Construction And Validation Of Two Independent Screening Models For Hypertrophic Cardiomyopathy Based On Electrocardiogram Features And Plasma Amino Acid Metabolomics

Project 1:Derivation and validation of the screening model for hypertrophic cardiomyopathy based on electrocardiogram features1.1 Background Hypertrophic cardiomyopathy（HCM）is a clinically heterogeneous genetic heart disease characterized by asymmetric cardiac hypertrophy,which was not solely explained by pressure afterload.It is widely distributed worldwide with a prevalence rate of1:200-500.HCM is an important cause of sudden cardiac death（SCD）especially in the youth,progression of heart failure（HF）,and thromboembolism attributable to atrial fibrillation（AF）.With the advancement of medical technology and the improvement of treatment approaches,the mortality rate of HCM has significantly decreased.Those HCM patients received standardized management nearly have a satisfactory expected survival similar to that of normal population.However,due to the hidden clinical symptoms,and the high dependence on advanced equipment and specific expertise for clinical diagnosis,the clinical diagnosis rate is seriously insufficient,approximately 80-90%HCM patients were unidentifiable.Therefore,it is of great significance to improve the detection rate of HCM in the general population.Furthermore,HCM is divided into obstructive HCM（HOCM）and nonobstructive HCM（HNCM）based on left ventricular outflow tract gradient（LVOTG）,which is vulnerable by many factors.And the clinical presentations,the treatments and prognosis between HOCM and HNCM are quite different.So,the discrimination of HOCM and HNCM is necessary and clinically meaningful.ECG is economical and accessible,widely used in medical care.More than 90%HCM patients show abnormal ECG presentations,which are diverse and distinct.Although with easy operation,it depends on professional ECG interpreters to make judgments and explanations,which limits its popularization in the primary hospital and community clinic.Nowadays,there is a lack of a pragmatic model that can easily and conveniently screen HCM based on diversely differential ECG variables,in order to improve the clinical detection rate of HCM in the general population,prevent adverse cardiac events,and improve long-term prognosis.1.2 Aims We aimed to develop and validate simple and convenient pre-diagnosis models based on ECG features for widespread screening of HCM,and initial assessing whether HOCM for those suspicious HCM.Population-based screening followed with timely referral and prompt identification of true cases is of utmost importance.1.3 Methods Between April 1^st and September 30^th,2020,423 consecutive participants（172HCM patients,including 70 HOCM and 102 HNCM;251 participants without left ventricular hypertrophy[Non-HCM]）from the International Cooperation Center for Hypertrophic Cardiomyopathy of Xijing Hospital were prospectively included in the training cohort.Between January 4^th and February 28^th,2021,163 participants（62 HCM patients,including 30 HOCM and 32 HNCM;101 Non-HCM participants）from the same center were included in the temporal validation cohort.External validation was performed using retrospectively collected ECG data from Xijing Hospital(3232 HCM ECG samples from January 1^st,2000,to March 31^st,2020;95184 Non-HCM ECG samples from January 1^st to December 31^st,2020).Several independent statistical algorithms including the least absolute shrinkage and selection operator（LASSO）,backward stepwise logistic regression（LR）,LASSO followed by backward stepwise LR,and LASSO followed by best subset,were used to screen candidate features from numerously differential variables.LR was used to construct the screening model for discriminating HCM from Non-HCM.The AUC was used to evaluate the discriminative ability of the model.Meanwhile,the ECG variables with statistical differences between HNCM and HOCM were included into a multivariate backward stepwise LR to construct a model to distinguish HOCM from HNCM.390 HOCM and 499 HNCM ECG samples with confirmed diagnosis were used as the external validation dataset to validate the HOCM model.The area under the receiver operator characteristic（ROC）curve（AUC）was also used to measure the discriminative performance of the model,and AUC>0.75 was considered to have clearly useful discriminative performance.The browser-based calculators of the two independent models were generated accordingly for clinical utilization.1.4 Results Among 30 ECG features examined,all except abnormal Q wave significantly differed between the HCM patients and Non-HCM comparators.After feature selection and model evaluation,we included only two common ECG features,T wave inversion（TWI）and the amplitude of S wave in lead V1（SV1）,in the HCM prediction model.The model showed a clearly useful discriminative ability in the training（AUC 0.857[0.818-0.896]）and temporal validation cohorts（AUC 0.871[0.812-0.930]）.In the external validation cohort,the AUC of the model was 0.833[0.825-0.841],all AUC>0.75.Further differential analysis between HOCM and HNCM subgroups showed that there were a total of 10 variables,including P wave interval,the amplitude of R wave in lead I and a VL（RI,Ra VL）,the amplitude of S wave in lead V1,V2,V5and V6（SV1,SV2,SV5 and SV6）,the sum amplitude of R wave in lead V5 and S wave in lead V1（RV5SV1）,the sum amplitude of R wave in lead I and S wave in lead III（RISIII）,and the sum amplitude of S wave in lead V1 and V2（SV1V2）,existed statistical differences,all P<0.05.Then,P wave interval and SV1 were further selected by backward stepwise LR among the 10variables with statistical differences to construct HOCM screening model,also with a clearly useful discriminative performance for differentiating HOCM from HNCM with an AUC of 0.786[0.718-0.854]in the training cohort,0.805[0.697-0.914]in the temporal validation cohort,and0.776[0.746-0.806])in the external validation cohort.（All AUC>0.75）1.5 Conclusion The pragmatic model established using TWI and SV1 may be clearly useful for assessing the probability of HCM and shows promise for use in population-based HCM screening.Furthermore,the HOCM model constructed by P wave interval and SV1 also had useful discriminative ability for HOCM and HNCM.Project 2:Derivation and validation of the screening model for hypertrophic cardiomyopathy based on plasma amino acid metabolomics2.1 Background HCM is characterized as glucose,lipids and amino acids（AAs）dysmetabolism.A recent work published in Circulation performed comprehensive multi-omics analysis of myocardium in HCM patients and revealed that the metabolic changes of AAs differed evidently between HCM and NC.The genes involving in AAs metabolism were also robustly changed.Which suggested that AAs dysmetabolism is one of the features in HCM.Circulating AAs are stable in vitro,not easily be degraded.In routine clinical practice,the detection of AAs is easy-operation,efficient and low-cost by AA chromatographic analyzer or liquid chromatography.So,circulating AAs metabolomics have the potential to be utilized in population-based screening and diagnosis.Nowadays,it is not clear whether the plasma AAs metabolism profile in HCM patients also changes correspondingly,and whether such changed AAs metabolites can be used for screening HCM.2.2 Aims This study aims to delineate plasma AAs profiles,and then construct and validate potential pre-diagnosis models based on AAs metabolomics for screening HCM and its obstructive subtype.2.3 Methods Between December 1^st 2019 and September 30^th 2020,fasting plasma samples were consecutively collected from 166 subjects,including 102 HCM patients（57 HOCM and 52HNCM）and 57 normal controls（NCs）,who first visited the International Cooperation Center for Hypertrophic Cardiomyopathy,Xijing Hospital,and then analyzed by high-performance liquid-chromatography mass spectrometry（HPLC–MS）based on targeted AAs metabolomics.Candidate AAs were screened based on a variety of feature selection algorithms.5-fold cross-validation by three separate algorithms,including logistic regression（LR）,random forest（RF）,and support vector machine（SVM）were performed to assess the models constructed by the potentially optimal screening panel.Two independent models were constructed by LR to distinguish HCM from NC,and HOCM from HNCM,respectively.The dataset is randomly divided into the training and validation dataset according to the ratio of 7:3,and ROC curve is performed to evaluate the discriminative ability of the model,and AUC>0.75 was considered to have clearly useful discriminative performance.2.4 Results The univariate analysis showed that a total of 10 AAs and derivatives（serine,glycine,proline,citrulline,glutamine,cystine,creatinine,cysteine,choline and aminoadipic acid）in HCM were significantly different from those in NC（all P<0.05）.Four AAs and derivatives（proline,glycine,cysteine,and choline）were screened by multiple feature selection algorithms for discriminating HCM from NC.The discriminative ability of the model evaluated by LR,SVM and LR is consistent.The ROC of the model constructed by LR yielded an AUC of 0.83[0.75-0.91]in the training dataset and 0.79[0.65-0.94]in the validation dataset.In HCM,plasma ornithine level was negatively correlated with the maximum wall thickness of left ventricular,while arginine,phenylalanine,tyrosine,proline,alanine,asparagine,creatine,tryptophan,ornithine and choline were correlated with the LVOTG（all P<0.05）.Moreover,among ten AAs and derivatives（arginine,phenylalanine,tyrosine,proline,alanine,asparagine,creatine,tryptophan,ornithine and choline）with statistical significance between HOCM and HNCM,three AAs（arginine,proline,and ornithine）were picked out to differentiate the two subgroups.The AUC in the training and validation datasets were 0.83[0.74-0.93]and 0.82[0.66-0.98],respectively.（All AUC>0.75）2.5 Conclusion The plasma AAs and derivatives were distinct between HCM and NC groups.Based on the differential plasma AAs profiles,the two established screening models have potential value in assisting HCM screening and identifying whether it is obstructive.