search

nhejazi / txshift

:package: :game_die: R/txshift: Efficient Estimation of the Causal Effects of Stochastic Interventions, with Corrections for Outcome-Dependent Sampling
https://codex.nimahejazi.org/txshift
Other
13 stars 4 forks source link

piecewise monotonicity of dose-response projections #66

Open nhejazi opened 2 years ago

nhejazi commented 2 years ago

The summarization procedure of projecting individual risk estimates onto a working linear MSM has been very useful, but it could potentially be improved somewhat by incorporating an option for montonicity. As is, we can fit either linear or piecewise linear working MSM summaries (no inference for the latter unless the knot point is pre-specified), but the individual risk estimates need not follow a monotonic pattern. In certain use-cases (e.g., VE curves), it may be useful to enforce monotonicity when it can be justified appropriately by background knowledge (e.g., observed biological responses).

nhejazi commented 6 days ago

this could be trivial implemented by just running isotonic regression as per https://academic.oup.com/jrsssb/article-abstract/82/3/719/7056122