Predicting Survival In Dilated Cardiomyopathy: Clinical Risk Models With Cardiac Magnetic Resonance

Objective:Dilated cardiomyopathy(DCM)is a common non-ischemic cardiomyopathy.Despite the therapeutic achievements,patients’ outcomes remain highly variable.The number of risk prediction models for heart failure is expanding,but DCM-specific tools are currently not available.We conducted this study based on the largest prospective DCM cohort with cardiac magnetic resonance(CMR)in Southwest China.The purpose of the study was threefold: 1)to provide a comprehensive description of the Chinese DCM population,and identify potential prognostic factors;2)to external validate existing heart failure prediction models and improve the models by CMR variables;3)to develop prediction models estimating the risk of mortality/heart transplantation for DCM patients.Methods:We prospectively enrolled patients with DCM who were referred to West China Hospital,Sichuan University,and underwent CMR between June 2012 and May 2020.The eligible patients were followed up consecutively until December 2020.The endpoint was a composite of all-cause mortality or heart transplantation.Baseline characteristics were compared between the patients who reached the endpoint(event group)and those without end-point events(event-free group).Cox-regression modelbased restricted cubic splines were used to investigate the association between significant continuous variables and all-cause mortality/transplantation.We then performed external validation for four well-established heart failure prediction models,including the BIOlogy Study to TAilored Treatment in Chronic Heart Failure(BIOSTAT-CHF)model,Gruppo Italiano per lo Studio della Streptochinasi nell’Infarto Miocardico-Heart Failure(GISSI-HF)model,MetaAnalysis Global Group in Chronic Heart Failure(MAGGIC)score and Seattle Heart Failure Model(SHFM).Discrimination and calibration(if baseline survival rate were accessible)were determined.Harrell’s C-statistics were calculated to evaluate the overall capacity to categorize patients by relative risk(discrimination);time-dependent C-statistics were calculated separately for 6-,12-,24-and 36-month discrimination.Brier’s scores were calculated to evaluate the accuracy of absolute survival prediction.Extended models with CMR variables were created and compared to the original models.Finally,we used the least absolute shrinkage and selection operator method to select key variables predictive of the endpoint.For extended use in clinic settings without CMR,we developed clinical-CMR and clinical-only prediction models.The models were constructed with different combinations of demographic,comorbidities,biochemical,therapeutic medications,and CMR variables.Results:Eight hundred and forty-eight patients with DCM aged 47.8±14.7 years,69.1% male,were prospectively enrolled.The mean left ventricular ejection fraction was25.9±12.2%.Angiotensin receptor neprilysin inhibitor,ACE-inhibitors/angiotensin receptor blockers,and β-blockers were administered in 16.4%,61.8%,and 81.6% of the population,respectively.Over 25.7(ranged 0.2 to 90.7)months of follow-up,139(16.4%)patients reached the composite endpoint,including 75 heart failure death,40 sudden deaths,9 non-cardiac deaths,and 15 heart transplants.Patients in event group(N=139)were older [50.5±16.5 vs.47.3±14.3,P=0.020],with long-lasting symptoms[24(Interquartile range,IQR: 2,60)vs.6(IQR: 2,24),P<0.001],advanced NYHA class [NYHA Ⅲ/Ⅳ: 79.1% vs.48.0%,P<0.001],and more prevalent atrial fibrillation[35.3% vs.16.7%,P<0.001].Gender and co-morbidities were comparable between the groups(P>0.05).After adjustment,age and the duration of symptoms had a U-shaped and J-shaped association with the composite event,respectively.Patients aged 45 harbored the lowest risk of the observed event.On external validation,all the models showed poor-to-moderate discrimination,with the best performance from MAGGIC [Harrell’s C-statistic: 0.72,95% CI: 0.68-0.77],followed by SHFM [0.71,95% CI: 0.66-0.76],GISSI-HF [0.67,95% CI: 0.62-0.72]and BIOST-CHF [0.60,95% CI: 0.55-0.66].MAGGIC was well calibrated for 12-month prediction of mortality/heart transplantation(Brier score: 0.06)and poor for 36-month(Brier score 0.16).Calibration plot demonstrated an underestimation of the absolute risks by MAGGIC.SHFM showed an overall poor calibration and risk overestimation;the brier score for 6-month,12-month,24-month and 36-month prediction were 0.06,0.10,0.16 and 0.20,respectively.Left ventricular end-diastolic volume index(LVEDVi)and late gadolinium enhancement(LGE)pattern were selected to create CMR-extended models.Adding CMR variables significantly improved the discrimination of all the models [Harrell’s C-statistic: BIOSSTAT-CHFCMR 0.65,95% CI: 0.61,0.70;GISSI-HF-CMR 0.73,95% CI: 0.68,0.78;SHFMCMR 0.74,95% CI: 0.69,0.78;MAGGIC-CMR 0.76,95% CI: 0.71,0.80].However,calibration remained suboptimal.We developed the clinical-CMR and the clinical-only prediction models among 848 DCM patients using significant independent predictors for all-cause mortality/transplantation.Eleven variables were included in the final clinical-CMR model: age,duration of symptoms,systolic blood pressure,atrial flutter/fibrillation,hemoglobin,N-terminal pro-brain natriuretic peptide,lactate dehydrogenase,β-blockers and inotropic prescription at baseline,LVEDVi,and LGE patterns.Nine variables were included in the final clinical-only model: age,duration of symptoms,NYHA class,systolic blood pressure,hemoglobin,N-terminal pro-brain natriuretic peptide,lactate dehydrogenase,β-blockers and inotropic prescription at baseline.Global discrimination of both models was excellent,with better performance from the clinical-CMR model [Harrell’s C index: 0.83(95% CI: 0.80-0.87)vs.0.80(95% CI:0.77-0.84),P=0.025].Discrimination remained excellent for 6-,12-and 24-month prediciton(time-dependent C index: 0.81-0.88),followed by slight decrease for 36month;the clinical-CMR model outperformed for 24-and 36-month prediction [24-month: 0.85(95% CI: 0.81-0.90)vs.0.81(95% CI: 0.76-0.87),P=0.049;36-month:0.82(95% CI: 0.80-0.88)vs.0.79(95% CI: 0.74-0.85),P=0.016].Calibration for each time point was comparable between the models;the brier score of 6-,12-,24-,and 36-month were 3%,4%-5%,8%,and 11%-12%,respectively.Conclusion:We reported the largest DCM cohort with CMR in Southwest China.Existing heart failure mortality models showed varied predictive power in DCM patients.MAGGIC and SHFM demonstrated superior yet modest discrimination and poor calibration.Adding CMR predictors substantially improved the discrimination of all the models,but the calibration of MAGGIC and SHFM remained suboptimal.We developed two DCM-specific models to predict mortality/heart transplantation: the clinical-CMR model and the clinical-only model.Both models provided comparably accurate estimates of survival using easily obtained variables,whereas CMR predictors favored the prediction of long-term survival.The models may serve as powerful tools in different clinic settings.However,external validation is warranted.