The main findings of this study are: 1) hsTNT was the only independent predictor of 1-year mortality after adjustment for other factors; 2) elevated levels of hsTNT were found in the majority of patients; 3) E/é was higher in non-survivors; 4) neither BNP nor echocardiographic LV systolic function parameters were predictive of 1-year mortality.
Early elevation of hsTNT is an independent predictor of 1-year mortality in critically ill patients with shock
Elevation of cTn is common among ICU patients for several reasons including myocardial infarction, sepsis and renal failure
 and is known to be predictive of mortality during shorter follow-up periods such as ICU- and hospital mortality
[14, 15, 41–44]. Even in critically ill patients where coronary artery disease has been excluded, elevated cTn is known to be associated with increased mortality
. In medical ICU patients, elevated cTn measured within 12 h of admission has been shown to be an independent risk factor for 30-day and 2-year mortality after adjustment for severity of illness
. The recent introduction of a new generation of high sensitivity assays for cTn, with a lower detection limit and sufficient analytical precision
[9, 10], allows the detection of elevated cTn in a considerably higher frequency compared to earlier essays (97% vs 76%) in patients with cardiovascular disease
[10, 11]. In general, renal insufficiency and cardiologically ill populations, detectable levels of hsTnT are associated with adverse outcomes
[13, 45, 46]. However in critically ill patients, information about the value of high sensitive cTn is scarce. In a study by Rosjö et al.
, hsTNT on inclusion was detectable in all patients with sepsis and septic shock. Further, hsTNT correlated to severity of disease and was significantly higher in hospital non-survivors but could not be identified as an independent predictor of mortality. Reynolds et al.
 showed that an increased cTnI concentration while in ICU was associated with increased mortality in hospital, after adjusting for admission characteristics, age, severity of illness at admission, organ support, and serum creatinine concentrations. These results are in line with our study, where hsTNT measured within 12 h was detectable in 100%, elevated in 92% of patients, and correlated significantly with critical illness scores. Further, hsTNT was identified as the only independent predictor of even longer-term (1-year) mortality. The findings are strengthened by the increased odds ratios for mortality for increasing quartiles of hsTNT identified in this population although the wide CIs indicate that larger studies are needed to support our findings.
We also note that median hsTnT in non-survivors was higher in our study compared to the study by Rosjö et al.: 168 vs 54 ng/l. As the frequency of cardiovascular disease was comparable in both studies (24% and 26% respectively) we speculate that this might be due to sicker patients in our non-survivor group (median SOFA 13 vs 9 in Rosjö et al.). Since this was an exploratory study, and due to the paucity of literature regarding accepted levels of hsTNT in the critically ill, we chose the cut-off level with the best balance of specificity and sensitivity. We do not know if this is adequate and we hope that future studies will inform us as to what to expect from different critically ill populations. We note that the cut-off point identified by our ROC analysis is much higher than that identified in non-critically ill patients. The reason for this and its relevance is unclear and deserves attention in future studies.
We investigated if there was a confounding relationship between hsTNT and pre-existing cardiac disease or renal insufficiency but found no significant association which is propably due to sample size. Additionally when entering these confounders into the multivariable model, hsTNT could still be identified as the sole significant predictor of mortality.
Diastolic but not systolic function parameters are associated with mortality
Impairment of myocardial function in patients with shock is often masked by concurrent elevations in cardiac index and a low systemic vascular resistance, making parameters such as LVEF often unreliable for monitoring LV systolic function and as a prognostic marker
[1, 47]. Additionally, markers of LV systolic function are frequently described to be normal or near normal
[3, 4, 24] in these patients. This is in line with our results, where all echocardiographic parameters of LV systolic function were normal or mildly reduced and none were predictive of 1-year mortality. LV diastolic dysfunction with increased filling pressures is known to be predictive of mortality in cardiac patients
 but has shown conflicting results regarding prognosis in ICU patients
[3, 4, 24, 25]. In our study E/é and La volume, both surrogates of LV filling pressure, were predictive of mortality but other diastolic parameters were not. E/é did, as expected, correlate significantly with age but not with hsTNT, BNP, APACHE II, SOFA score or lactate. These results were to some extent unexpected, as E/é has been shown in previous studies to be correlated to the severity of critical illness (3, 47).
Recent studies allude to the importance of E/é to prognosis in critically ill patients with shock
[3, 25]. Sturgess et al.
 identified E/é as an independent predictor of hospital mortality, although with a considerably higher cut-off value than in our study (E/é = 14.5). This might be attributable to a higher percentage of pre-existing cardiac disease (43% in that study vs 24% in our study) and narrower inclusion criteria (septic shock vs shock). An É/é ratio < 8 cm/s is usually associated with normal filling pressure and E/é ratio > 15 cm/s commonly with increased filling pressure
. In our study median E/é was 10.1, thus representing a level between 8–15 where filling pressures might be elevated according to international guidelines
. Our study population represents a group of patients with increased age, pre-existing cardiovascular disease as well as acute critical illness. All can affect diastolic function and thus filling pressures. As we do not know to what extent E/é is affected by either of these different entities, we cannot separate their cardiac effects. Although E/é was only mildly increased it was still predictive irrespective of the underlying cause.
BNP is not a valuable marker of 1-year mortality in this population
The role of natriuretic peptides as prognostic markers is well described in patients with cardiovascular disease
. Even in ICU patients several outcome studies refer to their usefulness
[21, 22]. Nevertheless the prognostic value of natriuretic peptides has been described as questionable
, as they can be elevated due to a variety of reasons in critically ill patients
. Age, gender, pre-existing or critical illness associated renal and myocardial impairment as well as inflammatory states such as sepsis or septic shock all affect BNP
[20, 22]. In our study, elevated BNP was seen in most patients (98%) but did not correlate with critical illness (APACHE II) or organ dysfunction (SOFA), nor discriminated survivors from non-survivors. As LV systolic function overall was near normal and no patient had acute heart failure as the sole diagnosis, we speculate that elevated BNP due to other factors than heart failure is of little prognostic value.
Our study group containing the sickest of ICU patients with hemodynamic instability, implying cardiovascular impairment, is prone to bias and our results could have been completely different in another set of ICU patients. Therefore there is a risk of bias that could have led to overestimation of the prognostic ability of hsTNT. We have not excluded patients with known heart failure or atrial fibrillation, nor have we excluded patients with new onset of atrial fibrillation during their critical illness, which might have influenced our results. We did not record the absence or presence of ischemic ECG changes. This could have been of additional value to the echocardiographic examination in interpreting the likely cause of hsTNT elevations although this was not the aim of this study. In this observational study our intention was to investigate a group of critically ill patients with shock knowing that increased age, the likelihood of pre-existing cardiovascular disease and critical illness induced cardiac abnormalities such as atrial fibrillation, ischemic and/or cardiomyopathy probably would influence our results. Excluding patients with pre-existing cardiac disease would probably have reduced the cardiological impact of the regression model but would also have made the sample less representative of the population of critically ill patients. Since coronary angiography was not a possibility in this study, and since patients did not have pre-morbid echocardiograms, it is possible that we may have identified a subpopulation of critically ill patients with cardiac disease. We maintain however, that in this general group of very ill patients with shock, only hsTNT was indentified as a predictor of 1-year mortality, regardless of aetiology and background co-morbidity. A larger study stratifying BNP by age and gender might have yielded different results. TDI measurements were only done at the septal mitral annulus whereas current recommendations include both the septum and lateral wall
. Further a blinded assessment of echocardiographic data would have been desirable. Finally, the sample size is small; hence only limited variables could be used for the model, increasing the likelihood of confounding. We have attempted to reduce this by using univariate analysis to identify probable predictors, limiting the number of predictors and including these in the multivariate model. The results were congruent for different models that showed some consistency over the outcome predictor.
The strength of this study is that both LV systolic and diastolic echocardiographic measurements together with cardiac biomarkers were analysed as predictors of longer-term outcome.