Interpretable prediction of 3-year all-cause mortality in patients with chronic heart failure based on machine learning

BackgroundDiastolic heart failure (DHF) and type 2 diabetes mellitus (T2DM) often coexist, causing increased mortality rates. Glycaemic variability (GV) exacerbates cardiovascular complications, but its impact on outcomes in patients with DHF and T2DM remains unclear. This study examined the relationships between GV with mortality outcomes, and developed a machine learning (ML) model for long-term mortality in these patients.MethodsPatients with DHF and T2DM were included from the Medical Information Mart for Intensive Care IV, with admissions (2008–2019) as primary analysis cohort and admissions (2020–2022) as external validation cohort. Multivariate Cox proportional hazards models and restricted cubic spline analyses were used to evaluate the associations of GV with 90-day, 1-year, and 3-year all-cause mortality. The primary analysis cohort was split into training and internal validation cohorts, then developing ML models for predicting 1-year all-cause mortality in training cohort, which were validated using the internal and external validation cohorts.Results2,128 patients with DHF and T2DM were included in primary analysis cohort (meidian age 71.0years [IQR: 62.0–79.0]; 46.9% male), 498 patients with DHF and T2DM were included in the external validation cohort (meidian age 75.0years [IQR: 67.0–81.0]; 54.0% male). Multivariate Cox proportional hazards models showed that high GV tertiles were associated with higher risk of 90-day (T2: HR 1.45, 95%CI 1.09–1.93; T3: HR 1.96, 95%CI 1.48–2.60), 1-year (T2: HR 1.25, 95%CI 1.02–1.53; T3: HR 1.54, 95%CI 1.26–1.89), and 3-year (T2: HR 1.31, 95%CI: 1.10–1.56; T3: HR 1.48, 95%CI 1.23–1.77) all-cause mortality, compared with lowest GV tertile. Chronic kidney disease, creatinine, potassium, haemoglobin, and white blood cell were identified as mediators of GV and 1-year all-cause mortality. Additionally, GV and other clinical features were pre-selected to construct ML models. The random forest model performed best, with AUC (0.770) and G-mean (0.591) in internal validation, with AUC (0.753) and G-mean (0.599) in external validation.ConclusionGV was determined as an independent risk factor for short-term and long-term all-cause mortality in patients with DHF and T2DM, with a potential intervention threshold around 25.0%. The ML model incorporating GV demonstrated strong predictive performance for 1-year all-cause mortality, highlighting its importance in early risk stratification management of these patients.

Interpretable prediction of 3-year all-cause mortality in patients with heart failure caused by coronary heart disease based on machine learning and SHAP

BackgroundIdiopathic pulmonary fibrosis (IPF) is an interstitial lung disease characterized by chronic inflammation and progressive fibrosis. The blood urea nitrogen-to-albumin ratio (BAR) is a comprehensive parameter associated with inflammation status; however, it is unknown whether the BAR can predict the prognosis of IPF.MethodsThis retrospective study included 176 patients with IPF, and 1-year all-cause mortality of these patients was recorded. A receiver operating characteristic (ROC) curve was used to explore the diagnostic value of BAR for 1-year all-cause mortality in IPF patients, and the survival rate was further estimated using the Kaplan–Meier survival curve. Cox proportional hazards regression model and forest plot were used to assess the association between the BAR and 1-year all-cause mortality in IPF patients.ResultsThe BAR of IPF patients was significantly higher in the non-survivor group than in the survivor group [0.16 (0.13–0.23) vs. 0.12 (0.09–0.17) mmol/g, p = 0.002]. The area under the ROC curve for predicting 1-year all-cause mortality in IPF patients was 0.671, and the optimal cut-off value was 0.12 mmol/g. The Kaplan–Meier survival curve showed that the 1-year cumulative survival rate of IPF patients with a BAR ≥0.12 was significantly decreased compared with the patients with a BAR <0.12. The Cox regression model and forest plot showed that the BAR was an independent prognostic biomarker for 1-year all-cause mortality in IPF patients (HR = 2.778, 95% CI 1.020–7.563, p = 0.046).ConclusionThe BAR is a significant predictor of 1-year all-cause mortality of IPF patients, and high BAR values may indicate poor clinical outcomes.

Exploring sex-specific patterns of mortality predictors among patients undergoing cardiac resynchronization therapy: a machine learning approach

Aims: We aimed to characterise the complex relationships of body roundness index (BRI) and triglyceride-glucose index (TyG) with 5-year all-cause mortality in patients with diabetes and comorbid hypertension. Methods: 5,728 patients from the 1999–2014 US National Health and Nutrition Examination Survey (NHANES) cycles and 3,456 from the 2005–2010 China Kailuan cycles were included. BRI was calculated as 364.2−365.5×√(1−(waist circumference in centimeters/2π) 2 ÷(0.5×height in centimeters) 2 ). TyG was calculated as the logarithmic product of the fasting triglyceride and glucose concentrations. The cut-off values of BRI and TyG were based on median values. Results: The prevalence of 5-year all-cause mortality was 8.4% in the NHANES database and 9.2% in the Kailuan cohort. Multivariable Cox regression demonstrated that a BRI of &gt;6.2 (HR: 2.53, 95% CI: 1.71–2.96) and TyG of &gt;9.5 (HR: 1.64, 95% CI: 1.39–2.72) were independent predictors of 5-year all-cause mortality in diabetic patients with hypertension after adjusting forage, gender, marital status, education level, smoking status, history of myocardial infarction, stroke, urinary protein and CKD-EPI eGFR. These results were reconfirmed in the Kailuan cohort for BRI (HR: 2.91, 95% CI: 1.93–3.57) and TyG (HR: 2.13, 95% CI: 1.86–3.02). TyG was found to mediate the association between BRI and all-cause mortality, being responsible for 24.1% (16.8%-28.5%) in the NHANES database and 27.5% (22.7%-30.1%) in the Kailuan cohort. No significant additive interactions were found between BRI and TyG on 5-year all-cause mortality. Significant multiplicative effects were identified between TyG and BRI in the NHANES database alone (Additive: RERI = 0.05, 95% CI – 3.21–1.97; Multiplicative, HR = 1.27, 95% CI 1.13–1.95 in the NHANES database; Additive: RERI = 0.27, 95% CI – 2.17–5.29; Multiplicative, HR = 1.02, 95% CI 0.85–2.21 in the Kailuan cohort). Conclusions: BRI and TyG were independent risk factors for 5-year all-cause mortality in patients with diabetes and comorbid hypertension. TyG was found to mediate a considerable proportion of the effect of BRI on all-cause mortality in cohorts from both the United States and China. Interventions aimed at improving obesity and abnormal fat distribution might reduce the all-cause mortality risk associated with insulin resistance.

The C-reactive protein (CRP)-albumin-lymphocyte (CALLY) index is a newly developed biomarker that combines measurements of CRP, serum albumin, and lymphocyte count. This index provides a thorough assessment of a patient's inflammation level, nutritional condition, and immunological function. The objective of this study is to examine the correlation between the CALLY index and all-cause mortality in COPD patients. We calculated the CALLY index using data from the National Health and Nutrition Examination Survey (NHANES) for the 2007-2008 and 2009-2010 cycles, extracted from the participants' peripheral blood samples. The study utilized Kaplan-Meier curves, restricted cubic spline (RCS) curves, and Cox regression analysis to evaluate the relationship between the CALLY index and the risk of all-cause mortality in COPD patients. To assess the predictive accuracy of the CALLY index, we calculated the area under the receiver operating characteristic (ROC) curve (AUC). The study included 1,048 participants and found a significant negative correlation between the CALLY index and all-cause mortality in patients with COPD. The CALLY index was a major predictor of survival in COPD patients [fully adjusted model: in the 3rd quartile, HR = 1.61, 95% CI: 1.02-2.52, p = 0.039; in the 2nd quartile, HR = 2.11, 95% CI: 1.22-3.65, p = 0.008; in the 1st quartile, HR = 3.12, 95% CI: 2.00-4.85, p < 0.001]. The RCS curves demonstrated a non-linear association between the CALLY index and all-cause mortality in COPD patients. The areas under the curve (AUC) in predicting 5- and 10-year all-cause mortality were 0.693 and 0.656. The CALLY index has a strong relationship with all-cause mortality in patients with COPD in the US and could serve as a prognostic biomarker for these patients.

BackgroundDepression, as the primary contributor to the global disease burden caused by mental disorders, necessitates the urgent discovery of new biomarkers to predict patient prognosis, thereby facilitating early treatment. The triglyceride glucose index (TyG), which has a close relationship with insulin resistance (IR), systemic inflammation, and other factors, could potentially serve as a valuable biomarker for determining depression severity. However, the association of the TyG index and its changes with the long-term prognosis of depression remains unexplored. This article aims to evaluate whether the levels and changes in the TyG index can predict long-term mortality in patients suffering from depression.MethodsA retrospective cohort study was conducted based on the MIMIC-IV (Medical Information Mart in Intensive Care IV) database, involving 1388 patients. Among them, 1120 patients had only one TyG index value during hospitalization, while 266 patients had two or more TyG index values, all of which were included in the analysis. The primary endpoint was 5-year all-cause mortality rate. Restricted cubic spline analysis was also used to evaluate any potential nonlinear correlations. Propensity score matching was performed to reduce any potential baseline bias. Cox proportional hazards analyses were used to adjust for confounders. The Kaplan-Meier method was utilized to compute the cumulative curve.ResultsThe entire cohort exhibited a 5-year all-cause mortality rate of 25.8% (358/1388), with rates of 23.8% (267/1120) and 34% (91/268) in the TyG and TyGVR groups, respectively. According to a multivariate Cox regression model, an increase in TyG elevated the 5-year all-cause mortality risk among depression patients (HR, 1.31; 95% CI, 1.1-1.56; P=0.002). When TyG was treated as a categorical variable, Quartile 4 demonstrated a 61% higher risk of 5-year all-cause mortality in depression patients compared to Quartile 1 (HR, 1.61; 95% CI, 1.15-2.25; P=0.005). Restricted cubic spline analysis revealed a linear relationship between TyG and 5-year all-cause mortality across both the entire and matched cohorts (P values for non-linearity were > 0.05). The Kaplan-Meier curves indicated that patients with elevated TyG experienced significantly higher 5-year all-cause mortality rates in both the entire and matched cohorts (P=0.00041, P=0.035, respectively). A linear association (non-linear P=0.953) was also observed between TyGVR and 5-year all-cause mortality risk. The analysis results across the subgroups were consistent, with no interactions detected (interaction P-values > 0.05).ConclusionsThe TyG index is linked with a 5-year all-cause mortality risk among patients suffering from depression. The dynamic fluctuations in the TyG index could potentially offer more substantial insights in identifying patients at high risk for all-cause mortality.

Chronic obstructive pulmonary disease (COPD) is characterized by pulmonary and systemic inflammation. The peripheral blood (neutrophil + monocyte)/lymphocyte ratio (NMLR) can predict the clinical outcomes of several inflammatory diseases. However, its prognostic value in COPD remains unknown. This retrospective study included 870 patients with COPD due to acute exacerbation, and the 5-year all-cause mortality of these patients was recorded. The Kaplan-Meier method was used to compare the mortality risk of these patients according to their NMLR value. Multivariable COX hazard regression and restricted cubic spline model were used to assess the relationship between the NMLR and 5-year all-cause mortality of patients with COPD. The NMLR values of non-surviving patients with COPD were significantly increased compared to the survivors [3.88 (2.53-7.17) vs 2.95 (2.08-4.89), P=0.000]. The area under the NMLR receiver operating characteristic curve for predicting the 5-year all-cause mortality of COPD patients was 0.63. Kaplan-Meier survival curves showed that the 5-year all-cause mortality of COPD patients was significantly increased when the admission peripheral blood NMLR was ≥ 5.90 (27.3% vs 12.4%, P=0.000). The COX regression model showed that NMLR was an independent predictor of 5-year all-cause mortality in COPD patients (hazard ratio=1.84, 95% confidence interval: 1.28-2.64, P=0.001). Moreover, the restricted cubic spline model showed a non-linear relationship between NMLR and COPD death risk (Pnon-linear < 0.05). The admission peripheral blood NMLR is a significant predictor of 5-year all-cause mortality in patients with COPD, and high NMLR values may indicate a poor clinical prognosis.

Objective Machine learning (ML) model has not been developed specifically for ischemic heart failure (HF) patients. Whether the performance of ML model is better than the MAGGIC risk score and NT-proBNP is unknown. The current study was to apply ML algorithm to build risk model for predicting 1-year and 3-year all-cause mortality in ischemic HF patient and to compare the performance of ML model with the MAGGIC risk score and NT-proBNP. Method Three ML algorithms without and with feature selection were used for model exploration, and the performance was determined based on the area under the curve (AUC) in five-fold cross-validation. The best performing ML model was selected and compared with the MAGGIC risk score and NT-proBNP. The calibration of ML model was assessed by the Brier score. Results Random forest with feature selection had the highest AUC (0.742 and 95% CI: 0.697–0.787) for predicting 1-year all-cause mortality, and support vector machine without feature selection had the highest AUC (0.732 and 95% CI: 0.694–0.707) for predicting 3-year all-cause mortality. When compared to the MAGGIC risk score and NT-proBNP, ML model had a comparable AUC for predicting 1-year (0.742 vs 0.714 vs 0.694) and 3-year all-cause mortality (0.732 vs 0.712 vs 0.682). Brier scores for predicting 1-year and 3-year all-cause mortality were 0.068 and 0.174, respectively. Conclusion ML models predicted prognosis in ischemic HF with good discrimination and well calibration. These models may be used by clinicians as a decision-making tool to estimate the prognosis of ischemic HF patients.

BackgroundIn this study, we evaluated the predictive utility of neutrophil percentage-to-albumin ratio (NPAR) for all-cause mortality in patients with chronic heart failure (CHF).MethodsPatients diagnosed as CHF enrolled in this retrospective cohort study were from Beijing Chaoyang Hospital, capital medical university. Admission NPAR was calculated as neutrophil percentage divided by serum albumin. The endpoints of this study were defined as 90-day, 1-year and 2-year all-cause mortality. Multivariable Cox proportional hazard regression model was performed to confirm the association between NPAR and all-cause mortality. Receiver operating characteristics (ROC) curves were used to evaluate the ability for NPAR to predict all-cause mortality.ResultsThe 90-day (P = 0.009), 1-year (P < 0.001) and 2-year (P < 0.001) all-cause mortality in 622 patients with CHF were increased as admission NPAR increased. Multivariable Cox regression analysis found the higher NPAR value was still independently associated with increased risk of 90-day (Group III versus Group I: HR, 95% CI: 2.21, 1.01–4.86, P trend = 0.038), 1-year (Group III versus Group I: HR, 95% CI:2.13, 1.30–3.49, P trend = 0.003), and 2-year all-cause mortality (Group III versus Group I: HR, 95% CI:2.06, 1.37–3.09, P trend = 0.001), after adjustments for several confounders. ROC curves revealed that NPAR had a better ability to predict all-cause mortality in patients with CHF, than either albumin or the neutrophil percentage alone.ConclusionsNPAR was independently correlated with 90-day, 1-year, and 2-year all-cause mortality in patients with CHF.

Background and ObjectivesAccumulating evidence suggests that oxidative stress is involved in the development of chronic obstructive pulmonary disease (COPD) and its progression. Activity of extracellular superoxide dismutase (ecSOD), the only extracellular enzyme eliminating superoxide radicals, has been reported to decline in acute exacerbations of COPD (AECOPD). However, the association between serum ecSOD activity and 1-year all-cause mortality in AECOPD patients remains unclear. The objective of our study was to explore the usefulness of ecSOD activity on admission in AECOPD as an objective predictor for 1-year all-cause mortality.MethodsWe measured serum ecSOD activity in AECOPD patients on admission in a prospective cohort study. We also recorded their laboratory and clinical data. Multivariate Cox regression was used to analyze the association between ecSOD activity and the risk of 1-year all-cause mortality. Restricted cubic spline curves were used to visualize the relationship between ecSOD activity and the hazard ratio of 1-year all-cause mortality.ResultsA total of 367 patients were followed up for 1 year, and 29 patients died during a 1-year follow-up period. Compared with survivors, the non-survivors were older (79.52 ± 8.39 vs. 74.38 ± 9.34 years old, p = 0.004) and had increased levels of tobacco consumption (47.07 ± 41.67 vs. 33.83 ± 31.79 pack-years, p = 0.037). Having an ecSOD activity ≤ 98.8 U/ml was an independent risk factor of 1-year all-cause mortality after adjustment for baseline differences, clinical variables and comorbidities [hazard ratio = 5.51, 95% confidence interval (CI): 2.35–12.95, p < 0.001].ConclusionLower serum ecSOD activity was a strong and independent predictor of 1-year all-cause mortality in AECOPD patients.

To investigate the value of body composition indices derived from pre-procedural computed tomography (CT) in predicting 1-year mortality among patients who underwent transcatheter aortic valve replacement (TAVR). We assessed consecutive patients who underwent TAVR between June 2016 and December 2021 at a single academic medical center. Skeletal muscle and subcutaneous fat area at the T4, T12, and L3 levels on pre-procedural CT were measured. The association between body composition and 1-year mortality was evaluated using Cox proportional hazard regression analysis. Finally, 408 patients were included (185 men and 223 women; mean age, 81.7 ± 5.1 years; range, 62-98 years). Post-procedural death occurred in 13.2% of patients. The muscle-height index and fat-height index at the L3 level were more strongly correlated with those at the T12 level (r = 0.765, p < 0.001 and r = 0.932, p < 0.001, respectively) than with those at the T4 level (r = 0.535, p < 0.001 and r = 0.895, p < 0.001, respectively). The cumulative 1-year mortality rate was highest for patients with both sarcopenia and adipopenia (26%), followed by those with adipopenia only(17%), those with sarcopenia only (12%), and those with neither sarcopenia nor adipopenia (8%, p = 0.002). Multivariable analysis revealed that body composition at the T12 level was an independent risk factor for 1-year mortality (hazard ratio: 4.09, 95% confidence interval: 2.01-8.35) in patients with both sarcopenia and adipopenia (p < 0.001). Sarcopenia or adipopenia assessed with CT at the thoracic level may be valuable for stratifying 1-year all-cause mortality in patients who undergo TAVR. Skeletal muscle and subcutaneous fat mass indices at the level of T12, measured on pre-procedural CT, have value for risk stratification of 1-year all-cause mortality in patients who undergo transcatheter aortic valve replacement. Sarcopenia and adipopenia are associated with the prognosis of patients undergoing transcatheter aortic valve replacement. Body composition at the T12 level was an independent risk factor for 1-year all-cause mortality. Sarcopenia or adipopenia assessed at T12 with pre-procedural CT is valuable for risk stratification.

Background: The CHA₂DS₂-VASc score has been widely used to predict stroke in patients with atrial fibrillation (AF). Recently, it was reported that the CHA₂DS₂-VASc score helps predict cardiovascular disease (CVD) or all-cause mortality in patients with or without AF. However, few reports have examined the association between this score and mortality in hemodialysis (HD) patients. Methods: We analyzed 557 consecutive patients who initiated HD at our facilities between February 2005 and October 2017. The CHA₂DS₂-VASc score was calculated at the time of initiation of HD. Patients were then categorized into three groups according to their CHA₂DS₂-VASc scores: 0–1 (low), 2–3 (intermediate), and 4–9 (high). Multivariate Cox proportional hazards analysis was used to assess independent risk factors for 3-year all-cause mortality. Results: During the 3-year follow-up period, 153 (27.5%) patients died (cardiovascular death: n = 88). According to multivariate analysis, serum albumin (hazard ratio [HR] 0.60, 95% confidence interval [CI] 0.43–0.85, p = 0.003), creatinine (HR 0.91, 95% CI 0.84–0.99, p = 0.049), and CHA₂DS₂-VASc score (HR 1.33, 95% CI 1.20–1.46, p < 0.001) were associated with 3-year all-cause mortality. Compared with patients in the low CHA₂DS₂-VASc score group, those in the intermediate- and high-score groups had a higher risk for all-cause and CVD mortality (all-cause mortality: HR 1.77, 95% CI 1.23–2.55, p = 0.002 and HR 2.94, 95% CI 1.90–4.53, p < 0.001, respectively; CVD mortality: HR 1.82, 95% CI 1.27–2.59, p = 0.001 and HR 2.85, 95% CI 1.88–4.31, p < 0.001, respectively). Conclusion: The CHA₂DS₂-VASc score is a valuable predictor of 3-year all-cause and CVD mortality in incident HD patients.

The precursor peptide of atrial natriuretic peptide (MR-proANP) has a physiological role in fluid homeostasis and is associated with mortality and adverse clinical outcomes in heart failure patients. Little is known about the prognostic potential of this peptide for long-term mortality prediction in community-dwelling patients. We evaluated associations of MR-proANP levels with 10-year all-cause mortality in patients visiting their general practitioner for a respiratory tract infection. In this post-hoc analysis including 359 patients (78.5%) of the original trial, we calculated cox regression models and area under the receiver operating characteristic curve (AUC) to assess associations of MR-proANP blood levels with mortality and adverse outcome including death, pulmonary embolism, and major adverse cardiac or cerebrovascular events. After a median follow-up of 10.0 years, 9.8% of included patients died. Median admission MR-proANP levels were significantly elevated in non-survivors compared to survivors (80.5 pmol/L, IQR 58.6-126.0; vs. 45.6 pmol/L, IQR 34.2-68.3; p<0.001) and associated with 10-year all-cause mortality (age-adjusted HR 2.0 [95% CI 1.3-3.1, p=0.002]; AUC 0.79). Results were similar for day 7 blood levels and also for the prediction of other adverse outcomes. Increased MR-proANP levels were associated with 10-year all-cause mortality and adverse clinical outcome in a sample of community-dwelling patients. If diagnosis-specific cut-offs are confirmed in future studies, this marker may help to direct preventive measures in primary care.

Metabolic dysfunction-associated steatotic liver disease (MASLD) increases the risk of cardiovascular disease and existing evidence indicates that MASLD affects the cardiovascular system through systemic inflammation. Our aim was to assess the association of hematological biomarkers of inflammation with the 10-year risk of major adverse cardiovascular events (MACE) and all-cause mortality in MASLD patients. A total of 1858 MASLD participants from the Atherosclerosis Risk in Communities cohort study at visit 2 (1990-1992) were included. A total of 1338 non-MASLD participants were also included in the comparison. At baseline, hematological biomarkers of inflammation such as leukocytes, neutrophils, lymphocytes, monocytes, and C-reactive protein (CRP) were measured. Participants were followed up for MACE and all-cause mortality for a period of 10 years. Multivariate adjusted Cox models were used to estimate hazard ratios (HR). The 10-year MACE was higher in MASLD participants than in non-MASLD participants (20.8% vs 9.3%). Monocytes (HR 1.114, [95% CI, 1.022-1.216] per 1-SD, P=0.015) and CRP (HR 1.109 [95% CI, 1.032-1.190] per 1-SD, P=0.005) were associated with an increased 10-year risk of MACE, independent of other cardiovascular risk factors. This association was specific to the MASLD population. None of these hematological biomarkers demonstrated a significant association with 10-year all-cause mortality. Increased levels of monocytes and CRP were associated with an increased 10-year risk of MACE in the MASLD population. Hematological biomarkers of inflammation may help identify MASLD populations at higher risk for cardiovascular events.