Category Archives: Prognostication

Transient Hypotension in the Emergency Department

An interesting technicality in the use of the PERC rule to rule out pulmonary embolism is the tachycardia component — it asks not whether the patient is tachycardic at the time of the application of the rule, or whether tachycardia was sustained throughout the emergency department stay, but instead whether the patient had (as described by Jeff Kline in his great review article on PE diagnosis and risk stratification): “3. Pulse <100 beats/min during entire stay in ED”.  Meaning, even transient tachycardia may suggest a life-threatening diagnosis, even if it resolves while the patient is in the emergency department, and we’re probably PERCing out a whole bunch of patients inappropriately, at least according to Kline (who, notably, testifies a whole bunch as an expert witness in cases of missed pulmonary emboli).

I recently had a handful of patients in whom concerning blood pressures were measured and documented, which then resolved when vital signs were re-checked or after a small quantity of fluid or repositioning. I was wondering whether anyone had looked at the prognostic significance of ED hypotension, and whether these momentary dips in blood pressure should be something that concerns me. I did a quick search and found two studies that addressed this question in two different populations:

First we have, from the Rick Bukata school of title writing: “Emergency department hypotension predicts sudden unexpected in-hospital mortality: A prospective cohort study.”  This study, by Alan Jones and Jeff Kline out of (and formerly out of) Carolinas, prospectively enrolled 4,790 adult ED patients admitted to the hospital for reasons other than trauma. Patients were divided into those with and without systolic BPs below 100 mmHg at any time during their ED visit and followed through their hospitalization for the primary outcome of in-hospital mortality. Secondary outcomes included “sudden and unexpected death”, the relationship between the degree and the duration of hypotension measured and mortality, and the test characteristics of hypotension as a test for predicting in-hospital mortality.

Their conclusions are illustrated well in this graph:


As they concisely summarize in the article’s conclusion:

Patients exposed to hypotension had a threefold increased risk of in-hospital death and a 10-fold increased risk of sudden, unexpected in-hospital death. Patients with any one SBP < 80 mm Hg had a sixfold-increased incidence of in-hospital death, and patients with a SBP < 100 mm Hg for > 60 min had almost a threefold-increased incidence of in-hospital death.

The second article from the same group echoes this conclusion in a different population of patients. This article, “The significance of non-sustained hypotension in emergency department patients with sepsis” is a secondary analysis of the above data set which looks specifically at the prognostic value of non-sustained hypotension defined as one or more occurrence of SBP < 100 mmHg in patients with sepsis as defined by the receipt of antibiotics in the ED + at least two SIRS criteria.

774 patients met their inclusion criteria for sepsis, and after 74 were excluded for “overt shock” (sustained hypotension or use of pressors). They examined the remaining patients for a primary outcome of in-hospital death.  They found, as one might expect, that hypotension predicts worse outcomes in this sub-population of patients — including when patients had non-sustained hypotension. Again, there seemed to be a “dose-dependent” relationship, with an inverse relationship between the nadir of the ED SBP and the frequency of in-hospital death, as shown here:


Another important finding (though taken in context of a fairly small sample) was the statistically similar incidence of the primary outcome in both the groups with transient and sustained hypotension. Both groups of patients had a 2.5-3x higher risk of in-hospital mortality when compared to patients without any hypotension.

Without belaboring the point, these two studies underscore the prognostic significance of even transient hypotension in the undifferentiated emergency department patient, and (as is better known to have implications in terms of severity) in patients diagnosed with sepsis. Like the previous post regarding lactate, or the well-known pearl about tachycardia at discharge, this is a number that should get your attention and which demands evaluation and possible intervention / escalation of care.


Marchick MR1, Kline JA, Jones AE. The significance of non-sustained hypotension in emergency department patients with sepsis. Intensive Care Med. 2009 Jul;35(7):1261-4. PMID: 19238354. [PubMed] [Read by QxMD]
Holler JG1, Bech CN1, Henriksen DP2, Mikkelsen S3, Pedersen C4, Lassen AT1. Nontraumatic hypotension and shock in the emergency department and the prehospital setting, prevalence, etiology, and mortality: a systematic review. PLoS One. 2015 Mar 19;10(3):e0119331. PMID: 25789927. [PubMed] [Read by QxMD]

D-Dimers & Dissection

A recent patient I saw in the emergency department was a fifties year-old woman with a family history of aortic dissections presenting with “chest pain” per the triage note. On my history and exam, she more endorsed vague neck and epigastric discomfort (which had now resolved), and had no other classic findings for a dissection (e.g. hemodynamic instability, asymmetric pulses or blood pressures, abnormal neurologic findings, etc.). She also had a a normal chest x-ray and a negative initial workup for ACS, including a normal ECG and undetectable troponin. In terms of other life-threatening diagnoses, she did not PERC out, and had a Wells score that suggested the D-Dimer would be an appropriate test to rule out pulmonary embolism.

When I discussed with her the potential utility of getting a CT scan of her chest to evaluate for an aortic dissection — she asked me about how much radiation exposure this involved, and shared her (valid and very appropriate) concerns about getting too much radiation. She had many CT scans for various reasons over the years she felt, and did not want any additional unnecessary radiation.

I talked to her more about this and tried to start some shared decision making by sharing a favorite infographic of mine about radiation amounts in diagnostic imaging, and (to myself) pondered a clinical question: If the D-dimer test was low, did that along with the low-ish pretest probability, safely decrease the likelihood of dissection enough to forego a CT scan? There is an emerging literature on the use of dimer testing to rule out aortic dissections, but how good is it? Do you use the same cut-off as in pulmonary embolism? Should that cutoff be age adjusted? And what are the test characteristics in this context? I had no idea, so that’s what today’s post-didactics reading was about.

I read through “A Systematic Review and Meta-analysis of D-dimer as a Rule-out Test for Suspected Acute Aortic Dissection” by Asha et al., which reviews the work of 30 studies and combines the data for 4 studies using a standard cutoff of 0.50 μg/mL to estimate sensitivity, specificity, and positive and negative likelihood ratios of a D-dimer. As the abstract conclusion reads:

“Overall, sensitivity and negative likelihood ratio were 98.0% (95% confidence interval [CI] 96.3% to 99.1%) and 0.05 (95% CI 0.03 to 0.09), respectively. These measurements had little statistical heterogeneity. Specificity (41.9%; 95% CI 39.0% to 44.9%) and positive likelihood ratio (2.11; 95% CI 1.46 to 3.05) showed significant statistical heterogeneity. When applied to a low-risk population as defined by the American Heart Association (prevalence 6%), the posttest probability for acute aortic dissection was 0.3%.”

So there you have it. Obviously, there’s more to it, and the actual paper is worth reading — it discusses some of the drawbacks of the included studies, specifically unanswered questions about bias and the generalizability to ED populations given the high prevalence of disease in the included cohorts. Limitations aside, basic conclusion that was in low risk patients, a negative D-dimer confers an even lower risk of acute aortic dissection, and it may be reasonable (don’t you love that phrase?) to consider using this result to inform your decision-making regarding the utility of imaging. Of course, one must also consider the rate of false positives, and the potential harms of resultant downstream testing as has been discussed regarding testing for PE.

I think that one of the more important (and potentially easily-overlooked, as when it comes to all clinical decision tools or supports, or anything that serves as a Bayesian modifier) points I took away from this review is that while this is a potentially useful test in this context, pretest probability matters. As the abstracts of some of the included studies say: “When applied to a low-risk population…”, “…in patients with low likelihood of the disease”,  “…the presence of ADD risk score 0 or ≤ 1 combined with a…” and so on. You should only really hang your hat on a negative dimer assay when you think the probability is low in the first place. Another question to consider though, is how low is the pre-test probability to suggest you *shouldn’t* order a dimer to r/o dissection? And how many people with potential dissections that might be caught and thereby managed earlier PERC’d out of receiving a test that might reveal this diagnosis (though might also subject them to an unnecessary scan for PE)?

As the full text of the article states:

It would be pertinent to comment on the many case reports of patients with confirmed acute aortic dissection but a negative D-dimer result. It should first be recognized that these cases did not have a risk-stratification applied and also that no test, no matter how good, including the reference standards for the disease, has 100% accuracy. These cases mostly represent a subgroup of patients with a thrombosed false lumen or an intramural hematoma who seem particularly likely to have a lower or negative D-dimer result. The studies in this meta-analysis included such patients, which means that the high sensitivity and excellent negative likelihood ratio were achieved with the inclusion of these problematic cases.

It is always worth remembering that rare diseases are rare, and that in a patient with a low pretest-probability of having a disease, any test can be construed as to have a high sensitivity when applied to the wrong population. For instance, I can figure out who is low risk for aortic dissection in most chest pain patients with the “Bryan” rule — I just ask if their name is Bryan, spelled with a y. If negative, they are extremely unlikely to have an aortic dissection. Of course, if they do, my test will likely miss them but the point remains. In the patient described above, even though the D-dimer was negative, this patient was not low risk by the fairly-conservative AHA acute aortic dissection risk score (pictured below), and therefore the sensitivities and specificities cited in the articles presented in this meta-analysis don’t apply to their case.


In cases where acute aortic dissection is suspected as a likely potential diagnosis, a D-dimer is probably not an appropriate test to replace definitive diagnostic imaging of the aorta–  specifically, as stated by previous guidelines from the AHA: computed tomography (CT), magnetic resonance imaging (MRI), or transesophageal echocardiography. Let this inform your discussions of shared decision making in the emergency department, and document accordingly, and hopefully you’ll be able to adopt a strategy to help everyone sleep better at night.


Nazerian P1, Morello F2, Vanni S1, Bono A3, Castelli M1, Forno D3, Gigli C1, Soardo F3, Carbone F3, Lupia E3, Grifoni S1. Combined use of aortic dissection detection risk score and D-dimer in the diagnostic workup of suspected acute aortic dissection. Int J Cardiol. 2014 Jul 15;175(1):78-82. PMID: 24838058. [PubMed] [Read by QxMD]

Post-Arrest Prognostication

While I want to focus this blog on things relevant to practice in the Emergency Department, I have an academic interest (and maybe a career interest long-term) in critical care. I also feel that cardiac arrest is a particular area in critical care should be something that EPs are expert in — it’s also an area in which there is considerable nihilism which may lead in sub-optimal patient care, or early withdrawal of efforts before such withdrawal is justifiable.

What do I mean by nihilism? I mean that we in the ED rarely see good outcomes in out-of-hospital cardiac arrest (OOHCA) patients (and when we do, they’re often comatose and whisked away to the ICU, which means that even if they *do* have a good clinical outcome we do not see it happen and rarely even hear about it), and this leads to a sentiment that any cardiac arrest patient is bound for either death or a meaningless life due to neurologic injury.

Everyone in in the department, from patient care assistants and techs and medical students to the attendings, puts a lot of energy and effort into running codes and trying to resuscitate these patients. People care a lot and do some of their best work in these stressful contexts. But at the same time, I sometimes wonder whether we would focus more on improving our process and quality of care– and perhaps thereby do even better– if we had a better sense that our interventions translated into patients who could again be alive and well because of them. This sense is difficult to come by if many of the patients that you successfully attain ROSC on have features that many people associate with a very low likelihood of meaningful recovery.

This pair of recent review articles focused on prognostication in post-cardiac arrest patients — findings on clinical exam, imaging, and other methods to try to suss out who will go onto do well and who is unlikely to ever regain meaningful neurologic function. As ICU bed availability dwindles and the incidence of cardiac arrest and survival thereof continues to increase, this will be of increasing relevance to ED docs, intensivists, and those working with these patients.

So what is a “good outcome”? As the article says, “Experts in coma prognositication defined outcome by cerebral performance categories (CPCs; CPC 1 back to baseline, CPC 2 moderate impairment, CPC 3 severe impairment, CPC 4 vegetative or comatose, CPC 5: dead).” They bifurcate these into either a good (CPC 1 or 2) or poor (CPC 3-5) outcome. Obviously the difference between “moderate” and “severe” impairment is somewhat subjective, but there are additional tools used to help with this distinction.

The old standard was clinical assessment of brainstem reflexes, the response to pain, and the absence or presence of myoclonus during the first 72 hours post-arrest. In the TTM era, this becomes trickier because temperature management and the required sedation can alter these features, though the bedside exam still has significant prognostic significance. Absence of pupillary reflexes at 72 hours is the best bedside predictor of a bad outcome, with a false positive rate (FPR) of only 0.5% — presence of pupillary reflexes however, does not confer a good outcome, given that it only has a PPV of 61% (95% CI 50-71).

What about earlier? In the first 24 hours post-arrest, particularly in hypothermic patients, ~ 8% of patients without pupillary reflexes will go on to have a good recovery — so don’t count them out. In terms of corneal reflexes, the reliability is less than that of pupillary reflexes but their absence still correlates with a poor prognosis, with an FPR of 5%.

Motor response is the most affected by sedatives, opiates, and neuromuscular blockade — all common in patients undergoing TTM, and absent or extensor responses to painful stimuli at 72 hours had a FPR of 24%. To reliably utilize this for prognostication, you need exclusion of residual effects of sedation, which can be extended beyond when the drips are simply turned off secondary to the effects of TTM and also the effects of reduced clearance due to shock liver, renal dysfunction, or both.

In terms of myoclonus, which is classically associated with poor outcomes, ~ 9% of patients with myoclonus may survive, according to the data presented here. As the article states, myoclonus is somewhat of a nebulously defined entity — “Not all so-called twitches have the same prognostic implication, rather their usefulness in predicting prognosis depends on semiology, duration, and associated EEG findings.”

I’ll skip EEG and ERPs because this is already too long, but suffice to say they’re useful after hypothermia and for ruling out sub-clinical status epilepticus, which is something we really want to avoid happening in our post-arrest patients, but is very common. More to come on this, which I feel is of particular relevance to us in the ED. Same goes for biomarkers such as neuron specific enolase and Serum S-100B, which can both be measured and trended as the “troponins of the brain”, so to speak.

In terms of imaging — CT scan of the head is recommended in patients in whom there is not another obvious cause of cardiac arrest, to evaluate for a bleed or ischemic stroke. Evaluation of gray:white ratios can predict poor outcomes, but is less reliable than clinical exam and EEG, and this is true for MRI as well, though again MRI does not add very much prognostic capability beyond what can be achieved with bedside tests and the logistics and cost associated with MRI scans of every comatose survivor of cardiac arrest make this somewhat limited in utility.
PPV for Neuro Findings

So what’s the takeaway from all this? Basically, reliable prognostication after cardiac arrest is hard, but at the same time, it isn’t– don’t do it right away, and if you do, it shouldn’t necessarily be based on your bedside neurologic exam. There are tools that can give us useful information, but rarely certainty, to guide conversations with family. And the reality is that none of them are accurate enough inside the first 48-72 hours, especially in patients who are being cooled. There is a very powerful desire to be able to give families hope, or to caution against hope in a way that changes outcomes before they’ve happened — in my very early-in-development opinion, all you can really tell them is something I heard one of my mentors say to families whose children were in the Pediatric ICU: “Prepare for the worst, and hope for the best.”

I also take away from this that nihilism is an un-useful form of prognsotication in these patients — I have seen patients myself who had unreactive pupils or myoclonic jerks, who went onto walk out of the hospital, fairly neurologically intact. This is even more true if the arrest was witnessed, was a shockable rhythm such as VT or VF, and if the patient received high-quality chest compressions and early defibrillation, preventing lengthy low/no-flow states to the brain.

The message not to take away from this post that I believe in any sense that there is no ability to meaningfully make predictions about the likely outcome of cardiac arrest patients, whether or not you’ve gotten ROSC — there are many other variables not considered in the above article  that predict do reliably predict outcomes such as comorbidities, age of the patient, how long they were down for, the initial rhythm, and an often-overlooked variable in the literature (because it’s tough to quantify): consideration of their quality of life before they suffered a cardiac arrest. I also think that the pragmatic realities of cardiac arrest care — an emotionally charged event where patients are often teetering along a line between life and death, and where decisions have real and immediate impacts on that outcome– may require a sense of somewhat-morbid realism when the outcomes are often so dismal. I just hope that when people are making decisions about termination of efforts (or withdrawal of care post-ROSC) they’re considering all of these things and more, beyond just what their clinical gestalt is.

More to come, I’m sure — I’m especially interested in what happens moving forward in terms of biomarkers, cerebral oximetry, and near-infrared brain imaging to try to determine cerebral oxygenation and metabolism without having to move patients out of the ICU.


Rossetti AO1, Rabinstein AA2, Oddo M3. Neurological prognostication of outcome in patients in coma after cardiac arrest. Lancet Neurol. 2016 Mar 23. PMID: 27017468. [PubMed] [Read by QxMD]
Sivaraju A1, Gilmore EJ, Wira CR, Stevens A, Rampal N, Moeller JJ, Greer DM, Hirsch LJ, Gaspard N. Prognostication of post-cardiac arrest coma: early clinical and electroencephalographic predictors of outcome. Intensive Care Med. 2015 Jul;41(7):1264-72. PMID: 25940963. [PubMed] [Read by QxMD]

Pulmonary Embolism in Pregnancy

The diagnosis of pulmonary embolism in pregnant patients is one made difficult by many factors, including a normal elevation in serum d-dimer levels (see below) as well as the additional concern regarding exposure of a developing fetus to the high levels of radiation and contrast associated with CT pulmonary angiography. It is well-known that exogenous estrogen is a risk factor for thromboembolic disease, and while it seems from the data discussed below that pregnancy is not as scarily-high-risk for PE as we might think, we certainly know that pregnancy is a time when homones are running high Add to this the fact that in pregnancy, women are both tachypnic and tachycardic due to normal changes in cardiovascular and respiratory physiology — making a clinical diagnosis that much more difficult.

In these sequentially-published review articles by the PE guru Jeff Kline et al., the authors review the diagnostic dilemma presented by these patients and present the following algorithm:

Microsoft Word - jem_10231_JEM10231.edt

Note the inclusion of the trimester-stratified quantitative d-dimer for patients without a high pretest probability who are PERC negative — this goes against the conventional wisdom that the d-dimer is a worthless test in pregnant women due to the normal elevation found intrapartum. Similar to the way we have begun “age-adjusting” the threshold value of the quantitative d-dimer in non-pregnant patients, they propose that the threshold be “adjusted according to the trimester of pregnancy, as follows: first trimester, 750 ng/mL; second trimester, 1000 ng/mL; third trimester, 1250 ng/mL (assuming a standard cutoff of 500 ng/mL). If the patient has a non-high-pretest probability, has no high-risk features, is PERC negative, and the bilateral ultrasound is negative, and the D-dimer is below the trimester-adjusted values, PE can be ruled out to a reasonable degree of medical certainty.”

They acknowledge the limitations of this approach, including that it hasn’t been prospectively validated, and they do not present any data showing its performance as they’ve been using it, but in cases like this expert opinion is the best we have (so far). He discussed this approach on an episode of ER Cast, and explains it a little bit more in terms of the integration into clinical practice, as well as the role that gestalt can play in risk stratification. 

What I found interesting about this was the idea that the post-partum period is the most risky period of time for women in terms of pulmonary embolism — this echoes what we know about cardiovascular disease in the post-partum period, i.e. when women are autotransfused and their cardiopulmonary physiology is rapidly and massively altered, this presents the highest risk in terms of women with heart failure, valvular abnormalities, or disease entities like peripartum cardiomyopathy. According to the data presented by Kline et al, while the risk increases throughout a pregnancy, 70% of all peripartum PEs occur post partum, and the risk during pregnancy is low (OR 0.4-0.8, depending on trimester) — though, as the authors note, this may not actually reflect that pregnancy is protective against PE but instead suggest that we overtest women for pulmonary embolism during pregnancy, perhaps because of the clinical changes described above. The also cite a large meta-analysis of 23 epidemiologic studies that found PE occuring in only 3 of 10,000 pregnancies.

Another thing that stood out to me while reviewing this article was that for a patient to PERC out of these algorithms, their vital signs must be normal throughout their entire ED stay — normalization of vital signs during an ED visit does not lower the risk of PE, as specifically stated by the authors.



Kovac M1, Mikovic Z, Rakicevic L, Srzentic S, Mandic V, Djordjevic V, Radojkovic D, Elezovic I. The use of D-dimer with new cutoff can be useful in diagnosis of venous thromboembolism in pregnancy. Eur J Obstet Gynecol Reprod Biol. 2010 Jan;148(1):27-30. PMID: 19804940. [PubMed] [Read by QxMD]
Kline JA1, Williams GW, Hernandez-Nino J. D-dimer concentrations in normal pregnancy: new diagnostic thresholds are needed. Clin Chem. 2005 May;51(5):825-9. PMID: 15764641. [PubMed] [Read by QxMD]

Duration of symptoms of respiratory tract infections in children

From the BMJ, we have a very interesting systematic review evaluating the duration of symptoms in children seen in the ED (or A&E, if you will) for fairly minor complaints: otitis media, acute cough, sore throat, and common cold. In my time in the pediatric ED, I’ve noticed that a not-insignificant number of visits are repeat visits for persistent symptoms in well-appearing children who were seen and discharged from the ED within the last week or so. The parents are often concerned that the cough has still not gone away, or that the child’s breathing at night still sounds funny to them — these are not different symptoms than the child was originally evaluated for, but I thought it was possible that better anticipatory guidance in terms of the duration of symptoms parents could expect might result in fewer of these “bounce

So what did the authors at BMJ find? In 90% of children, earache was resolved by seven to eight days, sore throat between two and seven days, croup by two days, bronchiolitis by 21 days, acute cough by 25 days, common cold by 15 days, and non-specific respiratory tract infections symptoms by 16 days.

21 days of cough for bronchiolitis and 25 days for non-bronchiolitis URIs? That is way longer than what I hear when parents are being discharged — I am no less guilty of underselling the duration of symptoms than others. It’s a tough question to answer, right? “How much longer will this last?” — Prognostication is always the hardest part of medicine, whether you’re talking to the dying cancer patient or to the parents of the child with the perpetually stuffy nose and inflamed upper airways. Well, I personally intend to try to provide parents with a more evidence-based answer for the rest of this season– something along these lines: “Longer than you can possibly imagine. Most kids will have a cough for three weeks or more, and many will seem like they go the entire winter without getting better. But as long as they’re eating, drinking, pooping, peeing, moving about and more or less acting like a slightly-more-congested-and-therefore-irritable version of themselves, that’s okay!”

It’s a tough balance. You wouldn’t want to dissuade parents from seeking medical attention (ideally from their PMD) if the child doesn’t get better in a reasonable amount of time, but it’s very difficult knowing what that time is for them. Moral of the story: encourage that follow up visit with the PMD, and make sure to give thorough and explicit return precautions accounting for the myriad reasons we *do* need to see these patients back ASAP.


Thompson M1, Vodicka TA, Blair PS, Buckley DI, Heneghan C, Hay AD; TARGET Programme Team. Duration of symptoms of respiratory tract infections in children: systematic review. BMJ. 2013 Dec 11;347:f7027. PMID: 24335668. [PubMed] [Read by QxMD]