# Maximum Bilateral Grip Strength Task Card Task id: `maximum_bilateral_grip_strength` Evidence status: grounded Evidence reviewed: 2026-07-29 ## Target Maximum of the first valid left- and right-hand attempts, requiring both sides and `piera_hand_grip_status == success`. Equal opportunities are compared for every participant; the maximum need not be the dominant hand. Later-introduced second attempts are excluded, so this is not a clinical best-of-three measure or a one-to-one HealthFormer parameter. Implementation facts: - target fields: `hand_grip_strength_left`, `hand_grip_strength_right` - target dataset: `hand_grip` - target source: None (loaded through HPP `hpp_data_loader`) - default research stage: `00_00_visit` - valid range: 1-90 kg - primary metric: R2 ## Why This Matters Grip strength is a low-cost measure of muscle function and a core sarcopenia/frailty component. Lower strength is associated with adverse outcomes, but this Task supports forecasting, not diagnosis, prognosis, or causality. ## Related HPP Measurements EWGSOP2 pairs strength with muscle quantity, making DXA appendicular lean mass a grounded companion. Plasma metabolomics and objective activity provide independent biological and behavioral views of muscle function. ## Clinically Meaningful Variants Raw kg remains primary. Age and sex belong in the demographic floor and stratified residuals; normalizing Y would create a reference-dependent Task. A future low-strength variant may test EWGSOP2 cutoffs below 27 kg for men and 16 kg for women after calibration on HPP's later two-attempt subset. ## Baselines And Ceilings `NoFeaturesPredictor` contributes no modality features; `DemographicFloorStrategy` fits age, sex, and BMI. Last observation carried forward (LOCF) copies prior Y. Ordinary least squares (OLS) and gradient-boosted trees (GBDT) fit prior Y plus demographics. HealthFormer reported LOCF and per-modality linear baselines, not trees or the demographic floor. Compare representations on one cohort. ## Benchmark-Track Evidence The frozen v0 tracks use the fixed nine-field V1 Nightingale amino-acid panel (`Ala`, `Gln`, `Gly`, `His`, `Ile`, `Leu`, `Val`, `Phe`, `Tyr`) plus V1 age, sex, and BMI for both V1 and V2 grip targets. literature and literature support plasma metabolomics and branched-chain amino acids as independent muscle-mass and strength signals. Every hand-grip and DXA field is excluded. Gait Fusion is not v0-benchmark-ready. Raw Newton skeleton files exist on S3, but no canonical operational source contract, checksum-bound embeddings, or accessible checkpoint was found. The canonical V1 engineered-gait cohort overlaps 345 V1 and zero V2 eligible grip targets, so gait remains a future blocked candidate. ## Caveats - Age and sex explain substantial variance and must remain visible. - First-attempt-only semantics improve visit comparability but can understate maximal strength. - Pain, handedness, effort, and device technique contribute noise. ## Metrics R2 is primary. Also report RMSE and MAE in kg, Pearson r, Spearman rho, same-cohort deltas against LOCF, and sex- and age-stratified residuals. ## References - Lutsker et al. (2026), HealthFormer, . - HPP `hand_grip` dataset documentation for attempt fields and Piera QC. - Cruz-Jentoft et al. (2019), EWGSOP2 consensus, . - Reijnierse et al. (2017), grip protocol comparison, . - Gabet et al. (2026), HPP gait foundation model, . ## Open Questions - Should a later deep task model side-specific trajectories or asymmetry?