Abstract
Background Sepsis is a common admission diagnosis in the intensive care unit (ICU). The Sepsis-3 consensus associates sepsis diagnosis with acute organ dysfunction. In these patients troponin elevation is a well-established phenomenon, but its clinical significance is not settled, as no systematic review has addressed the prognostic significance of the increasingly prevalent high-sensitivity troponin assays in acute organ dysfunction setting.
This study aims to clarify the association between early serum troponin levels in high-sensitivity assays with short-term mortality risk in septic patients with acute organ dysfunction.
Methods We will systematically search PubMed, Scopus and Embase for original articles; additionally, a manual search will be carried out through relevant literature. Generally, studies will be deemed eligible for inclusion if they evaluate the association between high-sensitivity troponin in the first 24 hours of admission and ICU, 30-days, or In-hospital mortality; in patients with septic shock or sepsis related to acute organ dysfunction. Two reviewers will independently select studies and extract the data. A meta-analysis for mortality outcome will be performed for comparative data regarding two effect measures: Odd ratios and Standardized Mean differences.
Discussion This study will provide further evidence about the role of high-sensitivity troponin assays in predicting mortality in septic patients; potentially helping to guide further research and yielding valuable information for patient assessment.
Conclusion about the certainty of evidence will be presented in a ‘Summary of finding’ table.
PROSPERO registration: (CRD42024468883)
1. Background
Sepsis is the most common mortality cause in Intensive Care Unit (ICU) patients in the United States. A fourth of patients diagnosed with sepsis will die within their hospitalization, rising to a half within the septic shock group (1). Furthermore, sepsis-related deaths have been on the rise in the US and worldwide (1–3), and are expected to continue to do so.
Sepsis definitions have varied throughout the years. Sepsis-3 consensus, agreed in 2016, defines sepsis as a life-threatening organ dysfunction (assessed by SOFA score) in behalf of an infection; disposing of the former category of severe sepsis while preserving septic shock (4). Thus, compared with previous sepsis definitions, patients diagnosed with sepsis-3 criteria are exposed to an increased mortality and adverse events risk (5).
Sepsis is associated with complications and dysfunction of several systems (6). At the cardiac level, sepsis is well-documented to be associated with acute myocardial injury independent of coronary perfusion abnormality (7). This phenomenon is primarily attributed to inflammatory and cardio-depressant factors that act indirectly and directly on the cardiomyocyte (7,8). Furthermore, when the myocardial injury ensues in diastolic and/or systolic dysfunction, it is considered Septic Cardiomyopathy (7). Although there is no consensus on the criteria for this entity, commonly accepted cardinal features are: The absence of an underlying coronary etiology, the acute presentation, and the reversibility of the dysfunction (7–9).
Cardiac troponins are widely used in the diagnosis of acute coronary syndrome and other cardiac pathologies presenting with myocardial injury (10), including sepsis. High-sensitivity cardiac troponin assays (hs-cTn) are usually defined by their capacity to detect troponin in at least 50% of healthy individuals (10,11). In recent years hs-cTn have increasingly replaced conventional assays since they have both a lower limit of detection, and a lower and categorically defined normality cut-off. Thus, a hs-cTn level above the 99th percentile for the healthy population allows for a precise differentiation between normal and elevated groups (11).
Troponin elevation in septic patients is well-established but debated as a prognostic risk factor for clinical outcomes (7,12). Three meta-analyses have been published on the issue (13–15), all showed a significant association of troponin elevation with increased mortality, but suffer from many problems of published data on troponin and mortality in sepsis: Lack of strict or similar criteria for sepsis diagnosis, heterogeneity of sampling time, usage of conventional rather than hs-cTn essays, and absence of controlling of confounders (7). Moreover, the emergence and consolidation of Sepsis-3 definition, and the rise of hs-cTn assays could impact on the stratification of patients, modifying the prognostic association of the biomarkers with mortality risk (16). Consequently, it is currently unknown if elevation in hs-cTn levels is associated with mortality risk in patients with sepsis by Sepsis-3 consensus.
This study aims to assess the association of early high-sensitivity serum troponin levels with short-term mortality in patients admitted to ICU because of sepsis with acute organ dysfunction.
1. Methods
2.1. Registration
The protocol was registered on the International Prospective Register of Systematic reviews (CRD42024468883). This protocol aligns with Preferred Reporting Items for Systematic review and Meta-Analysis Protocols (PRISMA-P) statement (17), and follows Rileýs guide for systematic review of prognostic factors (18). Deviations from protocol will be recognized in the final article.
2.2. Literature search
With the help of a knowledgeable librarian a search strategy was created: Pubmed, Scopus and Embase will be searched for primary articles published in English and involving humans, with no date restrictions. The prepared search strings combine free text and MeSH/EMTREE terms for all words related to troponin and sepsis (see Table 1). No additional filters besides the ones described in the strings will be utilized.
Subsequently, we will manually scrutinize references of included studies and previous reviews on the topic to retrieve missing pertinent papers.
Searches will be re-run just before the final analyses to check for new studies that meet the inclusion criteria.
2.3. Study selection
In accordance to CHARMS guidance (19), the research question was formulated with the Population/Index prognostic factor/Comparator prognostic factors/Outcome/Timing/Setting (PICOTS) system, a modified version of the traditional PICO, where the P and O remains the same as the original, but the I stands for Index prognostic factors and C for other prognostic factors that can be considered as Comparators (18). (See table 2 for details on the defined scope of the systematic review). For the determination of inclusion and exclusion criteria table 3 was constructed.
Based on this criteria, after deleting duplicates the corpus of studies will be scanned in Rayyan (20) by two independent reviewers, discrepancies will be solved by a third reviewer. Studies will be discarded based on 1 criterion only.
Studies will be first analyzed based on their title and abstract. Papers that pass this primary scrutiny, or whenever it seems insufficient, will be subject to full text assessment by the same reviewers for final selection.
Basically, interventional and observational primary studies will be deemed eligible if they reported some form of association between troponin levels and mortality in a general clinical context. We will only include studies that report on patients diagnosed either with sepsis by Sepsis-3 criteria (4), with Severe sepsis by Sepsis-1 or Sepsis-2 (21,22), or Septic shock by any of them (4,21,22). Selection will be restricted to studies that specify the utilization of any hs-cTn assay in the first 24 hours after hospital admission, or whose reported limit of detection we consider high-sensitivity (i.e. < 9 ng/L)(10). In the case that the latter information about the assays is omitted, we will exclude studies if they fail to meet a normality cutoff of 20 ng/L or lower, since they will be regarded as having a cutoff over the 99th percentile of the healthy population and therefore considered as non hs-cTn assay, based on European Society of Cardiology guidelines (10).
The study selection process will be reported using the PRISMA flowchart, providing a clear visual representation of the inclusion and exclusion criteria application (17).
2.4. Data extraction
Based on CHARMS-PF checklist (18) we designed a standardized data extraction form, containing the relevant characteristics of studies to be included (table 4). Data will be extracted by two reviewers. Types of reported associations suitable to be extracted for this table are: 1) “dose-response”: data reported as odds, risk, or hazard ratio per unit increase in exposure; 2) “category/quantile based”: numbers (ideally 2x2 tables) or ratios comparing groups as defined by quantiles or categories of exposure; 3) “means”: data reported as means or mean differences in exposure, comparing those presenting and not presenting the event (23). If a study presents more than one of these associations, we will extract all of them.
We will collect extracted effect measures together with their standard errors (SE). If SEs are not explicitly available, we will derive them from metrics such SDs, exact p-values, or confidence intervals (24). We will collect adjusted and unadjusted effect measures separately. If many adjusted models are presented, we will extract the one which fits better the core adjustment factors selected (see Risk of Bias section). If effect measures for several time points are presented, we will prioritize ICU and 28-mortality (in that order) over in-hospital mortality. If effect measures for several hs-cTn samples are presented, we will prioritize the sample extracted closest to ED/ICU admission. Data only presented as figures will be extracted with WebPlotDigitizer software (25).
Authors will be contacted to request critical missing or not reported data regarding associations of interest. In case of no response, the article will be excluded from this study.
We will present a summarized version of this table in the final review.
2.5. Risk of bias assessment
Bearing in mind the prognostic nature of the review and taking into account previously published meta-analysis (13–15), we can expect that mainly observational studies will be included.
Two authors will independently assess the risk of bias (RoB) applying the QUIPS (quality in prognostic factor studies) tool (26), and then discuss together their assessments in order to decide the final score. In QUIPS RoB assessments are made within a set of 6 domains: (1) Study Participation, (2) Study Attrition, (3) Prognostic Factor Measurement, (4) Outcome Measurement, (5) Study Confounding and (6) Statistical Analysis and Reporting. We defined Study participation, Prognostic Factor Measurement and Study Confounding as the key domains for our assessment, and we specified criteria for each of these domains. For the confounder domain RoB evaluation, we will look for 3 core adjustment factors in each study: severity of sepsis, age, and comorbidities (any cardiac or renal).
Classification for each RoB item could be ‘low’, ‘moderate’, or ‘high’. Moreover, this assessment will include an overall judgment concerning the total RoB quality of each study, which will be visualized in the review.
2.5. Statistical analysis
We will investigate the mortality outcome with two effect measures: Mortality risk and Mean difference in troponin distribution.
1) Regarding mortality risk analysis: Whenever feasible a 2x2 contingency table will be gathered from available data, with one axis for survivor vs non-survivor groups, and other for elevated vs normal troponin level group, according to the cut-off defined by every study for each specific hs-cTn assay. If the sample is divided into more than 2 groups or categories (e.g. quantiles) it will be dichotomized around the closest match to a normality cutoff of 15 ng/L (27). We will add 0.5 to each cell in any table that contains one or more zero values (24). We will compute an Odds Ratio (OR) with its SE from this contingency table for each study. For studies not presenting sufficient data for a contingency table we will extract explicitly reported unadjusted ORs with their SEs. ORs from studies handling hs-cTn as a continuous variable (i.e. dose-response) will be scaled to an OR per 50 ng/L of increment. This magnitude was considered representative of the difference between the group usually referred to as “normal” compared to the “elevated” troponin level group –based on the reported troponin distributions of pertinent studies known to us and in accordance with our clinical expert (28–30).
Results from studies reporting Hazard Ratio (HR) or Relative Risk (RR) will be presented separately. Only if the event probability in the control group is <0.2, and the HR or RR are below 2.5, the effect measures will be interpreted as numerically equivalent to an OR and reported together with the other studies (31).
With these effect measures we will perform a structured quantitative analysis (meta-analysis) for the pooled unadjusted mortality effect.
ORs of adjusted (i.e. multivariate) regressions will be included in a separate model for the adjusted mortality effect regardless of the set of confounders controlled for. We will report variables incorporated in the regression for each study in a table.
2) For difference in troponin levels analysis: Studies presenting mean difference data in troponin level by survivor status will be meta-analyzed in a single model using Hedgeś g Standardized Mean Differences (SMD).
All models will be generated using R-4.3.1 package metafor (32), utilizing the random-effects model and the inverse variance weighting method (24).
Heterogeneity will be assessed and reported by I2 statistic, where >50% and a p-value < 0.10 are of concern and will be further explored (24).
Studies not presenting minimum data (e.g. missing SE) will be included in the narrative synthesis and compared to the results of the meta-analysis to the extent possible.
For the sake of replicability all formulas, raw data, and calculations shall be included either in the article itself or an online supplement.
2.6. Subgroup and sensitivity analysis
Models will undergo a comprehensive heterogeneity assessment. Since reviews of prognostic studies usually can result in high I2 regardless of similar point estimates (33), we will favor visual inspection over statistical criteria.
If in any of the model’s heterogeneity is deemed substantial, we will run the following Subgroup analysis:
1. Subgroups by shock status: To assess whether effect measures and heterogeneity is significantly dependent on the degree of organic dysfunction, we will proceed with a subgroup analysis between studies restricting inclusion criteria to septic shock only, versus studies that encompass patients with sepsis, severe sepsis and septic shock.
2. Subgroup by type of troponin: Despite the fact that troponin essays have longed being considered equivalent in the acute coronary syndrome setting (34), emergency evidence supports that hs-cTnT and hs-cTnI outcome prognostic value might differ in other scenarios (35,36). To assess if such a difference exists in this case, we will group studies by the declared essay type (i.e. hs-cTnI, hs-cTnT, hs-cTnC).
3. Subgroup by type of association: In order to assess whether heterogeneity may be due to the type of analysis employed to derive a dichotomous effect size, a subgroup analysis will be performed segregating studies reporting dose-response data (i.e. regressions handling hs-cTn as a continuous dependent variable) versus category/quantile-based data (i.e. contingency tables and/or regressions handling hs-cTn as a dichotomous dependent variable). This subgrouping is not applicable for the SMD model.
Furthermore, to assess whether missing adjustment for relevant confounders might be impacting the heterogeneity and pooled effect size of the mortality adjusted model: We will undertake a sensitivity analysis, meta-analyzing only effect measures of studies adjusting for (at least) the 3 predefined core confounders.
To avoid p-value overreliance, the credibility of the differential effect of statistically significant Subgroup analysis will be critically appraised with the ICEMAN questionnaire (37).
If we suspect a small study effect in any model, we will run a sensitivity analysis with a fixed effect model.
2.8. Publication bias
For assessment of possible small-study effect, we will generate funnel plots for each model plotting the logarithm of the obtained odds ratios against their standard error (38).
We will visually inspect the plots for signs of publication bias (namely, asymmetry)(18,39). If more than 10 (for the sake of statistical power) studies are included in the model, we will run an Eggeŕs test (40).
3. Discussion and Conclusions
The prognostic significance of troponin for sepsis outcomes is not yet settled. Although previous works have consistently shown a significant association between troponin and mortality, there is a need for more research shedding light on the prognostic value of such biomarker for the most common current clinical setting in the ICU. This study will provide further evidence about the role of hs-cTn assays in predicting mortality in septic patients; potentially helping to guide further research and yielding valuable information for patient assessment.
We will evaluate the certainty of evidence for each investigated outcome with the ‘Grading of Recommendations Assessment, Development and Evaluation’ (GRADE) system guideline appropriate for synthesis of prognostic factors evidence (41,42). A “Summary of findings’ table will communicate with standardized statements the certainty of this evidence
Data Availability
No datasets were generated or analysed during the current study. All relevant data from this study will be made available upon study completion.
Supporting information
S1 Checklist. PRISMA-P (Preferred Reporting Items for Systematic review and Meta-Analysis Protocols) 2015 checklist: Recommended items to address in a systematic review protocol+. (DOC1)
Author Contributions
Conceptualization: Santiago Ferrière-Steinert, Abraham I. J. Gajardo Cortez, José Ramos-Rojas
Formal analysis: Santiago Ferrière-Steinert, Joaquín Valenzuela Jiménez, Sebastián Heskia Araya, Thomas Kouyoumdjian Carvajal
Project administration: Santiago Ferrière-Steinert, Abraham I. J. Gajardo Cortez
Resources: Abraham I. J. Gajardo Cortez
Supervision: Abraham I. J. Gajardo Cortez
Validation: Santiago Ferrière-Steinert, Abraham I. J. Gajardo Cortez
Writing – original draft: Santiago Ferrière-Steinert, Joaquín Valenzuela Jiménez, Sebastián Heskia Araya, Thomas Kouyoumdjian
Writing – review & editing: Abraham I. J. Gajardo Cortez, José Ramos-Rojas