POMPE-C tool for pulmonary embolism mortality in cancer

Clinical background

Patients with active cancer are at markedly increased risk of pulmonary embolism and at the same time have a worse prognosis after the embolism than patients without cancer. This confronts the clinician with two difficult decisions: how aggressively should a patient with advanced cancer be treated for the embolism, and which patients can safely be managed as outpatients? The generic prognostic instrument PESI, used for the general pulmonary embolism population, performs poorly in cancer patients. In the derivation cohort, PESI classified only 0.8 per cent of cancer patients as low risk, and discrimination fell to an AUC of 0.68 [1]. The POMPE-C tool was developed to fill this gap and to provide a quantitative basis for precisely this decision about the level of treatment in patients with active cancer and pulmonary embolism.

Calculating the POMPE-C tool

POMPE-C is a logistic regression model with eight variables, three continuous and five dichotomous. The model calculates a logit $L$ that is then converted into a percentage mortality risk:

$$L = 3{.}718 + 1{.}55171 \times \text{DNR} + 0{.}79961 \times \text{respiratory distress} + 0{.}73433 \times \text{swelling} + 1{.}47345 \times \text{altered mental status} + 1{.}02789 \times \text{HR>100} + 0{.}04422 \times \text{respiratory rate} - 0{.}063 \times \text{SpO}_2 - 0{.}01161 \times \text{weight}$$

$$\text{Mortality }$$

The variables are: body weight in pounds (lbs), respiratory rate in breaths per minute, oxygen saturation in per cent, heart rate >100/min (dichotomous), respiratory distress (dichotomous), altered mental status (dichotomous), do-not-resuscitate (DNR) status (dichotomous) and unilateral limb swelling (dichotomous). A higher weight and a higher oxygen saturation lower the calculated risk, while the other variables raise it.

The model was derived as a pre-planned secondary analysis of the EMPEROR registry, a prospective multicentre study at 22 US emergency departments with data collection from January 2005 to December 2008 [1]. Of 1,880 patients with pulmonary embolism, 408 were identified as having active cancer, defined as metastatic disease or ongoing oncological care. Of these, 51 (12.5 per cent) died within 30 days. Twenty-five candidate variables were reduced by CART analysis and bivariate testing to the eight retained in the final model. All eight were significant predictors in the logistic regression.

Interpretation in practice

POMPE-C generates a percentage estimate of 30-day mortality, not a categorical risk class. In the derivation study, two thresholds for clinical action were proposed [1]:

The threshold of $\leq$ 5 per cent was validated in the derivation study's independent validation sample: of 50 patients with an estimate of $\leq$ 5 per cent, none died within 30 days (0 per cent, 95 per cent CI 0 to 7 per cent), corresponding to a sensitivity of 100 per cent for identifying those who die [1]. Specificity at this threshold was low, however, at 32 per cent, meaning that many patients who survive fall above the threshold. At the threshold of > 50 per cent, 10 of 13 patients died (77 per cent, 95 per cent CI 46 to 95 per cent).

In the Chinese validation cohort, POMPE-C classified 30.9 per cent of patients as low risk, with an observed mortality of 3.5 per cent in this group [3]. This indicates that the $\leq$ 5 per cent threshold holds externally as well, but with an observed mortality somewhat above zero.

Validation and performance

In the derivation study, POMPE-C was validated in six independent prospective cohorts from the USA, Europe and New Zealand, comprising 182 patients with complete data, of whom 27 died within 30 days (mortality 15 per cent) [1]. The AUC in the validation sample was 0.86 (95 per cent CI 0.78 to 0.93), comparable to the derivation set's 0.84 (95 per cent CI 0.78 to 0.89). The AUC for PESI in the same cancer population was 0.68 (95 per cent CI 0.60 to 0.76).

Weeda et al. carried out an external validation in 124 patients with active cancer and pulmonary embolism at a US university hospital [2]. Thirty-day mortality was 20.2 per cent. POMPE-C classified 32 to 43 per cent of patients as low risk, with a sensitivity of 88 to 96 per cent and a specificity of 38 to 53 per cent for the cancer-specific instruments taken together. Generic instruments such as PESI and Hestia had higher sensitivity (>96 per cent) but classified fewer than 19 per cent as low risk and had very low specificity (PESI 6.1 per cent).

Li et al. validated POMPE-C in 460 patients with active cancer and pulmonary embolism in China [3]. The AUC for POMPE-C was 0.78 (95 per cent CI 0.74 to 0.81), compared with 0.74 (95 per cent CI 0.70 to 0.78) for PESI and a similar figure for Hestia. POMPE-C and RIETE both classified 30.9 per cent as low risk, with a low mortality in this group (3.5 per cent for POMPE-C, 1.4 per cent for RIETE). Generic instruments performed less well: PESI classified only 9.1 per cent as low risk and Hestia 65.4 per cent, but with a mortality above 5 per cent in the low-risk group.

A meta-analysis of eight studies and ten clinical prediction rules found that cancer-specific instruments generally identified a larger proportion of low-risk patients with a sensitivity comparable to that of generic instruments [4]. The authors noted, however, that the evidence base is limited and recommended that the prediction rules be used together with patient-specific clinical factors, not as the sole basis for a decision.

Limitations

POMPE-C was derived and validated for patients with active cancer and symptomatic pulmonary embolism diagnosed in the emergency department. It has not been validated specifically for incidentally detected pulmonary embolism at cancer staging, a population that may differ considerably in symptomatology and prognosis [5]. A British study of 234 cancer patients with incidental pulmonary embolism found a 30-day mortality of only 3.4 per cent, considerably lower than in the POMPE-C cohorts, and instead developed its own prognostic model based on performance status and symptoms [5].

Weight is entered in pounds (lbs), which requires conversion when the tool is used in clinical practice outside the USA. A patient weighing 70 kg should be entered as approximately 154 lbs. An incorrect weight unit affects the estimate directly, since weight is a continuous variable with a negative coefficient.

The variable do-not-resuscitate (DNR) status is culturally and legally context-dependent. In the derivation cohort, patients receiving comfort care only were excluded, and DNR status referred to patients in whom a decision had been made but active treatment was still being pursued. In other health systems the DNR category may cover a different patient population, which may distort the estimate. In the European validation cohort, explicit data on DNR status and altered mental status were missing, and these variables were assumed to be absent in all patients [1], which may have affected the estimate of performance.

The validation cohorts are small. The largest external validation comprises 460 patients [3], and several cohorts have fewer than 200. The confidence intervals for the AUC are therefore wide. No study has yet validated POMPE-C in a randomised trial in which the estimate actually guides treatment decisions, and the instrument therefore lacks demonstrated clinical utility in the strict sense.

POMPE-C estimates mortality only and takes no account of bleeding risk, of the cancer prognosis in terms of expected survival, or of patient preferences. A low POMPE-C estimate does not automatically justify outpatient treatment if the patient has a high bleeding risk, impaired renal function or social circumstances that make follow-up difficult.

References

1. Kline JA et al. Derivation and validation of a multivariate model to predict mortality from pulmonary embolism with cancer: the POMPE-C tool. Thromb Res 2012;129(5):e194-9. PMID: 22475313
2. Weeda ER et al. External Validation of Generic and Cancer-Specific Risk Stratification Tools in Patients With Pulmonary Embolism and Active Cancer. J Natl Compr Canc Netw 2017;15(12):1476-1482. PMID: 29223985
3. Li X et al. Comparison of Different Clinical Prognostic Scores in Patients with Pulmonary Embolism and Active Cancer. Thromb Haemost 2021;121(6):834-844. PMID: 33450779
4. Nguyen E et al. Clinical prediction rules for mortality in patients with pulmonary embolism and cancer to guide outpatient management: a meta-analysis. J Thromb Haemost 2018;16(2):279-292. PMID: 29215781
5. Bozas G et al. Prognostic assessment for patients with cancer and incidental pulmonary embolism. Thromb J 2018;16:1. PMID: 29445314