Prognosis and Antiplatelet Strategies for Patients With PCI and High Bleeding Risk:A Study Protocol

Prognosis and Antiplatelet Strategies for Patients With PCI and High Bleeding Risk:A Study Protocol

Percutaneous coronary intervention (PCI) is an important treatment strategy for patients with coronary artery disease. Combined bleeding after PCI significantly increases the risk of death in patients. The search for prognostic predictors and optimal antiplatelet therapy for patients with high bleeding risk (HBR) after PCI has been a hot topic in cardiovascular research. There is no accepted prognostic model or recommended antiplatelet therapy for patients with PCI-HBR. In this project, based on retrospective data extraction and prospective database building, we used artificial intelligence (AI) to analyze the adverse prognostic predictors of PCI-HBR patients, observe the types of antiplatelet drugs and duration of dual antiplatelet therapy in PCI-HBR patients, and compare the safety and feasibility of different antiplatelet regimens and treatment courses. The safety and feasibility of different antiplatelet regimens and regimens were compared.

This trail is a single center investigator-initiated prospective registry. PPP-PCI aims to observe the characteristics and prognosis of the PCI-HBR population and to explore appropriate antiplatelet therapy regimens to provide a basis for intervention guidance for patients with PCI-HBR. This project will help to further improve the existing bleeding prediction models and improve the efficiency of treating PCI-HBR patients.

Patients' baseline information is based on the latest test before PCI procedure. Basic information include age, gender, systolic blood pressure, diastolic blood pressure, body mass index, smoking status, smoking volume, positive family history of cardiovascular disease, hyperlipidemia, hypertension, diabetes, stroke history, peripheral artery disease, etc. Real-time update features include DAPT sessions, MACE event records, symptom records, sign records, test results, diagnosis, medical advice, real-time sign monitoring equipment data, ECG abnormalities, etc. Tests include a full set of lipid levels (including triglycerides, cholesterol, HDL, LDL, etc.), biochemical parameters (including creatinine, glomerular filtration rate, uric acid, etc.), hemoglobin, glucose, glycated hemoglobin, homocysteine, and lipoprotein(a) level. Imaging and functional testing data include coronary angiography images, intervention-related parameters, and target vessel lesion characteristics. The patient data is correlated with the visit intensity. The imaging images are used for deep learning to build unstructured classification models. The non-imaging data are used for machine learning to build a structured classification model. Pre-processing of the data includes image normalization, correction and normalization of irregular values, detection and removal of outliers and anomalies, interpolation and rejection of null values, removal of multicollinearity, and data normalization.

For the imaging images, a deep learning model was constructed using convolutional neural network to dichotomize the coronary vascular lesions and functional conditions contained in the coronary angiography images. For the non-imaging image data, Embedded method was used as the top-level method, and logistic regression, random forest, and gradient boosting tree were used as the bottom-level algorithms, and the key factors affecting the occurrence of MACE in the PCI-HBR population were extracted by fusing the feature weights through integrated learning. Based on the extracted key factors, a binary machine learning discriminative model was established, and SVM, XGBoost, random forest, and artificial neural network were used to complete the evaluation of multiple models, and the best model was selected as the machine learning classification model.

The deep learning model and machine learning model structures are weighted and fused to output the final results. Then the data collected by the future model is passed back to the training dataset for incremental learning to correct the model.

This trial will provide new insights and evidence on optimal antiplatelet therapy for a high bleeding risk patient cohort which is frequently encountered in real-world practice.

Inclusion Criteria:

PCI patients >18 years of age and meeting 1 major criterion or 2 minor criteria of the ARC-HBR The ARC-HBR major criteria included:

• long-term use of oral anticoagulants;
• severe or end-stage chronic kidney disease [eGFR <30 ml/(min*1.73m2 )];
• hemoglobin <11 g/dl, spontaneous bleeding requiring hospitalization or transfusion within the past 6 months or at any time;
• chronic bleeding constitutional;
• cirrhosis with portal hypertension spontaneous bleeding requiring hospitalization or transfusion within the past 6 months or at any time;
• moderate to severe baseline thrombocytopenia (platelets <100×10^9/L); chronic bleeding constitutional;
• cirrhosis with portal hypertension;
• active malignancy within the past 12 months (excluding non-melanoma skin cancer);
• previous spontaneous brain hemorrhage (at any time);
• traumatic brain hemorrhage within the past 12 months;
• within the past 6 months moderate or severe ischemic stroke within the past 6 months;
• the presence of cerebral arteriovenous malformation;
• recent major surgery or major trauma within 30 days prior to PCI;
• and major non-delayable surgery during DAPT.

Secondary criteria included:

• age ≥75 years;
• moderate chronic kidney disease [30 ml/(min*1.73m2 ) ≤ eGFR ≤ 59 ml/(min*1.73m2 )];
• 11 g/dl ≤ hemoglobin < 13 g/dl in men and 11 g/dl ≤ hemoglobin < 12 g/dl in women; - spontaneous bleeding requiring hospitalization or blood transfusion in the past 6 months to 12 months;
• long-term use of oral NSAIDs or steroids
• Ischemic stroke of any duration not covered by the primary criteria.

Exclusion Criteria:

• Patients who were already bleeding at the time of baseline inclusion
• Patients who could not be followed up (including previously reserved phone changes, etc.) to obtain MACE events.