Why We Do What We Do: Limitations of the PERC Score in PE – Dr. Jason West

Home / Respiratory / Why We Do What We Do: Limitations of the PERC Score in PE – Dr. Jason West

When the PERC Rule Fails      

  Kline et al (1) developed a clinical decision tool based on parameters that could be obtained from a brief initial assessment to reasonably exclude the diagnosis of pulmonary embolism (PE) without the use of D-dimer in order to prevent unnecessary cost and the use of medical resources.  Even our interns have used the PERC rule by now, but we should be clear on what it includes: The PERC rule includes age < 50, HR < 100, oxygen saturation on RA > 94%, no prior history of DVT or PE, no recent trauma or surgery, no hemoptysis, no exogenous estrogen, and no clinical signs suggestive of DVT. 

 A review and meta-analysis published in the Annals in 2012 (2) found 12 qualifying studies evaluating the PERC rule and ultimately determined that the pooled sensitivity to rule out pulmonary embolism is 97.2%, which the authors concluded to be a low, but acceptable sensitivity to rule out PE without further testing.  The overall proportion of missed PEs was 0.32% (44 of 13,855 total cases).

 So who are the CT-PE or V/Q positive patients who could have been falsely “PERC ruled out?”

 In other words…when does the PERC rule fail?

 Kline reworked the data from a previous paper showing the outcomes of patients who presented to the ED and were diagnosed with PE (3), and used it as a dataset to determine the characteristics of patient who received an ED diagnosis of PE, but could have been included in the PERC clinical decision tool.  The initial study was used to determine that the overall morality attributed to PE was 1%, the mortality from hemorrhage was 0.2%, and the all cause 30 day mortality was 5.4%.  In the reworking of this dataset of 1,880 patients, Kline et al (4) found that 114 would have been included in the PERC rule should it have been applied. 

 Of these patients, they found that only 3 variables that demonstrated a true difference in the proportions between those who would have been included within the PERC rule and those who would have not been able to be “PERC ruled out:” pleuritic chest pain, pregnancy, and postpartum status.  Specifically, pleuritic chest pain, which is not included in any clinical decision rule to risk stratify potential PE patients, was found in 56% of the patients where the PERC rule would have failed. Although the N numbers were small for pregnancy and postpartum status, they concluded that the PERC rule should not be used in isolation to rule out PE in patients who are either pregnant or postpartum.  Again, it should be noted that pleuritic chest pain is not a component of either the Wells PE or revised Geneva score for PE. 

 And the reason for its absence in these scores could be considered questionable.  A large study including nearly 8,000 patients of whom 7.2% had PE by Courtney et al published in the Annals in 2010 (5) was designed to study the variables commonly believed to modify the pretest probability of PE and those already within the existing pretest probability scores.  The odd ratio (OR) for pleuritic chest pain in patients diagnosed with PE was 1.53, which seems weak in comparison to the ORs for history of PE and unilateral leg swelling, which are 2.9 and 2.6, respectively.  However, the ORs for hemoptysis and tachycardia (defined in this study as a pulse of > 94) are 0.78 and 1.52.  Both of these factors are included in the Wells PE score and the revised Geneva score.  Excluding hemoptysis and tachycardia, however, all variables used in the Well PE score have higher ORs than pleuritic chest pain.  The next closest OR of the variables included in the Wells PE score is immobilization with an OR of 1.72.  The authors also found that the other two variables not included in clinical decision rules with useful ORs were a personal history of non-cancer related thrombophilia (OR 1.99) and a family history of PE (OR 1.51).

 It is important to note that the PERC rule was never intended to be applied to anything but a low risk group of patients determined either by clinical gestalt or by the Wells PE score, and this point has been stressed in commentary (6). Only after knowing and applying the Wells PE score, an alternative method of risk stratification, or your clinical gestalt should you consider the PERC Rule in a patient you believe is at low risk for PE.  As our Monte Chairmen, Dr. Gallagher, has pointed out several times, if we believe that our patient population in the “Bx boro” has a higher prevalence of both DVT and PE than the general population in which these rules were derived, then our use of these decision rules, however well-validated in the literature, should be employed with some hesitance.  In fact, the meta-analysis found some heterogeneity in the PERC rule sensitivity to exclude PE.   Two studies from European populations with a prevalence of PE ranging from 21-30% (7,8) found that a negative PERC rule combined with the low risk Revised Geneva Score only reduced the prevalence of PE in the studied patients to 6%.  Only in one of these studies (7), did the PERC rule combined with clinical gestalt reduce the prevalence of PE down to nearly zero.

 Take home points:

  •  The PERC rule cannot be a substitute for gestalt.
  •  Gestalt or some form of risk stratification should be employed first before using the PERC rule.
  • The only evidence we have about PERC rule inclusive CT-PE or V/Q positive patients suggests that 56% of those will have pleuritic chest pain, which is not in a validated clinical decision rule despite having a higher OR for PE than hemoptysis and recent immobilization.
  •  The PERC rule should not be used in isolation to rule out PE in pregnant or postpartum patients.

  Jason West MD


 1). Kline JA, Mitchell AM, Kabrhel C, et al. Clinical criteria to prevent unnecessary diagnostic testing in emergency department patients with suspected pulmonary embolism. J Thromb Haemost. 2004;2:1247-1255.

 2). Singh, et al. Diagnostic Accuracy of Pulmonary Embolism Rule-Out Criteria: A Systematic Review and Meta-analysis. Ann Emerg Med. 2012;59:517-520.3).

 3). Pollack CV, et al. Clinical Characteristics, Management, and Outcomes of Patients Diagnosed With Acute Pulmonary Embolism in the Emergency Department Initial Report of EMPEROR (Multicenter Emergency Medicine Pulmonary Embolism in the Real World Registry). J Am Coll Cardiol. 2011 Feb 8;57(6):700-6

 4). Kline JA, et al. Clinical Features of Patients With Pulmonary Embolism and a Negative PERC Rule Result. Ann Emerg Med. 2013 January 60(1): 122-124.

 5). Courtney DM, et al. Clinical features from the history and physical examination that predict the presence or absence of pulmonary embolism in symptomatic emergency department patients: results of a prospective, multi-center study. Ann Emerg Med. 2010 April 55(4): 307–315.



  • alvarezzy

    there’s a nice debate between david newman (smartEM) and scott weingart (EMcrit) regarding PERC alone and well’s/geshtalt + PERC + d-dimer, etc… http://emcrit.org/blogpost/a-debate-on-pe-decision-rules/

    Also, in June 2013, Jeff Tabas (UCSF) has a great segment on EMrap, and reminds us which pretest probability cohort Kline (PERC) used: “researchers define low risk as up to 15% chance of PE. Moderate risk is 15-­‐40%.” This is clearly not the same as our generally acceptable risk tolerance in ED, which is 2% miss rate. If you use MDcalc to calculate the perc rule, they also quote this 15%. We should be more liberal with applying PERC rule to the intended low risk patients (15% chance of having PE) and not only to the what we commonly would attribute as “low risk”.


  • Thanks for the commenting, LA.

    I look forward to listening to the EM-RAP post.

    However, I disagree with you that at 15% pre-test probability is not the same thing “as our generally acceptable risk tolerance in ED, which is a 2% miss rate.” In fact, I’d argue that it’s the SAME thing.

    If you bust out your Bayes Nomogram, you should see that a 15% pre-test probability would need around a negative likelihood ratio of about 0.2 or so to help you get down to a post-test probability of less than 2%. If you believe the meta-analysis is correct in identifying the sensitivity of the PERC rule, then you should also believe in its negative likelihood ratio. The meta-analysis by Singh et al I referenced and another by Carpenter of Wash U both determined the pooled negative likelihood ratio to be a little less than 0.2. I’d say the “low probability” as defined by 15% pre-test probability is an excellent match for the PERC in the general setting.


    • alvarezzy

      hey jason..

      good point, though i doubt anyone is grabbing the bayes nomogram when they’re deciding their choice of diagnostic tests.

      i think we’re ultimately talking about the same thing, but I do welcome the counterargument. I do enjoy the brief review of biostatistics.

      what i was merely highlighting is the failure to stop the workup for PE (and therefore end up doing MORE CTA thorax) is because clinicians are reluctant to use the PERC test because what they perceive as low prob is not the same as what the initial researchers of the PERC rule used, which is 15% (and as you pointed out, have been well validated). The same goes with the d-dimer and it’s NPV.

      If you don’t believe the above statement that the “general acceptable miss rate of 2%” is NOT the same as the 15% that researchers use for low pretest prob, in the next week, ask around the ED (monte and jacobi) and talk to your attendings. Statistically, if you use the bayes nomogram, in this discussion, the NPV should still move your post test prob to a very low number, and I agree with you. But for an exercise of “Why we do what we do,” try asking different attendings around and see if you get the same response (without you initially busting out the bayes nomogram).

      I’m sure you also know that your “seasoned” attending’s pretest probability will be different from yours, simply because of their experience and the number of patients they’ve seen/better feel of the prevalence of the disease in the area.


Leave a Reply