Optical Coherence Tomography Predicts The Prognosis Of Coronary Percutaneous Coronary Intervention

Background:Coronary angiography(CAG)guided percutaneous coronary intervention(PCI)is the main treatment for patients with coronary artery disease(CAD).CAG uses 2D images to display the 3D structure of blood vessels,but has limitations in evaluating the degree of stenosis,plaque composition,lesion length,and ischemic status.The correspondence between the stenosis judged by CAG and myocardial ischemia often shows inconsistency.Fractional flow reserve(FFR)and optical coherence tomography(OCT)can provide more accurate vascular functional information and imaging characteristics,respectively,which have become two major ways to compensate for the limitations of CAG.However,it is currently impossible to use both FFR and OCT simultaneously in routine clinical practice,considering surgical efficiency,intraoperative safety,and medical expenditure.Optical flow ratio(OFR)is an Artificial intelligence(AI)-based OCT image analysis which has good consistency with FFR,and is expected to replace FFR as a new indicator for guiding PCI treatment.However,the clinical application of OFR is hindered by the lack of evidence-based medicine.Therefore,there is an urgent need to conduct large-sample,multicenter,multi-ethnic clinical studies to evaluate the clinical application value of OFR.Furthermore,because most clinical medical centers’ OCT systems are not equipped with OFR modules,OCT imaging features cannot be converted into OFR values,which limits the clinical application of OCT data.Therefore,constructing an evaluation model for OFR using OCT imaging features would be an effective alternative solution for the application of OCT data without an OFR module.Objective(1)To evaluate the predictive value of OFR and OCT imaging features for clinical prognosis in CAD patients undergoing PCI treatment.(2)Construct an OFR evaluation model based on OCT imaging features and compare its predictive efficacy for PCI prognosis with the OFR values provided by the ODR functional module,in order to provide an alternative approach for OCT imaging analysis system personnel without the equipped OFR functional module.(3)Incorporate OFR,clinical characteristics,and OCT imaging features into a model using machine learning algorithms(ML),and compare the differences between the model and OFR in predicting PCI clinical prognosis,in order to further improve the accuracy of clinical prognosis prediction for PCI-treated patients.MethodsFrom June 2019 to May 2021,a total of 871 cases underwent CAG and OCT examinations at the Second Affiliated Hospital of Guangzhou Medical University and Guangdong Provincial People’s Hospital,among which 437 cases underwent PCI treatment.The 437 patients who underwent PCI treatment were followed up for more than 12 months.The impact of OFR on major adverse cardiovascular events(MACE)was evaluated using binary univariate and multivariate COX regression methods.The Youden index was calculated using receiver operating characteristic curve(ROC)and the critical value(CV)of OFR was determined.The 437 patients who underwent PCI treatment were divided into high OFR(> Cv)and low OFR(≤ Cv)subgroups based on the CV of OFR.A logistic regression algorithm was used to construct an OFR evaluation model that incorporated different OCT imaging feature variables,and the predictive efficacy of different models for OFR values in predicting clinical prognosis of CAD patients undergoing PCI was evaluated.The random survival forests(RSF)algorithm was used to incorporate clinical characteristics,OCT imaging features,and OFR to calculate patient RSF risk scores(RS),and to compare the differences between RSF-RS and OFR in predicting PCI clinical prognosis.The predictive performance of all models was evaluated using the area under the ROC curve(AUC).Results(1)The results of binary multivariate COX regression analysis showed that OFR was an independent predictive factor for MACE after PCI(HR: 0.272,95%CI: 0.105-0.459,P < 0.001);ROC analysis showed that OFR had good predictive efficacy for PCI postoperative MACE(AUC: 0.771,95%CI:0.684-0.857),and OFR=0.883 was the optimal CV.(2)Based on whether OFR was greater than 0.88,437 CAD patients treated with PCI were divided into two subgroups: high OFR(H-OFR)and low OFR(L-OFR),with 327 cases in the H-OFR group.COX regression analysis showed that the total lipid plaque volume(HR: 1.020,95%CI:1.002-1.038,P=0.027)and maximum lipid plaque volume(HR: 1.028,95%CI: 1.000-1.056,P=0.047)were independent predictive factors for MACE in the H-OFR subgroup.ROC analysis showed that the accuracy of predicting MACE events in the H-OFR subgroup by the total lipid plaque volume and maximum lipid plaque volume was 0.778(95%CI: 0.641-0.915)and 0.791(95%CI: 0.665-0.918),respectively.(3)Most OCT systems are not equipped with OFR calculation modules,making it impossible to evaluate PCI prognosis through OFR.We analyzed OCT imaging features affecting OFR values,among which target vessel lesion,minimum lumen area,distal reference vessel diameter,lesion length,and plaque burden were independent influencing factors for OFR.We included these 5 variables to construct a logistic regression model(OFR-LGM)to evaluate OFR values.The consistency between the predicted OFR categories(high or low)by OFR-LGM and the OFR values given by the machine module reached 94%(AUC: 0.940,95%CI:0.924-0.956),indicating that the OFR-LGM constructed in this study can be used for OCT system OFR evaluation.The accuracy of predicting PCI postoperative MACE risk using the OFR categories given by OFR-LGM was 0.713(AUC: 0.713,95%CI: 0.631-0.782).(4)In order to construct the optimal OFR evaluation model,we compared the logistic regression machine learning model incorporating all OCT imaging features(OFR-LG-ML)with the Lasso machine learning model after feature selection(OFR-Las-ML).ROC analysis and Delong test were used to evaluate the predictive efficacy of OFR-LG-ML and OFR-Las-ML for OFR in predicting MACE after PCI.The results showed that the accuracy of prediction by OFR-LG-ML was 0.750(95%CI:0.670-0.809),and that of OFR-Las-ML was 0.735(95%CI: 0.655-0.799),slightly better than the OFR-LGM model incorporating 5 OCT feature variables.There was no significant difference in the predictive accuracy between OFR-LG-ML and OFR-Las-ML.Considering that OFR-Las-ML only included 9 OCT feature variables and the model was simpler,OFR-Las-ML was the optimal model for OFR evaluation.(5)Considering that clinical characteristics and OCT imaging features may affect PCI clinical prognosis,we calculated the RSF-RS of patients based on OFR and the above two types of feature variables.ROC analysis and Delong test were used to compare the predictive efficacy of RSF-RS and OFR for PCI postoperative MACE.The results showed that RSF-RS was significantly better than OFR.Conclusion(1)OFR and OCT imaging features have good predictive value for clinical prognosis in CAD patients undergoing PCI treatment;(2)A machine learning model based on OCT imaging features for evaluating OFR has good predictive value for PCI prognosis.Its predictive accuracy is slightly lower than that provided by the specialized OFR function module,which can provide accurate OFR assessment for OCT systems without OFR calculation capabilities.The OFR-Las-ML that we constructed is an optimal model for OFR evaluation.(3)RSF-RS has better predictive efficacy for MACE than OFR and may become a substitute indicator for OFR.