Time-dependent analysis in CHF follow-up
========================================

* Murray Finkelstein

Justin Ezekowitz and colleagues1 have concluded that patients with congestive heart failure who are followed by specialists and family physicians (FPs) experience better survival than patients who are followed by FPs alone; however, their analysis is not internally consistent.

In the Methods section they state, quite appropriately, that “[a] time-dependent analysis is essential when examining the effect of physician follow-up because patients' outcomes can determine their exposure.” Nevertheless, apart from a brief paragraph at the end of the Results section, all of their findings (in Tables 2 and 3 and in Fig. 1) are presented in terms of an inappropriate time-independent analysis that ignores any change in the provision of care during follow-up.

At the moment of discharge, all patients will have had no cardiovascular follow-up, and they will remain in that category until the first physician visit, at which time their status will change. Should that visit be to an FP, they will move into the FP-only category. Should they subsequently visit a specialist, they will move from the FP-only category to the combined (FP and specialist) category. From a methodologic point of view, these patients will leave behind the days at risk they experienced while in each of the preceding categories. A time-dependent Cox regression will assign them to the appropriate category in the risk set formed at the time of each death in the cohort. Neither the log-rank analysis of Fig. 1 nor the multiple logistic regression analysis of Table 3 make this correct comparison.

It is also not clear that the time- dependent Cox analysis mentioned in the last paragraph of the Results section has been done correctly. The authors state that the model was adjusted for “cumulative days spent in hospital within 1 year after discharge.” However, in a Cox analysis, the characteristics of subjects who died are compared with the characteristics of subjects still alive at the time of death of each case subject. The relevant variable would thus be time spent in hospital up to that time. Use of cumulative days within 1 year of discharge requires the use of future information. This is logically untenable.

I conclude that the authors' results cannot be accepted at face value because their methods were inappropriate for their study design. I encourage them to compute the appropriate time-dependent models to answer this important question about management of congestive heart failure.

## Reference

1.  1. Ezekowitz JA, van Walraven C, McAlister FA, Armstrong PW, Kaul P. Impact of specialist follow-up in outpatients with congestive heart failure. CMAJ 2005;172(2):189-94.
    
    [Abstract/FREE Full Text](http://www.cmaj.ca/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoiY21haiI7czo1OiJyZXNpZCI7czo5OiIxNzIvMi8xODkiO3M6NDoiYXRvbSI7czoyMjoiL2NtYWovMTczLzMvMjM4LjMuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9)