Factors influencing the predictive power of models for predicting mortality and/or heart failure hospitalization in patients with heart failure

A machine learning model to predict the risk of 30-day readmissions in patients with heart failure: a retrospective analysis of electronic medical records data

BACKGROUND

Heart failure is one of the leading causes of hospitalization in the United States. Advances in big data solutions allow for storage, management, and mining of large volumes of structured and semi-structured data, such as complex healthcare data. Applying these advances to complex healthcare data has led to the development of risk prediction models to help identify patients who would benefit most from disease management programs in an effort to reduce readmissions and healthcare cost, but the results of these efforts have been varied. The primary aim of this study was to develop a 30-day readmission risk prediction model for heart failure patients discharged from a hospital admission.

METHODS

We used longitudinal electronic medical record data of heart failure patients admitted within a large healthcare system. Feature vectors included structured demographic, utilization, and clinical data, as well as selected extracts of un-structured data from clinician-authored notes. The risk prediction model was developed using deep unified networks (DUNs), a new mesh-like network structure of deep learning designed to avoid over-fitting. The model was validated with 10-fold cross-validation and results compared to models based on logistic regression, gradient boosting, and maxout networks. Overall model performance was assessed using concordance statistic. We also selected a discrimination threshold based on maximum projected cost saving to the Partners Healthcare system.

RESULTS

Data from 11,510 patients with 27,334 admissions and 6369 30-day readmissions were used to train the model. After data processing, the final model included 3512 variables. The DUNs model had the best performance after 10-fold cross-validation. AUCs for prediction models were 0.664 ± 0.015, 0.650 ± 0.011, 0.695 ± 0.016 and 0.705 ± 0.015 for logistic regression, gradient boosting, maxout networks, and DUNs respectively. The DUNs model had an accuracy of 76.4% at the classification threshold that corresponded with maximum cost saving to the hospital.

CONCLUSIONS

Deep learning techniques performed better than other traditional techniques in developing this EMR-based prediction model for 30-day readmissions in heart failure patients. Such models can be used to identify heart failure patients with impending hospitalization, enabling care teams to target interventions at their most high-risk patients and improving overall clinical outcomes.

Validation of U.S. mortality prediction models for hospitalized heart failure in the United Kingdom and Japan

AIMS

Prognostic models for hospitalized heart failure (HHF) were developed predominantly for patients of European origin in the United States of America; it is unclear whether they perform similarly in other health care systems or for different ethnicities. We sought to validate published prediction models for HHF in the United Kingdom (UK) and Japan.

METHODS AND RESULTS

Patients in the UK (n =894) and Japan (n =3158) were prospectively enrolled and were similar in terms of sex (∼60% men) and median age (∼77 years). Models predicted that British patients would have a higher mortality than Japanese, which was indeed true both for in-hospital (4.8% vs. 2.5%) and 180-day (20.7% vs. 9.5%) mortality. The model c-statistics for the published/derivation (range 0.70-0.76) and Japanese (range 0.75-0.77) cohorts were similar and higher than for the UK (0.62-0.75) but models consistently overestimated mortality in Japan. For in-hospital mortality, the OPTIMIZE-HF model performed best, providing similar discrimination in published/derivation, UK and Japanese cohorts [c-indices: 0.75 (0.74-0.77); 0.75 (0.68-0.81); and 0.77 (0.70-0.83), respectively], and least overestimated mortality in Japan. For 180-day mortality, the c-statistics for the ASCEND-HF model were similar in published/derivation (0.70) and UK [0.69 (0.64-0.74)] cohorts but higher in Japan [0.75 (0.71-0.79)]; calibration was good in the UK but again overestimated mortality in Japan.

CONCLUSION

Calibration of published prediction models appears moderately accurate and unbiased when applied to British patients but consistently overestimates mortality in Japan. Identifying the reason why patients in Japan have a better than predicted prognosis is of great interest.

Predictive biomarkers for death and rehospitalization in comorbid frail elderly heart failure patients

Background

Heart failure (HF) is associated with a high rate of readmissions within 30 days post-discharge and in the following year, especially in frail elderly patients. Biomarker data are scarce in this high-risk population. This study assessed the value of early post-discharge circulating levels of ST2, NT-proBNP, CA125, and hs-TnI for predicting 30-day and 1-year outcomes in comorbid frail elderly patients with HF with mainly preserved ejection fraction (HFpEF).

Methods

Blood samples were obtained at the first visit shortly after discharge (4.9 ± 2 days). The primary endpoint was the composite of all-cause mortality or HF-related rehospitalization at 30 days and at 1 year. All-cause mortality alone at one year was also a major endpoint. HF-related rehospitalizations alone were secondary end-points.

Results

From February 2014 to November 2016, 522 consecutive patients attending the STOP-HF Clinic were included (57.1% women, age 82 ± 8.7 years, mean Barthel index 70 ± 25, mean Charlson comorbidity index 5.6 ± 2.2). The composite endpoint occurred in 8.6% patients at 30 days and in 38.5% at 1 year. In multivariable analysis, ST2 [hazard ratio (HR) 1.53; 95% CI 1.19–1.97; p = 0.001] was the only predictive biomarker at 30 days; at 1 year, both ST2 (HR 1.34; 95% CI 1.15–1.56; p < 0.001) and NT-proBNP (HR 1.19; 95% CI 1.02–1.40; p = 0.03) remained significant. The addition of ST2 and NT-proBNP into a clinical predictive model increased the AUC from 0.70 to 0.75 at 30 days (p = 0.02) and from 0.71 to 0.74 at 1 year (p < 0.05). For all-cause death at 1 year, ST2 (HR 1.50; 95% CI 1.26–1.80; p < 0.001), and CA125 (HR 1.41; 95% CI 1.21–1.63; p < 0.001) remained independent predictors in multivariable analysis. The addition of ST2 and CA125 into a clinical predictive model increased the AUC from 0.74 to 0.78 (p = 0.03). For HF-related hospitalizations, ST2 was the only predictive biomarker in multivariable analyses, both at 30 days and at 1 year.

Conclusions

In a comorbid frail elderly population with HFpEF, ST2 outperformed NT-proBNP for predicting the risk of all-cause mortality or HF-related rehospitalization. ST2, a surrogate marker of inflammation and fibrosis, may be a better predictive marker in high-risk HFpEF.

Electronic supplementary material

The online version of this article (10.1186/s12877-018-0807-2) contains supplementary material, which is available to authorized users.

Performance of the MAGGIC heart failure risk score and its modification with the addition of discharge natriuretic peptides

Aims

Predictive models for heart failure patients are widely used in the clinical practice to stratify patients' mortality and enable clinicians to tailor and intensify their approach. However, such models have not been validated internationally. In addition, biomarkers are now frequently measured to obtain prognostic information, and the implications of this practice are not known. In this study, we aimed to validate the model performance of the Meta‐analysis Global Group in Chronic Heart Failure (MAGGIC) score in a Japanese acute heart failure registry and further explore the incremental prognostic value of discharge B‐type natriuretic peptide (BNP) level.

Methods and Results

In this study, we evaluated the registered data of 2215 consecutive acute HF patients (with 694 119 person‐years follow‐up) from a prospective multicentre registry (the West Tokyo Heart Failure) conducted in Japan from April 2006 to August 2016. The mean age was 73.0 ± 13.0, and 61.2% were male. The MAGGIC score demonstrated modest discrimination (c‐index = 0.71, 95% confidence interval 0.67–0.74) and good calibration (R² value = 0.97); there was constant overestimation for 1 year mortality. However, when the BNP level was added to the original MAGGIC variables, the model demonstrated good discrimination (c‐index = 0.74, 95% confidence interval 0.70–0.78) with adequate calibration (R² value = 0.91). The modified MAGGIC BNP score was externally validated in a separate Japanese registry (NaDEF) and demonstrated moderate discrimination (c‐index = 0.69, 95% confidence interval 0.65–0.73) and calibration (R² value = 0.85).

Conclusion

The original MAGGIC score performed modestly in Japanese patients, but the addition of discharge BNP level enhanced model performance. The addition of objective biomarkers may result in effective modification of preexisting internationally recognized risk models and aid in multinational comparisons of heart failure patients' outcomes.

Predictive modeling of inpatient mortality in departments of internal medicine

Despite overwhelming data on predictors of inpatient mortality, it is unclear which variables are the most instructive in predicting mortality of patients in departments of internal medicine. This study aims to identify the most informative predictors of inpatient mortality, and builds a prediction model on an individual level, given a constellation of patient characteristics. We use a penalized method for developing the prediction model by applying the least-absolute-shrinkage and selection-operator regression. We utilize a cohort of adult patients admitted to any of 5 departments of internal medicine during 3.5 years. We integrated data from electronic health records that included clinical, epidemiological, administrative, and laboratory variables. The prediction model was evaluated using the validation sample. Of 10,788 patients hospitalized during the study period, 874 (8.1%) died during admission. We find that the strongest predictors of inpatient mortality are prior admission within 3 months, malignant morbidity, serum creatinine levels, and hypoalbuminemia at hospital admission, and an admitting diagnosis of sepsis, pneumonia, malignant neoplastic disease, or cerebrovascular disease. The C-statistic of the risk prediction model is 89.4% (95% CI 88.4-90.4%). The predictive performance of this model is better than a multivariate stepwise logistic regression model. By utilizing the prediction model, the AUC for the independent (validation) data set is 85.7% (95% CI 84.1-87.3%). Using penalized regression, this prediction model identifies the most informative predictors of inpatient mortality. The model illustrates the potential value and feasibility of a tool that can aid physicians in decision-making.

Red blood cell distribution width in heart failure: A narrative review

The red blood cell distribution width (RDW) is a simple, rapid, inexpensive and straightforward hematological parameter, reflecting the degree of anisocytosis in vivo. The currently available scientific evidence suggests that RDW assessment not only predicts the risk of adverse outcomes (cardiovascular and all-cause mortality, hospitalization for acute decompensation or worsened left ventricular function) in patients with acute and chronic heart failure (HF), but is also a significant and independent predictor of developing HF in patients free of this condition. Regarding the biological interplay between impaired hematopoiesis and cardiac dysfunction, many of the different conditions associated with increased heterogeneity of erythrocyte volume (i.e., ageing, inflammation, oxidative stress, nutritional deficiencies and impaired renal function), may be concomitantly present in patients with HF, whilst anisocytosis may also directly contribute to the development and worsening of HF. In conclusion, the longitudinal assessment of RDW changes over time may be considered an efficient measure to help predicting the risk of both development and progression of HF.

In-hospital mortality risk score for very old patients hospitalized with decompensated chronic heart failure

OBJECTIVE

To derive a risk score to predict in-hospital mortality for very old patients with decompensated chronic heart failure (DCHF).

METHODOLOGY

Retrospective cohort study that included patients ≥ 80 years admitted to a Geriatric Acute Care Unit with DCHF between January 2012 and December 2014. We analyzed 70 candidate risk factors and in-hospital mortality. We derived a risk model using multivariate logistic regression model and constructed a scale for scoring risk. We used bootstrapping techniques for the internal validation.

RESULTS

We included 629 patients with mean age of 90 (SD5) years, 470 (73.1%) being women. Eighty-six (13.7%) patients died during the hospitalization. Factors included in the final risk model were NYHA class III-IV, severe functional dependence (Katz activities of daily living index < 2), infection as cause of exacerbation of heart failure, number of medications ≥ 8, albumin < 3 mg/dL, glomerular filtration rate < 60 mL/min, level of potassium in blood > 5.5 mEq/L and red blood cell distribution width (RDW) > 17%. In-hospital mortality in risk groups was 3.0, 4.6, 9.5, 15.1 and 36.3%, respectively. The area under ROC curve risk for score after bootstrapping was 0.77 (95%: CI 0.70-0.83).

CONCLUSION

This risk score could be useful for stratifying risk for in-hospital mortality among very old patients admitted to hospital for DCHF.

Factors associated with health-related quality of life in stable ambulatory congestive heart failure patients: Systematic review

Background Since improved treatment of congestive heart failure has resulted in decreased mortality and hospitalisation rates, increasing self-perceived health-related quality of life (HRQoL) has become a major goal of congestive heart failure treatment. However, an overview on predictieve factors of HRQoL is currently lacking in literature. Purpose The aim of this study was to identify key factors associated with HRQoL in stable ambulatory patients with congestive heart failure. Methods A systematic review was performed. MEDLINE, Web of Science and Embase were searched for the following combination of terms: heart failure, quality of life, health perception or functional status between the period 2000 and February 2017. Literature screening was done by two independent reviewers. Results Thirty-five studies out of 8374 titles were included for quality appraisal, of which 29 were selected for further data extraction. Four distinct categories grouping different types of variables were identified: socio-demographic characteristics, clinical characteristics, health and health behaviour, and care provider characteristics. Within the above-mentioned categories the presence of depressive symptoms was most consistently related to a worse HRQoL, followed by a higher New York Heart Association functional class, younger age and female gender. Conclusion Through a systematic literature search, factors associated with HRQoL among congestive heart failure patients were investigated. Age, gender, New York Heart Association functional class and depressive symptoms are the most consistent variables explaining the variance in HRQoL in patients with congestive heart failure. These findings are partly in line with previous research on predictors for hard endpoints in patients with congestive heart failure.

Basal functional status predicts one-year mortality after a heart failure hospitalization in elderly patients - The RICA prospective study

BACKGROUND

Dependence for basic activities of the daily living (ADL) relates to adverse outcomes in elderly acute heart failure (AHF) patients.

METHODS

We evaluated patients ≥75years admitted because of AHF, divided according to preadmission Barthel Index (BI) category: severe (BI 0-60), moderate (BI 61-90) and slight dependence or independence for basic ADL (BI 91-100). We compared their baseline characteristics and used logistic regression models to determine whether a BI≤60 confers higher one-year mortality risk.

RESULTS

We included 2195 patients, mean age 83years; 57% women, Charlson Index 3, 65% with preserved left ventricular ejection fraction. Their median preadmission BI was 90 (65-100); 21.7% had BI≤60. Patients with BI≤60 were older, more often females, with higher comorbid and cognitive burden and more likely to be institutionalized. 560 patients (26%) died within the follow-up period. A preadmission BI≤60 was significantly associated with higher risk of 12-month mortality (HR 1.42, 95% CI 1.14-1.77) together with male sex (1.27, 1.04-1.54), valve disease (1.49, 1.20-1.83), worse preadmission NYHA class (1.44, 1.20-1.73), stage IV chronic kidney disease (1.70, 1.35-2.15), pulmonary edema (1.33, 1.01-1.76), no family support (1.47, 1.06-2.06), and higher Charlson Comorbidity Index (1.09, CI 1.05-1.13) and Pfeiffer cognitive screening questionnaire scores (1.10, 1.05-1.14).

CONCLUSION

Among elderly AHF patients, the presence of severe (BI≤60) preadmission dependence for basic ADL confers a significant and independent risk of one-year post-discharge mortality.

Transcoronary concentration gradients of circulating microRNAs in heart failure

AIMS

Circulating levels of microRNAs (miRNAs) are emergent promising biomarkers for cardiovascular disease. Altered expression of miRNAs has been related to heart failure (HF) and cardiac remodelling. We measured the concentration gradients across the coronary circulation to assess their usefulness to diagnose HF of different aetiologies.

METHODS AND RESULTS

Circulating miRNAs were measured in plasma samples simultaneously obtained from the aorta and the coronary venous sinus in patients with non-ischaemic HF (NICM-HF, n = 23) ischaemic HF (ICM-HF, n = 41), and in control patients (n = 11). A differential modulation of circulating levels of miR-423, -34a, -21-3p, -126, -199 and -30a was found across the aetiology groups. Interestingly, a positive transcoronary gradient was found for miR-423 (P < 0.001) and miR-34a (P < 0.001) only in the ICM-HF group. On the contrary, a positive gradient was found for miR-21-3p (P < 0.001) and miR-30a (P = 0.030) only in the NICM-HF group. Finally, no significant variations were observed in the transcoronary gradient of miR-126 or miR-199.

CONCLUSIONS

The present findings suggest that circulating levels of miRNAs are differentially expressed in patients with HF of different aetiologies. The presence of a transcoronary concentration gradient suggests a selective release of miRNAs by the failing heart into the coronary circulation. The presence of aetiology-specific transcoronary concentration gradients in HF patients might provide important information to better understand their role in HF, and suggests they could be useful biomarkers to distinguish HF of different aetiologies.

Sex-Related Aspects of Biomarkers in Cardiac Disease

Biomarkers play an important role in the clinical management of cardiac care. In particular, cardiac troponins (cTn) and natriuretic peptides are the cornerstones for the diagnosis of acute myocardial infarction (AMI) and for the diagnosis of heart failure (HF), respectively. Current guidelines do not make a distinction between women and men. However, the commonly used "one size fits all" algorithms are topic of debate to improve assessment of prognosis, particularly in women. Due to the high-sensitivity assays (hs-cTn), lower cTn levels (and 99th percentile upper reference limits) were observed in women as compared with men. Sex-specific diagnostic thresholds may improve the diagnosis of AMI in women, though clinical relevance remains controversial and more trials are needed. Also other diagnostic aspects are under investigation, like combined biomarkers approach and rapid measurement strategies. For the natriuretic peptides, previous studies observed higher concentrations in women than in men, especially in premenopausal women who might benefit from the cardioprotective actions. Contrary to hs-cTn, natriuretic peptides are particularly incorporated in the ruling-out algorithms for the diagnosis of HF and not ruling-in. Clinical relevance of sex differences here seems marginal, as clinical research has shown that negative predictive values for ruling-out HF were hardly effected when applying a universal diagnostic threshold that is independent from sex or other risk factors. Apart from the diagnostic issues of AMI in women, we believe that in the future most sex-specific benefits of cardiac biomarkers can be obtained in patient follow-up (guiding therapy) and prognostic applications, fitting modern ideas on preventive and personalized medicine.

Contractile reserve as a predictor of prognosis in patients with non-ischaemic systolic heart failure and dilated cardiomyopathy: a systematic review and meta-analysis

Objective

Patients with non-ischaemic systolic heart failure (HF) and idiopathic dilated cardiomyopathy (DCM) are a heterogenous group with varied morbidity and mortality. Prognostication in this group is challenging. We performed a systematic review and meta-analysis to examine the significance of the presence of contractile reserve as assessed via stress imaging on mortality and hospitalisation.

Methods

A search for studies that non-invasively assessed contractile reserve in patients with DCM or non-ischaemic HF with reduced ejection fraction, stress imaging with follow-up data comparing outcomes. A range of imaging modalities and stressors were included. We examined primary endpoints of mortality and secondary endpoints of combined cardiovascular events including HF progression or hospitalisation. Our analysis compared endpoints in patients with contractile reserve and those without it.

Results

Nine prospective cohort studies were identified describing a total of 787 patients. These studies are methodologically but not statistically heterogenous (I² = 31%). Using a random effect model, the presence of contractile reserve was associated with a significantly lower risk of mortality and cardiovascular events odds ratios of 0.20 (CI 0.11, 0.39) (P < 0.00001) and 0.13 (CI 0.04, 0.40) (P = 0.0004), respectively.

Conclusion

Regardless of stressor and imaging modality and despite the significant methodological heterogeneity within the current data (imaging techniques and parameters), patients with non-ischaemic cardiomyopathy and reduced EF who demonstrate contractile reserve have a lower mortality, and lower events/hospitalisations. The presence of contractile reserve therefore offers a potential positive prognostic indicator when managing these patients.

Impact of Frailty and Disability on 30-Day Mortality in Older Patients With Acute Heart Failure

The objectives were to determine the impact of frailty and disability on 30-day mortality and whether the addition of these variables to HFRSS EFFECT risk score (FBI-EFFECT model) improves the short-term mortality predictive capacity of both HFRSS EFFECT and BI-EFFECT models in older patients with acute decompensated heart failure (ADHF) atended in the emergency department. We performed a retrospective analysis of OAK Registry including all consecutive patients ≥65 years old with ADHF attended in 3 Spanish emergency departments over 4 months. FBI-EFFECT model was developed by adjusting probabilities of HFRSS EFFECT risk categories according to the 6 groups (G1: non frail, no or mildly dependent; G2: frail, no or mildly dependent; G3: non frail, moderately dependent; G4: frail, moderately dependent; G5: severely dependent; G6: very severely dependent).We included 596 patients (mean age: 83 [SD7]; 61.2% females). The 30-day mortality was 11.6% with statistically significant differences in the 6 groups (p < 0.001). After adjusting for HFRSS EFFECT risk categories, we observed a progressive increase in hazard ratios from groups G2 to G6 compared with G1 (reference). FBI-EFFECT had a better prognostic accuracy than did HFRSS EFFECT (log-rank p < 0.001; Net Reclassification Improvement [NRI] = 0.355; p < 0.001; Integrated Discrimination Improvement [IDI] = 0.052; p ;< 0.001) and BI-EFFECT (log-rank p = 0.067; NRI = 0.210; p = 0.033; IDI = 0.017; p = 0.026). In conclusion, severe disability and frailty in patients with moderate disability are associated with 30-day mortality in ADHF, providing additional value to HFRSS EFFECT model in predicting short-term prognosis and establishing a care plan.

Prognostic value of circulating microRNAs on heart failure-related morbidity and mortality in two large diverse cohorts of general heart failure patients

AIMS

Small studies suggested circulating microRNAs (miRNAs) as biomarkers for heart failure (HF). However, standardized approaches and quality assessment for measuring circulating miRNAs are not uniformly established, and most studies have been small, so that results are inconsistent. We used a standardized data handling protocol, optimized for circulating miRNA qPCRs to remove noise and used it to assess which circulating miRNAs robustly add prognostic information in patients with HF.

METHODS AND RESULTS

We measured 12 miRNAs in two independent cohorts totalling 2203 subjects. Cohort I (Barcelona) comprised 834 chronic HF patients. Cohort II (Detroit) comprised 1369 chronic HF patients. Each sample was measured in duplicate, and normalized to a very abundant and stable miRNA (miR-486-5p). We used a multistep algorithm to distinguish false amplification signals and thus classify each miRNA measurement as 'valid', 'undetectable' or 'invalid'. Higher levels of miR-1254 and miR-1306-5p were significantly associated with risk of the combined endpoint of all-cause mortality and HF hospitalization in both cohorts, with hazard ratios ranging from 1.11 to 1.21 per log increase (P-values 0.004 to 0.009). However, adding these miRNAs to established predictors (age, sex, haemoglobin, renal function, and NT-proBNP) did not further augment the c-statistic beyond 0.69 (cohort I) or 0.70 (cohort II).

CONCLUSION

We used a stringent quality assessment for miRNA testing, and were able to replicate the association of miR-1254 and miR-1306-5p with risk of death and HF hospitalization in HF patients of two independent cohorts. However, these two circulating miRNAs failed to improve prognostication over established predictors.

Cardiac Computed Tomography Angiographic Findings as Predictors of Late Heart Failure in an Asymptomatic Diabetic Cohort: An 8-Year Prospective Follow-Up Study

OBJECTIVES

Predictive models for heart failure (HF) in heterogeneous populations have had limited success. We examined cardiac computed tomography angiography (CTA) predictors of HF or cardiovascular death (HF-CVD) in a prospective study of asymptomatic diabetics undergoing baseline assessment by CTA.

METHODS

The subjects (n = 735, aged 55-74 years, 51.2% women) had no clinical history of cardiovascular disease at study entry. Coronary artery calcium (CAC) score, CTA-defined coronary atherosclerosis, cardiac chamber volumes, and clinical data were collected and late outcome events recorded over 8.4 ± 0.6 years (range 7.3-9.3).

RESULTS

HF-CVD occurred in 41 (5.6%) subjects, with HF occurring mostly (19/23, 82.6%) in subjects without preceding myocardial infarction. Baseline univariate clinical outcome predictors of HF-CVD included older age (p = 0.027), the duration of diabetes (p = 0.004), HbA1c (p < 0.0001), microvascular disease (retinopathy, microalbuminuria) (p < 0.0001), and systolic blood pressure (p = 0.035). Baseline univariate CTA predictors included CAC score (p = 0.004), coronary stenosis (p = 0.047), and a CTA-defined left/right atrial (LA/RA) volume ratio >1 (p < 0.0001). Independent predictors were an LA/RA volume ratio >1, microvascular disease, and systolic blood pressure (model C-statistic 0.792, 95% CI 0.758-0.824). Measures of the extent of coronary artery disease (CAD) were not independent predictors of HF-CVD.

CONCLUSIONS

In a low- to moderate-risk asymptomatic diabetic population, CTA LA enlargement (LA/RA volume ratio) but not the extent of CAD had independent prognostic value for HF-CVD in addition to the clinical variables.

Community burden and prognostic impact of reduced kidney function among patients hospitalized with acute decompensated heart failure: The Atherosclerosis Risk in Communities (ARIC) Study Community Surveillance

Background

Kidney dysfunction is prevalent and impacts prognosis in patients with acute decompensated heart failure (ADHF). However, most previous reports were from a single hospital, limiting their generalizability. Also, contemporary data using new equation for estimated glomerular filtration rate (eGFR) are needed.

Methods and results

We analyzed data from the ARIC Community Surveillance for ADHF conducted for residents aged ≥55 years in four US communities between 2005–2011. All ADHF cases (n = 5, 391) were adjudicated and weighted to represent those communities (24,932 weighted cases). The association of kidney function (creatinine-based eGFR by the CKD-EPI equation and blood urea nitrogen [BUN]) during hospitalization with 1-year mortality was assessed using logistic regression. Based on worst and last serum creatinine, there were 82.5% and 70.6% with reduced eGFR (<60 ml/min/1.73m²) and 37.4% and 26.6% with severely reduced eGFR (<30 ml/min/1.73m²), respectively. Lower eGFR (regardless of last or worst eGFR), particularly eGFR <30 ml/min/1.73m², was significantly associated with higher 1-year mortality independently of potential confounders (odds ratio 1.60 [95% CI 1.26–2.04] for last eGFR 15–29 ml/min/1.73m² and 2.30 [1.76–3.00] for <15 compared to eGFR ≥60). The association was largely consistent across demographic subgroups. Of interest, when both eGFR and BUN were modeled together, only BUN remained significant.

Conclusions

Severely reduced eGFR (<30 ml/min/1.73m²) was observed in ~30% of ADHF cases and was an independent predictor of 1-year mortality in community. For prediction, BUN appeared to be superior to eGFR. These findings suggest the need of close attention to kidney dysfunction among ADHF patients.

Multistate Model to Predict Heart Failure Hospitalizations and All-Cause Mortality in Outpatients With Heart Failure With Reduced Ejection Fraction: Model Derivation and External Validation

BACKGROUND

Outpatients with heart failure (HF) who are at high risk for HF hospitalization and death may benefit from early identification. We sought to develop and externally validate a model to predict both HF hospitalization and mortality that accounts for the semicompeting nature of the 2 outcomes and captures the risk associated with the transition from the stable outpatient state to the post-HF hospitalization state.

METHODS AND RESULTS

A multistate model to predict HF hospitalization and all-cause mortality was derived using data (n=3834) from the HEAAL study (Heart Failure Endpoint evaluation of Angiotensin II Antagonist Losartan), a multinational randomized trial in symptomatic patients with reduced left ventricular ejection fraction. Twelve easily and reliably obtainable demographic and clinical predictors were prespecified for model inclusion. Model performance was assessed in the SCD-HeFT cohort (Sudden Cardiac Death in Heart Failure Trial; n=2521). At 1 year, the probability of being alive without HF hospitalization was 94% for a typical patient in the lowest risk quintile and 77% for a typical patient in the highest risk quintile and this variability in risk continued through 7 years of follow-up. The model c-index was 0.72 in the derivation cohort, 0.66 in the validation cohort, and 0.69 in the implantable cardiac defibrillator arm of the validation cohort. There was excellent calibration across quintiles of predicted risk.

CONCLUSIONS

Our findings illustrate the advantages of a multistate modeling approach, providing estimates of HF hospitalization and death in the same model, comparison of predictors for the different outcomes and demonstrating the different trajectories of patients based on baseline characteristics and intermediary events.

CLINICAL TRIAL REGISTRATION

URL: http://www.clinicaltrials.gov. Unique identifiers: NCT00000609 and NCT00090259.

Early in-hospital variation of red blood cell distribution width predicts mortality in patients with acute heart failure

BACKGROUND

Some studies showed that the value of red blood cell distribution width (RDW) at admission may predict clinical outcomes in patients with acutely decompensated heart failure (ADHF). Therefore, this study was planned to investigate whether in-hospital variations of RDW may also predict mortality in this condition.

METHODS

The final study population consisted of 588 patients admitted to the local Emergency Department (ED), who were hospitalized for ADHF. The RDW was measured at ED admission and after 48h and 96h of hospital stay. In-hospital variations from admission value, expressed as absolute variation (DeltaRDW) or percent variation (Delta%RDW), were then correlated with 30- and 60-day mortality.

RESULTS

Overall, 87 (14.8%) and 118 (20.1%) patients with ADHF died at 30 or 60days of follow-up. Delta%RDW after 96h of hospital stay independently predicted 30-day mortality (odds ratio, 1.12; 95% CI, 1.07-1.18). An increase >1% of Delta%RDW after 96h of hospital stay independently predicted both 30-day (odds ratio, 2.86; 95% CI, 1.67-4.97) and 60-day (odds ratio, 3.06; 95% CI, 1.89-4.96) mortality. A similar trend was observed for DeltaRDW, since an increase after 96h of hospital stay was associated with a nearly 4-fold higher 30-day mortality (odds ratio, 3.65; 95% CI, 2.02-6.15).

CONCLUSION

Despite it remains unclear whether RDW is a real risk factor or an epiphenomenon in ADHF, these results suggest that more aggressive management may be advisable in ADHF patients with increasing anisocytosis during the first days of hospitalization.

ANMCO/ELAS/SIBioC Consensus Document: biomarkers in heart failure

Biomarkers have dramatically impacted the way heart failure (HF) patients are evaluated and managed. A biomarker is a characteristic that is objectively measured and evaluated as an indicator of normal biological or pathogenic processes, or pharmacological responses to a therapeutic intervention. Natriuretic peptides [B-type natriuretic peptide (BNP) and N-terminal proBNP] are the gold standard biomarkers in determining the diagnosis and prognosis of HF, and a natriuretic peptide-guided HF management looks promising. In the last few years, an array of additional biomarkers has emerged, each reflecting different pathophysiological processes in the development and progression of HF: myocardial insult, inflammation, fibrosis, and remodelling, but their role in the clinical care of the patient is still partially defined and more studies are needed before to be well validated. Moreover, several new biomarkers have the potential to identify patients with early renal dysfunction and appear to have promise to help the management cardio-renal syndrome. With different biomarkers reflecting HF presence, the various pathways involved in its progression, as well as identifying unique treatment options for HF management, a closer cardiologist-laboratory link, with a multi-biomarker approach to the HF patient, is not far ahead, allowing the unique opportunity for specifically tailoring care to the individual pathological phenotype.

A multimarker multi-time point-based risk stratification strategy in acute heart failure: results from the RELAX-AHF trial

AIMS

We evaluated the added prognostic value of a multi-time point-based multimarker panel of biomarkers in patients with acute heart failure (AHF).

METHODS AND RESULTS

Seven circulating biomarkers [NT-proBNP, high sensitivity cardiac troponin T (hs-cTnT), soluble ST2 (sST2), growth differentiation factor 15 (GDF-15), cystatin-C, galectin-3, and high sensitivity C-reactive protein (hs-CRP)] were measured at baseline and on days 2, 5, 14, and 60 in 1161 patients enrolled in the RELAX-AHF trial. Patients with BNP ≥350 ng/L or NT-proBNP ≥1400 ng/L, mild to moderate renal impairment, and systolic blood pressure >125 mmHg were included in the trial. Time-dependent Cox regression analysis was utilized to evaluate the incremental value of serial measurement of biomarkers. Added value of individual biomarkers and their combination, on top of a pre-specified baseline model, was quantified with the gain in the C-index. Serial biomarker evaluation showed incremental predictive value over baseline measurements alone for the prediction of 180-day cardiovascular mortality except for galectin-3. While a repeat measurement as early as day 2 was adequate for NT-proBNP and cystatin-C in terms of maximizing discriminatory accuracy, further measurements on days 14 and 60 provided added value for hs-cTnT, GDF-15, sST2, and hs-CRP. Individual biomarker additions on top of the baseline model showed additional prognostic value. The greatest prognostic gain was, however, attained with the combination of NT-proBNP, hs-cTnT, GDF-15, and sST2, which yielded 0.08 unit absolute increment in the C-index to 0.87 (95% confidence interval 0.83-0.91].

CONCLUSION

In patients with AHF and mild to moderate renal impairment, a multimarker approach based on a panel of serially evaluated biomarkers provides the greatest prognostic improvement unmatched by a single time point-based single marker strategy.

Analysis of Machine Learning Techniques for Heart Failure Readmissions

BACKGROUND

The current ability to predict readmissions in patients with heart failure is modest at best. It is unclear whether machine learning techniques that address higher dimensional, nonlinear relationships among variables would enhance prediction. We sought to compare the effectiveness of several machine learning algorithms for predicting readmissions.

METHODS AND RESULTS

Using data from the Telemonitoring to Improve Heart Failure Outcomes trial, we compared the effectiveness of random forests, boosting, random forests combined hierarchically with support vector machines or logistic regression (LR), and Poisson regression against traditional LR to predict 30- and 180-day all-cause readmissions and readmissions because of heart failure. We randomly selected 50% of patients for a derivation set, and a validation set comprised the remaining patients, validated using 100 bootstrapped iterations. We compared C statistics for discrimination and distributions of observed outcomes in risk deciles for predictive range. In 30-day all-cause readmission prediction, the best performing machine learning model, random forests, provided a 17.8% improvement over LR (mean C statistics, 0.628 and 0.533, respectively). For readmissions because of heart failure, boosting improved the C statistic by 24.9% over LR (mean C statistic 0.678 and 0.543, respectively). For 30-day all-cause readmission, the observed readmission rates in the lowest and highest deciles of predicted risk with random forests (7.8% and 26.2%, respectively) showed a much wider separation than LR (14.2% and 16.4%, respectively).

CONCLUSIONS

Machine learning methods improved the prediction of readmission after hospitalization for heart failure compared with LR and provided the greatest predictive range in observed readmission rates.

Prognostic impact of comorbidities in hospitalized patients with acute exacerbation of chronic heart failure

BACKGROUND

To assess the impact of comorbidities on long-term all-cause mortality in patients hospitalized with exacerbated signs/symptoms of previously chronic stable HF (AE-CHF).

METHODS

1119 patients admitted for AE-CHF and with NT-proBNP levels >900pg/mL were enrolled. Univariable and multivariable Cox analyses were performed to assess the association of age, gender, hypertension, diabetes, obesity, atrial fibrillation, coronary heart disease (CHD), chronic obstructive pulmonary disease, previous cerebrovascular accidents, chronic liver disease (CLD), thyroid disease, renal impairment (RI), and anemia with 3-year all-cause mortality.

RESULTS

During the follow-up, 441 patients died and 126 underwent heart transplantation (HT) or ventricular assist device (VAD) implantation. 45.8% of the fatal events and 52.4% of HT/VAD implantations occurred within 180days after admission. Increasing age (p=.012), obesity (p=.037), atrial fibrillation (p=.030), CHD (p=.015), CLD (p=.001), RI (p<.001), and anemia (p<.001) were independently associated with 3-year all-cause mortality. Most of the prognostic impact of CHD, took place within the first 180days after admission. Male gender was associated with mortality beyond 180days. Compared with normal weight, obesity was associated with better overall survival. Obese patients, however, had significantly lower NT-proBNP concentrations and less frequently presented with hypotension, hyponatremia, and severe left ventricular systolic dysfunction, despite a similar prevalence of severe dyspnea at admission.

CONCLUSIONS

Several comorbidities are associated with long-term risk of death in hospitalized patients with worsening HF, although the nature of this association does appear to be complex. Our data may help to raise awareness about the clinical relevance of comorbid conditions.

Changes in Left Ventricular Ejection Fraction Predict Survival and Hospitalization in Heart Failure With Reduced Ejection Fraction

BACKGROUND

Left ventricular remodeling, as commonly measured by left ventricular ejection fraction (LVEF), is associated with clinical outcomes. Although change in LVEF over time should reflect response to therapy and clinical course, serial measurement of LVEF is inconsistently performed in observational settings, and the incremental prognostic value of change in LVEF has not been well characterized.

METHODS AND RESULTS

The β-Blocker Evaluation of Survival Trial measured LVEF by radionuclide ventriculography at baseline and at 3 and 12 months after randomization. We built a series of multivariable models with 16 clinical parameters plus change in LVEF for predicting 4 major clinical end points, including the trial's primary end point of all-cause mortality. Among 2484 patients with at least 1 follow-up LVEF, change in LVEF was the second most significant predictor (behind baseline creatinine) of all-cause mortality (adjusted hazard ratio for improvement in LVEF by ≥5 U responder versus nonresponder [95% confidence intervals] for all-cause mortality=0.62 [0.52-0.73]). Other end points, including heart failure hospitalization or the composite of all-cause mortality and heart failure hospitalization, yielded similar results. LVEF change ≥5 U was associated with a modest increase in discrimination when added to traditional predictors and was predictive of outcomes in both the bucindolol and placebo treatment groups. LVEF change as a predictor of outcomes was affected by sex and race, with evidence that LVEF improvement is associated with less survival benefit in African Americans and women.

CONCLUSIONS

Serial evaluation for LVEF change predicts both survival and heart failure hospitalization and provides a dynamic/real-time measure of prognosis in heart failure with reduced LVEF.

CLINICAL TRIAL REGISTRATION

URL: http://www.clinicaltrials.gov. Unique identifier: NCT00000560.

Comorbid Depression and Heart Failure: A Community Cohort Study

Objective

To examine the association between depression and clinical outcomes in heart failure (HF) in a community cohort.

Patients and Methods

HF patients in Minnesota, United States completed depression screening using the 9-item Patient Health Questionnaire (PHQ-9) between 1^st Oct 2007 and 1^st Dec 2011; patients with PHQ-9≥5 were labelled “depressed”. We calculated the risk of death and first hospitalization within 2 years using Cox regression. Results were adjusted for 10 commonly used prognostic factors (age, sex, systolic blood pressure, estimated glomerular filtration rate, serum sodium, ejection fraction, blood urea nitrogen, brain natriuretic peptide, presence of diabetes and ischaemic aetiology). Area under the curve (AUC), integrated discrimination improvement (IDI) and net reclassification improvement (NRI) compared depression as a predictor against the aforementioned factors.

Results

425 patients (mean age 74, 57.6% males) were included in the study; 179 (42.1%) had PHQ-9≥5. The adjusted hazard ratio of death was 2.02 (95% CI 1.34–3.04) and of hospitalization was 1.42 (95% CI 1.13–1.80) for those with compared to those without depression. Adding depression to the models did not appreciably change the AUC but led to statistically significant improvements in both the IDI (p = 0.001 and p = 0.005 for death and hospitalization, respectively) and NRI (for death and hospitalization, 35% (p = 0.002) and 27% (p = 0.007) were reclassified correctly, respectively).

Conclusion

Depression is frequent among community patients with HF and associated with increased risk of hospitalizations and death. Risk prediction for death and hospitalizations in HF patients can be improved by considering depression.

Heart Failure: Diagnosis, Management and Utilization

Despite the advancement in medicine, management of heart failure (HF), which usually presents as a disease syndrome, has been a challenge to healthcare providers. This is reflected by the relatively higher rate of readmissions along with increased mortality and morbidity associated with HF. In this review article, we first provide a general overview of types of HF pathogenesis and diagnostic features of HF including the crucial role of exercise in determining the severity of heart failure, the efficacy of therapeutic strategies and the morbidity/mortality of HF. We then discuss the quality control measures to prevent the growing readmission rates for HF. We also attempt to elucidate published and ongoing clinical trials for HF in an effort to evaluate the standard and novel therapeutic approaches, including stem cell and gene therapies, to reduce the morbidity and mortality. Finally, we discuss the appropriate utilization/documentation and medical coding based on the severity of the HF alone and with minor and major co-morbidities. We consider that this review provides an extensive overview of the HF in terms of disease pathophysiology, management and documentation for the general readers, as well as for the clinicians/physicians/hospitalists.

Long-term serial kinetics of N-terminal pro B-type natriuretic peptide and carbohydrate antigen 125 for mortality risk prediction following acute heart failure

AIM

Baseline values of N-terminal pro B-type natriuretic peptide (NT-proBNP) and carbohydrate antigen 125 (CA125) predict all-cause mortality in acute heart failure (AHF). However, there is limited information about the added prognostic benefit of using longitudinal values, and how this predictive ability is modified when modelling together. The aim of this study was to determine the mutually-adjusted association between the longitudinal trajectories of NT-proBNP and CA125 with all-cause mortality after an episode of AHF.

METHODS AND RESULTS

We included 946 consecutive patients discharged for AHF. NT-proBNP and CA125 were measured at each physician-patient encounter (median (interquartile range (IQR)):3 (2-4)). The effect on mortality (time-dependent modelling) was assessed using joint modelling (JM) and multi-state Markov. The mean age was 71±11 years and 51% exhibited left ventricular systolic dysfunction. At a median follow-up of 2.64 years (IQR=1.20-5.36), 498 patients died (52.6%). The observed trajectories of both biomarkers markedly differed over survival status, with sustained higher values in patients who died. After being adjusted by established risk factors and by each other, the baseline absolute change in CA125 and NT-proBNP were significantly associated to mortality (hazard ratio (HR)=1.05 (1.01-1.09); p=0.011 (area under the curve (AUC)=0.76) and HR=1.04 (1.02-1.06); p<0.001 (AUC=0.75), respectively). After merging the binary version of NT-proBNP (⩾1000 pg/ml) and CA125 (>35 U/ml) into a four-level variable, we found the highest risk when both were elevated, intermediate risk when either one was low, and lowest risk when both were low.

CONCLUSION

The combination of long-term longitudinal trajectories of CA125 and NT-proBNP improves risk stratification for all-cause mortality after a hospitalization for AHF.

Importance of Abnormal Chloride Homeostasis in Stable Chronic Heart Failure

BACKGROUND

The aim of this analysis was to determine the long-term prognostic value of lower serum chloride in patients with stable chronic heart failure. Electrolyte abnormalities are prevalent in patients with chronic heart failure. Little is known regarding the prognostic implications of lower serum chloride.

METHODS AND RESULTS

Serum chloride was measured in 1673 consecutively consented stable patients with a history of heart failure undergoing elective diagnostic coronary angiography. All patients were followed for 5-year all-cause mortality, and survival models were adjusted for variables that confounded the chloride-risk relationship. The average chloride level was 102 ± 4 mEq/L. Over 6772 person-years of follow-up, there were 547 deaths. Lower chloride (per standard deviation decrease) was associated with a higher adjusted risk of mortality (hazard ratio 1.29, 95% confidence interval 1.12-1.49; P < 0.001). Chloride levels net-reclassified risk in 10.4% (P = 0.03) when added to a multivariable model (with a resultant C-statistic of 0.70), in which sodium levels were not prognostic (P = 0.30). In comparison to those with above first quartile chloride (≥ 101 mEq/L) and sodium (≥ 138 meq/L), subjects with first quartile chloride had a higher adjusted mortality risk, whether they had first quartile sodium (hazard ratio 1.35, 95% confidence interval 1.08-1.69; P = 0.008) or higher (hazard ratio 1.43, 95% confidence interval 1.12-1.85; P = 0.005). However, subjects with first quartile sodium but above first quartile chloride had no association with mortality (P = 0.67).

CONCLUSIONS

Lower serum chloride levels are independently and incrementally associated with increased mortality risk in patients with chronic heart failure. A better understanding of the biological role of serum chloride is warranted.

Serial galectin-3 and future cardiovascular disease in the general population

BACKGROUND

Lifetime risk for cardiovascular (CV) disease is high but predicting incident events on an individual level remains difficult. Single measurements of galectin-3, a marker of tissue fibrosis, predict mortality and new-onset heart failure (HF). Persistently elevated levels may indicate a clinically silent disease process.

OBJECTIVES

Our aim was to establish the value of serial galectin-3 measurements to predict CV outcomes in the general population.

METHODS

Plasma galectin-3 was measured in the Prevention of REnal and Vascular ENd-stage Disease (PREVEND) study at baseline and after ∼4 years. Changes in serial galectin-3 were expressed as categorical changes or absolute change from baseline and were related to subsequent outcome.

RESULTS

Serial galectin-3 was measured in 5958 subjects (mean age 49±12 years; 49% female). The median duration of follow-up was 8.3 years. Persistently elevated galectin-3 (defined as highest quartile at baseline and highest quartile during visit 2, n=757 subjects) was associated with a higher risk for new-onset HF, CV mortality, all-cause mortality, new-onset atrial fibrillation and CV events, compared with subjects with non-persistently elevated galectin-3. After multivariable adjustments for baseline characteristics, serial galectin-3 remained an independent predictor of new-onset HF (HR 1.85 (1.10-3.13); p=0.02) but not for other outcomes. Serial measurements provided more accurate prognostic value to predict new-onset HF, compared with a single baseline measurement (Harrell's C: 0.72 (0.68-0.75) vs 0.68 (0.65-0.72); p=0.002, respectively) with significant net reclassification.

CONCLUSIONS

Persistently elevated galectin-3 predicts new-onset HF after adjustment for covariates, and serial measurements provide more accurate prognostic information compared with single determination of galectin-3. This may help to identify individuals who are at risk for incident HF and might provide a measure to monitor interventions.

Incremental utility of prognostic variables at discharge for risk prediction in hospitalized patients with acutely decompensated chronic heart failure

OBJECTIVES

To assess the incremental prognostic utility of discharge serum creatinine (SCr), systolic blood pressure (SBP), and NT-proBNP and sodium concentrations in hospitalized patients with acutely decompensated chronic heart failure.

BACKGROUND

Whether key prognostic variables at discharge provide incremental prognostic information beyond that provided by a model based on admission variables (referent) remains incompletely defined.

METHODS

The primary outcome was a composite of death, urgent heart transplantation, or ventricular assist device implantation at 1 year. The gain in predictive performance was assessed using C index, Bayesian Information Criterion, and Net Reclassification Improvement.

RESULTS

The best fit was obtained when discharge NT-proBNP was added to the referent model. No interaction between admission and discharge NT-proBNP was found. Discharge SCr, SBP, and sodium did not improve goodness-of-fit.

CONCLUSIONS

Admission and discharge NT-proBNP provide complementary and independent prognostic information; as such, they should be taken into account concurrently.