All-Cause Mortality and Trimethylamine N-Oxide Levels in Patients with Cardiovascular Disease

• The predictive value of trimethylamine N-oxide (TMAO) levels is highly limited by other factors. Therefore, mortality in high-risk patients with cardiovascular disease (CVD) is determined by a combined effect of multiple factors that significantly affect not only the risk of death but also the TMAO levels themselves. This should be considered when assessing the prognosis of patients with CVD.

Trimethylamine N-oxide (TMAO) originates from the enzymatic oxidation of the gut microbiota metabolite trimethylamine (TMA). It is also absorbed directly from food [1]. Serum TMAO levels depend on the dietary intake of TMA precursors [2], the quantitative and qualitative composition of the gut microbiota [3], and kidney function [4, 5].

Studies conducted to date have confirmed the involvement of TMAO in the pathophysiological processes such as atherogenesis [6], platelet hyperreactivity [7], and disorders of cholesterol and bile acid metabolism [8]. The adverse effects of TMAO on cardiac hypertrophy and fibrosis [9]were also reported. In the last decade, there have been numerous studies assessing the effect of increased TMAO levels on the risk of adverse cardiovascular events, including death. They mainly included populations with coronary artery disease [10‒13] and heart failure (HF) [14, 15]. A few studies investigated these associations also in patients with atrial fibrillation (AF) [16].

The study included consecutive patients from the Department of Cardiology at University Hospital in Wrocław, Poland, who were hospitalized between March 2013 and November 2015 and who gave written informed consent to provide a serum sample for laboratory tests. Samples were collected always in the morning after night rest and at least 12 h fasting. Samples were collected one–two days after patients’ admittance to the hospital. The samples obtained from patients were immediately frozen at a temperature of −80°C and stored at BioBank of the Łukasiewicz Research Network – PORT Polish Center for Technology Development in Wroclaw.

A simple and rapid method of liquid chromatography with tandem mass spectrometry was used to determine TMAO levels in the serum [17]. Samples were extracted by adding d9-TMAO (Cambridge Isotope Laboratories, Tewksbury, MA, USA) to acetonitrile:methanol 1:1 (vol/vol) mixture at a ratio of 1:4. Chromatographic separation was performed on a Luna Silica analytical column (3 µm, 100 Å, 150 × 2 mm; Phenomenex, Torrance, CA, USA) using the Ultimate 3000 UPLC system (Dionex, Sunnyvale, CA, USA). An isocratic elution of the mobile phase consisting of 0.1% formic acid in acetonitrile and water at a ratio of 60:40 (vol/vol) was applied at a flow rate of 0.3 mL/min (total run time, 5 min). A multiple reaction monitoring mode was applied for mass-spectrometric detection using ESI-Q-TOF (Bruker Daltonics, Bremen, Germany) in a positive-ion mode. A calibration curve was constructed in a range of 0.125–25 µM from TMAO working solution (Sigma-Aldrich, Saint Louis, MO, USA). All samples were measured in 2 biological and 2 technical replicates. The peak area ratio (analyte/internal standard) was used for the calculation of mean TMAO levels in the samples.

Data on anthropometric parameters, current diagnosis, comorbidities, laboratory test results, and medication use were obtained from the medical records of the index hospitalization. Mortality was assessed on the basis of data from the registry of the Polish National Health Fund (as of February 19, 2020).

The statistical analysis was performed for the whole population, and the subgroups of patients classified according to the main clinical diagnosis of cardiovascular disease (CVD): (1) patients with acute coronary syndrome (ACS) during the index hospitalization; (2) patients with stable coronary syndrome (SCS); (3) patients with chronic HF without ACS; and (4) patients with AF. Depending on data distribution, quantitative data were presented as mean (SD) and median with interquartile range (IQR). The rates of CVD and medication use were presented as percentage of patients. The normality of data distribution was assessed using the Kolmogorov-Smirnov and Lilliefors tests. The Kendall tau test was used to assess correlations between variables. The mean values of 2 independent variables were compared using the nonparametric Mann-Whitney test, while the Kruskal-Wallis analysis of variance was used to compare more than 2 independent variables. The effect of continuous variables on mortality in study subgroups was assessed using the Cox proportional hazards model. The proportional hazards assumption was tested using the proportional test and presented graphically using the plots of scaled Schoenfeld residuals. The goodness of fit was tested by calculating the coefficient of determination (R²). The Cox proportional hazards model with interactions was used to assess the independence of variables for mortality prediction. Survival curves for individual variables were generated using the Kaplan-Meier estimator. The receiver operating characteristic (ROC) curve analysis was used to assess the predictive value of selected variables. The cutoff values for the prediction of mortality were calculated using the Youden index. The areas under the ROC curve and their significance were also calculated. A P value of less than 0.05 was considered significant. All analyses were performed using TIBCO Software Inc. (2017) (Statistica version 13; https://www.tibco.com/).

The characteristics of the study population are presented in Table 1. The study included a total of 1,036 patients. The 5-year-mortality rate in the whole population was 16.5% (171 deaths). Most clinical and biochemical parameters differed between deceased patients and survivors. Deceased patients were older, were more often male, and had higher levels of TMAO, high-sensitivity C-reactive protein, and glycated hemoglobin HbA_1c. Moreover, they had lower total cholesterol, high-density lipoprotein, and triglyceride levels. They also had lower estimated glomerular filtration rate (eGFR) and left ventricular ejection fraction. They more often used acetylsalicylic acid, oral anticoagulation, β-blockers, statins, loop diuretics, and angiotensin receptor blockers and more often presented with ACS, SCS, chronic HF, AF, diabetes, hypertension, and a history of smoking than survivors. Finally, deceased patients more often had previous percutaneous coronary intervention and coronary artery bypass grafting.

The univariate Cox proportional hazards regression model revealed that higher TMAO levels predicted 5-year mortality without clinical significance (hazard ratio [HR], 1.01; 95% CI: 1.006–1.018; p < 0.0001). However, in the multivariate model adjusted for mortality risk factors listed in Table 1 (i.e., age, sex, comorbidities, and laboratory parameters), even this weak correlation was no longer significant (HR: 1.01; 95% CI: 0.99–1.04; p = 0.19).

For a more detailed analysis, TMAO levels were divided into quartiles and compared using the Kaplan-Meier method. A significant difference was noted in survival between patients with TMAO levels in the highest quartile versus those with TMAO levels in the remaining quartiles (χ² = 49.61; p < 0.00001; Fig. 1).

A separate Cox proportional hazards regression analysis was performed for TMAO levels according to quartiles. In the univariate analysis, TMAO levels in the highest quartile were a significant predictor of 5-year mortality in the study population (HR: 1.55; 95% CI: 1.34–1.79; p < 0.0001). However, in the multivariate model adjusted for mortality risk factors listed in Table 1, the predictive value of TMAO levels in the highest quartile was no longer significant (HR: 1.13; 95% CI: 0.87–1.46; p = 0.35).

In the ROC curve analysis for the whole population, TMAO levels higher than 5.94 µM identified patients at greater risk of death at 5 years, with low sensitivity of 48% and a specificity of 80% (area under the ROC curve: 0.67; 95% CI: 0.62–0.71; p < 0.0001; Youden index, 0.28; Fig. 2). For the cutoff value determined in the ROC curve analysis, the Kaplan-Meier survival probability curves were plotted with a significant difference of probability (log-rank test, 7.07; p < 0.00001; Fig. 2).

To obtain a more detailed insight into the effect of higher TMAO levels on mortality in patients with CVD, we divided patients into subgroups with ACS, SCS, chronic HF, and AF. The levels of TMAO in the study subgroups according to survival at 5 years are presented in Table 2. The univariate Cox proportional hazards analysis indicated a significant association between higher TMAO levels and mortality in patients with SCS, chronic HF, and AF. However, in the multivariate analysis adjusted for mortality risk factors (Table 1), TMAO levels were no longer a significant predictor of 5-year mortality (Table 3).

The ROC curve analysis determined significant cutoff values for TMAO levels for the subgroups of patients with SCS, chronic HF, and AF, in which TMAO levels were shown to be a significant predictor of mortality in the univariate Cox proportional hazards analysis (Table 4; Fig. 3). For the cutoff values determined in the ROC curve analyses for these subgroups, the Kaplan-Meier survival probability curves were plotted with a significant difference of probability for a given cutoff value (Fig. 3).

For quantitative variables, TMAO levels were positively correlated (p < 0.05) with age (r_s = 0.28), the levels of high-sensitivity C-reactive protein (r_s = 0.12), HbA_1c (r_s = 0.27), and triglycerides (r_s = 0.09), while inversely with left ventricular ejection fraction (r_s = −0.12), eGFR (r_s = −0.42), and high-density lipoprotein levels (r_s = −0.14). For dichotomous variables, TMAO levels were positively correlated with the presence of diabetes (r_s = 0.24), SCS (r_s = 0.1), chronic HF (r_s = 0.24), arterial hypertension (r_s = 0.17), and AF (r_s = 0.19). The correlations were weak or very weak. The strongest correlation was an inverse correlation with eGFR.

Considering the high prevalence of comorbidities in the study population, we investigated an association between TMAO levels and multimorbidity. We focused on such comorbidities as ACS, SCS, chronic HF, AF, arterial hypertension, and arrhythmias or conduction disorders. The presence of 1 comorbidity was noted in 130 patients; 2 comorbidities in 260 patients; 3 in 338 patients; 4 in 191 patients; 5 in 85 patients; and all 6 comorbidities were present in 32 patients. Multimorbidity was positively correlated with TMAO levels (Kendall Tau = 0.20; p < 0.05). There were also significant differences in TMAO levels between the subgroups, particularly between patients with 1 or 2 comorbidities and the remaining subgroups.

The univariate analysis revealed that increased TMAO levels were associated with only 1% increase in the relative risk of death at 5-year follow-up. However, even this clinically insignificant predictive value was no longer statistically significant in the multivariate analysis. At the same time, patients with TMAO levels in the highest quartile showed a 55% higher risk of death at 5 years compared with those with TMAO levels in the lowest quartile, although the association was no longer significant in the multivariate analysis. The cutoff value of TMAO levels determined in the ROC curve analysis (5.94 µM) had a low sensitivity and an acceptable specificity for predicting a higher risk of death in the whole study population.

Compared with the results of meta-analyses [18, 19] on various populations of patients with higher cardiovascular risk, our univariate analysis showed clinically insignificant increase in the risk of mortality. However, for TMAO levels in the highest quartile, the relative risk of mortality in our study was similar to the results of meta-analyses that applied tertile, quartile, and quintile comparisons [18, 19]. The relative risk reported by meta-analyses ranged from 1.466 (95% CI: 1.291–1.665); p < 0.001 [18] to 1.91 (95% CI: 1.40–2.61); p < 0.0001 [19].

In our univariate analysis, higher TMAO levels were a significant predictor of mortality in patients with SCS, chronic HF, and AF. However, the increase in the relative risk of mortality by 1–2% in these subpopulations was no longer significant in the multivariate analysis adjusted for other mortality risk factors. In the ROC analysis, the cutoff values of TMAO levels showed low sensitivity and a maximum specificity of 80% for predicting mortality at 5 years.

In previous studies of patients with coronary artery disease, higher TMAO levels predicted mortality independently of traditional risk factors at 3-year [10] and 5-year follow-up [11]. The median TMAO levels were 3.7 µM (IQR, 2.4–6.20) [10] and 3.8 (IQR, 2.5–6.5) µM, respectively [11], which is lower than in our patients with SCS (4.29 µM [IQR, 2.94–6.67]). Additionally, in our study, median TMAO levels were higher in deceased patients with SCS (6.26 µM [IQR, 2.70–10.1]; p < 0.00001). At the same time, the cutoff value of 6.41 µM is close to the upper quartile reported by Senthong et al. [15]. Despite the higher TMAO levels than in the above studies [10, 11], there was clinically insignificant increase in the mortality risk of patients with SCS (1–2%), and after adjustment for other mortality risk factors, TMAO levels lost their statistical significance.

Considering the methodology adopted in our study, we cannot conclude that TMAO levels are an independent predictor of mortality in patients with SCS. However, in the analysis by the quartiles of TMAO levels, the Kaplan-Maier curves revealed significant differences in survival.

Importantly, the mortality rate in the study by Senthong et al. [11] was 15.1% as compared with 21% in our patients with SCS. This suggests the potential influence of other factors in our patients. Similarly, our results for the population of patients with ACS do not correspond with previous studies including patients with ST-segment elevation myocardial infarction [12] and acute myocardial infarction [13], which reported high TMAO levels to be an independent predictor of death.

In the multivariate analysis, increased TMAO levels were not significantly associated with mortality in patients with chronic HF, which is in line with the results reported by Troseid et al. [14]. Similar to our study, they included age, kidney function, and diabetes in the multivariate analysis, among other factors. Also, Tang et al. [15] revealed a significant association of TMAO levels with mortality, even after adjustment for eGFR. However, as the authors admit themselves, most of their patients had normal kidney function.

To our knowledge, no previous studies have assessed the effect of TMAO on mortality in patients with AF. In our study, deceased patients with AF showed significantly higher TMAO levels than survivors. However, TMAO levels were no longer an independent predictor in the multivariate analysis. Previous studies described a correlation between higher TMAO levels with thrombus formation in patients with AF [16]. Higher TMAO levels seem to be linked with a higher risk of thromboembolic events, but their ultimate effect on the risk of death may also depend on such factors as age, comorbidities, and concomitant medical therapy.

The strongest correlation between TMAO levels and clinical parameters was shown for eGFR, age, and the presence of diabetes. Previous studies investigated the cause-and-effect relationship between the increasing TMAO levels and kidney damage, including also in patients with diabetes [4]. This relationship probably works in both directions, and the possible underlying pathophysiological mechanism is the impaired renal clearance of TMAO as well as the common coexistence of chronic CVD and kidney disorders. It also seems reasonable to consider the deterioration of kidney function with age.

Lee et al. [5] provided a valuable insight into the relationship between TMAO levels, kidney function, and the risk of atherosclerotic CVD (ASCVD) [5]. The authors revealed that the risk of newly diagnosed ASCVD in healthy individuals at 15 years was greater only in patients with higher TMAO levels and impaired kidney function (eGFR <60 mL/min/1.73 m²). In patients with preserved kidney function, the increase in the risk was not significant. However, in patients with previously diagnosed ASCVD, the risk of recurrent ASCVD increased with an increase in TMAO levels independently of kidney function.

We also observed the association between higher TMAO levels and multimorbidity in our subpopulations of patients with CVD. This is in line with a study by Montrucchio et al. [20], who revealed that higher TMAO levels were associated with multimorbidity in patients with HIV. While this is a different population with a different underlying disease, their findings underline the uncertainties as to the exact role of TMAO in CVD. The pathophysiological processes associated with higher TMAO levels such as atherogenesis, cardiac remodeling, and arrhythmia may coexist at varied levels of severity and are influenced by numerous external time-related factors. Therefore, it is challenging to determine TMAO as either a causative factor of these conditions or merely a contributing factor.

Our study included a relatively large population of patients and had a long follow-up. Therefore, our results for the overall study group as well as patient subgroups may be considered as statistically representative of the population of patients with CVD. Moreover, as the study included consecutive patients, our findings may be extrapolated to the general population of patients treated at cardiac departments. Finally, as we assessed numerous biochemical and anthropometric parameters as well as data from medical records, we were able to perform a multivariate analysis including traditional risk factors.

Our study has several limitations. First, this was a single-center retrospective observational study. The retrospective design does not exclude the presence of other uncontrolled factors that might have affected clinical outcomes and the incidence of death, despite the use of multivariate regression. Second, data on the causes of death were lacking. As the study had a long follow-up, we may suspect that not all deaths were from cardiovascular causes. Third, the results are based on a single measurement of TMAO levels, which are liable to change over time and depend on multiple other factors. Most importantly, we did not have access to data on antibiotic therapy prior to serum collection, the levels of TMAO precursors, gut microbiota composition, and dietary habits of participants before hospitalization. During follow-up, we did not record data on subsequent hospitalizations, significant cardiovascular events, subsequent comorbidities, changes in medical therapy, and lifestyle modifications by patients. These factors might have affected TMAO levels and mortality rates after the index hospitalization. Finally, the lack of a control group made it impossible to compare the effect of higher TMAO levels between patients with CVD and healthy individuals. As our study included patients with high mortality risk due to the underlying CVD, the results cannot be generalized to a healthy population.

In conclusion, although our study revealed a very weak positive correlation between increased TMAO levels and a higher risk of mortality, the use of TMAO as an independent mortality predictor is limited, and numerous other factors should be included in the assessment of patient prognosis. Further research is needed to establish the exact role of TMAO as either a causative or a contributing factor of mortality in patients with high cardiovascular burden. Interventional studies assessing the benefits of reducing TMAO levels may provide additional important insights.

The samples were provided by the Biobank of Łukasiewicz Research Network – PORT Polish Center for Technology Development.

The study protocol was approved by the local Ethics Committee of Wroclaw Medical University (No. 163/2019; February 28, 2019). The study was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent to participate in the study.

The authors have no conflicts of interest to declare.

The study was funded by a statutory grant of the Institute for Heart Diseases.

First author: Radosław Konieczny. Radosław Konieczny, Wiktor Kuliczkowski, and Andrzej Mysiak – conceived the concept for the study and designed the research. Radosław Konieczny, Ewa Żurawska-Płaksej, Konrad Kaaz, and Hanna Czapor-Irzabek – acquisition the data. Radosław Konieczny, Wiktor Kuliczkowski, Wojciech Bombała – analysis and interpretation of data. Radosław Konieczny, Wiktor Kuliczkowski, Ewa Żurawska-Płaksej, Konrad Kaaz, and Hanna-Czapor-Irzabek – drafting of the manuscript. Radosław Konieczny, Wiktor Kuliczkowski, Ewa Żurawska-Płaksej,Wojciech Bombała, and Andrzej Mysiak – revising for important intellectual content. All the authors have read and accepted the manuscript in its final form, including the authorship list.

All data generated or analyzed during this study are included in this article. Further inquiries can be directed to the corresponding author.