anthropometric data analytics: a portuguese case study
TRANSCRIPT
Anthropometric Data Analytics: a Portuguese Case Study
António Barata1, Lucília Carvalho2, Francisco M Couto1
1 LaSIGE, Faculdade de Ciências da Universidade de Lisboa, Portugal
2 Hospital de Egas Moniz, Centro Hospitalar de Lisboa Ocidental, Portugal
PACBB, June 21-23, 2017
Porto Portugal
Goal
• Estimation of Gestational Age (GA)
– For correct diagnosis
– treatment of disease
– tool for parental counselling
– plan for appropriate perinatal care
– prime requisite for foetal autopsy
• which variables should we use?
Anthropometric Data Analytics
• Previous studies measured the accuracy of different foetal parameters in GA, in particular:– head circumference
– crown-heel length
– crown-rump length
– foot length
– hand length
• Regression analysis and model fittingImage Source: http://slideplayer.biz.tr/slide/3068759/
Case study
• a dataset of 450 individuals
– central-southern region of Portugal
– Hospital de Egas Moniz – CHLO
– ages of 13 and 42 inclusive
– comprised of 24 foetal parameters
Foetal Variables Collected
1. Gestational Age (GA) (weeks)
Weights (g)2. Adrenals3. Body weight (Body)4. Kidneys5. Liver6. Lungs7. Spleen8. Thymus
Distances (cm)9. abdominal circumference (AC)10. chest circumference (CC)11. crown-heel length (CHL)12. crown-rump length (CRL)13. foot length (FL)14. hand length (HL)15. head circumference (HC)16. inner canthal distance (ICD)17. intercommissural distance (ID)18. left ear length (LEL)19. left palpebral fissure width (LPFW)20. middle finger length (MFL)21. outer canthal distance (OCD)22. philtrum length (PL)23. right ear length (REL)24. right palpebral fissure width (RPFW)
Collection of data
• Usage of optical character recognition (OCR) software
• But each report had to be manually inserted
• Stored as a relation database
Methodology
• Principal Component Analysis (PCA)
• Multiple Linear Regression (MLR)
• Polynomial Regression (PR)
PCA
Multiple Linear Regression Models
• Standardized β-weights and variables selected by each regression algorithm method
• R2 ≈ 0.953
Polynomial Regression Models
• 1st to 5th degree; 0.363 ≤ R2 ≤ 0.945
R2 Clusters• 5th degree polynomial regression
Conclusions
• CHL, CRL, and FL are the most appropriate for GA estimation
– Consistent with previously published work
• Body weight, HC, HL, and ear length are also noteworthy candidate variables
• Data and Software available at:https://github.com/BarataAP/Anthropometric-Data-Analytics-Portugal/
Future Work