Disruption of insulin-like growth factor-I (IGF-I) increases health and life span in animal models, though this is unconfirmed in humans. If IGF-I stability indicates homeostasis, the absolute level of IGF-I may be less clinically relevant than maintaining an IGF-I setpoint. Participants were 945 U.S. community-dwelling individuals aged ≥65 years enrolled in the Cardiovascular Health Study with IGF-I levels at 3-6 timepoints. We examined the association of baseline IGF-I level, trajectory slope, and variability around the trajectory with mortality. There were 633 deaths over median 11.3 years of follow-up. Lower IGF-I levels, declining or increasing slope, and increasing variability were each individually associated with higher mortality (all p < .001). In an adjusted model including all three trajectory parameters, baseline IGF-I levels <70 ng/mL (hazard ratio [HR] 1.58, 95% CI 1.28-1.96 relative to IGF-I levels of 170 ng/mL), steep declines and steep increases in trajectory slope (HR 2.22, 1.30-3.80 for a 15% decline; HR 1.40, 1.07-1.84 for a 10% decline; HR 1.80, 1.12-2.89 for a 15% increase; HR 1.31, 1.00-1.72 for a 10% increase, each vs no change), and variability ≥10% (HR 1.59, 1.09-2.32 for ≥ 30%; HR 1.36, 1.06-1.75 for 20%; and HR 1.17, 1.03-1.32 for 10% variability, each vs 0%) in IGF-I levels were independently associated with mortality. In contrast to data from animal models, low IGF-I levels are associated with higher mortality in older humans. Irrespective of the actual IGF-I level, older individuals with stability of IGF-I levels have lower mortality than those whose IGF-I levels fluctuate over time.