Abstract
Background
Predicting survival in patients with advanced heart failure (HF) remains difficult and prognostic scores such as the Seattle Heart Failure Model (SHFM) are cumbersome to use. Alternative approaches like the Surprise Question (SC) or the number of HF hospitalisations within the last year (NoH) could simplify prognostication.
Purpose
We assessed the prognostic utility of the SHFM, SC and NoH for predicting one-year survival status in patients with advanced HF.
Methods
A secondary analysis of a multisite, single-blinded cluster-randomized, controlled trial to test whether a structured intervention of educational content and automated reminders increased the likelihood of ICD deactivation conversations and ICD deactivation. The study was performed within the advanced HF practices at six US academic medical centers, between September 2011 to February 2016. Patient eligibility criteria included advanced HF, an implantable cardiac defibrillator and a high risk of death, with complete data on SHFM, SC, NoH and one-year survival status. SHFM survival was calculated from baseline variables; the SC (“Would you be surprised if the patient were to die within one year?”) was answered by cardiologists; and the NoH was extracted from medical records. For prediction of survival status, cut-offs for predicted survival per SHFM and NoH were chosen empirically by means of receiver operating characteristic (ROC) curve analysis maximising Youden's index. The resulting binary prediction models were assessed based on area under the ROC curve (AUC), sensitivity and specificity.
Results
Of the 535 subjects in our sample, 82 (15.3%) had died after one-year of follow-up. For the SHFM and the NoH, optimal cut-offs were found to be a predicted survival <86% and ≥2 hospitalisations, respectively.
Performance metrics of prognostic models are detailed in Table 1. The SHFM yielded an AUC of 0.65 (0.60–0.71 95% confidence interval [CI]), a sensitivity of 0.76 (0.65–0.84 95% CI), and a specificity of 0.55 (0.50–0.60 95% CI). The SC demonstrated a comparable AUC 0.58 (0.54–0.63 95% CI), similar sensitivity 0.84 (0.74–0.91 95% CI), but lower specificity 0.33 (0.28–0.37 95% CI) compared to the SHFM. The NoH demonstrated a comparable AUC 0.56 (0.50–0.62 95% CI), similar sensitivity 0.56 (0.45–0.67 95% CI), and similar specificity 0.56 (0.51–0.61 95% CI) compared to the SHFM. The combination of positive SC and NoH ≥2 showed significantly higher specificity compared to the SHFM (0.68 [0.64–0.73 95% CI]).
Conclusion
The SC and NoH are clinically feasible bedside alternatives to the more complex SHFM model, yet yield similar overall prognostic utility for one-year survival status among advanced HF patients.
FUNDunding Acknowledgement
Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Heart, Lung, and Blood Institute
An Optimal Model of Financial Distress Prediction: A Comparative Study between Neural Networks and Logistic Regression