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

Developing EHR-driven heart failure risk prediction models using CPXR(Log) with the probabilistic loss function

Computerized survival prediction in healthcare identifying the risk of disease mortality, helps healthcare providers to effectively manage their patients by providing appropriate treatment options. In this study, we propose to apply a classification algorithm, Contrast Pattern Aided Logistic Regression (CPXR(Log)) with the probabilistic loss function, to develop and validate prognostic risk models to predict 1, 2, and 5year survival in heart failure (HF) using data from electronic health records (EHRs) at Mayo Clinic. The CPXR(Log) constructs a pattern aided logistic regression model defined by several patterns and corresponding local logistic regression models. One of the models generated by CPXR(Log) achieved an AUC and accuracy of 0.94 and 0.91, respectively, and significantly outperformed prognostic models reported in prior studies. Data extracted from EHRs allowed incorporation of patient co-morbidities into our models which helped improve the performance of the CPXR(Log) models (15.9% AUC improvement), although did not improve the accuracy of the models built by other classifiers. We also propose a probabilistic loss function to determine the large error and small error instances. The new loss function used in the algorithm outperforms other functions used in the previous studies by 1% improvement in the AUC. This study revealed that using EHR data to build prediction models can be very challenging using existing classification methods due to the high dimensionality and complexity of EHR data. The risk models developed by CPXR(Log) also reveal that HF is a highly heterogeneous disease, i.e., different subgroups of HF patients require different types of considerations with their diagnosis and treatment. Our risk models provided two valuable insights for application of predictive modeling techniques in biomedicine: Logistic risk models often make systematic prediction errors, and it is prudent to use subgroup based prediction models such as those given by CPXR(Log) when investigating heterogeneous diseases.

Depression and Multiple Rehospitalizations in Patients With Heart Failure

BACKGROUND

There have been few studies of the effect of depression on rehospitalization in patients with heart failure (HF), and even fewer on its role in multiple rehospitalizations.

HYPOTHESIS

Depression is an independent risk factor for multiple readmissions in patients with HF.

METHODS

A cohort of 662 patients with HF who were discharged alive after hospitalization were interviewed to evaluate symptoms of depression and were followed for 1 year. All-cause readmissions were documented by chart review. A marginal proportional rates model was used to model the effect of depression on the rate of rehospitalization with adjustment for known predictors of HF outcomes.

RESULTS

Depression symptoms predicted multiple readmissions (adjusted hazard ratio [HR]: 1.08, 95% confidence interval [CI]: 1.03-1.13, P = 0.0008). Compared with patients without depression, those who met the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria for major depression at index were at the highest risk for multiple rehospitalizations (HR: 1.51, 95% CI: 1.15-1.97, P = 0.003).

CONCLUSIONS

Depression is an independent risk factor for multiple all-cause readmissions in patients with HF.

Validation of the Get With The Guideline-Heart Failure risk score in Japanese patients and the potential improvement of its discrimination ability by the inclusion of B-type natriuretic peptide level

BACKGROUND

Detailed characteristics of patients with acute heart failure (AHF) in Japan have not been elucidated. Furthermore, international application of risk models obtained in the United States has not been validated.

METHODS

We evaluated the Get With The Guidelines-Heart Failure (GWTG-HF) risk score performance in AHF patients enrolled in the West Tokyo Heart Failure registry, a large, ongoing, prospective, multicenter cohort registry. Variables required for the GWTG-HF risk score were race, age, systolic blood pressure, heart rate, blood urea nitrogen level, sodium concentration, and presence of chronic obstructive pulmonary disease. Score discrimination and calibration were evaluated by the c statistic, Hosmer-Lemeshow statistic, and visual plotting. We conducted additional analyses to determine whether other variables improved the performance of the score. The primary outcome was in-hospital mortality.

RESULTS

In total, 1,876 patients were admitted for AHF between April 2006 and August 2014. The patients were predominantly men (60.6%), with a mean age of 73.3 ± 13.6 years. Sixty-eight (3.6%) patients died during hospitalization. The GWTG-HF risk score showed acceptable discrimination; the c statistic for in-hospital mortality in this cohort was 0.763 (95% CI, 0.700-0.826). The calibration plot showed good conformance between the predicted and observed in-hospital mortality. Notably, addition of B-type natriuretic peptide level to the conventional GWTG-HF score significantly improved the discrimination (c statistic, 0.818; 95% CI, 0.771-0.865).

CONCLUSIONS

The GWTG-HF risk score can be applied in Japanese AHF patients with good discrimination and calibration. Furthermore, addition of B-type natriuretic peptide level improves discrimination and could be considered in future risk models.

Circulating microRNAs and Outcome in Patients with Acute Heart Failure

Background

The biomarker value of circulating microRNAs (miRNAs) has been extensively addressed in patients with acute coronary syndrome. However, prognostic performances of miRNAs in patients with acute heart failure (AHF) has received less attention.

Methods

A test cohort of 294 patients with acute dyspnea (236 AHF and 58 non-AHF) and 44 patients with stable chronic heart failure (CHF), and an independent validation cohort of 711 AHF patients, were used. Admission levels of miR-1/-21/-23/-126/-423-5p were assessed in plasma samples.

Results

In the test cohort, admission levels of miR-1 were lower in AHF and stable CHF patients compared to non-AHF patients (p = 0.0016). Levels of miR-126 and miR-423-5p were lower in AHF and in non-AHF patients compared to stable CHF patients (both p<0.001). Interestingly, admission levels of miR-423-5p were lower in patients who were re-admitted to the hospital in the year following the index hospitalization compared to patients who were not (p = 0.0001). Adjusted odds ratio [95% confidence interval] for one-year readmission was 0.70 [0.53–0.93] for miR-423-5p (p = 0.01). In the validation cohort, admission levels of miR-423-5p predicted 1-year mortality with an adjusted odds ratio [95% confidence interval] of 0.54 [0.36–0.82], p = 0.004. Patients within the lowest quartile of miR-423-5p were at high risk of long-term mortality (p = 0.02).

Conclusions

In AHF patients, low circulating levels of miR-423-5p at presentation are associated with a poor long-term outcome. This study supports the value of miR-423-5p as a prognostic biomarker of AHF.

Biomarkers and low risk in heart failure. Data from COACH and TRIUMPH

AIM

Traditionally, risk stratification in heart failure (HF) emphasizes assessment of high risk. We aimed to determine if biomarkers could identify patients with HF at low risk for death or HF rehospitalization.

METHODS AND RESULTS

This analysis was a substudy of The Coordinating Study Evaluating Outcomes of Advising and Counselling in Heart Failure (COACH) trial. Enrolment of HF patients occurred before discharge. We defined low risk as the absence of death and/or HF rehospitalizations at 180 days. We tested a diverse group of 29 biomarkers on top of a clinical risk model, with and without N-terminal pro-B-type natriuretic peptide (NT-proBNP), and defined the low risk biomarker cut-off at the 10th percentile associated with high positive predictive value. The best performing biomarkers together with NT-proBNP and cardiac troponin I (cTnI) were re-evaluated in a validation cohort of 285 HF patients. Of 592 eligible COACH patients, the mean (± SD) age was 71 (± 11) years and median (IQR) NT-proBNP was 2521 (1301-5634) pg/mL. Logistic regression analysis showed that only galectin-3, fully adjusted, was significantly associated with the absence of events at 180 days (OR 8.1, 95% confidence interval 1.06-50.0, P = 0.039). Galectin-3, showed incremental value when added to the clinical risk model without NT-proBNP (increase in area under the curve from 0.712 to 0.745, P = 0.04). However, no biomarker showed significant improvement by net reclassification improvement on top of the clinical risk model, with or without NT-proBNP. We confirmed our results regarding galectin-3, NT-proBNP, and cTnI in the independent validation cohort.

CONCLUSION

We describe the value of various biomarkers to define low risk, and demonstrate that galectin-3 identifies HF patients at (very) low risk for 30-day and 180-day mortality and HF rehospitalizations after an episode of acute HF. Such patients might be safely discharged.

AHEAD score--Long-term risk classification in acute heart failure

BACKGROUND

The role of co-morbidities in the prognosis of patients hospitalized for AHF was examined using the AHEAD (A--atrial fibrillation, H--haemoglobin<130 g/l for men and 120 g/l for women (anaemia), E--elderly (age>70years), A--abnormal renal parameters (creatinine>130 μmol/l), D--diabetes mellitus) scoring system.

METHODS

AHEAD--multicentre prospective Czech registry of AHF patients; GREAT registry--international cohort of AHF patients. Data from 5846 consecutive patients hospitalized for AHF (AHEAD registry; derivation cohort) were analysed to build the AHEAD score. Each risk factor of the AHEAD score was counted as 1 point. The model was validated externally using an international cohort of similar patients in the GREAT registry (6315).

RESULTS

Main outcome was one year all-cause mortality. The mean age of patients was 72±12 years, with 61.6% of patients aged >70 years; 43.4% were women. Atrial fibrillation was present in 30.7%, anaemia in 38.2%, creatinine>130 mmol/l (abnormal renal parameters) in 30.1%, and diabetes mellitus in 44.0%. The mean AHEAD score was 2.1. In patients with AHEAD scores of 0-5, the one-year mortality rates were 13.6%, 23.4%, 32.0%, 41.1%, 47.7%, and 58.2%, respectively (p<0.001), and the 90 month mortality rates were 35.1%, 57.3%, 73.5%, 84.8%, 88.0%, and 91.7%, respectively (p<0.001).

CONCLUSION

The AHEAD is a simple scoring system based on the analysis of co-morbidities for the estimation of the short and long term prognosis of patients hospitalized for AHF.

Fibroblast Growth Factor 23 Is an Independent and Specific Predictor of Mortality in Patients With Heart Failure and Reduced Ejection Fraction

BACKGROUND

Strategies to improve risk prediction are of major importance in patients with heart failure (HF). Fibroblast growth factor 23 (FGF-23) is an endocrine regulator of phosphate and vitamin D homeostasis associated with an increased cardiovascular risk. We aimed to assess the prognostic effect of FGF-23 on mortality in HF patients with a particular focus on differences between patients with HF with preserved ejection fraction and patients with HF with reduced ejection fraction (HFrEF).

METHODS AND RESULTS

FGF-23 levels were measured in 980 patients with HF enrolled in the Ludwigshafen Risk and Cardiovascular Health (LURIC) study including 511 patients with HFrEF and 469 patients with HF with preserved ejection fraction and a median follow-up time of 8.6 years. FGF-23 was additionally measured in a second cohort comprising 320 patients with advanced HFrEF. FGF-23 was independently associated with mortality with an adjusted hazard ratio per 1-SD increase of 1.30 (95% confidence interval, 1.14-1.48; P<0.001) in patients with HFrEF, whereas no such association was found in patients with HF with preserved ejection fraction (for interaction, P=0.043). External validation confirmed the significant association with mortality with an adjusted hazard ratio per 1 SD of 1.23 (95% confidence interval, 1.02-1.60; P=0.027). FGF-23 demonstrated an increased discriminatory power for mortality in addition to N-terminal pro-B-type natriuretic peptide (C-statistic: 0.59 versus 0.63) and an improvement in net reclassification index (39.6%; P<0.001).

CONCLUSIONS

FGF-23 is independently associated with an increased risk of mortality in patients with HFrEF but not in those with HF with preserved ejection fraction, suggesting a different pathophysiologic role for both entities.

A simple validated method for predicting the risk of hospitalization for worsening of heart failure in ambulatory patients: the Redin-SCORE

AIMS

Prevention of hospital readmissions is one of the main objectives in the management of patients with heart failure (HF). Most of the models predicting readmissions are based on data extracted from hospitalized patients rather than from outpatients. Our objective was to develop a validated score predicting 1-month and 1-year risk of readmission for worsening of HF in ambulatory patients.

METHODS AND RESULTS

A cohort of 2507 ambulatory patients with chronic HF was prospectively followed for a median of 3.3 years. Clinical, echocardiographic, ECG, and biochemical variables were used in a competing risk regression analysis to construct a risk score for readmissions due to worsening of HF. Thereafter, the score was externally validated using a different cohort of 992 patients with chronic HF (MUSIC registry). Predictors of 1-month readmission were the presence of elevated natriuretic peptides, left ventricular (LV) HF signs, and estimated glomerular filtration rate (eGFR) <60 mL/min/m(2) . Predictors of 1-year readmission were elevated natriuretic peptides, anaemia, left atrial size >26 mm/m(2) , heart rate >70 b.p.m., LV HF signs, and eGFR <60 mL/min/m(2) . The C-statistics for the models were 0.72 and 0.66, respectively. The cumulative incidence function distinguished low-risk (<1% event rate) and high-risk groups (>5% event rate) for 1-month HF readmission. Likewise, low-risk (7.8%), intermediate-risk (15.6%) and high-risk groups (26.1%) were identified for 1-year HF readmission risk. The C-statistics remained consistent after the external validation (<5% loss of discrimination).

CONCLUSION

The Redin-SCORE predicts early and late readmission for worsening of HF using proven prognostic variables that are routinely collected in outpatient management of chronic HF.

[Key messages for the initial management of the elderly patient with acute heart failure]

Acute heart failure is a high prevalence geriatric syndrome that has become one of the most frequent causes of visits to emergency departments, as well as hospital admission, and is associated with high morbidity, mortality and functional impairment. There has been an increasing amount of information published in recent years on the initial management of acute heart failure and the results of the short-term outcomes, as well as the natural history of the disease. The objective of this study is to provide several recommendations that should be taken into account in the initial management of the elderly patient with acute heart failure in the emergency departments, and to review the most interesting currently on-going clinical trials.

Multimarker proteomic profiling for the prediction of cardiovascular mortality in patients with chronic heart failure

Risk stratification of patients with systolic chronic heart failure (HF) is critical to better identify those who may benefit from invasive therapeutic strategies such as cardiac transplantation. Proteomics has been used to provide prognostic information in various diseases. Our aim was to investigate the potential value of plasma proteomic profiling for risk stratification in HF. A proteomic profiling using surface enhanced laser desorption ionization - time of flight - mass spectrometry was performed in a case/control discovery population of 198 patients with systolic HF (left ventricular ejection fraction <45%): 99 patients who died from cardiovascular cause within 3 years and 99 patients alive at 3 years. Proteomic scores predicting cardiovascular death were developed using 3 regression methods: support vector machine, sparse partial least square discriminant analysis, and lasso logistic regression. Forty two ion m/z peaks were differentially intense between cases and controls in the discovery population and were used to develop proteomic scores. In the validation population, score levels were higher in patients who subsequently died within 3 years. Similar areas under the curves (0.66 - 0.68) were observed for the 3 methods. After adjustment on confounders, proteomic scores remained significantly associated with cardiovascular mortality. Use of the proteomic scores allowed a significant improvement in discrimination of HF patients as determined by integrated discrimination improvement and net reclassification improvement indexes. In conclusion, proteomic analysis of plasma proteins may help to improve risk prediction in HF patients.

Analysis of BAG3 plasma concentrations in patients with acutely decompensated heart failure

BACKGROUND

BCL-2-associated athanogene 3 (BAG3) is a protein implicated in the cardiomyocyte stress response and genesis of cardiomyopathy. Extracellular BAG3 is measurable in patients with heart failure (HF), but the relationship of BAG3 with HF prognosis is unclear.

METHODS

BAG3 plasma concentrations were measured in 39 acutely decompensated HF patients; the primary endpoint was death at 1 year. Baseline characteristics were compared by vital status and median BAG3 concentration. Correlation of BAG3 with left ventricular ejection fraction (LVEF) and other biomarkers was performed. Prognostic value was assessed using Cox proportional hazards regression and Kaplan-Meier analysis.

RESULTS

At baseline, median BAG3 was significantly higher in decedents (N=11) than survivors (N=28; 1489 ng/mL versus 50 ng/mL; P=0.04); decedents also had worse renal function and higher median natriuretic peptide (NP) and sST2. BAG3 was not significantly correlated with NPs, mid-regional pro-adrenomedullin, sST2, or eGFR, however. Mortality was increased in patients with supra-median BAG3 (>336 ng/mL; 42.1% versus 15.0%, P=0.06). In age and LVEF-adjusted Cox proportional hazards, BAG3 remained a significant mortality predictor (HR=3.20; 95% CI=1.34-7.65; P=0.02); those with supra-median BAG3 had significantly shorter time-to-death (P=0.04).

CONCLUSION

The stress response protein BAG3 is measurable in patients with ADHF and may be prognostic for death.

Predicting short-term mortality in advanced decompensated heart failure - role of the updated acute decompensated heart failure/N-terminal pro-B-type natriuretic Peptide risk score

BACKGROUND

The first few months after admission are the most vulnerable period in patients with acute decompensated heart failure (ADHF).

METHODS AND RESULTS

We assessed the association of the updated ADHF/N-terminal pro-B-type natriuretic peptide (NT-proBNP) risk score with 90-day and in-hospital mortality in 701 patients admitted with advanced ADHF, defined as severe symptoms of worsening HF, severely depressed left ventricular ejection fraction, and the need for i.v. diuretic and/or inotropic drugs. A total of 15.7% of the patients died within 90 days of admission and 5.2% underwent ventricular assist device (VAD) implantation or urgent heart transplantation (UHT). The C-statistic of the ADHF/NT-proBNP risk score for 90-day mortality was 0.810 (95% CI: 0.769-0.852). Predicted and observed mortality rates were in close agreement. When the composite outcome of death/VAD/UHT at 90 days was considered, the C-statistic decreased to 0.741. During hospitalization, 7.6% of the patients died. The C-statistic for in-hospital mortality was 0.815 (95% CI: 0.761-0.868) and Hosmer-Lemeshow χ(2)=3.71 (P=0.716). The updated ADHF/NT-proBNP risk score outperformed the Acute Decompensated Heart Failure National Registry, the Organized Program to Initiate Lifesaving Treatment in Patients Hospitalized for Heart Failure, and the American Heart Association Get with the Guidelines Program predictive models.

CONCLUSIONS

Updated ADHF/NT-proBNP risk score is a valuable tool for predicting short-term mortality in severe ADHF, outperforming existing inpatient predictive models.

Procalcitonin and long-term prognosis after an admission for acute heart failure

BACKGROUND

Traditionally, procalcitonin (PCT) is considered a diagnostic marker of bacterial infections. However, slightly elevated levels of PCT have also been found in patients with heart failure. In this context, it has been suggested that PCT may serve as a proxy for underrecognized infection, endotoxemia, or heightened proinflammatory activity. Nevertheless, the clinical utility of PCT in this setting is scarce. We aimed to evaluate the association between PCT and the risk of long-term outcomes.

METHODS AND RESULTS

We measured at admission PCT of 261 consecutive patients admitted for acute heart failure (AHF) after excluding active infection. Cox and negative binomial regression methods were used to evaluate the association between PCT and the risk of death and recurrent rehospitalizations, respectively. At a median follow-up of 2years (IQR: 1.0-2.8), 108 deaths, 170 all-cause rehospitalizations and 96 AHF-rehospitalizations were registered. In an adjusted analysis, including well-established risk factors such as natriuretic peptides and indices of renal function, the logarithm of PCT was associated with a higher risk of death (HR=1.43, CI 95%: 1.12-1.82; p=0.004), all-cause rehospitalizations (IRR=1.22, CI 95% 1.02-1.44; p=0.025) and AHF-rehospitalizations (IRR=1.28, CI 95%: 1.02-1.61; p=0.032). The association with these endpoints persisted after adjustment for other inflammatory biomarkers such as white blood cells, C-reactive protein and interleukins.

CONCLUSION

In patients with AHF and no evidence of infection, PCT was independently and positively associated with the risk of long-term death and recurrent rehospitalizations.