[Skip to Content]
Access to paid content on this site is currently suspended due to excessive activity being detected from your IP address 34.204.52.4. Please contact the publisher to request reinstatement.
[Skip to Content Landing]
Figure.
Final Risk Score for Prediction of 30-Day Readmission or Death Among Heart Failure Patients
Final Risk Score for Prediction of 30-Day Readmission or Death Among Heart Failure Patients
Table.  
Prediction of 30-day Readmission or Death in Heart Failure
Prediction of 30-day Readmission or Death in Heart Failure
1.
Jencks  SF, Williams  MV, Coleman  EA.  Rehospitalizations among patients in the Medicare fee-for-service program.  N Engl J Med. 2009;360(14):1418-1428.PubMedGoogle ScholarCrossref
2.
Kociol  RD, Peterson  ED, Hammill  BG,  et al.  National survey of hospital strategies to reduce heart failure readmissions: findings from the Get With the Guidelines-Heart Failure registry.  Circ Heart Fail. 2012;5(6):680-687.PubMedGoogle ScholarCrossref
3.
Huynh  QL, Saito  M, Blizzard  CL,  et al; MARATHON Investigators.  Roles of nonclinical and clinical data in prediction of 30-day rehospitalization or death among heart failure patients.  J Card Fail. 2015;21(5):374-381.PubMedGoogle ScholarCrossref
4.
Pencina  MJ, D’Agostino  RB  Sr, D’Agostino  RB  Jr, Vasan  RS.  Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond.  Stat Med. 2008;27(2):157-172.PubMedGoogle ScholarCrossref
5.
Keenan  PS, Normand  SL, Lin  Z,  et al.  An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all-cause readmission rates among patients with heart failure.  Circ Cardiovasc Qual Outcomes. 2008;1(1):29-37.PubMedGoogle ScholarCrossref
6.
Saito  M, Negishi  K, Marwick  TH.  Meta-analysis of risks for short-term readmission in patients with heart failure.  Am J Cardiol. 2016;117(4):626-632.PubMedGoogle ScholarCrossref
Research Letter
June 2016

Predictive Score for 30-Day Readmission or Death in Heart Failure

Author Affiliations
  • 1Menzies Institute for Medical Research, University of Tasmania, Hobart, Australia
  • 2Baker IDI Heart and Diabetes Research Institute, Melbourne, Australia
 

Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.

JAMA Cardiol. 2016;1(3):362-364. doi:10.1001/jamacardio.2016.0220

Readmissions shortly after heart failure (HF) are common, expensive, and usually considered preventable.1 However, despite the use of several interventions, rates of readmission after HF remain stable.2 An effective risk score might permit the targeting of resource-intensive interventions (such as disease-management programs) specifically on high-risk patients. We sought to determine the combination of clinical and nonclinical factors that would have the best discriminatory power in predicting 30-day readmission or death in HF.

Methods
Study Design

We developed a score for likelihood of readmission or death from HF from a prospective Australia-wide study of 430 HF patients (median age 74 years), of which 275 patients (64%) were male, and validated it in a group of 161 HF patients (median age 78 years), of which 89 patients (55%) were male.

The primary outcome measure in the study was 30-day all-cause readmission or death. Data on readmission and death were collected from medical records. All patients provided written informed consent for participation in the study, which was approved by the Tasmanian Health and Medical Human Research Ethics Committee.

Potential Predictors

Clinical data included patient history, medications, physical measurements, blood tests, and findings on echocardiography. Nonclinical data included age, sex, language background, marital status, living alone or with others, education, socioeconomic status, remoteness index (differentiating residence in a metropolitan, rural, or remote area of Australia), medical insurance, and any home health care services provided. Questionnaires used for data collection included the Montreal Cognitive Assessment (MoCA), Patient Health Questionnaire (PHQ-9), and Generalized Anxiety Disorder (GAD-7).

Statistical Analyses

Logistic regression was used to determine the variables that served as the best predictors of readmission or death. Predictors were ranked by the change of deviance (G, the difference between null and residual deviance) that reflected the improvement in predictability provided by the univariable model as compared with the null model for each predictor.3 A predictor was included in the multivariable model if it contributed by 0.01 or more units to the area under the curve.4 Changes in standard errors when new variables were added were small (<10%), implying limited variance inflation in our models and no overfitting. The final model was internally validated through the use of 500 bootstrapped samples3 and externally validated by applying the intercept and regression coefficients to a cohort of 161 HF patients from Tasmania, Australia. Patients without any admissions for HF in the previous 6 months were recruited in the 2 largest public hospitals in Tasmania. Within 30 days of discharge, 44 of the 161 patients (27%) in the cohort either died or were readmitted. The claims-based prediction model developed by Keenan et al was applied to our study population by using the intercept and coefficients described in the original study.5

Results

The Table shows the patients’ characteristics that were typical of HF in Australia. Within 30 days of discharge, 38 of the 430 HF patients (9%) in the study cohort died and 92 of the 430 patients (21%) were readmitted. The univariable associations are shown in the Table, with predictors ranked by their predictability of the outcome. The final prediction model (C statistic = 0.82; 95% CI, 0.77-0.87) (Table) had very good discrimination when predicting 30-day death (C statistic = 0.83; 95% CI, 0.73-0.93) or readmission (C statistic = 0.80; 95% CI, 0.74-0.85). The Figure shows the association between score and outcome. The discriminatory power of the model was much higher than that of the claims-based model (C statistic = 0.56; 95% CI, 0.50-0.61).

The internal (C statistic = 0.82; 95% CI, 0.76-0.87) and external (C statistic = 0.80; 95% CI, 0.69-0.91) validation values demonstrated great stability and generalizability of our final prediction model. The model calibration across different risk categories showed a close association of predicted and observed outcomes.

Discussion

The short-term risks of death or readmission after HF remain very high. Effective targeting of disease management programs for HF is likely to reduce readmissions and save money. However, a systematic review of readmission risk scores showed that the strongest prediction models provided only poor discrimination (C statistic <0.6) in predicting readmissions among HF patients.6 This study optimized the predictive score of 30-day readmission or death by adding important determinants not included in previous models, including echocardiography, mental health, cognitive function, and individual socioeconomic status. The model developed in the study has excellent internal and external validation and calibration, and might be used to predict both short-term mortality and readmission for HF with very good discrimination. Further validation of the model in a larger sample of HF patients that can be generalized to other health systems is needed.

Back to top
Article Information

Correction: This article was corrected on May 18, 2016, to fix errors in the Table and text.

Corresponding Author: Thomas H. Marwick, MBBS, PhD, MPH, Baker IDI Heart and Diabetes Institute, 75 Commercial Rd, Melbourne, Vic 3004, Australia (tom.marwick@bakeridi.edu.au).

Published Online: April 20, 2016. doi:10.1001/jamacardio.2016.0220.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: Supported in part by a Partnership grant from the National Health and Medical Research Foundation (Canberra), Tasmania Medicare Local (Hobart), Department of Health and Human Services (Hobart), and National Heart Foundation of Australia (Canberra).

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

References
1.
Jencks  SF, Williams  MV, Coleman  EA.  Rehospitalizations among patients in the Medicare fee-for-service program.  N Engl J Med. 2009;360(14):1418-1428.PubMedGoogle ScholarCrossref
2.
Kociol  RD, Peterson  ED, Hammill  BG,  et al.  National survey of hospital strategies to reduce heart failure readmissions: findings from the Get With the Guidelines-Heart Failure registry.  Circ Heart Fail. 2012;5(6):680-687.PubMedGoogle ScholarCrossref
3.
Huynh  QL, Saito  M, Blizzard  CL,  et al; MARATHON Investigators.  Roles of nonclinical and clinical data in prediction of 30-day rehospitalization or death among heart failure patients.  J Card Fail. 2015;21(5):374-381.PubMedGoogle ScholarCrossref
4.
Pencina  MJ, D’Agostino  RB  Sr, D’Agostino  RB  Jr, Vasan  RS.  Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond.  Stat Med. 2008;27(2):157-172.PubMedGoogle ScholarCrossref
5.
Keenan  PS, Normand  SL, Lin  Z,  et al.  An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all-cause readmission rates among patients with heart failure.  Circ Cardiovasc Qual Outcomes. 2008;1(1):29-37.PubMedGoogle ScholarCrossref
6.
Saito  M, Negishi  K, Marwick  TH.  Meta-analysis of risks for short-term readmission in patients with heart failure.  Am J Cardiol. 2016;117(4):626-632.PubMedGoogle ScholarCrossref
×