We read with great interest the important paper by Riera-Mestre et al.  that reported an apparent increase in mortality among normotensive patients with acute pulmonary embolism (PE) who were treated with fibrinolytics. The authors used a propensity score to produce what is essentially a matched case–control study in which the cases received fibrinolytics and the controls did not. The method of matching employed a logistic regression equation with beta coefficients, reported in Table 4. After propensity matching, the authors found an increased risk of death among the cases receiving fibrinolytics (odds ratio [OR] 2.32; 95% confidence interval [CI] 1.15–4.68).
The multicenter Emergency Medicine Pulmonary Embolism in the Real world Registry (EMPEROR), is a registry of 1880 emergency department patients in the USA who were diagnosed with acute PE . This registry included 1740 patients who were normotensive at presentation. Forty-one of these 1740 patients received fibrinolytics, and four of the 41 (9.7%, 95% CI 2.7–23.7%) died within 30 days, as compared with 72 of 1699 (4.3%, 95% CI 3.4–5.4%) who died within 30 days but did not receive fibrinolytics.
Table 4 from Riera-Mestre et al. provides the coefficients for the logistic regression equation that they used to calculate the propensity score, and the authors kindly provided us with the y-intercept (intercept = − 3.2; personal email communication, D. Jimenez, 24 May 2012). With this information, we were able to reproduce the propensity score used by Riera-Mestra et al. We had two goals: (i) to assess the accuracy of this score in predicting the use of fibrinolytics in EMPEROR; and (ii) to calculate the OR for death at 30 days for normotensive patients in EMPEROR who received fibrinolytics, adjusted for the propensity score derived by Riera-Mestra et al.
For the first goal, we calculated the propensity score for each of the 1740 normotensive patients in EMPEROR by solving the logistic regression equation of Riera-Mestre et al., for the value P, the estimated probability of fibrinolytics being given We tested the ability of P to predict the use of fibrinolytics on the basis of receiver operating characteristic (ROC) curve analysis. The area under the ROC curve was 0.62 (95% CI 0.52–0.71). This finding indicates that the equation had a better ability than chance to predict the use of fibrinolytics in EMPEROR, and thus provides provisional evidence of the validity of the propensity score derived by Riera-Mestre et al.
For the second goal, we constructed a second logistic regression equation containing two independent variables: (i) the value P obtained from the propensity logistic regression equation; and (ii) a dichotomous variable indicating the presence or absence of fibrinolytics. The dependent variable for this equation was DEATH within 30 days. This equation yielded a propensity-adjusted OR for death after fibrinolysis equal to 2.66, with wide CIs (0.92–7.71) owing to the low number of deaths. Although not statistically significant, this finding tends to support the conclusions of Riera-Mestre et al.
Upon further analysis, we were surprised to find that all four normotensive PE patients given fibrinolytics who died in EMPEROR all died directly from PE, as opposed to hemorrhage. Indeed, none of these patients experienced any abnormal bleeding of any kind. Three of the four died in hospital, and the fourth died of PE after discharge. All four patients receiving fibrinolytics with systolic blood pressure (sBP) of > 100 mm Hg who died experienced respiratory failure requiring intubation before their death. Two who underwent echocardiography showed severe right ventricular strain. None of the four had venous ultrasonogaphy. All four patients who died had histories of hypertension, but only one had an sBP of > 140 mm Hg upon emergency department presentation.
These observations from EMPEROR lead us to speculate that factors not contained in the propensity score by Riera-Mestre et al. may have contributed to the patients’ increased risk of death independently of the effect of fibrinolysis. These factors could include the work of breathing resulting from respiratory distress, the size of the PE observed on imaging, or circulatory shock, as shown by hypotension relative to the patient’s baseline, a high blood lactate level, an elevated base deficit, or a low serum bicarbonate concentration. We could also speculate about a fibrinolytic-caused but non-hemorrhagic mechanism of death in patients with concomitant deep vein thrombosis (DVT). Clinicians may have been motivated to give fibrinolytics to patients with known DVT, recognizing this as an independent risk factor for death from PE . It could be reasonably hypothesized that, in the absence of a vena caval filter, fibrinolysis may increase the probability of detachment and embolization of residual, mobile DVT.
We respectfully ask Riera-Mestre et al. to report the cause and timing of death in their normotensive case and control patients, and how many had known concomitant DVT. If they find that most cases (patients treated with fibrinolytics) died of a non-hemorrhagic etiology, this finding would raise the question of what other factors contributed to the clinician’s decision to give fibrinolytics that may not be contained in the propensity analysis. Such unaccounted for factors may have not only led to the administration of fibrinolytics, but separately and significantly contributed to those patients’ deaths.