Troponin-Based Risk Stratification of Patients With Acute Nonmassive Pulmonary Embolism: Systematic Review and Metaanalysis FREE TO VIEW

Although most patients with acute pulmonary embolism (PE) have an uncomplicated clinical course while undergoing standard anticoagulation treatment, the overall 3-month mortality rate exceeds 15%.¹ Death from acute PE usually occurs before or soon after hospital admission.² Patients presenting with clear signs of shock have high morbidity and mortality rates. It is generally accepted that these patients should be considered for thrombolytic therapy.³ One area of controversy focuses on the extension of the indication for thrombolytic therapy to a subgroup of patients who appear stable at presentation but have impending right ventricular failure and a high risk of PE-related death. Thus, a major challenge is the identification of such potential candidates for thrombolytic therapy by a simple, rapid, and noninvasive method.

Studies of patients with acute PE have demonstrated an association between elevated serum levels of troponin and right ventricular dysfunction or adverse in-hospital outcome.⁴ However, the positive predictive value of an elevated troponin level is low, and the prognostic implications remain uncertain.⁵ Only a few studies have evaluated the prognostic significance of elevated levels of troponin in the subgroup of normotensive patients with acute PE in which alternatives to conventional anticoagulation (ie, thrombolysis) may be considered. In addition, some reports^6–8 have suggested that cardiac troponin levels have a high negative predictive value with regard to early death. Few studies have evaluated the prognostic significance of low levels of troponin in the subgroup of normotensive patients with acute PE in which home therapy of acute symptomatic PE may be considered.

A metaanalysis published in 2007⁹ and a large cohort study¹⁰ have come to conflicting conclusions and have prompted further debate about the usefulness of the measurement of troponin levels for the risk stratification of PE patients. The metaanalysis pooled data for normotensive and hemodynamically unstable patients. Three more recent studies^10–12 were not included in the previous metaanalysis. To further clarify how well troponin levels predict mortality in normotensive patients with acute symptomatic PE and its usefulness for treatment decision making, this study aimed to review the literature systematically and perform an updated metaanalysis.

All studies of patients with PE were considered eligible for the metaanalysis if they fulfilled the following criteria: original publication; inclusion of normotensive patients with an objectively confirmed diagnosis of acute symptomatic PE; measurement of cardiac-specific troponin levels (cardiac troponin T [cTnT] or cardiac troponin I [cTnI]) in nanograms per milliliter, or the reporting of proportions of patients in different categories surrounding a predefined cut point for elevation defined by the local laboratories; separate reporting within the cohort of normotensive patients; and analysis of all-cause mortality within the cohort of normotensive patients. Type of troponin assay, definition of normotensive status, study sample size, and duration of follow-up did not determine study eligibility. We applied no language restriction.

A computerized search of Medline and EMBASE from 1980 through April 2008 was conducted to identify eligible studies. Studies published in abstract form were excluded because of the concern that non-peer-reviewed data might introduce bias into the report. We used a sensitive search strategy for prognosis studies.^13,14 We used the terms “venous thrombosis” (MeSH) or “pulmonary embolism” (MeSH) or “PE” (text word); “troponin” (MeSH) or “cardiac biomarkers” (text word); “incidence” (MeSH) or “explode mortality” (MeSH) or “follow-up studies” (MeSH) or “mortality” (subheading) or “prognosis*” (text word) or “predict*” (text word) or “course” (text word). Full articles of all potentially appropriate abstracts were reviewed. Hand searching of cited bibliographies and investigator files complemented the literature search.

Two investigators (D.J. and D.M.) independently assessed the identified articles to determine study eligibility. Based on brief study details (eg, title and abstract), the reviewers excluded nonrelevant studies. For relevant studies, the reviewers independently carried out data extraction using a standardized form designed a priori. Consensus or discussion with a third reviewer resolved eligibility and data extraction discrepancies or uncertainties.

Although quality criteria for assessing randomized controlled trials exist, the literature lacks a generally accepted list of such criteria for observational studies.^15–17 Two investigators (D.J. and D.M.) independently assessed the quality of the eligible studies, as recommended by Hayden et al,¹⁸ by grading six potential sources of bias related to study participation, study attrition, prognostic factor measurement, outcome measurement, confounding measurement and control, and analysis. We classified the control of confounding as follows: (1) poor, if little or no attempt was made to control for known basic confounders (ie, age, sex, comorbidities, anticoagulant treatment, or prognostic indexes); (2) adequate, if analyses controlled for age and comorbidities; and (3) good, if analyses controlled for the majority of the confounders.

We a priori created a 2 × 2 table based on troponin level categories (positive ([high level] and negative [low level]) and overall death. For each study, we determined the incidence of short-term, all-cause mortality for the positive and negative troponin groups, and we calculated an odds ratio (OR) and its 95% CI. We pooled ORs across studies by using a DerSimonian and Laird random-effects model approach. Statistical heterogeneity between groups was measured using the Cochran Q statistic (p value < 0.1 [indicative of heterogeneity; χ² test]) and the Higgins I² statistic (heterogeneity was defined as low if < 25%, moderate if between 25% and 50%, or high if > 50%). We evaluated the effect of the different troponin cutoff levels for the prediction of death. We fitted a summary receiver operating characteristic curve and analyzed the relationship between sensitivity and specificity of troponin to predict overall mortality by means of the Spearman rank correlation coefficient. Although the positive and negative predictive values give outcome probabilities for particular test results, they depend on the prevalence of the outcome in the study sample, and they may be difficult to generalize beyond the study. Thus, we calculated likelihood ratios (LRs) to see if we could discern patients at high vs low risk for death. We separately analyzed studies that used troponin I and those that used troponin T, and studies that had good/adequate control of confounding and those with poor control of confounding. The Begg rank correlation method assessed publication bias. All analyses were carried out using statistical software package (Stata; Stata Corp; College Station, TX¹⁹; and Meta-DiSc, version 1.4; publicly available at http://www.hrc.es/investigacion/metadisc_en.htm).²⁰

Of the 596 articles screened, 28 appeared potentially eligible and were reviewed in depth.^{4–7,10–12,21–41} Nineteen studies were deemed ineligible^{4–7,21–35} (Fig 1), and 9 studies met the eligibility criteria.^{10–12,36–41} Agreement between the two reviewers for study eligibility was very high (κ = 0.9).

The year of publication of the nine eligible studies ranged from 2003 to 2008 (Table 1). Of the total sample size of 1,366 patients, the individual study samples ranged in size from to 28 to 458 patients. The largest study by Douketis et al³⁹ accounted for 33% of all patients in the metaanalysis, and the two largest studies^10,39 accounted for 56% of all patients.

Regarding study design, eight studies^{10–12, 36–38, 40, 41} were described as prospective cohort studies, and one study³⁹ utilized a cohort from a randomized controlled trial retrospective database analysis. For the survival (all-cause mortality) analyses, two studies^37,40 used the Cox proportional-hazards technique and four studies^10,36,39,40 used logistic regression.

The studies used tests for measuring levels of either cTnT or cTnI. Most of the studies used a predefined cutoff point to determine abnormal (elevation) vs normal (nonelevation) levels. Three of the studies used the same assay and the same cutoff (> 0.01 ng/mL) for abnormality, and all studies used a cutoff of no more than 0.5 ng/mL. Receiver operating characteristic analysis assessed the optimal cutoff value of cTnT for death and serious adverse event selection in one study.³⁷

In two of the studies,^40,41 the mean age of the patients was approximately 53 years, which was younger than the mean age of the other studies and large cohort studies of patients with PE.^1,42 Length of follow-up ranged from the in-hospital period^{11,12,36,38,40} to 3 months.^39,41 Reported losses to follow-up varied from 0 to 6%. The mortality rates varied from 1%⁴⁰ to 15%.³⁷

The definition of hemodynamic status differed among the studies. Two studies did not provide a definition for the term “normotensive.”^38,41 Jiménez et al¹⁰ and Douketis et al³⁹ excluded patients with massive PE associated with hypotension (systolic BP, < 90 mm Hg), cardiogenic shock, or respiratory failure in whom thrombolytic therapy or surgical thrombectomy might be considered. Kline et al⁴⁰ excluded patients with systolic BP < 100 mm Hg. Logeart et al¹² excluded patients with hemodynamic impairment, which they defined as shock, systolic BP < 90 mm Hg, or syncope on hospital admission. Pruszczyk et al³⁶ excluded patients with systolic BP at hospital admission of < 90 mm Hg, those requiring catecholamine infusion or ventilatory support, and those who had undergone cardiopulmonary resuscitation. Kostrubiec et al³⁷ excluded patients who presented with systolic BP < 90 mm Hg. Gallotta et al¹¹ excluded patients with shock or systolic BP < 90 mm Hg, or severe tachyarrhythmias or bradyarrhythmias.

In most studies, the primary outcome was all-cause mortality. PE-related death was the primary outcome in three studies.^10,32,36 For the purpose of this metaanalysis, we only analyzed all-cause mortality for all studies.

Regarding study quality assessment criteria (Table 2), the study participation was adequate and the baseline study sample was adequately described in six of the studies.^{10–12,37,39,40} In the studies of Pruszczyk et al³⁶ and Bova et al,³⁸ it was unclear whether only outpatients were eligible, and there was no information about the period of recruitment. Tulevski et al⁴¹ did not specify study eligibility criteria, and the source population was not adequately described for key characteristics. Only three studies^10,40,41 provided adequate information about patients lost to follow-up, whereas the other six studies^{11,12,36–39} did not. All studies adequately measured troponin levels, although the assays and the cutoffs for high/low levels varied, as described earlier and in Table 1. An independent blinded committee assessed the outcome criteria in one study.¹⁰ Important potential confounders were appropriately accounted for in six studies.^{10,11,36,37,39,40} The use of appropriate statistical analyses in six studies^{10,11,36,37,39,40} limited the potential for the incorporation of and presentation of invalid results, whereas the other three studies did not perform any predictive statistics.

Overall, 27.6% of normotensive patients with acute symptomatic PE had elevated troponin levels. Sixty of the 377 patients with elevated troponin levels died (15.9%; 95% CI, 12.2 to 19.6) compared with 34 of 989 with normal troponin levels (3.4%; 95% CI, 2.3 to 4.6). Summary receiver operating characteristic analysis showed a relationship between the sensitivity and specificity of troponin to predict overall mortality (Fig 2). Pooled results showed that elevated troponin levels, compared with nonelevated levels, were associated with a 4.26-fold increased odds of overall mortality (95% CI, 2.13 to 8.50; heterogeneity χ² = 12.64; degrees of freedom = 8; p = 0.125) [Fig 3] during short-term follow-up. The pooled estimate was dominated by the three larger studies,^10,39,40 which together provided about three-quarters of the total number of patients, and these studies had the most conservative (nonextreme) results. Five of the nine studies did not have statistically significant findings, although all studies showed the same trend. The result was consistent for either troponin I (OR, 2.65; 95% CI, 1.26 to 5.56) or troponin T (OR, 8.60; 95% CI, 2.72 to 27.22), and for high-quality studies (OR, 3.18; 95% CI, 1.56 to 6.45) or low-quality studies (OR, 15.29; 95% CI, 3.32 to 70.37). There was no evidence of publication bias using the Begg rank correlation method.

The metaanalysis showed the following: (1) a slight increase in the rate of elevation of troponin levels in patients who died, compared with the rate of elevated troponin levels in patients who survived (positive LR, 2.26; 95% CI, 1.67 to 3.07) [Fig 4]; and (2) a slight decrease in the rate of nonelevated troponin levels in patients who died, compared with the rate of nonelevated troponin in patients who survived (negative LR, 0.59; 95% CI, 0.39 to 0.88) [Fig 5] during the short-term follow-up. These nonextreme LRs do not significantly change the odds of death based on an elevated troponin level or the odds of survival based on a nonelevated troponin level.

In this metaanalysis of nine studies, which included 1,366 normotensive patients with acute symptomatic PE, patients with elevated troponin levels had a fourfold increased risk of short-term death compared with patients with nonelevated levels. Other studies of cohorts that consisted of both normotensive and unstable patients with acute PE showed similar results^6,23 and support these findings. Regardless of the varying assays employed or the cutoffs used to define an elevated troponin level, the studies showed relatively consistent results.

Risk stratification algorithms for patients presenting with acute PE could incorporate troponin level assessment.^5,42 Ideally, high troponin levels would identify hemodynamically stable patients who may benefit from thrombolytic treatment, and low troponin levels would identify patients who are at low risk of death or PE-related complications. Such low-risk patients might be the most appropriate patients to consider for full or partial out-of-hospital treatment of acute PE. Unfortunately, troponin by itself does not appear clinically to change significantly the pretest-to-posttest probabilities (ie, positive and negative LRs are not extreme) of risk in either scenario. For patients with acute symptomatic PE, a high troponin level in itself should not significantly drive the decision to administer thrombolytic therapy, and a low troponin level should not significantly drive the decision to treat at home. The evidence does not support the use of troponin levels for such decision making.

These metaanalytic results support the data from individual studies^26,27 that showed that elevated troponin level has higher predictive value when used in combination with other tests (eg, echocardiography) to identify patients who are at high risk for death after PE. We suggest that troponin measurement combined with other tools (eg, clinical prognostic scores and echocardiography) might better determine eligibility of patients for out-of-hospital treatment of acute PE.⁴³

Limitations of each of the included studies may have introduced significant biases into the estimates of the prognostic value of troponin levels this metaanalysis. Some of the studies did not clearly describe the referral patterns, and they may have not included consecutive patients. Also, some of the studies did not adjust for other important prognostic factors or types of treatment that could have been prescribed based on troponin levels.

A previous metaanalysis⁹ and a recent large cohort study¹⁰ have provided conflicting conclusions about the benefits of troponin measurement for risk stratification of patients with acute symptomatic PE. The metaanalysis⁹ suggested that elevated troponin levels identify patients with acute PE at high risk of short-term death, whereas the cohort study¹⁰ suggested that troponin testing does not identify patients with stable PE who are at very low risk of fatal medical outcomes. Unlike the previous metaanalysis,⁹ the current metaanalysis focused only on normotensive patients and it included three additional studies with 475 more patients. Because most of the included studies in both metaanalyses did not report whether or not an independent blinded committee assessed the outcome criteria, we aimed to minimize diagnostic suspicion bias by only assessing the outcome of all-cause mortality. The total sample size, the proportion of patients with an elevated troponin level, and the death rate all allowed for reasonable estimates of risk. The varied settings and patient characteristics improved the generalizability of the study. The prognostic value of troponin was consistent among high-quality and low-quality studies, and for both cTnI and cTnT. Moreover, the results of this metaanalysis showed a statistically significant correlation between different levels of elevated troponins and death in patients with PE.

In conclusion, the prognostic value of troponin levels in normotensive patients in whom acute PE had been diagnosed depends greatly on the cutoff points used, and the usefulness of basing therapeutic decision making solely on troponin levels does not appear warranted. Troponin levels are likely to be most useful when used in combination with echocardiographic or CT scan evidence of right ventricular strain and with clinical prognostic scores. If the combination of troponin measurement with those prognostic tools improves our ability to predict outcomes, we may be able to better determine eligibility for thrombolytic therapy and for out-of-hospital treatment of acute PE.

Author contributions: Drs. Jiménez, Uresandi, Otero, Lobo, Monreal, Martí, and Yusen contributed to the study concept and design. Drs. Jiménez, Zamora, Muriel, Monreal, Aujesky, and Yusen contributed to the acquisition of data, analysis and interpretation of data, and statistical analysis. Drs. Jiménez, Martí, Monreal, and Yusen contributed to the drafting of the manuscript. Drs. Jiménez, Uresandi, Monreal, Aujesky, and Yusen contributed to the critical revision of the manuscript for important intellectual content. Drs. Jiménez and Yusen contributed to the study supervision. Dr. Jiménez, the corresponding author, had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Financial/nonfinancial disclosures: The authors have reported to the ACCP that no significant conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article.

Other contributions: We thank Vittorio Palmieri for his unpublished data that were included in this analysis.