BACKGROUND AND OBJECTIVES: Longitudinal studies of health over time may be misleading if some people die. Self-rated health (excellent to poor) and the SF-36 profile scores have been transformed to incorporate death. We applied the same approaches to incorporate death into activities of daily living difficulties (ADLs), IADLs, mini-mental state examination, depressive symptoms, blocks walked per week, bed days, the timed walk, body mass index and blood pressure. STUDY DESIGN AND SETTING: The Cardiovascular Health Study of 5,888 older adults, was followed up to 9 years. Mean age was 73 at baseline, and 658 had an incident stroke during follow-up. METHODS: We recoded each variable as the probability of being healthy 1 year in the future (PHF), conditional on the current value of the variable. This was done for 11 health variables, using three definitions of healthy, and two estimation models. Deaths were set to zero, and mean PHF was plotted in the 3 years before and after an incident stroke. RESULTS: Analyses without the deaths were too optimistic. The effect of stroke was greatest on hospitalization, self-rated health, and IADLs. Alternative transformation approaches had slightly different results. CONCLUSION: These methods provide an additional approach for handling death in longitudinal studies.
BACKGROUND AND OBJECTIVES: Longitudinal studies of health over time may be misleading if some people die. Self-rated health (excellent to poor) and the SF-36 profile scores have been transformed to incorporate death. We applied the same approaches to incorporate death into activities of daily living difficulties (ADLs), IADLs, mini-mental state examination, depressive symptoms, blocks walked per week, bed days, the timed walk, body mass index and blood pressure. STUDY DESIGN AND SETTING: The Cardiovascular Health Study of 5,888 older adults, was followed up to 9 years. Mean age was 73 at baseline, and 658 had an incident stroke during follow-up. METHODS: We recoded each variable as the probability of being healthy 1 year in the future (PHF), conditional on the current value of the variable. This was done for 11 health variables, using three definitions of healthy, and two estimation models. Deaths were set to zero, and mean PHF was plotted in the 3 years before and after an incident stroke. RESULTS: Analyses without the deaths were too optimistic. The effect of stroke was greatest on hospitalization, self-rated health, and IADLs. Alternative transformation approaches had slightly different results. CONCLUSION: These methods provide an additional approach for handling death in longitudinal studies.
Authors: Paula H Diehr; Stephen M Thielke; Anne B Newman; Calvin Hirsch; Russell Tracy Journal: J Gerontol A Biol Sci Med Sci Date: 2013-05-10 Impact factor: 6.053
Authors: Mark D Neuman; Jeffrey H Silber; Jay S Magaziner; Molly A Passarella; Samir Mehta; Rachel M Werner Journal: JAMA Intern Med Date: 2014-08 Impact factor: 21.873
Authors: Anne B Newman; Barbara L Naydeck; Diane G Ives; Robert M Boudreau; Kim Sutton-Tyrrell; Daniel H O'Leary; Lewis H Kuller Journal: Am J Cardiol Date: 2008-01-15 Impact factor: 2.778
Authors: Emily Locke; Stephen Thielke; Paula Diehr; Anthony G Wilsdon; R Graham Barr; Nadia Hansel; Vishesh K Kapur; Jerry Krishnan; Paul Enright; Susan R Heckbert; Richard A Kronmal; Vincent S Fan Journal: COPD Date: 2013-07-02 Impact factor: 2.409