Mitral Annular Disjunction: Simple Anomaly or Barlow Signature?

Study background and objectives

Mitral annular disjunction (MAD) is increasingly recognized in the context of mitral valve prolapse (MVP), but its precise anatomical integration within different degenerative forms remains to be defined. This study analyzes a surgical cohort of 979 patients presenting with significant mitral regurgitation, classified according to etiology: Barlow's disease (BD) or fibroelastic deficiency (FED).

The main objective is to evaluate whether explainable supervised learning and unsupervised clustering (UMAP) allow for the identification of MVP phenotypes specifically associated with MAD. Using models such as Random Forest and XGBoost, researchers sought to characterize MAD beyond a simple binary observation.

The authors test the hypothesis that MAD is not an isolated anomaly, but part of a global remodeling of the annulo-valvular apparatus. The study postulates that precise anatomoclinical variables, such as the number of prolapsed scallops or the annular diameter, allow for the definition of a distinct phenotype in which MAD is predominant, particularly within Barlow's disease.

Methodology of computational analysis

This study is based on the retrospective analysis of a public surgical cohort of 979 patients presenting with mitral valve prolapse (MVP) associated with clinically significant mitral regurgitation. The etiology was classified into two categories: Barlow's disease (BD) and fibroelastic deficiency (FED).

The analytical protocol combined two distinct machine learning approaches:

• Supervised learning: Three models (logistic regression, Random Forest and XGBoost) were evaluated to predict the presence of mitral annular disjunction (MAD). The methodology included nested cross-validation with threshold optimization. The interpretability of the XGBoost model was analyzed via SHAP values.
• Unsupervised phenotyping: UMAP algorithm-based clustering was performed using anatomical and clinical variables, specifically excluding MAD and etiology from the input data to avoid classification bias.

Model performance was measured by AUROC (XGBoost: 0.695 ± 0.057), precision-recall (PR-AUC), Brier score (0.13–0.20), and a decision curve analysis to evaluate clinical net benefit. Sensitivity analyses validated the stability of the identified clusters.

Results of phenotypic and predictive analysis

Unsupervised clustering analysis (based on the UMAP algorithm) identified three distinct anatomical phenotypes within the cohort of 979 patients. The smallest cluster is distinguished by a specific morphological signature closely linked to mitral annular disjunction (MAD).

Characterization of the high MAD prevalence cluster

The smallest identified cluster (n = 91, or 9.3% of the cohort) presents the highest prevalence of MAD. The characteristics of this group reveal extensive annular and valvular remodeling:

Performance of supervised machine learning models

The tested models (Logistic Regression, Random Forest, XGBoost) showed moderate discrimination capacity for predicting the presence of MAD, suggesting that these tools are more explanatory than diagnostic in this clinical context:

• Discrimination (XGBoost): AUROC of 0.695 ± 0.057 and PR-AUC of 0.314 ± 0.066.
• Calibration: Brier score between 0.13 and 0.20, with calibration slopes below 1, indicating modest predictive performance.
• Net benefit: Decision curve analysis shows a limited clinical benefit for the binary identification of MAD alone.

Predictive factors and explainability (SHAP)

Explainability analysis via SHAP values identified the variables contributing most to the presence of MAD. Barlow's disease (BD) status emerged as the most powerful predictor, followed by:

• The number of prolapsed scallops ;
• Mitral ring diameters;
• Left ventricular (LV) volumes;
• The bileaflet nature of the prolapse.

These results confirm that MAD is not an isolated anomaly but is part of a global remodeling of the mitral apparatus, particularly marked in multisegmental forms of Barlow's disease.

Clinical significance of the identified phenotypes

This analysis of 979 patients demonstrates that mitral annular disjunction (MAD) is not an isolated anatomical anomaly, but is part of a continuum of complex annular and valvular remodeling. The unsupervised clustering approach identified a specific phenotype, although a minority (9.3% of the cohort), presenting a MAD prevalence of 36.3%, compared to approximately 12 to 16% in the other groups. This cluster is distinguished by an almost total predominance of Barlow's disease (95.6%), systematic bileaflet prolapse involving at least 4 scallops, and significantly wider mitral annular diameters, despite a lower frequency of chordal ruptures.

Model limits and performance

The study highlights the limitations of current algorithmic prediction: supervised models, such as XGBoost, achieve only moderate discriminative performance (AUROC 0.695). The authors specify that these tools should be interpreted as explanatory rather than diagnostic models. The modest calibration and limited net benefit mandate clinical caution. Furthermore, these results from a surgical cohort require prospective external validation incorporating long-term clinical outcome data.

Implications for practice

SHAP data confirm that Barlow disease status is the most powerful contributor to the presence of MAD, followed by the number of prolapsed scallops and left ventricular volumes. These findings suggest that MAD is structurally intertwined with the most extensive forms of Barlow disease. For the surgeon, the discovery of a bileaflet multi-scallop prolapse should immediately raise suspicion of underlying MAD, integrated into a global annular distension.

In concrete terms, for the practitioner:

• Targeted vigilance: In the presence of multi-segmental Barlow's disease involving at least 4 scallops, the risk of SAM is tripled; systematic screening using multimodal imaging is essential before any surgery.
• Anatomical evaluation: Integrate MAD not as an isolated binary anomaly, but as a marker of complex global annular remodeling associated with marked valve dilation.
• Diagnostic limitations: Keep in mind that current predictive models (AUROC 0.69) remain explanatory tools; they do not yet replace direct clinical expertise for definitive diagnosis in the operating theatre.

Technical lexicon of the study

Mitral Annular Disjunction (MAD): Anatomical detachment or separation between the mitral annulus and the left ventricular myocardium. In this cohort, its prevalence is particularly high (36.3%) in a specific phenotype of Barlow's disease.

Barlow's Disease (BD): A form of degenerative mitral valve prolapse characterized by excessive valvular tissue. The study identifies BD status as the primary predictor of the presence of MAD.

Fibroelastic deficiency (FED): Etiology of degenerative mitral regurgitation classified here in opposition to Barlow's disease, generally associated with less diffuse involvement of the leaflets.

SHAP (Shapley Additive Explanations): Explainable artificial intelligence method used to quantify the contribution of each variable (such as the number of prolapsed scallops) to the predictions of supervised models.

UMAP-based clustering: Unsupervised learning technique (Uniform Manifold Approximation and Projection) allowing the grouping of patients into phenotypic clusters based on anatomical and clinical variables.

Bileaflet prolapse: Simultaneous involvement of the anterior and posterior leaflets of the mitral valve, identified as a universal characteristic of the cluster presenting the highest prevalence of MAD.

Multiscallop involvement: Prolapse involving several segments (scallops) of the valve. The phenotype associated with MAD in the study is defined by the involvement of at least 4 scallops.

Information intended for healthcare professionals. This content may contain errors or truncated summaries. We recommend always verifying with the original source article. Delynov disclaims all responsibility regarding the use of this information. This document is not intended for patients or the general public.