A novel, data-driven conceptualization for critical left heart obstruction

Comput Methods Programs Biomed. 2018 Oct:165:107-116. doi: 10.1016/j.cmpb.2018.08.014. Epub 2018 Aug 20.


Background: Qualitative features of aortic and mitral valvar pathology have traditionally been used to classify congenital cardiac anomalies for which the left heart structures are unable to sustain adequate systemic cardiac output. We aimed to determine if novel groups of patients with greater clinical relevance could be defined within this population of patients with critical left heart obstruction (CLHO) using a data-driven approach based on both qualitative and quantitative echocardiographic measures.

Methods: An independent standardized review of recordings from pre-intervention transthoracic echocardiograms for 651 neonates with CLHO was performed. An unsupervised cluster analysis, incorporating 136 echocardiographic measures, was used to group patients with similar characteristics. Key measures differentiating the groups were then identified.

Results: Based on all measures, cluster analysis linked the 651 neonates into groups of 215 (Group 1), 338 (Group 2), and 98 (Group 3) patients. Aortic valve atresia and left ventricular (LV) end diastolic volume were identified as significant variables differentiating the groups. The median LV end diastolic area was 1.35, 0.69, and 2.47 cm2 in Groups 1, 2, and 3, respectively (p < 0.0001). Aortic atresia was present in 11% (24/215), 87% (294/338), and 8% (8/98), in Groups 1, 2, and 3, respectively (p < 0.0001). Balloon aortic valvotomy was the first intervention for 9% (19/215), 2% (6/338), and 61% (60/98), respectively (p < 0.0001). For those with an initial operation, single ventricle palliation was performed in 90% (176/215), 98% (326/338), and 58% (22/38) (p < 0.0001). Overall mortality in each group was 27% (59/215), 41% (138/338), and 12% (12/98) (p < 0.0001).

Conclusions: Using a data-driven approach, we conceptualized three distinct patient groups, primarily based quantitatively on baseline LV size and qualitatively by the presence of aortic valve atresia. Management strategy and overall mortality differed significantly by group. These groups roughly correspond anatomically and are analogous to multi-level LV hypoplasia, hypoplastic left heart syndrome, and critical aortic stenosis, respectively. Our analysis suggests that quantitative and qualitative assessment of left heart structures, particularly LV size and type of aortic valve pathology, may yield conceptually more internally consistent groups than a simplistic scheme limited to valvar pathology alone.

Keywords: Cluster analysis; Congenital aortic stenosis; Data science; Hypoplastic left heart syndrome.

MeSH terms

  • Aortic Valve / abnormalities
  • Aortic Valve / diagnostic imaging
  • Aortic Valve Stenosis / classification
  • Aortic Valve Stenosis / congenital
  • Aortic Valve Stenosis / diagnostic imaging
  • Cluster Analysis
  • Cohort Studies
  • Diagnosis, Computer-Assisted / methods*
  • Diagnosis, Computer-Assisted / statistics & numerical data
  • Echocardiography
  • Female
  • Heart Defects, Congenital / classification*
  • Heart Defects, Congenital / diagnostic imaging*
  • Heart Defects, Congenital / surgery
  • Heart Ventricles / diagnostic imaging
  • Humans
  • Hypoplastic Left Heart Syndrome / classification
  • Hypoplastic Left Heart Syndrome / diagnostic imaging
  • Infant, Newborn
  • Male
  • Models, Cardiovascular
  • Prospective Studies
  • Unsupervised Machine Learning
  • Ventricular Dysfunction, Left / classification
  • Ventricular Dysfunction, Left / diagnostic imaging