On the necessity of new decision-making methods for cancer-associated, symptomatic, pulmonary embolism

Highlights

• We describe why new decision-making methods are needed for cancer-associated, symptomatic, pulmonary embolism

• We have evaluated the performance of five prognostic scales and a clinical decision rule (CDR) to predict 30-day mortality

• None of the five models contributed to qualitative clinical judgment

• These methods were no better than the ECOG PS or the simple dichotomic classification based on altered vital signs (CDR)

• A clinically meaningful attempt to further refine the stratification is presented

Abstract

Background

Acute symptomatic pulmonary embolism (PE) varies in its clinical manifestations in patients with cancer and entails specific issues. The objective is to assess the performance of five scores (PESI, sPESI, GPS, POMPE, and RIETE) and a clinical decision rule to predict 30-day mortality.

Methods

This is an ambispective, observational, multicenter study that collected episodes of PE in patients with cancer from 13 Spanish centers. The main criterion for comparing scales was the c-indices and 95% confidence intervals (CIs) of the models for predicting 30-day mortality.

Results

585 patients with acute symptomatic PE were recruited. The 30-day mortality rate was 21.3 (95% CI; 18.2–24.8%). The specific scales (POMPE-C and RIETE) were equally effective in discriminating prognosis (c-index of 0.775 and 0.757, respectively). None of these best performing scales was superior to the ECOG-PS with a c-index of 0.724. The remaining scores (PESI, sPESI, and GPS) performed worse, with c-indexes of 0.719, 0.705, and 0.722, respectively. The dichotomic “clinical decision rule” for ambulatory therapy was at least equally reliable in defining a low risk group: in the absence of all exclusion criteria, 30-day mortality was 2%, compared to 5% and 4% in the POMPE-C and RIETE low-risk categories, respectively.

Conclusion

The accuracy of the five scales examined was not high enough to rely on to predict 30-day mortality and none of them contribute significantly to qualitative clinical judgment.

Introduction

Venous thromboembolism (VTE) is frequent in patients with cancer and constitutes the second leading cause of death, after the tumor itself. However, this risk is not equally distributed; rather, it depends on patients' baseline characteristics and those of the pulmonary embolism (PE) itself [1]. The clinical spectrum of PE ranges from potentially fatal events to incidental findings on computed tomography (CT) scans. PE entails specific issues in patients with cancer, such as greater risk of re-thrombosis or major bleeding, and therefore calls for a specific classification.

Previous studies suggest that risk can be stratified and that early ambulatory treatment must be limited to low-risk patients [2], [3]. Patient selection becomes the most critical part of this therapeutic strategy, and although a variety of selection methods have been developed in the past, our ability to predict clinical risk is still limited in oncologic patients.

In the general population, acute symptomatic PE can be classified into risk levels based on the probability of short-term death (e.g., at 30 days). Presently, the Pulmonary Embolism Severity Index (PESI) is the most widely validated tool to stratify risk [4]. Other models have been developed to simplify prognostic classification or to make it safer, e.g., an equally effective, yet simplified version (sPESI) or the Geneva Prognostic Scoring (GPS) [5], [6]. These scales have been used to complement eligibility criteria in clinical trials comparing standard hospitalization versus early ambulatory management [7], [8]. However, all of them include a history of cancer as a highly relevant predictor, in spite of the variability of PE in these patients [1]. The POMPE-C and the RIETE scales have been recently devised in an attempt to improve classifications for cancer patients with acute symptomatic PE [9], [10]. Although these models predict 30-day mortality more accurately than general scales, there have been few independent validations and performance comparisons. Consequently, it remains unclear whether these models can add or be extrapolated to different oncologic settings.

On the other hand, the effect of the five validated scales on daily practice beyond the qualitative clinical evaluation made by experienced physicians is uncertain [11], [12], [13], since most prospective cohort studies have used simpler clinical decision rules (CDR) to select patients for specific treatments [13], [14], [15], [16]. These CDR, similar to the Hestia study eligibility criteria [14], had also proven useful in decision-making, enabling patients to be stratified into high and low risk categories. Consequently, this study seeks to evaluate the performance of the five afore-mentioned scales and a CDR to predict 30-day mortality.