Immortal Time Bias - Deceased Control Patients Can’t Be Hospitalized, page-10

I did some more reading. The Joint Frailty Model returns two different hazard ratios - one for recurrent hospitalization and another for cardiac mortality. I really cannot understand why it would be valid to report on only one of these coefficients. They do in their DREAM-HF design paper note that time to first event is specified at FDA’s request as the key secondary endpoint to ‘verify that improvement in recurrent HF hospitalizations is not at the expense of worsening TCEs’. In our case it’s just the opposite - it seems equivalence in hospitalizations was due to fewer terminal cardiac events (TCEs). There is still hope. In general I find the assumptions that need to go into the JFM analysis such as the alpha factor to also be questionable - is hospitalization really positively correlated with frailty or is it possibly conditional on being healthy enough to seek care?