0
$\begingroup$

I am attempting to classify people as pregnant or non-pregnant. The business case requires that someone be considered in the model if they meet certain criteria (morning sickness, missed ovulation, etc.). Additionally, the business case requires that someone be classified as pregnant if that individual has a terminating event (delivery, miscarriage, etc.). If a member has a terminating event they are classified as pregnant. If a member does not have a terminating event they are classified as non-pregnant.

This leads to varying time periods during which classifiers for pregnancy (classifiers being diagnosis codes, procedure codes, prescriptions). These variable are used to train a model for identifying pregnancy.

All members that meet criteria are scooped into the model if thought to be pregnant. This leads to variable lengths for which a member is considered within the model -- some exit the model more quickly than others (miscarraige for instance). Members time out if they do not have a pregnancy terminating event with 40 weeks of be considered and therefore classified as non-pregnant.

The question is does the variable time period for which a member is considered present a statistical problem in classifying people as pregnant or non-pregnant?

In this case we are using a random forest, logistic regression, and boosting algorithm for prediction

$\endgroup$
1
  • $\begingroup$ This answer of mine might be of interest. $\endgroup$
    – Dave
    May 3, 2023 at 22:23

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.