Abstract
Abstract
Background
Handgrip strength is associated with cardiometabolic abnormalities, but whether it adds information once detailed adiposity and muscle mass are known is unclear.
Methods
We studied adults aged 40–59 years from NHANES 2011–2012 (development, n=508) and 2013–2014 (temporal validation, n=616). The outcome was at least two of four non-adiposity abnormalities (elevated blood pressure, fasting glucose or triglycerides, or low HDL cholesterol); waist circumference was not part of the outcome. A prespecified feature ladder added DXA body fat percentage and visceral adipose tissue to age, sex, BMI and waist circumference (M1), then appendicular skeletal muscle index (M2), then absolute (M3A, primary) or weight-normalised (M3B) grip strength. Logistic regression was primary; elastic-net and XGBoost were supportive. Development performance used nested cross-validation; frozen models were scored once in the validation cohort. Paired increments used 2000 stratified bootstrap resamples.
Results
The outcome occurred in 249 (49.0%) and 277 (45.0%) participants. Adding adiposity increased the logistic regression AUC by 0.0209 (95% CI 0.0024–0.0402) in development and 0.0209 (0.0032–0.0375) in temporal validation. Adding absolute grip strength to M2 changed the AUC by −0.0046 (−0.0107 to 0.0013) and 0.0045 (−0.0009 to 0.0098), respectively. All 12 grip-strength AUC increments across algorithms, definitions and cohorts had intervals that included zero. Calibration was unchanged and the incremental net benefit was not consistently positive. Model attributions to grip strength varied by algorithm.
Conclusions
In middle-aged adults, detailed adiposity, but not grip strength, added reproducible information for concurrent cardiometabolic risk stratification beyond conventional anthropometry and muscle mass.
Highlights
DXA adiposity added about 0.02 AUC in both development and temporal validation
Grip strength added no stable increment beyond adiposity and muscle mass
All 12 grip-strength AUC increments had 95% CIs that included zero
Grip strength received model attributions that did not improve performance
Frozen models were scored once in a later, temporally separated NHANES cycle