Effectiveness of congenital anomaly screening based on basic risk factors using supervised machine learning
In most of the developed countries the national prenatal screening policies for congenital abnormalities has resulted in the reduction of the prevalence rates. However, in most developing countries there is no national prenatal screening policy for congenital abnormalities. This study explores the e...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/102271/1/MominaTehreemKhattakMSBME2020.pdf.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In most of the developed countries the national prenatal screening policies for congenital abnormalities has resulted in the reduction of the prevalence rates. However, in most developing countries there is no national prenatal screening policy for congenital abnormalities. This study explores the effect of prenatal screening on the prevalence rates of live birth, fetal death, and termination of pregnancy (TOPFA) on Trisomies and Neural tube defects in Europe. Meanwhile, the prevalence and existing methods for prenatal screening in Malaysia are reviewed. The data used is from the European Surveillance of Congenital Anomalies (EUROCAT) and the Malaysian neonatal registries. The analysis of prevalence rates showed that a prenatal screening policy can reduce the Live Birth (LB) and Fetal Death (FD) prevalence for Trisomies by 77% and 80% respectively, while, for Neural Tube Defects (NTD) by 36% and 38.5%, respectively. The prevalence of Trisomy 21 (T- 21) and Neural Tube Defects has increased by 72% and 32% respectively over a period of four years in Malaysia. For this, a risk prediction model using only basic risk factors is developed. This thesis used different supervised machine learning techniques, i.e., logistic regression, random forests, and artificial neural networks (ANN) for the model. Moreover, we also used k-means clustering on our training data and used it to create a Euclidean distance based (supervised) prediction model. The best model according to the results is logistic regression, which can predict T-21 with a sensitivity of 79.75%, specificity of 41.16% and a Balanced Classification Rate (BCR) of 60.46. It is observed that the specificity is low at 41.16% but sensitivity is high which means detection rate is high. The best model for NTD is also logistic regression, which can predict neural tube defect (NTD) with a sensitivity of 68.35%, specificity of 45.32% and a BCR of 59.84%. The risk prediction model of congenital anomalies has a sensitivity of 80%, specificity of 45% and BCR of 63%. The risk prediction model will help the doctors point out the high-risk woman. The accuracy of the prediction model may be improved by adding more predictors, which do not require expensive tests. |
|---|