Consider an analysis using an IP weighted model. How would you check for violations of the assumption of constant treatment effects? If you found such a violation, how could the model be modified to...

Consider an analysis using an IP weighted model. How would you check for violations of the assumption of constant treatment effects? If you found such a violation, how could the model be modified to accommodate it? And in that case, how would you estimate that hazard ratio for the comparison of always-on versus always-off treatment patterns?