Add LTMLE about zepid HOT 5 OPEN

As a further note to self, this paper also describes a different approach to estimating the g-formula. It might be an easier approach to code and involves less model specification for users

Essentially, you go back through time fitting an outcome model at the last time point, generate your predictions then keep working backwards. This approach relies on a clever covariate to become doubly-robust and semi-parametric. https://www.jstor.org/stable/27645860?seq=1#metadata_info_tab_contents

This is useful background to consider for the LTMLE as well, since it uses a similar estimation process

from zepid.

Adapt code from g-formula sequential regression estimator part of #30 and #32

from zepid.

Bumping the LTMLE implementation timeline back. Will be added after 0.4.0 since it will take me some additional time to implement (and I would like to release 0.4.0 soon)

from zepid.

Estimation of g_t is IPTW for follow-up (cumulative product). Need to estimate the IPTW for each time point. Likely best if I fit a model to each time point for this estimation. This will make no parametric assumptions regarding time (however some models might not fit).

Maybe later create a work around that allows for parametric time?...

from zepid.

https://arxiv.org/abs/1802.05005

from zepid.