Jump to
Abstract
Background: Sepsis is frequently complicated by the release of cardiac troponin, but the clinical significance of this myocardial injury remains unclear. We studied the associations between troponin release during sepsis and 1-year outcomes.
Methods and Results: We enrolled consecutive patients with sepsis in 2 Dutch intensive care units between 2011 and 2013. Subjects with a clinically apparent cause of troponin release were excluded. High-sensitivity cardiac troponin I (hs-cTnI) concentration in plasma was measured daily during the first 4 intensive care unit days, and multivariable Cox regression analysis was used to model its association with 1-year mortality while adjusting for confounding. In addition, we studied cardiovascular morbidity occurring during the first year after hospital discharge. Among 1258 patients presenting with sepsis, 1124 (89%) were eligible for study inclusion. Hs-cTnI concentrations were elevated in 673 (60%) subjects on day 1, and 755 (67%) ever had elevated levels in the first 4 days. Cox regression analysis revealed that high hs-cTnI concentrations were associated with increased death rates during the first 14 days (adjusted hazard ratio, 1.72; 95% confidence interval, 1.14–2.59 and hazard ratio, 1.70; 95% confidence interval, 1.10–2.62 for hs-cTnI concentrations of 100–500 and >500 ng/L, respectively) but not thereafter. Furthermore, elevated hs-cTnI levels were associated with the development of cardiovascular disease among 200 hospital survivors who were analyzed for this end point (adjusted subdistribution hazard ratio, 1.25; 95% confidence interval, 1.04–1.50).
Conclusions: Myocardial injury occurs in the majority of patients with sepsis and is independently associated with early—but not late—mortality, as well as postdischarge cardiovascular morbidity.
Introduction
WHAT IS KNOWN
Sepsis is frequently complicated by the release of cardiac troponin (indicating myocardial injury).
Troponin release is associated with increased case fatality during sepsis, but its impact on long-term mortality and cardiovascular morbidity remains unclear.
WHAT THE STUDY ADDS
Systematic cardiac troponin measurements reveal that troponin release during sepsis is independently associated with increased short-term, but not long-term, death rates.
High troponin concentrations (>100 ng/L) do not convey additional mortality risk compared with moderately elevated levels (>26 ng/L).
Patients exhibiting troponin release during sepsis are at increased risk of acquiring new cardiovascular medication prescriptions during the first year after hospital discharge
Myocardial dysfunction affects approximately half of patients with sepsis admitted to intensive care units (ICUs).1 Cardiac abnormalities can involve both the left and right ventricles and may include both systolic and diastolic dysfunctions.2 Studies have shown that plasma concentrations of troponin, as biomarker of myocardial injury, correlate well with functional abnormalities seen on echocardiography.3 Importantly, troponin levels were independently associated with mortality after sepsis in 2 recent meta-analyses.4,5 Both studies, however, noted significant limitations of the existing literature. For example, many previous studies were hampered by small sample sizes,3,6–16 nonsystematic blood sampling for troponin measurement,12,17,18 and a general lack of adequate adjustment for potential confounders.6–9,13,15 In addition, most reports focused exclusively on short-term mortality, with possible associations between troponin release and long-term mortality and cardiovascular morbidity remaining largely unstudied.4 As a result, it is currently uncertain whether troponin release has significant impact on morbidity and mortality after sepsis.
To address these uncertainties, we performed daily measurements of high-sensitivity cardiac troponin I (hs-cTnI) plasma concentrations in a large cohort of patients presenting to the ICU with sepsis and determined their association with mortality and cardiovascular morbidity during the first year after hospital discharge.
Methods
Study Population
The current analysis was embedded within the MARS (Molecular Diagnosis and Risk Stratification of Sepsis) initiative, a prospective cohort study in the mixed ICUs of 2 tertiary care hospitals in The Netherlands.19 We included consecutive patients presenting to the ICU with sepsis between 2011 and 2013. Sepsis was prospectively defined according to the 1991 American College of Chest Physicians/Society of Critical Care Medicine consensus definitions.20 The post hoc likelihood of infection had to be rated at least probable, using criteria as described in detail previously.21 In addition, the clinical onset of infection had to occur in a time period ranging from 2 days before until 1 day after ICU admission. We excluded patients with inflammatory processes primarily involving the heart (ie, myocarditis, pericarditis, and endocarditis), patients after recent cardiac surgery, and those after cardiopulmonary resuscitation (because these conditions are all known to potentially result in significant troponin release unrelated to sepsis). An opt-out consent procedure was in place, as approved by the medical ethical committee of the University Medical Center Utrecht under protocol number 10-056C/15–232. The data that support the findings of this study are available from the corresponding author on reasonable request.
Data Collection
Trained observers prospectively collected detailed information on patient characteristics, comorbidities, and disease severity (including daily sequential organ failure assessment [SOFA] scores). In addition, heparin plasma was obtained from all patients on a daily basis and stored for analysis at −80°C within 4 hours of sampling. Subsequently, we measured troponin concentrations in plasma during the first 4 days in ICU, using a high-sensitivity troponin I assay (Abbott ARCHITECT STAT, Abbott Laboratories, Abbott park, IL). The upper reference limit for this test is 26 ng/L based on the 99th percentile of measurements reported in healthy men and women, and the coefficient of variation is 10% at the 3 ng/L level according to the manufacturer’s specification.22 Troponin measurements performed as part of this study were not made available to the treating clinicians.
Outcomes
The primary study outcome was all-cause mortality within the first year after ICU admission, for which vital status was retrieved through the Dutch municipal registry. As a secondary outcome, we assessed the occurrence of cardiovascular disease during the first year after hospital discharge. To this end, we combined clinical information as collected in the MARS cohort with reimbursement data as provided by a large Dutch health insurance company (Achmea, Zeist, The Netherlands). To ensure the privacy of patients, records were pseudonymized via a trusted third party (ZorgTTP, Houten, The Netherlands) before data were merged. The Achmea Health Database is representative of the urbanized Dutch population and contains all drug prescriptions delivered by pharmacists both before and after hospitalization for the sepsis episode.23 Anatomical Therapeutic Chemical (ATC) codes were used to determine whether patients received drugs from any of the following 6 major categories of cardiovascular medication:(1) β-blockers (ATC-code C07), (2) lipid-lowering drugs (ATC-code C10), (3) drugs acting on the renin–angiotensin system (ACT-code C09), (4) antithrombotic drugs (ATC-code B01A), (5) nitrates (ATC-code C01DA), and (6) diuretics (ATC-code C03).24 The secondary outcome new or worsened cardiovascular disease was subsequently defined as a ≥2 increase in the number of concurrent categories of cardiovascular medication (as listed above) used by the patient in the first year after hospital discharge as compared with preadmission chronic use. Medication use was based on whether a patient had filled at least 1 prescription for the drug at his/her pharmacy. As a sensitivity analysis, we required at least 2 prescriptions to fulfill the definition of medication use. New drug prescriptions have been used in the literature before as proxies for various cardiovascular events, including atrial fibrillation,25 ischemic heart disease,26 and congestive heart failure.27 A recent study assessing the accuracy of newly prescribed nitrates and antithrombotic drugs as a proxy for incident hospitalizations with major cardiovascular disease found a sensitivity of 71% and specificity of 94%.28
Statistical Analysis
Multivariable Cox regression analysis was used to study the relationship between troponin release during sepsis and mortality up to 1 year after ICU admission. We adjusted for potential confounders that were selected a priori based on their reported associations with both myocardial injury and death in the literature.3,18,29,30 These included age, sex, smoking status, admission type (surgical versus medical), site of infection (pulmonary, abdominal, and other), and a history of chronic renal insufficiency, diabetes mellitus, hypertension, peripheral vascular disease, myocardial infarction, chronic obstructive pulmonary disease, and congestive heart failure (for full definitions of chronic comorbidities, see Table I in the Data Supplement). In addition, we adjusted for acute disease severity at ICU admission by adding the Acute Physiology and Chronic Health Evaluation IV score and vasopressor use as confounders. We deliberately did not adjust for other factors that occur during ICU stay, such as acute kidney injury, because this may be an intermediate on the causal pathway from myocardial injury to death. For the secondary outcome, we used multivariable Cox regression analysis with competing risks (new or worsened cardiovascular disease and death). Cause-specific hazard ratios (HRs) were first estimated for both (competing) end points. A Fine and Gray model was then used to estimate a subdistribution HR reflecting the association between troponin release and new or worsened cardiovascular disease during the first year after sepsis while accounting for competing events. Potential confounders for this analysis included all demographic and chronic comorbidities as listed above, as well as the number of cardiovascular drug categories prescribed before ICU admission, and whether a troponin measurement ordered by the treating clinicians during ICU stay was elevated. The latter element was included to minimize bias because of misclassification of the outcome (ie, spuriously attributing drug treatment to new or worsened cardiovascular disease when in fact it was started in response to clinically recognized troponin release).
For the primary analysis, we used the hs-cTnI concentration measured at ICU admission as the determinant of interest. For our secondary outcome, follow-up time started at the moment of hospital discharge, and for this reason, we used the peak hs-cTnI concentration measured during the first 4 days in ICU as determinant. All hs-cTnI levels were transformed using a natural logarithm before model fitting. We subsequently assessed linearity of the relationship between predictor and outcome for all continuous variables using restricted cubic splines. If a spline fit significantly improved the model (using a likelihood ratio test), then a nonlinear relationship was assumed to be present. In addition, the proportional hazards assumption was tested by visual inspection of scaled Schoenfeld residuals and by testing for statistically significant interactions of covariates with time. In case the proportional hazards assumption appeared to be violated, we divided follow-up time into discrete segments and subsequently obtained separate HR for each time period. It is important to note that period-specific HRs have the limitation that they only represent the effect of exposure for patients who have at least survived until that time point.31 In addition, HRs provide a measure of relative effect and thus no information about absolute survival probabilities. Therefore, we also constructed adjusted survival curves, using inverse probability weighting to adjust for confounding.32 Details on this methodology can be found in the Data Supplement.
We performed multiple imputation for missing data under the assumption of data being missing at random (see Data Supplement). All analyses were performed using SAS version 9.4 (SAS institute, Cary, NC) or R version 3.2.2 (R foundation for Statistical Computing, 2015).
Results
During the study period, a total of 1256 patients were admitted to the participating ICU’s for presumed sepsis with a post hoc infection likelihood rated at least probable, of whom 1124 (89%) were eligible for study inclusion (Figure 1). Among these, 673 (60%) subjects had a troponin level above the upper reference limit of the test (26 ng/L) on day 1, and an additional 82 (7%) developed raised concentrations within the first 4 days in the ICU. In this group of 755 (67%) patients (ever) having troponin release, median hs-cTnI plasma concentrations on days 1 through 4 were 109 (interquartile range [IQR], 39–394), 103 (IQR, 38–449), 79 (IQR, 31–281), and 82 (IQR, 32–253) ng/L, respectively. Of note, 568 (84%) of the 673 patients who had raised troponin on day 1 also had elevated levels on all subsequent days. Patients with elevated troponin concentrations were older, more severely ill, and had more comorbidities than patients with normal troponin levels (Table 1).
Characteristics of Patients Admitted to the Intensive Care Unit With Sepsis (n=1124)
Patient inclusion flow chart. hs-cTnI indicates high-sensitivity cardiac troponin I; and ICU, intensive care unit.
Association With Mortality
Crude mortality rates in the ICU, in hospital, and after 1 year were higher in patients having elevated compared with normal hs-cTnI plasma concentrations at admission (Table 1). Initial multivariable Cox regression revealed that the proportional hazards assumption for this analysis was violated (see Figure I in the Data Supplement). In addition, the use of restricted cubic splines for hs-cTnI significantly improved model fit (P value for likelihood ratio test, 0.04). For these reasons, we performed cox regression analyses while splitting follow-up time into distinct periods to accommodate nonproportionality and incorporating spline functions for hs-cTnI to accommodate the nonlinear association.
Figure 2 shows the relationship between hs-cTnI plasma concentrations and mortality rate for 2 distinct time periods after sepsis onset (ie, days 0–14 and 14–365). There was a steady increase in the hazard of dying during the first 2 weeks for rising troponin concentrations up to ≈100 ng/L, yet further troponin release did not lead to a higher mortality rate during this time period (Figure 2A). These findings were confirmed in a multivariable Cox regression analysis using categorization (rather than spline transformation) of troponin values, revealing that both moderate (100–500 ng/L) and high (>500 ng/L) hs-cTnI levels were associated with significantly increased—but very similar—death rates (adjusted HR, 1.72; 95% confidence interval [CI], 1.14–2.59 and adjusted HR, 1.70; 95% CI, 1.10–2.62; Table II in the Data Supplement). In contrast, neither mild, moderate, nor high levels of troponin were associated with increased death rates during the time period ranging from day 14 to 365 (Figure 2B; Table II in the Data Supplement).
Adjusted hazard ratio plots for the effect of troponin on mortality across different time periods (n=1124). A and B, The blue bands represent 95% confidence intervals. Hazard ratio for high-sensitivity cardiac troponin I (hs-cTnI) is adjusted for age, sex, chronic renal insufficiency, peripheral vascular disease, diabetes mellitus, hypertension, prior myocardial infarction, chronic obstructive pulmonary disease, active smoker, congestive heart failure, site of infection, Acute Physiology and Chronic Health Evaluation IV score, and vasopressor requirement.
To gain a deeper understanding of the relationship between troponin release and early sepsis mortality, we analyzed SOFA scores at the time of death in patient who died within the ICU in the first 14 days (n=160). Patients having elevated hs-cTnI at admission (129 of 160; 81%) more often died with renal failure than subjects having normal hs-cTnI concentrations (renal SOFA score 4 [IQR, 2–4] versus 0 [IQR, 0–2]; P<0.001) while no differences in other organ systems or total SOFA score were observed (data not shown).
Figure 3 shows the adjusted survival curves for patients with normal hs-cTnI, as well as mild, moderate, and high levels of hs-cTnI release at ICU admission. After 1 year, there was no statistically significant difference in estimated survival between normal (59%; 95% CI, 54%–65%) and mild (51%; 95% CI, 45%–58%; P value, 0.07), moderate (55%; 95% CI, 48%–63%; P value, 0.40), or high hs-cTnI concentrations (53%; 95% CI, 44%–64%; P value, 0.29).
Adjusted survival curves for different categories of troponin release on intensive care unit admission (n=1124). hs-cTnI indicates high-sensitivity cardiac troponin I.
Association With Cardiovascular Morbidity
Among the 1124 study participants, 787 survived until hospital discharge. A subgroup of 226 (29%) of these were matched to the Achmea Health Database; 23 patients were excluded because they already used medication comprising ≥5 cardiovascular drug categories before the onset of sepsis (and were thus not at risk of developing the outcome), and 3 additional patients were excluded because of missing prescription data, leaving 200 patients for this secondary analysis. Patients carrying health insurance by Achmea had similar characteristics compared with those who were insured by other companies (Table III in the Data Supplement). The 122 patients of this subgroup who had troponin release during sepsis received more cardiovascular drugs in the first year after hospital discharge than before the sepsis episode while the 78 patients having normal hs-cTnI plasma concentrations during their first 4 days in the ICU did not (Table 2; Figure II in the Data Supplement). As a result, 20 (16%) subjects with troponin release met predefined criteria for new or worsened cardiovascular disease, compared with only 4 (5%) patients with hs-cTnI ≤26 ng/L (P value, 0.02). This association remained after multivariable adjustment for confounders (adjusted cause-specific HRs, 1.25; 95% CI, 1.00–1.55; P value, 0.045). A Fine-Gray model, accounting for the competing end point of death, showed a similar relationship (adjusted subdistribution HR, 1.25; 95% CI, 1.05–1.50; P value, 0.015). A sensitivity analysis requiring at least 2 prescriptions filled at the pharmacy to establish cardiovascular medication use yielded similar results (data not shown).
Cardiovascular Drug Use Before and After Admission in Hospital Survivors of Sepsis (n=200)
Discussion
We performed a detailed analysis of the relationship between myocardial injury and mortality in a large cohort of patients with sepsis and found that troponin release at ICU admission was associated with early, but not late, mortality. This association was nonlinear, indicating that progressive troponin release beyond hs-cTnI plasma concentrations of ≈100 ng/L does not convey an additional mortality risk. Furthermore, raised hs-cTnI levels during the first 4 days in the ICU were associated with increased use of new cardiovascular medication after hospital discharge.
Although an association between troponin release and short-term mortality after sepsis has been reported before, no studies evaluated the possibility of a nonlinear relationship.4,5 Prior studies either categorized troponin concentrations or simply assumed a linear outcome relation, and these approaches may lead to loss of information or even erroneous conclusions.33–35 We observed a clear nonlinear relationship with mortality, yet it remains uncertain whether this represents causality. However, there are several pathophysiological mechanisms that could explain a possible pathogenic link. For example, troponin release is known to be a marker of poor left and right ventricular functions during sepsis,2,3 and reduced cardiac output may impair tissue perfusion and increase the risk of death.36 Analysis of SOFA scores at the time of death revealed that patients with hs-cTnI release on admission were more likely to subsequently develop renal failure, which fits the theory that myocardial injury might lead to reduced cardiac output, thereby increasing the risk of renal injury. However, it should be noted that both the incidence of chronic kidney injury was already dissimilar at baseline. In our study, extreme troponin peaks were not associated with additional deaths relative to milder troponin levels. Perhaps, high hs-cTnI concentrations no longer accurately reflect deterioration of cardiac function, or perhaps active medical treatment blunts the effects of underlying myocardial dysfunction. Nonetheless, increased short-term mortality has now repeatedly been reported across various clinical settings, and it is thus essential that future studies attempt to uncover the underlying pathogenesis of myocardial injury during sepsis. This may lead to therapies (such as heart rate control or antiplatelet therapy) aimed at improving (short-term) survival.
Most investigations thus far have only reported on short-term mortality after troponin release, leaving long-term outcomes largely unstudied. However, a cohort study in patients with end-stage renal disease suggested that troponin elevation was associated with mortality up to 4 years after sepsis onset.14 In contrast, we did not observe an association with death rates beyond day 14, perhaps because our cohort included patients having a greater burden of acute disease severity, yet who (mostly) did not have chronic renal disease. This notion is supported by another study that also did not observe a relationship between cardiac troponin T concentrations and late mortality in a cohort of sepsis patients admitted to ICU who survived the first 30 days after disease onset.18 Although we found no association with late mortality, we did observe an association between hs-cTnI concentrations and the increased use of new cardiovascular medications after hospital discharge. Sepsis survivors are known to be at increased risk for cardiovascular events after hospital discharge compared with matched population controls,37,38 and we found that troponin release during ICU stay might identify patients especially prone to this complication. Our findings are corroborated by a recent study in patients who were admitted to the hospital with community-acquired pneumonia, which also showed that high troponin levels on admission were associated with the development of cardiovascular events during follow-up.39 Regardless of whether this association is causal or merely prognostic, troponin release during sepsis could act as an important alarm signal for physicians, identifying patients at high risk of cardiovascular complications, and future studies should evaluate whether they may benefit from preventative strategies.
Our study has several limitations. First, we chose to accommodate nonproportional hazards during follow-up by splitting time into 2 distinct periods. Because the cutoff point for defining these time periods was selected based on a plot of the Schoenfeld residuals, the resulting Cox model should be viewed as a post hoc analysis. We opted to categorize follow-up time in favor of other methods of dealing with nonproportional hazards, such as Gray’s survival model,40 because it allowed us to better depict the nonlinear associations (Figure 2). Despite the post hoc nature of these decisions, the large study cohort (including >1000 patients with sepsis) provided ample statistical power to explore nonlinear and time-dependent relationships. Second, our definition of new or worsened cardiovascular disease was based on newly prescribed cardiovascular medications. Use of a proxy, rather than direct observation of the outcome of interest, may lead to misclassification. Furthermore, some of the new cardiovascular drug prescriptions filled at the pharmacy shortly after discharge might in fact already have been initiated during hospital stay. Thus, these prescriptions may not reflect true new morbidity but rather be related to secondary cardiovascular prophylaxis after clinically recognized troponin release in the ICU. However, we included clinically recognized troponin events as a confounder in our analysis to limit potential bias by such misclassification. Nonetheless, because the number of patients included in our analysis of the secondary end point was low relative to the many potential confounders in the model, we regard this analysis as exploratory.
Conclusions
Myocardial injury occurs in a majority of patients with sepsis and is independently associated with early—but not late—mortality. Furthermore, patients sustaining myocardial injury during their ICU stay are at increased risk of developing new or worsened cardiovascular disease during the first year after hospital discharge.
Acknowledgments
We thank all members of the MARS (Molecular Diagnosis and Risk Stratification of Sepsis) consortium and trial nurses for their participation in data collection. We are grateful to Tesy Merkx and Ron Stokwielder for their assistance in laboratory measurements. In addition, we thank Achmea Health Database for the use of the insurance data and ZorgTTP for the pseudonymization of the data. Dr Frencken, Dr Donker, Dr Spitoni, M.E. Koster-Brouwer, Dr Soliman, Dr Ong, Dr Poll, Dr Horn, Dr Klei, Dr Bonten, and Dr Cremer substantially contributed to the conception and design of the study. Dr Frencken and M.E. Koster-Brouwer acquired the data. Dr Frencken had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Frencken, Donker, Spitoni, Bonten, and Cremer were involved in the interpretation of the data. Drs Frencken and Cremer drafted the manuscript, and all authors revised it critically for important intellectual content. All authors gave final approval of this version to be submitted.
Sources of Funding
This study is supported by the Center for Translational Molecular Medicine (http://www.ctmm.nl), MARS (Molecular Diagnosis and Risk Stratification of Sepsis) project (grant 04I-201).
Disclosures
None.
Appendix: MARS Consortium
Members of the MARS (Molecular Diagnosis and Risk Stratification of Sepsis) consortium: Friso M. de Beer, MD; Lieuwe D.J. Bos, PhD; Gerie J. Glas, MD; Roosmarijn T.M. van Hooijdonk, MD, PhD; Laura R.A. Schouten, MD; Marleen Straat, MD; Esther Witteveen, MD; Luuk Wieske, MD, PhD; Lonneke A. van Vught, MD, PhD; Maryse Wiewel, MD, PhD; Arie J. Hoogendijk, PhD; Mischa A.Huson, MD, PhD; Brendon Scicluna, PhD; Marcus J. Schultz, MD, PhD (Academic Medical Center, University of Amsterdam, Amsterdam, the Netherlands); Peter M.C. Klein Klouwenberg, MD, PhD; Kirsten van de Groep, MD; Diana Verboom, MD (University Medical Center Utrecht, Utrecht, the Netherlands).
Footnotes
The Data Supplement is available at http://circoutcomes.ahajournals.org/lookup/suppl/doi:10.1161/CIRCOUTCOMES.117.004040/-/DC1.
Circ Cardiovasc Qual Outcomes is available at http://circoutcomes.ahajournals.org.
- Received June 23, 2017.
- Accepted November 30, 2017.
- © 2018 American Heart Association, Inc.
References
This Issue
Jump to
Article Tools
Share this Article
- Myocardial Injury in Patients With Sepsis and Its Association With Long-Term Outcome
