We conducted an observational cohort study using prospectively collected data from the RIETE (Registro Informatizado Enfermedad Tromboembólica) registry (ClinicalTrials.gov identifier: NCT02832245), an ongoing international registry of patients with PE or deep vein thrombosis (DVT) that includes more than 200 centers across 28 countries. The registry design has been described previously [16]. Patients with objectively confirmed PE, as demonstrated by contrast-enhanced CT, were consecutively enrolled. All patients provided written informed consent in accordance with the requirements of the local ethics committees.

We included patients enrolled in RIETE between January 1st, 2001, and May 31st, 2025, who were diagnosed with hemodynamically stable, symptomatic acute PE, either in the emergency department or during hospitalization. Central PE was defined as the presence of thrombus in the main pulmonary artery (PA) or in the left or right main PA, whereas peripheral PE was defined as thrombus limited to the remaining pulmonary arterial territories [17]. Baseline data included demographic characteristics, clinical presentation, ECG information, and diagnostic test results.

An abnormal ECG was defined as the presence of at least 1 of the following prespecified findings: heart rate <50 or >100beats/min, any rhythm other than sinus rhythm, right bundle branch block (RBBB), S1Q3T3 pattern, or T-wave inversion in the precordial leads. All patients were followed up for 30 days to assess clinical outcomes.

Eligible patients met all the following inclusion criteria: (1) age ≥18 years; (2) imaging-confirmed, hemodynamically stable acute symptomatic PE; and (3) ECG performed within 24h of PE diagnosis. We excluded patients with (1) incidental or asymptomatic PE; (2) hemodynamic instability at presentation, defined as systolic blood pressure <90mm Hg for >15min not attributable to hypovolemia, sepsis, or arrhythmia; (3) no ECG within 24h of diagnosis; or (4) missing variables required to calculate the original sPESI. All patients were followed up for 30 days to assess clinical outcomes.

The original sPESI assigns 1 point for each of the following variables: age >80 years, past medical history of cancer, chronic cardiopulmonary disease, heart rate ≥110beats/min, systolic blood pressure <100mm Hg, and arterial oxygen saturation <90% [4]. Patients with a score of 0 points were classified as low risk.

A modified version using the same variables but with a heart rate threshold of 100beats/min has previously shown enhanced prognostic performance when combined with a CT-assessed RV/LV ratio ≥1.0; this model is referred to as the modified sPESI plus CT-assessed RV/LV ratio model [5] (Supplementary Table S1).

ECG interpretation was performed by the treating physicians and recorded in the RIETE case report form. A normal ECG was defined as sinus rhythm, upright P waves in leads I, II, and aVF, heart rate between 50 and 100beats/min, QRS duration ≤120ms, and absence of predefined abnormalities. ECGs were dichotomized as normal vs abnormal. This composite ECG definition was intended to identify the absence of any predefined ECG abnormality rather than to represent a single pathophysiological mechanism.

We constructed an sPESI-ECG score by assigning 1 point for each of the following variables: age >80 years, cancer, chronic cardiopulmonary disease, abnormal ECG, systolic blood pressure <100mm Hg, and arterial oxygen saturation <90%. The heart rate variable in the original sPESI was replaced, rather than supplemented, by the dichotomized ECG variable to avoid double counting. Thus, sPESI-ECG should be considered a modified risk classification strategy rather than the original validated sPESI score.

In addition, we assessed a combined model that incorporated the variable CT-assessed RV/LV ratio ≥1.0, assigning 1 additional point. This model was termed the sPESI-ECG plus CT-assessed RV/LV ratio model. Patients with a score of 0 in either model were classified as very low risk (Supplementary Table S1).

The primary endpoint was 30-day all-cause mortality. Secondary endpoints included 30-day PE-related mortality and a composite endpoint comprising 30-day all-cause mortality, early hemodynamic deterioration, defined as systolic blood pressure <90mm Hg for >15min not attributable to hypovolemia, sepsis, or arrhythmia and occurring within 48h of PE diagnosis, or need for escalated therapy, including rescue systemic thrombolysis, surgical or mechanical thrombectomy, extracorporeal membrane oxygenation, or vasopressor/inotropic support. Outcomes were analyzed as dichotomous variables. In addition, a predefined subanalysis was performed in patients with available CT-assessed RV/LV ratio data to determine whether this finding further enhanced risk stratification.

Categorical variables were expressed as frequencies and percentages. Continuous variables were reported as mean±SD or median (IQR), as appropriate. Normality was assessed using the Kolmogorov–Smirnov test. Comparisons were performed using the Student t test or analysis of variance for normally distributed variables and the Mann–Whitney U or Kruskal–Wallis test for nonnormally distributed variables. We calculated the proportion of patients classified as low risk by each strategy and estimated diagnostic performance metrics, including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and likelihood ratios, for each outcome.

Sensitivity was defined as the proportion of patients with the outcome correctly identified as high risk, and specificity was defined as the proportion of patients without the outcome correctly classified as low risk. Pairwise comparisons between strategies were performed using exact binomial tests. Confidence intervals for absolute differences were calculated using the Agresti–Min method.

Logistic regression analyses were performed to assess associations with outcomes. Univariable logistic regression estimated crude ORs with 95%CIs. Variables selected a priori included male sex, syncope, elevated cardiac troponin, and renal insufficiency. Multivariable logistic regression was performed using stepwise selection, and adjusted ORs were reported for variables meeting inclusion criteria, defined as P<.10 in univariable analysis. Multicollinearity was assessed using the variance inflation factor (VIF), with values >5 considered suggestive of problematic multicollinearity.

For regression models, covariables had <5% missing data, and complete-case analysis was performed, assuming data were missing completely at random, consistent with prior RIETE analyses [5]. Variables required for sPESI calculation with >20% missingness were not included as independent predictors, and patients with missing sPESI components were excluded from the corresponding analyses.

Two-sided P values <.05 were considered statistically significant. Analyses were conducted using IBM SPSS version 26 and R version 4.2.2 (packages binom, car, DescTools, and DTComPair).

A total of 32,113 hemodynamically stable patients with acute symptomatic PE were included in the analysis (Fig. 1). The mean (SD) age was 68 (16) years, and 52.8% were women. According to the original sPESI, 9771 patients (30.4%) were classified as low risk at presentation.

Fig. 1.

Flowchart of inclusion criteria.

A normal ECG was observed in 11,852 patients (36.9%), whereas 20,261 patients (63.1%) had an abnormal ECG. Among these patients, 28.1% had a heart rate >110beats/min, 12.2% had atrial fibrillation (AF), 25.8% had RBBB, and 24.5% had an S1Q3T3 pattern. Patients with abnormal ECG findings were older and more frequently presented with cardiopulmonary comorbidities, including chronic obstructive pulmonary disease and chronic heart failure (Table 1).

Patients with abnormal ECG findings presented with a more severe clinical profile. Compared with those with normal ECG findings, they more frequently presented with syncope (17.0% vs 10.3%), dyspnea (86.5% vs 78.8%), hypoxemia (oxygen saturation <90%: 29.7% vs 16.5%), and systolic blood pressure between 90 and 100mm Hg (6.5% vs 2.9%). Central PE on computed tomography was more common among patients with abnormal ECG findings (49.0% vs 33.9%). Markers of RV dysfunction were also more prevalent in patients with abnormal ECG findings, including elevated cardiac troponin levels (29.6% vs 15.9%) and CT-assessed RV/LV ratio ≥1.0 (57.3% vs 33.9%) (Table 1).

During follow-up, 30-day all-cause mortality occurred in 1462 patients (4.5%). The incidence was nearly twice as high in patients with abnormal ECG findings as in those with normal ECG findings (5.4% vs 2.9%). Similarly, 30-day PE-related mortality was higher among patients with abnormal ECG findings (1.7% vs 0.6%), and the composite endpoint occurred more than twice as frequently (10.1% vs 4.2%).

Early hemodynamic deterioration occurred in 65 of 4226 patients (1.5%), and 1119 of 5714 patients (19.6%) required therapy escalation, with a higher rate among patients with abnormal ECG findings (Table 1).

For 30-day overall mortality, the original sPESI demonstrated high sensitivity (94.5%) but limited specificity (31.6%), with an NPV of 99.2%. Incorporation of ECG findings into sPESI increased sensitivity to 97.7% and NPV to 99.4% (95%CI, 99.1–99.6), at the expense of lower specificity (17.4%).

Similar increases in sensitivity and NPV were observed for 30-day PE-related mortality and for the composite endpoint, with an NPV of 98.0% (95%CI, 97.6–98.3). The increase in NPV for the primary and secondary endpoints was statistically significant (P<.01) (Table 2). Positive predictive values remained low across all outcomes for both scores.

Receiver operating characteristic curve analysis showed comparable discrimination between strategies. For 30-day all-cause mortality, the C statistic was 0.73 (95%CI, 0.72–0.74) for both the original sPESI and sPESI-ECG. For 30-day PE-related mortality, discrimination was similar: 0.73 (95%CI, 0.71–0.76) vs 0.74 (95%CI, 0.72–0.76). For the composite endpoint, the C statistic was 0.66 (95%CI, 0.64–0.67) for the original sPESI and 0.66 (95%CI, 0.65–0.67) for sPESI-ECG.

In univariable analyses, syncope, central PE, transient risk factors, renal dysfunction, and elevated troponin levels were significantly associated with adverse outcomes. For 30-day all-cause mortality, sPESI-ECG yielded an adjusted OR, 10.16 (95%CI, 4.51–22.91), compared with OR, 7.04 (95%CI, 4.36–11.38) for the original sPESI. Similar patterns were observed for 30-day PE-related mortality (adjusted OR, 7.91 vs 4.40) and for the composite endpoint (adjusted OR, 3.39 vs 2.13) (Table 3).

The original sPESI classified 9771 patients (30.4%) as low risk, whereas sPESI-ECG identified a smaller low-risk group of 5359 patients (16.7%). Patients classified as low risk by sPESI-ECG had lower observed event rates. Thirty-day all-cause mortality was 0.62% in the sPESI-ECG low-risk group compared with 0.82% in the original sPESI low-risk group (RR, 1.33; P=.19). Similar reductions were observed for 30-day PE-related mortality (0.11% vs 0.24%; RR, 2.10; P=.14) and for the composite endpoint (2.00% vs 3.33%; RR, 1.67; P<.001) (Table 4).

ECG abnormality as a composite variable was independently associated with the primary outcome after adjustment for the original sPESI and relevant covariates.

To explore whether specific ECG abnormalities individually accounted for the association between abnormal ECG findings and outcomes, we evaluated each ECG component separately. For 30-day all-cause mortality, nonsinus rhythm, bradycardia, heart rate of 101 to 110 beats/min, and heart rate >110beats/min were significantly associated with the outcome. In contrast, ECG signs typically related to RV overload, including RBBB, S1Q3T3, and negative precordial T waves, were not significantly associated with all-cause mortality, either individually or when combined. For the composite endpoint, however, most ECG components were significantly associated with events, including RBBB, S1Q3T3, and negative precordial T waves. These findings suggest that RV overload patterns on ECG may be more closely related to early clinical deterioration than to all-cause mortality alone. Detailed estimates are shown in Table S2.

When each ECG component was individually added to the original sPESI, none achieved the sensitivity or NPV observed with the composite sPESI-ECG model, nor did any reduce event rates in the low-risk group to the same extent (Table S2). These findings suggest that the prognostic performance of the sPESI-ECG model was not solely explained by the heart rate component.

Among the 4943 patients with available CT-assessed RV/LV ratio data, 2353 (47.6%) had RV/LV ratio ≥1.0. Baseline characteristics are summarized in Table S3. In this subgroup, strategies incorporating ECG findings and/or RV/LV ratio ≥1.0 showed progressively higher sensitivity and negative predictive value for 30-day all-cause mortality and the composite endpoint, with a corresponding decrease in specificity.

For 30-day all-cause mortality, the original sPESI and sPESI-ECG showed similar discriminative ability, with a C statistic of 0.73. The addition of CT-assessed RV/LV ratio ≥1.0 yielded no meaningful improvement in discrimination, with C statistics of 0.71 (95%CI, 0.67–0.74) when added to the original sPESI and 0.70 (95%CI, 0.66–0.73) when added to the sPESI-ECG model. Discrimination for the composite endpoint was modest for clinical scores alone, with a C statistic of 0.62 (95%CI, 0.59–0.64) for both the original sPESI and sPESI-ECG, and increased slightly with the addition of CT-assessed RV/LV ratio ≥1.0, with a C statistic of 0.66 for both original sPESI and sPESI-ECG.

Event rates in the low-risk groups decreased stepwise with the addition of ECG findings and RV/LV ratio ≥1.0, supporting the incremental prognostic value of these variables in terms of risk reclassification rather than overall discrimination. Across all evaluated strategies, sPESI-ECG showed the strongest independent association with 30-day all-cause mortality and the composite endpoint in multivariable models, with adjusted ORs of 13.57 and 2.33, respectively (Tables S4–S6).