Pulmonary embolism (PE) is a disease which presents with a wide spectrum of clinical manifestations, with varying prognoses and treatment.1 Prognostic stratification of patients with PE is essential for: (a) selecting the best treatment: patients with greater risk of secondary complications from the PE itself may benefit from specific treatments for recanalization of the obstructed arteries and (b) the treatment location (domiciliary care for very low risk patients and intensive care units for higher risk patients).

In recent years, various prognostic tools for patients with acute symptomatic PE have been validated, including clinical scores, imaging studies and biological markers.2 Among these, prognostic models have been shown to be especially useful for identifying low risk patients who may be managed with a shortened hospital stay or even outpatient care.

Uresandi et al.3 derived a clinical prognostic score (Spanish score) in 681 consecutive outpatients diagnosed with PE enrolled in a prospective multicenter study. The score includes the following variables: recent severe bleeding, metastatic cancer, non-metastatic cancer, serum creatinine values >2mg/dl, immobilization due to a recent medical condition, absence of surgery in the past 2 months and age >60 years. The score classified 47.8% of patients as low risk. The area under the curve for the prognostic efficacy of the score was 0.75 (95% confidence interval [CI], 0.67–0.83). This score was not validated in an independent patient cohort.

Aujesky et al.4 derived a clinical prognostic model, the Pulmonary Embolism Severity Index (PESI) in 10354 patients admitted with a diagnosis of PE in 186 hospitals in Pennsylvania (US). The model includes 11 clinical parameters evaluated at the time of presentation of symptoms: age, sex, cancer, heart failure, chronic lung disease, heart rate≥110min–1, systolic blood pressure <100mmHg, respiratory rate≥30min–1, temperature <36°C, altered mental state and oxygen saturation <90%. In the original series, 2% of patients in classes I and II (very low and low risk) died, compared to 14% of patients in classes III–V (intermediate to very high risk) within 30 days after the diagnosis of PE. Jiménez et al.5 derived and validated a simplified PESI score (sPESI). In a cohort of 995 patients, the authors found 6 variables independently associated with 30-day mortality: age, cancer, chronic cardiopulmonary disease, tachycardia (heart rate≥110min–1), hypoxemia (oxygen saturation <90%) and hypotension (systolic blood pressure <100mmHg). Patients with any of these variables were included in the high risk group. This score was validated by the same authors in a cohort of 7106 patients with PE from the RIETE registry. The score classified 36.1% of patients as low risk and 30-day mortality was 1.1% (95% CI, 0.7%–1.5%).

The aim of this study was to validate the Spanish score and compare it with the sPESI in an independent cohort of patients diagnosed with acute symptomatic PE.

During the study period, 7109 patients with suspected acute symptomatic PE were evaluated and the diagnosis was confirmed in 1564 (22%). Of these, 19 (1.2%) refused to sign the informed consent and 98 (6.3%) were lost to follow-up, so the final sample consisted of 1447 patients. Approximately 60% of the patients (856/1447) were diagnosed by chest CT-angiogram, 659 (45%) by high probability ventilation/perfusion scan and 73 (5%) had negative chest results and DVT confirmed by lower limb ultrasonography. Some patients had several positive diagnostic tests simultaneously.

The characteristics of the patients enrolled are shown in Table 1. The number of patients treated with inferior vena cava filters was small (1.8%; 28 of 1447 patients). Thrombolytics were administered to 4% of the patients (64 of 1447). Of the 1447 patients, 138 (9.5%; 95% CI, 8.0–11.1) died during the first 30 days of follow-up. Sixty-two patients (45%) died due to the PE itself, 7 (5.1%) due to bleeding and 69 (50%) due to other causes. A secondary episode occurred in 47 patients; 11 patients had an objectively confirmed recurrent thromboembolism, 38 patients had non-fatal major bleeding and 2 patients had both bleeding and recurrent thromboembolism.

The discrimination of the Spanish score and the sPESI for 30-day mortality was 0.74 (95% CI, 0.70–0.78) and 0.72 (95% CI, 0.68–0.76), respectively (Fig. 1). Fig. 1 shows the area under the curve obtained from the ROC curves for 30-day mortality with each of the risk models. In contrast, the discriminatory capacity of the two risk models for predicting the secondary episode was poorer, since the C statistics were lower than 0.60. Fig. 1 shows the results of the Hosmer–Lemeshow goodness-of-fit test, which evaluated the calibration of the Spanish score and the sPESI for predicting the risk of death.

Fig. 1.

30-Day Mortality ROC Curves for Each of the Study Models. (A) sPESI Model. Area Under the Curve (AUC) (C Statistic): 0.72; 95% CI, 0.68–0.76; P<.001; Hosmer–Lemeshow Test, 8.30; P=.09. (B) Spanish Score. AUC (C Statistic): 0.74; 95% CI, 0.70–0.78; P<.001; Hosmer–Lemeshow Test, 4.02; P=.40.

(0.11MB).

The sPESI classified a significantly lower proportion of patients as low risk (32% [459/1447], 95% CI, 29%–34%) compared to the Spanish score (62% [900/1.447], 95% CI, 60%–65%) (P<.001) (Table 2). Compared with the Spanish score low risk group, the mortality in the sPESI low risk group was lower (4.2% [38/900] vs 1.1% [5/459) (Table 2). The high risk group according to sPESI had a somewhat lower mortality than the Spanish score high risk group (13.5% [133/987] vs 18.3% [100/546]). The rate of non-fatal episodes was similar in the low risk group of both models (2.2% [10/459] vs 2.3% [21/900]).

For the prediction of 30-day mortality, the sensitivity, negative predictive value and negative likelihood ratio were better for the sPESI model than for the Spanish score (Table 3). When the 30-day non-fatal episodes are considered, the negative predictive value for the sPESI was 97.8%, compared to 97.7% for the Spanish score (Table 3).

This study compares the ability of the Spanish score and the sPESI for predicting the risk of events after an episode of acute symptomatic PE. In our series, both scores showed a similar discriminatory capacity for predicting 30-day mortality. However the sPESI was more reliable in identifying PE patients with a low risk of early complications.

Risk stratification of clinical episodes is a fundamental aspect in the management of patients with acute symptomatic PE. Clinical prognostic models were developed to identify PE patients with a low risk of complications, who could benefit from shortened hospital stays or even outpatient treatment of their disease.16,17 Although the Spanish score was used in a clinical trial for the outpatient treatment of low-risk PE,16 it has never been validated or compared directly with the sPESI.

Our results indicate that, although the general functioning of the two models is similar, the sPESI is significantly better than the Spanish score in the high sensitivity area. Indeed, the negative predictive value and the negative likelihood ratio are significantly better for the sPESI. A sensitive model is of interest for predicting mortality, since a false negative has a very high cost; this gives the sPESI an advantage. This difference between the two risk models appears to be explained by the differences in the variables that compose them. Unlike the Spanish model, the sPESI includes blood pressure, O2-hemoglobin saturation and heart rate, variables which have always been associated with acute PE survival.18–20

In our series, the usefulness of both scores for predicting the risk of non-fatal adverse events was poorer. There may be various reasons for this low predictive power. The most important is that these predictive models were conceived for predicting mortality, so it is not surprising that their capacity for predicting other events is lower. In any case, the negative predictive value of both models for the prediction of non-fatal adverse events remains high, so they continue to be excellent tools for identifying patients suitable for early discharge or outpatient therapy.

When comparing predictive models, a balance must be established between their discriminatory capacity and their simplicity in terms of clinical applicability. The sPESI consists of 6 variables with a simple scoring system (the presence of any of the variables indicates a high risk patient). However, the Spanish score consists of 7 variables, one of which is a laboratory parameter (serum creatinine) and the system for allocating scores is more complex.

This study has some limitations. Firstly, it is an analysis of a historical cohort composed of patients from a single center and is susceptible to the type of bias inherent in this type of study. However, it represents a non-selected population of PE patients in our usual clinical practice setting, making it an ideal sample for testing the external validity of the risk models analyzed. Secondly, the main limitation of the risk models lies in the fact that they were designed for making group predictions, rather than individual predictions. Although their discriminatory capacity is high, it can never be perfect. For this reason, the use of these models in a specific patient can only be orientative, since they allow a certain risk to be estimated but cannot definitively predict the final clinical outcome.21

In conclusion, our results show that the sPESI is more reliable than the Spanish score for identifying PE patients with a low risk of early complications. Well-designed studies are needed for determining if clinical scores are useful tools for identifying low-risk patients with PE who may be treated in an outpatient setting.

The authors declare that they do not have any conflict of interests.