Transcript Analysis of IVF Data

DATA MINING:
INVESTIGATING
IVF MODELS
1
Why IVF?
9%
of married couples in the US are
medically infertile (age 15-44) = 2.3 million
couples.
An
additional 19% of couples have
impaired ability to have children.
Total
= ~7.2 million couples
From 1998 National report from Center for Disease Control
2
IVF Procedure:
3
Many diagnoses:
4
Concerns:
• Less than 1/3 of IVF cycles are successful.
5
Concerns:
• Less than 20% of cycles result in single pregnancy.
6
Concerns:
• Usually many embryos are transferred.
7
Concerns:
• Some result in multiple births.
8
Concerns:
• How to predict individual patient results.
9
The Data:
• 12 months of statistics from 678 cycles.
• Over 100 attributes for each record.
Technology
Patient
Procedural
Patient
Procedural
Technology
10
The Data:
• Procedural includes info on drug protocols,
#oocytes, #blasts, embryo transfer, trauma and
other cycle specific data.
11
The Data:
• Patient info includes patient specific data such as
diagnosis, BMI, and age.
12
The Data:
• Technology information includes incubators,
doctors, ICSItech, and catheter type among others.
ICSI (intracytoplasmic sperm injection)
13
Data Mining:
• A cyclical procedure.
14
Data Understanding & Cleaning:
• # > or = 6 Cell
• Diagnoses
• Day of Embryo Transfer
15
Sensitivity analysis using entropy
reduction:
Entropy formula:
 pik log2 ( pik )
k
16
17
Model
Assessment:
18
DATA MINING:
INVESTIGATING
IVF MODELS
19