Abstract
ABSTRACT
Background
Abdominal CT‐based assessments of skeletal muscle may provide important prognostic information for all‐cause mortality in aging adults. We aimed to evaluate whether AI‐segmented muscle area and muscle density predict long‐term survival in a large, retrospective adult population.
Methods
This retrospective study included 151 141 adult patients who underwent an abdominal CT examination for any indication between 2000 and 2021. A validated automated AI‐based algorithm measured L3‐level muscle area (cm
2
) and muscle density in Hounsfield units (HU). Demographic and clinical data (age, sex, BMI and date of death) were extracted from electronic health records. Survival analyses included Kaplan–Meier curves and multivariable hazard ratio (HR) models to determine the relationship between these muscle metrics and mortality.
Results
Among the 138 535 adults (66 468 men and 72 067 women) included, 28 489 died over the 20‐year period post‐CT, yielding an overall 20‐year survival rate of 0.620 (95% CI: 0.611–0.629). Of these, 9343 deaths occurred within the first year [survival rate: 0.933 (95% CI: 0.932–0.934)]. Mean and median follow‐up time was 6.4 and 4.9 years, respectively. Lower muscle density significantly predicted higher mortality when each patient's measurement was expressed by its percentile within the age and sex‐matched population (HR up to 3.5 in women and 4.0 in men), with decreasing mortality throughout the higher percentiles. Lower muscle area showed a more modest effect on mortality in both sexes when expressed in percentiles. Individuals with high muscle density demonstrated the most favourable survival and those with low density demonstrated the worst survival. Low muscle density significantly predicted mortality across all age groups and both sexes. Conversely, muscle area predicted mortality in all age groups in men, albeit to a lesser degree and did not predict mortality in any age group among women.
Conclusions
Automated CT‐based measurements of muscle density are superior to muscle area in predicting all‐cause mortality in a large, heterogeneous adult population. Incorporating AI‐driven muscle density assessments into routine clinical practice could substantially improve patient risk stratification and management, of particular relevance for aging and sarcopenic patients.