Reference: Kline, J. A., Mitchell, A. M., Kabrhel, C., Richman, P. B., & Courtney, D. M. (2004). Clinical criteria to prevent unnecessary diagnostic testing in emergency department patients with suspected pulmonary embolism. Journal of Thrombosis and Haemostasis, 2(8), 1247-1255.
Clinical Question: Can the Pulmonary Embolism Rule Out Criteria (PERC) be used to rule out pulmonary embolism (PE) at the bedside without additional ancillary testing such as D-dimer or computerized tomography (CT) imaging?
Derivation and Validation
Derivation: The goal was to create a set of criteria that, when all were absent, would confer a pretest probability so low that further testing (including D-dimer) would be more likely to cause harm than benefit. In other words, the risks would outweigh the benefits.
Population: 3148 emergency department patients who underwent evaluation for possible PE from 10 different hospitals.
21 variables extracted from the initial population:
6 were excluded in the initial logistic regression due to nonsignificant relationships.
2 more variables were excluded due to low interobserver reliability.
The logistic regression was re-executed using the remaining 13 dependent variables; at which time gender, dyspnea, and systolic blood pressure were excluded for non-significance.
Cough and asthma/COPD/wheezing were excluded due to being negative predictors (so if they were absent then it would increase the pretest probability which is the opposite of what they wanted).
8 variables were included that, when absent, should create a pretest probably so low that the risks of further testing outweigh the benefits.
Validation: Kline et al. tested the derived criteria in two ED populations: low-risk patients and very low-risk patients. They used low-risk patients to generate a sensitivity and very low-risk patients to test the feasibility (is the pretest probability ACTUALLY low enough to rule out PE without further testing?).
Population: 1,427 low-risk patients and 382 very lower-risk patients
The low-risk group were the patients who could have PE excluded with a D-dimer based on clinician gestalt.
The very low-risk group were the patients who could have PE excluded with no testing based on clinician gestalt.
Results:
Low-risk population: 1427 patients had a D-dimer ordered to rule out PE
362 patients were PERC negative.
5 had a PE.
357 did not have a PE.
False negative rate of 1.4%.
Sensitivity of 96%.
Very low-risk population.
57 were PERC negative.
None of the PERC negative patients had a PE.
False negative rate of 0%.
Sensitivity of 100%.
9 Patients were PERC positive and all had a PE.
Author Conclusion: “When all eight factors are negative, the pretest probability of pulmonary embolism is likely to be so low that D-dimer testing for pulmonary embolism will not yield a favorable risk–benefit ratio.”=
Hot take(s):
• Kline et al. literally write that the PERC rule can only have a false-negative rate below 1.0% in populations where the prevalence of VTE is less than 6%. HOW are we supposed to know what the prevalence of VTE is in our population?? They also conclude ‘it should be emphasized that the underlying prevalence of VTE may be too high in other ED populations for this strategy to be able to safely rule out PE.’ They literally suggest that a pilot-tested observational quality assurance study should be conducted prior to implementing the PERC rule in your ED. Have you done this?
• Hottest take: the original cohort of patients were classified as low-risk or very low-risk based on clinical gestalt. These patients were found to almost never have a PE (5 out of 419 patients). This confers a false negative rate of 1.2% and a sensitivity of 99% just using gestalt.
Practice your gestalt as a resident but recognize that the gestalt used in the study came from board board-certified emergency medicine physician.
Do not use PERC to convince yourself that it’s not a PE. Use your gestalt to rule out a PE, then apply PERC. The purpose of PERC was to show that the gestalt of a seasoned clinician is valuable.
Take Away: The PERC criteria can safely be used to generate a pretest probably for PE of less than 1% if all the criteria are negative in populations where the prevalence of PE is 6% or less.
Evidence-Based Medicine
Logistic Regression: Used to measure the strength of an association between a factor and an outcome. For the derivation of PERC, Kline et al. measured the strength between 21 variables (such as age, hemoptysis, syncope, etc) and the diagnosis of PE in 3,148 patients.
Logistic regressions can ‘adjust’ for confounding factors, which may distort the effect of the factors being measured. For example, if you are measuring the effect of alcohol consumption and lung cancer, you’d have to adjust for smoking because it is known to cause lung cancer.
Example:
In Table 1 from Kline et al., used for the derivation of PERC, you can see that (1) pleuritic chest pain, (2) substernal chest pain, (3) syncope, (4) current smoker, (5) prior or current malignancy, and (6) pregnancy or post-partum had p-values less than .05. They were then excluded from PERC because of this. This model was meant to identify low-risk and very-low-risk patients based on the provider’s gestalt. The 6 factors mentioned above essentially excluded patients from being low-risk or very low-risk. That is, if a patient had pleuritic chest pain, substernal chest pain, syncope, were current smokers, had prior or current malignancy or were pregnant/postpartum they were NOT LOW-RISK PATIENTS. Thus, they were not included in the PERC model.
It's important to note that the PERC rule CANNOT be applied to patients with these factors as they are NOT low-risk patients.
Summary: Kline et al. used a logistic regression to predict the outcomes of future patients based on their sample’s predictor variables.