clinician does not recognize that all of those positive studies are false-positive studies, then those patients will unnecessarily be subjected to anticoagulation therapy -- a very unsafe situation!)

Based on the results of my previous calculations, it is my belief that there are three main attitudes that clinicians can adopt with respect to D-dimer testing in low probability patients.

1. A risk-tolerant clinician could use a moderatedly sensitive, moderatedly specific D-dimer test (eg. SimpliRed test) if he is absolutely convinced that the suspected PE patient has a very low probability of PE (<10% and preferably <8%) -- in order to decrease the number of false-positive tests, while maintaining a borderline acceptable ability to achieve a NPV of >98% if the test is negative. Although the use of a moderatedly sensitive D-dimer test is not as safe as the use of a highly sensitive D-dimer test, the use of moderatedly sensitive D-dimer tests is actually endorsed by ACEPs clinical policy committee [3], and their official statement reads "A negative whole blood qualitative D-dimer assay in conjunction with a Wells score of less than 2 excludes PE in low probability patients (level B evidence). I am really surprised that ACEP's clinical policy committee endorsed that position -- because its validity depends on the results of a single study, which had an extremely low prevalence of PE overall (9%) and a very low prevalence of PE in the low probability subgroup (3.6%). It is well known that it is not sensible to attempt to determine the true efficacy of a diagnostic test in a population of patients that has a very low prevalence of disease, because the test will perform much better in that very low prevalence population than it would under normal "real life" clinical practice conditions where the prevalence of disease is significantly higher. In fact, the authors of ACEP's clinical policy noted that when Sanson tested the Wells clinical prediction rule in a mulicenter trial, that low probability patients had a prevalence of PE of 28% -- therefore, it is very questionable whether this recommendation can be regarded as safe.

2. A less risk-tolerant clinican should use a highly sensitive D-dimer test (eg. rapid ELISA test) in order to minimize the number of false-negative tests, if he thinks that the probability of PE in low probability patients could be anywhere between 0-20% -- based on the concordant results of two clinical prediction rules or a strong conviction based on empirical clinical assessment + one clinical prediction rule having consonant results for low probability, OR if he uses two clinical prediction rules that are not concordant for low probability, which implies that the patient may actually have a low-moderate probability of PE (20-40% probability) -- but then only if he is willing to accept a maximum NPV of 95% and a posterior probability of PE of <5% (rather than <1-2%) as an acceptable exclusionary endpoint.

3. A risk-averse clinician, who is not confident in his ability to accuratedly estimate the pretest probability of PE, or who does not have faith in the safety of D-dimer testing, should forgo D-dimer testing and proceed directly to a screening diagnostic study (V/Q scan or CT scan).

(* A negative rapid ELISA test may also be useful in the situation of a low-moderate probability patient who has a non-diagnostic V/Q scan result and a negative leg ultrasound result => a negative D-dimer test essentially excludes PE, while a positive D-dimer test suggests the need for further testing eg. serial leg ultrasound studies or PA)

What is the best screening study for low probability patients with a positive D-dimer test, or for all suspected PE patients -- irrespective of their estimated pretest probability of PE?

Note that Kearon chose a V/Q scan as the first test in his diagnostic algorithm. The V/Q scan has been traditionally used as the primary screening test for the past two decades, but it has many disadvantages.

The V/Q scan is time-consuming test and it is often not readily available after hours. Most importantly, the study may be non-diagnostic in about 60-70% of cases, and a non-diagnostic study does not significantly change the pretest probability of PE. A high probability scan implies that PE is present in moderate-high probability patients. However, a high probability scan in a patient with a low probability of PE is associated with a false-positive rate of 40-60% => further testing (PA or CT scan) is required to exclude the significant possibility of a false-positive test result. A normal scan excludes PE in low probability patients (2% rate of subsequent PE) and essentially excludes a PE in moderate-high probability patients (<6% rate of subsequent PE). Although a non-diagnostic V/Q scan is associated with a  roughly 20-30% prevalence of PE (16% in low probability scans and 40% in intermediate probability scans), clinical outcome studies suggest that if a subsequent leg ultrasound study is negative, and the patient is categorized as being a low-moderate probability patient who has good cardiopulmonary reserves and no signs of severe PE, that there is a <2% incidence of VTE during 3-6 months of follow-up if the patient is not treated and serial outpatient ultrasounds remain negative (see my previous argument). However, note that a V/Q scan could still be regarded as the best screening test if the chest X-ray is normal and the estimated pretest probability of PE is very low (<10%), because a normal scan result virtually excludes clinically significant PE. A V/Q scan is also the best screening test if contrast dye allergy, or renal failure, precludes the use of  sCT angiography as a screening test.

A spiral CT scan is probably a much better test to use as the first screening test. First of all, a sCT scan is usually available 24 hours-per-day. Secondly, it can be quickly performed, and there is only a <10% likelihood of a non-diagnostic test result (usually due to technical problems). There are also less interobserver disagreements with respect to sCT scan interpretations as compared to V/Q scan interpretations. Another great advantage of a sCT scan is that it can identify alternative diagnoses in a high percentage of non-PE cases. A sCT scan is also much more useful than a V/Q scan in patients with abnormal chest X-rays, because most patients with an abnormal chest X-ray will have a non-diagnostic V/Q scan result. A positive sCT scan has a high positive predictive value for central PE in moderate-high probability patients because the test has a high specificity (94% for multi-detector CT scanners). A sCT scan also has a high negative predictive value for central PE in low-moderate probability patients because it has a high sensitivity (94% for multi-detector CT scanners). With respect to peripheral PE, the latest generation of multidetector sCT scanners with 1.25mm-thick sections potentially allows accurate analysis of peripheral pulmonary arteries down to the fifth order, which should enable an experienced radiologist to positively identify a substantial percentage of isolated peripheral PEs. Finally, CT angiography can be combined with a delayed CT venous study of the thigh and pelvis in a single study, thus obviating the need to perform a leg ultrasound study if the sCTscan result is negative.

A large number of clinicians are still concerned that a sCT scan does not have sufficient sensitivity to totally exclude PE if the test is negative, and they often quote sensitivity figures of 60-90% from the medical literature to justify their concern. However, one should note that those sensitivity figures are derived from studies performed in the early-mid 1990s when single-detector sCT scanners were used. Also, the accuracy of some of those studies is highly questionable -- click here if you want to read my analysis of the Perrier study, which I think is highly flawed. The latest generation of multi-detector sCT scanners probably have a sensitivity of >95% for central PE, which is similar to the results of a PA. What about the problem of isolated subsegmental PE, and the inability of sCT scanners to detect all those peripheral PE? It is probably true that a sCT scan has a much lower sensitivity for peripheral PE, and that the sCT scan's sensitivity for peripheral PE is probably <60%. However, there are two additional EBM evidentiary facts to consider. First of all, there is increasing evidence from clinical outcome studies that a low-moderate probability patient with a negative sCT scan has a very low rate of subsequent PE (<2% risk during the subsequent 3-6 months of followup) and that the expected PE-rate is probably not higher than the PE-rate in a patient with a negative PA (1.6% during the subsequent year). The fact that a sCT scan misses a substantial number of isolated PEs (because the test has a low sensitivity) and the fact that clinical outcome studies shows a <2% incidence of subsequent PEs, suggests that most undetected peripheral PE must resolve spontaneously. It may therefore not be necessary to obsessively pursue the diagnosis of small peripheral PE in healthy patients with good cardiopulmonary reserves. The second point is that even if one wanted to diagnose all subsegmental PEs, there is no conclusive evidence that a PA has a better sensitivity for diagnosing isolated peripheral PEs than the newer generation multi-detector CT scanners, and both tests will likely miss a substantial number of isolated peripheral PEs. Note that it is also questionable if there is an advantage to ordering a PA if a sCT scan is negative in a high probability patient patient, or moderate probability patient with poor cardiopulmonary reserves and/or severe signs of PE -- the patient's high risk situation or clinically vulnerable status may give a clinician a strong impetus to definitively diagnose a PE, but there is no definite evidence that a PA has a better sensitivity/specificity for diagnosing a PE than the latest generation of multidetector CT scanners. One needs to determine the local radiologists' particular skill at performing and interpreting a PA, compared to his ability to interpret a high-resolution sCT scan, before making a final decision about the value of ordering a PA in a sCT scan negative patient. Interestingly, Kearon recommends a PA if a sCT scan shows a subsegmental intraluminal filling defect in a patient with a high probability of PE. I presume that he recommends a PA to ensure that the positive sCT scan finding is a true-positive result rather than a false-positive result. However, considering the fact that a PA is associated with a <66% intraobserver agreement with respect to the diagnosis of subsegmental PE, a negative PA does not necessarily mean that the sCT finding is a false-positive result. It could be that the subsequent PA result is a false-negative result, and the sCT scan result a true-positive result. I strongly suspect that the belief that a PA should be regarded as the "gold standard" test in this particular scenario will fall into disfavor with the passage of time. Finally, note that a PA is firmly recommended if a sCT scan study is non-diagnostic (indeterminate) because a non-diagnostic test does not really change the pre-test probability of disease,  and further testing is required to either drive the posterior probability of PE above the test-treat threshold (PE-confirmed) or below a do not treat-test threshold (PE-excluded).

(* Good review article -- Imaging of acute pulmonary thromboembolism: should spiral computed tomography replace the ventilation-perfusion scan? Powell T - Clin Chest Med - 2003 Mar; 24(1); 29)

What is the value of a leg ultrasound study in patients with a negative sCT scan or non-diagnostic V/Q scan?

It is common practice for clinicians to order a leg ultrasound study if an initial screening test is negative (sCT scan) or non-diagnostic (V/Q scan). How useful is a leg ultrasound study in suspected PE patients with asymptomatic legs? Note that in Kearon's diagnostic algorithm, that a leg ultrasound was performed in 410 patients and that it was only positive in 37 patients (9% of those tested). Also, note that only 76% of those positive leg ultrasound studies were true positives, and that 24% were false-positives. Why does a leg ultrasound have such a low yield, and why are there so many false-positive test results? The answer is because a leg ultrasound study has a low sensitivity of only 60% in patients with asymptomatic legs, in contrast to its sensitivity of >90% in patients with symptomatic legs.

Using the online Bayesian calculator, these are the calculated number of true-positive and false-positive test results if a leg ultrasound study (sensitivity 60% specificity 90%) is performed on 1,000 patients with varying prevalence of PE.
 

Prevalence of PE	True-positives	False-positives
10%	60	90
20%	120	80
30%	180	70
40%	240	60
60%	360	40

Note that if one is dealing with a low probability patient (<20% probability of disease), that the number of false-positive results relative to true-positive results is very high. Therefore, one should strongly consider performing a confirmatory test (eg. venography) to ensure that any positive leg ultrasound study result is a true-positive result, before treating a low probability patient with anticoagulation therapy.

My recommended PE diagnostic algorithm

This diagnostic algorithm is a summation of all my previously expressed opinions and concerns. Note, that in contrast to a number of standard diagnostic algorithms, there are many recommended options at different branch points -- individual clinicians are advised to make independent choices that are consonant with the patient's particular circumstances, the availability of diagnostic resources, and the skill of local radiologists in interpreting particular studies.

Note that the safety of the above algorithm is based on two presumptions -- i) that one is using a newer generation multi-detector sCT scanner; and ii) that local radiologists are skilled in interpreting those high resolution sCT scans.

Note that I do not offer firm recommendations for ordering a PA in sCT scan negative patients (point 5 in the algorithm) -- because it depends on local facilities and local expertise (local availability of a newer generation multi-detector sCT scanner and local radiologic skill in interpreting high resolution sCT scans, compared to local radiologic skill in performing and interpreting PAs). The only firm recommendation for a PA is when the sCT scan result is non-diagnostic (indeterminate) -- point 2 in the algorithm.

If a V/Q scan is used instead of a spiral CT scan, a normal scan excludes a PE. A high probability scan confirms the presence of a PE, except in a low probability patient who may need confirmatory testing (sCT scan or PA) to exclude the possibility of a false-positive result. A non-diagnostic scan can be regarded as "equivalent" to a negative spiral CT scan, and should be followed by further testing  -- using the same protocol that is recommended for sCT scan negative patients in the above algorithm. However, note that a sCT scan can replace a PA at point 5 in the algorithm -- as an alternative choice.

Also, note that Wells [1] suggests that a negative D-dimer test in a moderate probability patient, who has a non-diagnostic V/Q scan (or negative CT scan) and negative lower limb ultrasound study, may negate the need for serial ultrasound testing. Wells [1] also suggests that a negative D-dimer test in a high probability patient, who has a non-diagnostic V/Q scan (or negative CT scan) and a negative leg ultrasound study, implies a lower risk of PE, and he recommends serial ultrasound testing instead of a PA.

Jeff Mann.
April 2003.

References:

1. Diagnosis of pulmonary embolism: when is imaging needed? Wells P - Clin Chest Med - 2003 Mar; 24(1); 13

2. Diagnosis of pulmonary embolism. Kearon C - CMAJ - 21-Jan-2003; 168(2): 183-94

3. Clinical policy: critical issues in the evaluation and management of adult patients presenting with suspected pulmonary embolism. - Ann Emerg Med - 01-Feb-2003; 41(2): 257-70

4. Comparison of two clinical prediction rules and implicit assessment among patients with suspected pulmonary embolism. Chagnon I - Am J Med - 01-Sep-2002; 113(4): 269-75.

Appendix:

Wells clinical prediction rule
 

Variable	Points
Clinical signs and symptoms of DVT	3.0
An alternative diagnosis less likely than PE	3.0
Heart rate > 100 bpm	1.5
Immobilization or surgery in the previous 4 weeks	1.5
Previous DVT/PE	1.5
Hemoptysis	1.0
Malignancy (treatment ongoing or within past 6 months)	1.0

Pre-test probability based on total number of points

High -- >6 points
Moderate -- 2-6 points
Low -- <2 points

Wicki clinical prediction rule
 

Criteria	Points
Age 60-79 years	1
Age >79 years	2
Prior DVT/PE	2
Recent surgery	3
Heart rate >100 bpm	1
PaCO2, mmHg
<36	2
36-39	1
PaO2, mmHg
<49	4
49-60	3
>60-71	2
>71-82	1
Chest radiographic findings
Plate atelectasis	1
Elevation of hemidiaphragm	1

Pre-test probability of PE based on score range

High probability -- 9-12
Moderate probability -- 5-8
Low probability -- 0-4

Commentary, controversy and criticism:

Insightful comments by readers will be included in this section.