I am really excited to announce that we are starting a new format on the EMCases Journal Jam podcast series. The plan is to take important questions about everyday clinical practice and do a deep dive into the literature to provide you with the most evidence based answer we can come up with. Most importantly, we are adding new (or, I suppose you might argue, our first ever) talent to the series: the brilliant @EMNerd_ himself, Dr. Rory Spiegel!
The first episode has just gone live. Our question: Is there a role for D-dimer testing in the workup of aortic dissection in the emergency department? I would suggest the podcast as the more entertaining and succinct way to get an answer, but because of time restraints and our desire not to bore you to death, we did not discuss every single study in the podcast. What follows are my notes for the podcast – a brief review of all the original literature I could find on the topic, arranged by year of publication.
Weber T, Högler S, Auer J. D-dimer in acute aortic dissection. Chest. 123(5):1375-8. 2003. PMID: 12740250
Although this is a prospective cohort, they only looked at 10 patients with what they say is suspected aortic dissection. However, later in the text they say that all of these patients had confirmed aortic dissection, so either they identified only confirmed patients or they had a 100% rule in rate. They mention that the D-dimer was positive in all 10 patients. They looked at 35 patients without aortic dissection, and D-dimer was positive in 31% of them. Not much of value can be taken from this tiny cohort.
Eggebrecht H, Naber CK, Bruch C. Value of plasma fibrin D-dimers for detection of acute aortic dissection. Journal of the American College of Cardiology. 44(4):804-9. 2004. PMID: 15312863 [free full text]
This study looks at 16 patients with a diagnosis of aortic dissection, 16 with MI, 16 with PE, 16 with noncardiac chest pain, and 32 asymptomatic patients with chronic aortic dissection. They make no mention of how this sample was drawn, but it is clearly not a representative sample. We have no idea what sized cohort these patients were drawn from or why the D-dimer was ordered, so although they report a sensitivity of 100% and a specificity of 67%, I don’t think you can actually make those calculations in any meaningful way from this data.
I guess this is a paper on the topic, but you can probably just throw it out. It is a retrospective look at 156 patients with a diagnosis of acute aortic dissection, but out of those 156, only 7 had a D-dimer drawn. They report that all D-dimers were positive for what it’s worth (not much).
Akutsu K, Sato N, Yamamoto T. A rapid bedside D-dimer assay (cardiac D-dimer) for screening of clinically suspected acute aortic dissection. Circulation journal : official journal of the Japanese Circulation Society. 69(4):397-403. 2005. PMID: 15791032 [free full text]
This is a group of 78 consecutive patients in whom “acute aortic dissection was suspected or not ruled out” being admitted to a coronary care unit and who also had a D-dimer drawn. “Suspected” and “not ruled out” are very different things in my mind. Either way, I don’t think you can actually think about these patients as consecutive in any real world way. Their rule in rate was 38%. It’s not clear if the physicians were blinded to D-dimer result, or whether every patient received the gold standard CT. They report a sensitivity of 100% and specificity of 54% (with no confidence intervals reported, which is a recurring theme in these papers). Clearly, the n is too small to claim 100% sensitivity, and this is a highly select group of patients.
Hazui H, Fukumoto H, Negoro N. Simple and useful tests for discriminating between acute aortic dissection of the ascending aorta and acute myocardial infarction in the emergency setting. Circulation journal. 69(6):677-82. 2005. PMID: 15914945 [free full text]
This is a retrospective look at 29 consecutive patients with type A dissections and 49 patients with acute MI from a single center. They don’t say how the decision to order a D-dimer was made, but it sounds like they just do it on everyone. They look at the area under the curve for a lot of different markers, so the number of possible combinations is very large, and any result is likely to be over-fit to this specific data set. Using a D-dimer cut-off of 800ng/mL they report a sensitivity of 93.1%. There were 2 patients missed, both with thrombosed false lumens. They combine the D-dimer with a chest x ray M-ratio to get better numbers, which could be an interesting approach for future research, but is definitely not supported by this data set.
Ohlmann P, Faure A, Morel O. Diagnostic and prognostic value of circulating D-Dimers in patients with acute aortic dissection. Critical care medicine. 34(5):1358-64. 2006. PMID: 16557157
This is a retrospective look at 94 consecutive patients admitted to a single institution with a confirmed diagnosis of aortic dissection. Although they call them consecutive, they had to have a D-dimer drawn, and it is not well described in the paper which patients had D-dimers measured. 93 (99%) had a positive D-dimer using a cutoff of 400mg/mL. The one miss was a patient with a thrombosed false lumen. They looked at 62 matched controls, and the D-dimer was positive in 66%. They don’t report the confidence intervals around the sensitivity, but the negative likelihood ratio (0.03) has a 95% confidence interval that includes 0.2, so it’s not great.
Hazui H, Nishimoto M, Hoshiga M. Young adult patients with short dissection length and thrombosed false lumen without ulcer-like projections are liable to have false-negative results of D-dimer testing for acute aortic dissection based on a study of 113 cases. Circulation journal. 70(12):1598-601. 2006. PMID: 17127806 [free full text]
A second paper by the same group of authors. I am pretty sure this includes the patients in the above study and just adds a few more years of data. This is a retrospective look at 113 adult patients diagnosed with acute aortic dissection at a single center in Japan where D-dimer is “routinely measured” (they don’t say on whom) and therefore is available for all 113 patients. Overall, the sensitivity was 92%. They compared patients with and without a thrombosed false lumen, and found a sensitivity of 86.4% with a thrombosed false lumen and 98.1% without.
Xue C, Li Y. Value of D-Dimers in patients with acute aortic dissection. Journal of Nanjing Medical University. 21(2):86-88. 2007. [article] (Paper not PubMed indexed)
This is a prospective look at 43 consecutive patients for whom acute aortic dissection was “suspected or not ruled out” in either the emergency department or the coronary care unit over 4 years at a single site. There was a 37% rule in rate. Like to Gorla study below, the mean D-dimer level in the non-aortic dissection group is well above the cutoff value. The D-dimer was positive in 60% of all patients tested. They report a sensitivity of 100% (with no confidence intervals) and a specificity of 66%. There were only 16 patients, so the confidence intervals would be wide. Like most of these studies, they make no mention of whether the clinicians were blinded to the D-dimer result when ordering imaging or selecting patients into the study. Unfortunately, this study doesn’t tell us anything about the patients who were missed.
Paparella D, Malvindi PG, Scrascia G. D-dimers are not always elevated in patients with acute aortic dissection. Journal of cardiovascular medicine. 10(2):212-4. 2009. PMID: 19377387
This is a look at 80 consecutive patients at a single center with the diagnosis of aortic dissection. 61 of the patients had D-dimers available. 11 of the 61 (18%) had D-dimers below 400ug/L, which would be equivalent to a sensitivity of 82%.
Suzuki T, Distante A, Zizza A. Diagnosis of acute aortic dissection by D-dimer: the International Registry of Acute Aortic Dissection Substudy on Biomarkers (IRAD-Bio) experience. Circulation. 119(20):2702-7. 2009. PMID: 19433758 [free full text]
This is a large, prospective, multi-center study across 14 hospitals in the United States, Europe, and Japan. They identified 220 patients with acute onset symptoms (within 24 hours) where the physician had a suspicion of acute aortic dissection and ordered imaging to rule it out. (They don’t report how long the study ran for, but given that it was 14 centers, that equates to less than 16 patients per site, so it is probably a pretty select group of patients.) 87 patients (40%) were diagnosed with aortic dissection. All patients had a D-dimer measured. Using the standard 500 ng/mL cutoff, D-dimer had a sensitivity of 96.6% (95% CI, 90.3 to 99.3) and a specificity of 46.6% (95% CI, 37.9 to 55.5). I think it is telling that the median D-dimer was above 500 ng/mL for all diagnoses except stable angina, and the mean D-dimer was above 500 ng/mL for all diagnoses. This is probably the best data set we have when it comes to deciding whether to use D-dimer in the emergency department, but I’m not sure that’s saying much. This is a highly select group of patients. I am not getting a positive dissection in anywhere close to 40% of the CTs that I order, and I don’t order that many. Even with the 96.6% sensitivity they report, the post test probability of a negative D-dimer in this population would be 6%! And that doesn’t account for the fact that their 95% confidence intervals allow for a sensitivity as low as 90%. On the other hand, if we applied this in lower risk populations (which we all would), the specificity of 46% and positive likelihood ratio of 1.81 would result in a lot more imaging, when the only reason to use D-dimer would be to limit imaging.
Fan QK, Wang WW, Zhang ZL. Evaluation of D-dimer in the diagnosis of suspected aortic dissection. Clinical chemistry and laboratory medicine. 48(12):1733-7. 2010. PMID: 20704542
This is a prospective cohort of 260 patients admitted to a cardiac specific hospital in China with suspected aortic dissection. (They excluded 1276 patients because a final diagnosis was not established, which gives you a sense of how select this cohort is. Eliminating patients because you don’t know the final diagnosis is a problem. We obviously still need to work these patients up. We might order D-dimers. This introduces bias into the results.) The rule in rate is high, at 48%. They present a table with a number of different D-dimer cut off values that would have different test characteristics for their data. No cut off was 100% sensitive. If you use a cutoff of 0.17 ug/mL you get a sensitivity of 99.2% (95%CI 95.6-99.9%) and a specificity of 30.9% (95%CI 23.2-39.4%). A more traditional cutoff of 0.49 ug/mL gets you a sensitivity of 93.5% (95%CI 87.7-97.2%), a specificity of 83.1% (95%CI 75.7-89.0%), a negative likelihood ratio of 0.09 and a positive likelihood ratio of 3.18. The one significant miss was a descending aortic dissection with an intramural hematoma and a D-dimer level of 0.04 ug/mL. This cohort had a rule in rate of 48%. If I am going to see a dissection on every second CT I order, I am going to order the CT. How well the D-dimer works in lower risk patients where we could actually avoid the CT is the more important question. This study indicates that in order to achieve a high sensitivity, we will need to use a low cut-off with a very low specificity, which (like in PE workups) will actually result in more imagining rather than less.
Nazerian P, Morello F, Vanni S. Combined use of aortic dissection detection risk score and D-dimer in the diagnostic workup of suspected acute aortic dissection. International journal of cardiology. 175(1):78-82. 2014. PMID: 24838058
This study is the most representative of emergency department patients. It is a retrospective look at data collected prospectively at 2 emergency departments. A total of 1455 patients were evaluated for suspected aortic dissection (over 5 years) and this study looks at the 1035 who had a D-dimer drawn. The rule in rate was 22.5%. For all comers, the sensitivity and specificity of D-dimer (500 ng/mL) were 98.3% (95%CI 95.7-99.5%) and 35.9% (95%CI 32.6-39.3%) respectively. This correlates with negative and positive likelihood ratios 0.05 (95%CI 0.02–0.13) and 1.53 (95%CI 1.45–1.62) respectively. When combined with an ADD (aortic dissection detection score) of 0, the sensitivity was 100%, but you would only avoid CT in 9% of this cohort. There were 4 misses in this cohort: 1 intramural hematoma, 1 type B, and 2 type As. This illustrates that it is not just the intramural hematomas and penetrating ulcers that D-dimer will miss. A point that applies to all these papers: they only include typical presentations of aortic dissection. Would the D-dimer be as sensitive in atypical presentations?
Gorla R, Erbel R, Kahlert P. Accuracy of a diagnostic strategy combining aortic dissection detection risk score and D-dimer levels in patients with suspected acute aortic syndrome. European heart journal. Acute cardiovascular care. 2015. PMID: 26185259
This is a retrospective cohort of 376 patients at what sounds like a speciality hospital, “The West German Heart and Vascular Center”. They only looked retrospectively at patients who had had a D-dimer drawn (522 patients) and of those, 376 had chest pain and so were included. 85 patients (22.6%) had a final diagnosis of acute aortic dissection. (14.5% of the population had PE, which obviously affects the specificity numbers.) The aortic dissection group did have a higher mean D-dimer level (13.1 mg/L vs 1.9 mg/L), but it is interesting to note that the mean value in the non-aortic dissection group was well above the cutoff for the test (0.5 mg/L). In the overall group, the sensitivity of D-dimer was 97.6% and the specificity was 63.2%. The 2 patients that D-dimer missed in this study both had penetrating aortic ulcers. They were able to combine the D-dimer with the “aortic dissection detection score” to achieve a sensitivity of 100%, but this resulted in a positive predictive value of only 1.6%.
Asha SE, Miers JW. A Systematic Review and Meta-analysis of D-dimer as a Rule-out Test for Suspected Acute Aortic Dissection. Annals of emergency medicine. 66(4):368-78. 2015. PMID: 25805111
I included this systematic review in the November 2015 articles of the month. Here is what I said: This is a systematic review and meta-analysis looking to determine the diagnostic accuracy of D-dimer as a rule out test of aortic dissection. In total they found 5 studies including a total of 1600 patients. My first point of concern is that 1035 of those patients came from a single study, which could potentially dominate a meta-analysis, and that study was not designed to test the accuracy of D-dimer. In fact, the study enrolled 1455 patients, but only 1035 were counted in this meta-analysis, because the other patients never even had a D-dimer drawn. [This is the Nazerian study above.] The results they present are pretty impressive, with a pooled sensitivity of 98% (95%CI 96-100%), specificity of 42% (95%CI 39-45%), negative likelihood ratio of 0.05 and positive likelihood ratio of 2.11. However, I would be very careful interpreting those results. Not only are the majority of the patients from a registry where D-dimer didn’t have to be drawn, but these were almost all patients admitted to CCUs, so very different from our ED population. Finally, although you would be using this test to try to avoid CTs, the poor specificity in a lower risk population could actually paradoxically lead to increased CT usage, much like D-dimer for PE. Bottom line: This study isn’t enough to support D-dimer to rule out aortic dissection in the ED.
Overall Bottom Line
The easiest bottom line is that based on the current literature D-dimer should not be used in the evaluation of aortic dissection. Although the numbers look promising in some studies, methodologic issues mean that both sensitivity and specificity are probably overestimated. Even without the methodologic issues, the low specificity and imperfect sensitivity mean that it is unclear if D-dimer use would benefit patients. A prospective RCT would be the only way to accurately sort out the true harms and benefits.
I think the data we have is highly likely to be biased. However, let’s consider the implications this data assuming that is it accurate. Using the Asha 2015 meta-analysis numbers, D-dimer has a negative likelihood ratio of 0.05, which is below the cutoff of 0.1 we like to see in tests we are using to rule out disease. But what does that number actually mean? In these studies, the rule in rate for aortic dissection was high – often between 25 and 50%. If we assume a 25% pre-test probability, the negative likelihood ratio of 0.05 results in a post-test probability of 2%. Is that good enough? Are we willing to miss 2% of aortic dissections? Even more concerning, the positive likelihood ratio of 2.1 means that for every 1000 patients we see, there will be 348 false positives using D-dimer.
However, most of us aren’t seeing rule-in rates anywhere close to 25% for aortic dissection. Let’s look at how the same numbers apply to a population with a pretest probability of 5%. In this population the D-dimer would effectively rule out the disease (with a post test probability of 0.2%), although we would still miss 1 dissection for every 1000 patients we see. However, in this population there would be 440 false positives for every 1000 patients, meaning that we have to image a lot of patients. If applied to patients you weren’t already going to image, imaging rates would increase.
Even with those numbers, it is not clear that D-dimer testing will benefit patients, and I would want to see a RCT before implementing the practice clinically. However, we have every reason to believe that the numbers I used are unrealistically good. These numbers come from studies with highly selected populations – either in coronary care units or with very high rule in rates. Furthermore, these numbers are almost certainly inflated due to verification bias, in that not all patients received the gold standard CT to confirm their diagnosis. A number of the studies here (Hazui 2005, Hazui 2006, Paparella 2009) demonstrated significantly lower sensitivities. Although it is not clear why these populations were different, they could hint at the selective workups and verification bias present in the other studies.
There is one key distinction between the use of D-dimer in pulmonary embolism and aortic dissection: the base rate of disease. This incidence of aortic dissection is approximately 0.5-5/100,000 people/year. Compare that to PE at 100/100,000 people/year and the order of magnitude difference completely changes the game. PE is common enough that we can develop a gestalt about the pretest probability. It’s not clear that the same is true for aortic dissection. We know humans aren’t good at handling small numbers and our bias is to significantly overestimate low probabilities. (Admit it – you have bought a lottery ticket in the past.) More importantly, the incredibly low base rate means that the chance of harm from an increase in imagining (when the only reason we want to use D-dimer is to avoid CT) is significantly higher.
Reading through this literature I am left with the distinct impression that aortic dissection is incredibly easy to diagnose, when it is diagnosable. What I mean by that is that there seem to be two patient populations: the classic presentation (easily diagnosable, as evidenced by the incredibly high rule in rates of these studies) and the atypical presentation. An atypical presentation of a very rare disease simply can’t be diagnosed given our current knowledge and technology. Or more accurately, it can’t be diagnosed without causing more harm than good. Sure we could CT everyone with a chest, and order serial D-dimers, but because this disease is so rare to being with, the harms of testing and false positives would vastly outweigh the rare true positives.
This probably isn’t the last we’ll hear about this, but for now I am convinced that there is currently no role for D-dimer in the work up of aortic dissection.