OBJECTIVES: This study presents a model of the mechanisms affecting how time since baseline affects the correlation between mortality and commonly used predictors. METHODS: In 1986, 421 persons (aged 75 years or older) in a Swedish community were interviewed. Fifteen-year mortality rates were analyzed by using hazard regressions. Rather than using average risk over the whole follow-up time, this study looks at temporal differences in predictor strength. RESULTS: All studied health variables, living conditions, and life satisfaction were much stronger predictors of mortality during the first 1 or 2 years of follow-up than during later years. Gender, social contacts, and mental status were about equally correlated to mortality throughout the period. DISCUSSION: Of the presented mechanisms affecting predictive strength, results suggest the importance of the instability of predictors over time. Especially in old populations, predictors that can change rapidly (e.g., health) are strongest for the short term, revealing a lower average mortality risk for longer follow-ups. Rather stable variables (e.g., gender or social contacts) are not affected by the length of follow-up. When average risk is studied over a longer follow-up, insignificant results may hide significant effects during a part of the follow-up. These findings are relevant for studies that examine any kind of outcome after a follow-up.
OBJECTIVES: This study presents a model of the mechanisms affecting how time since baseline affects the correlation between mortality and commonly used predictors. METHODS: In 1986, 421 persons (aged 75 years or older) in a Swedish community were interviewed. Fifteen-year mortality rates were analyzed by using hazard regressions. Rather than using average risk over the whole follow-up time, this study looks at temporal differences in predictor strength. RESULTS: All studied health variables, living conditions, and life satisfaction were much stronger predictors of mortality during the first 1 or 2 years of follow-up than during later years. Gender, social contacts, and mental status were about equally correlated to mortality throughout the period. DISCUSSION: Of the presented mechanisms affecting predictive strength, results suggest the importance of the instability of predictors over time. Especially in old populations, predictors that can change rapidly (e.g., health) are strongest for the short term, revealing a lower average mortality risk for longer follow-ups. Rather stable variables (e.g., gender or social contacts) are not affected by the length of follow-up. When average risk is studied over a longer follow-up, insignificant results may hide significant effects during a part of the follow-up. These findings are relevant for studies that examine any kind of outcome after a follow-up.