Differential progression of coronary atherosclerosis according to plaque composition: a cluster analysis of PARADIGM registry data

Yeonyee E. Yoon, Lohendran Baskaran, Benjamin C. Lee, Mohit Pandey, Benjamin Goebel, Sang‐Eun Lee, Ji Min Sung, Daniele Andreini, Mouaz H. Al‐Mallah, Matthew J. Budoff, Filippo Cademartiri, Kavitha M. Chinnaiyan, Jung Hyun Choi, Eun Ju Chun, Edoardo Conte, Ilan Gottlieb, Martin Hadamitzky, Yong‐Jin Kim, Byoung Kwon Lee, Jonathon Leipsic, Erica Maffei, Hugo Marques, Pedro de Araújo Gonçalves, Gianluca Pontone, Sanghoon Shin, Jagat Narula, Jeroen J. Bax, Fay Y. Lin, Leslee J. Shaw, Hyuk‐Jae ChangScientific 2021 · DOI

TL;DR

This study uses unsupervised machine learning (K-means clustering) to categorize patients with coronary atherosclerosis based on their plaque composition. The findings reveal four distinct patient clusters with significantly different plaque progression patterns and risks of adverse cardiac events, suggesting that personalized risk assessment and treatment strategies could be improved by considering plaque composition.

What Problem Does This Paper Address?

Coronary atherosclerosis, the buildup of plaque in the arteries supplying the heart, is a leading cause of cardiovascular disease. While we can measure the extent of plaque (e.g., stenosis), understanding its composition (e.g., necrotic core, fibrous tissue, calcification) is crucial for predicting its progression and the risk of future events like heart attacks. However, coronary atherosclerosis is often viewed as a continuum, making it difficult to define distinct patient phenotypes based on plaque composition. This lack of clear categorization hinders personalized risk assessment and the development of targeted preventive treatments. This paper aims to address this by using unsupervised clustering to identify distinct patient groups based on their baseline plaque composition and then examining how these groups differ in terms of plaque progression and clinical outcomes.

Methods

The researchers employed K-means clustering, a common unsupervised machine learning algorithm, to group patients. This algorithm partitions data points into a pre-defined number of clusters (in this case, four) based on their similarity. The input features for clustering were the percentage volumes of four plaque components: necrotic core, fibro-fatty plaque, fibrous plaque, and calcified plaque, all measured from baseline coronary computed tomography angiography (CCTA) scans. The number of clusters (four) was determined using established methods like the Calinski-Harabasz Index, Average Silhouette Width, and the elbow method. After clustering, the researchers compared clinical characteristics, baseline CCTA findings, and changes in plaque volume and composition over time (median follow-up of 3.3 years) across the identified clusters. Multivariable analyses were used to assess the risk of plaque progression and major adverse cardiac events (MACE) associated with each cluster.

Datasets

The study utilized data from the PARADIGM (Progression of AtheRosclerotic PlAque DetermIned by Computed TomoGraphic Angiography Imaging) registry. This was a prospective, multi-national observational registry that collected serial CCTA data from 2,252 patients. For this specific analysis, 947 patients with coronary atherosclerotic plaque and available baseline and follow-up CCTA data were included. The dataset is characterized by its serial nature, allowing for the observation of plaque changes over time.

Key Insights and Ideas

  • Plaque Composition Defines Distinct Phenotypes: The clustering revealed that patients with coronary atherosclerosis can be meaningfully categorized into four distinct groups based on their baseline plaque composition. This moves beyond a simple "more plaque is worse" paradigm to a more nuanced understanding of disease heterogeneity.
  • Vulnerable vs. Stable Plaque Progression: The clusters clearly differentiate between patients with more "vulnerable" plaque (higher necrotic core and fibro-fatty components) who showed regression of these components and progression primarily through fibrous plaque, versus those with more "stable" plaque (higher calcified components) who experienced significant calcified plaque progression.
  • Composition Dictates Progression Rate: The rate and type of plaque progression varied significantly across the clusters. Some clusters showed minimal progression or even regression of certain components, while others exhibited rapid and substantial plaque buildup, particularly in the calcified component.
  • Composition Impacts Clinical Risk: Crucially, the distinct plaque composition clusters were associated with significantly different risks of major adverse cardiac events (MACE). This highlights the clinical relevance of plaque phenotyping beyond just the degree of stenosis.
  • Statin Use Modulates Risk: The study suggests that the increased risk of plaque progression and MACE observed in certain clusters might be more pronounced in patients taking statins, hinting at a complex interplay between plaque composition, statin therapy, and outcomes.

Key Results

  • Four Clusters Identified: K-means clustering identified four distinct patient clusters with significantly different plaque compositions (p<0.001 for all components).
  • Cluster 1 (n=52): Predominantly fibro-fatty plaque with a significant necrotic core. Showed the least total plaque volume (PV) progression (+23.3 mm³) and regression of necrotic core and fibro-fatty PV.
  • Cluster 2 (n=219): Largely fibro-fatty and fibrous plaque. Showed fibro-fatty PV regression (-2.4 mm³).
  • Cluster 3 (n=376): Mostly fibrous and calcified plaque. Showed increasing calcified PV progression (+21.4 mm³).
  • Cluster 4 (n=300): Predominantly calcified plaque. Demonstrated the greatest total PV increase (+50.7 mm³), mainly due to calcified PV increase (+35.9 mm³).
  • Progression Differences: Total PV progression increased from Cluster 1 to Cluster 4 (23.3 mm³ to 50.7 mm³). Necrotic core and fibro-fatty PV regression was evident in Cluster 1 and weakened in other clusters. Calcified PV progression was lowest in Cluster 1 and increased significantly in Clusters 3 and 4.
  • Risk of Progression: Multivariable analysis showed a higher risk of plaque progression in Clusters 3 and 4 compared to Cluster 1.
  • MACE Risk: The incidence of MACE was significantly lower in Cluster 1 (6.1%) compared to Clusters 2 (23.0%), 3 (19.9%), and 4 (22.6%). Multivariable Cox regression confirmed a higher risk of MACE in Clusters 2, 3, and 4 compared to Cluster 1.

Limitations and Future Work

  • Study Population Bias: The PARADIGM registry primarily included patients with low-to-moderate risk who underwent serial CCTAs. This might mean the findings are more representative of earlier stages of coronary atherosclerosis and may not fully generalize to higher-risk populations or those with more rapid progression who might not attend follow-up scans.
  • Subjectivity of Cluster Number: While methods were used to determine the optimal number of clusters (four), the choice of cluster number in K-means can have some subjectivity. However, the authors note that the visualization and clinical intuition supported the chosen number.
  • External Validation: The study lacked an external validation dataset, which is a common limitation in registry-based studies due to the rarity of similar large datasets with serial quantitative CCTA data.
  • Future Work: The authors suggest that further research is needed to determine the utility of patient-specific plaque phenotyping in personalized risk assessment and preventive treatment. They also propose that randomized controlled trials integrating CCTA with targeted prevention strategies are necessary to evaluate the benefits of this approach for guiding statin use and other interventions. The findings also underscore the need for a more personalized assessment of cardiovascular risk and the deployment of preventive strategies based on patient-specific plaque phenotyping.

Summary updated

Abstract

Patient-specific phenotyping of coronary atherosclerosis would facilitate personalized risk assessment and preventive treatment. We explored whether unsupervised cluster analysis can categorize patients with coronary atherosclerosis according to their plaque composition, and determined how these differing plaque composition profiles impact plaque progression. Patients with coronary atherosclerotic plaque (n = 947; median age, 62 years; 59% male) were enrolled from a prospective multi-national registry of consecutive patients who underwent serial coronary computed tomography angiography (median inter-scan duration, 3.3 years). K-means clustering applied to the percent volume of each plaque component and identified 4 clusters of patients with distinct plaque composition. Cluster 1 (n = 52), which comprised mainly fibro-fatty plaque with a significant necrotic core (median, 55.7% and 16.0% of the total plaque volume, respectively), showed the least total plaque volume (PV) progression (+ 23.3 mm<sup>3</sup>), with necrotic core and fibro-fatty PV regression (- 5.7 mm<sup>3</sup> and - 5.6 mm<sup>3</sup>, respectively). Cluster 2 (n = 219), which contained largely fibro-fatty (39.2%) and fibrous plaque (46.8%), showed fibro-fatty PV regression (- 2.4 mm<sup>3</sup>). Cluster 3 (n = 376), which comprised mostly fibrous (62.7%) and calcified plaque (23.6%), showed increasingly prominent calcified PV progression (+ 21.4 mm<sup>3</sup>). Cluster 4 (n = 300), which comprised mostly calcified plaque (58.7%), demonstrated the greatest total PV increase (+ 50.7mm<sup>3</sup>), predominantly increasing in calcified PV (+ 35.9 mm<sup>3</sup>). Multivariable analysis showed higher risk for plaque progression in Clusters 3 and 4, and higher risk for adverse cardiac events in Clusters 2, 3, and 4 compared to that in Cluster 1. Unsupervised clustering algorithms may uniquely characterize patient phenotypes with varied atherosclerotic plaque profiles, yielding distinct patterns of progressive disease and outcome.
BibTeX
@inproceedings{yoon2021differential,
  author = {Yeonyee E. Yoon and Lohendran Baskaran and Benjamin C. Lee and Mohit Pandey and Benjamin Goebel and Sang‐Eun Lee and Ji Min Sung and Daniele Andreini and Mouaz H. Al‐Mallah and Matthew J. Budoff and Filippo Cademartiri and Kavitha M. Chinnaiyan and Jung Hyun Choi and Eun Ju Chun and Edoardo Conte and Ilan Gottlieb and Martin Hadamitzky and Yong‐Jin Kim and Byoung Kwon Lee and Jonathon Leipsic and Erica Maffei and Hugo Marques and Pedro de Araújo Gonçalves and Gianluca Pontone and Sanghoon Shin and Jagat Narula and Jeroen J. Bax and Fay Y. Lin and Leslee J. Shaw and Hyuk‐Jae Chang},
  title = {{Differential progression of coronary atherosclerosis according to plaque composition: a cluster analysis of PARADIGM registry data}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-021-96616-w},
}
Figure