# AI4M Week 1 ITE Pair distance + censoring

I have a 2-part question regarding the C-for-benefit calculation:

1. Is there a threshold of minimum ITE distance between pairs to consider a pair a good match? The method explains that one should rank both the treatment and control group independently based on ITE, but one could imagine that potential matched pairs may have very different ITEs (ie. 0.8 vs 0.4). Is it OK to just match them based on the given order, or should we only consider pairs that fall within an ITE difference?

2. What do we do in case of censored patients? How do we evaluate the ITEs for a given method (s, t) if we have censored observations in either arm? Are they just considered as outcome=0?

Hey @J_Zam

1. In the C-for-benefit calculation, there is no specific threshold for minimum ITE distance between pairs to consider a pair a good match. The method aims to find pairs with the closest ITE values possible, but it does not require that the pairs have identical or very similar ITEs. Therefore, it is acceptable to match pairs based on the given order, even if they have different ITE values.
It is recommended to use common sense and consider the clinical significance of the differences in ITE values when matching pairs. If the differences in ITE values are large enough to affect the treatment effect estimate, it may be necessary to use more stringent matching criteria or consider other matching methods.

2. Censored observations in either arm can be handled in different ways depending on the specific method used to estimate the ITE. One possible approach is to consider censored observations as missing data and use imputation methods to estimate the missing values.

For example, one can use multiple imputation techniques to impute the missing outcomes for censored observations based on the observed outcomes and covariate information. Alternatively, one can use survival analysis methods to model the time-to-event outcomes and estimate the ITE based on the survival curves.

In either case, it is important to carefully consider the assumptions underlying the imputation or survival analysis method and to evaluate the sensitivity of the results to different assumptions and methods.