Question
Biomechanics vs symptoms: IMU‑estimated knee adduction moment (KAM) and pain. In older adults (≥70y) with knee OA, does an inertial‑measurement‑unit (IMU) estimate of knee adduction moment (KAM; peak and impulse, normalized by body weight and height) explain more variance in WOMAC pain than radiographic severity (Kellgren–Lawrence grade), and do brace/footwear interventions that reduce KAM produce proportional short‑term pain reductions?
Answer (protocol‑grade plan + expected benchmarks)
1) Background & definitions
- KAM is a surrogate for medial tibiofemoral compartment load during stance. Higher KAM → higher medial load.
- Normalize peak KAM to body weight × height: KAMnorm = KAMpeak / (BW·Ht), units: %BW·Ht.
- IMU KAM uses thigh + shank (±foot) IMUs to estimate frontal‑plane knee kinetics via sensor fusion + learned mapping.
2) What to measure (domains & instruments)
| Domain | Instrument / Protocol | Metric | Notes |
|---|
| Biomechanics | IMUs on distal thigh & proximal shank (bilateral); 100 Hz; 10 m walkway or treadmill @ self‑selected speed | Peak KAMnorm (early stance), KAM impulsenorm (area) | 5–10 strides per limb; average across strides |
| Symptoms | WOMAC Pain (0–20), Pain NRS (0–10) | Higher = worse | Recall 48–72 h |
| Structure | Radiographs (fixed‑flexion), KL grade 0–4 | Ordinal | Blinded readers |
| Covariates | Age, sex, BMI, varus/valgus alignment (goniometer or long‑leg film), gait speed, depression (PHQ‑8), pain catastrophizing (PCS), analgesic use | — | Pre‑specified |
| Intervention subset | Valgus unloader brace or medial‑wedge footwear for 2 weeks | ΔKAMnorm, ΔWOMAC pain | Adherence logged; brace wear ≥2 h/day |
3) IMU → KAM estimation pipeline
Calibration
Static T‑pose + neutral standing to align sensor frames; estimate segment lengths from anthropometrics.
Gait segmentation
Detect heel‑strike/toe‑off from shank angular velocity; select steady‑state strides.
Feature extraction
Frontal‑plane knee adduction angle/velocity, hip adduction, shank varus angle, step width, cadence, stance time.
Mapping model
Pre‑trained regression (e.g., elastic net/XGBoost) to map features → KAMnorm; bias‑correct with speed and alignment.
# Pseudocode (per stride)
features = [knee_add_angle_peak, knee_add_vel_peak, hip_add_angle, shank_varus, step_width, cadence, stance_time, speed, alignment]
KAM_norm_pred = dot(beta, features) + bias
4) Study design
- Primary analysis: Cross‑sectional, n = 220 adults ≥70 with medial knee OA.
- Sub‑study: n = 120 randomized to brace vs footwear wedge for 2 weeks (open‑label); repeated IMU and pain measures.
- Exclusions: Lateral OA predominant, inflammatory arthritis, recent IA steroid (<4 weeks), unstable gait disorders.
5) Statistical analysis plan (SAP)
- Primary model: Hierarchical regression for WOMAC Pain with blocks: (1) demographics (age/sex/BMI) + speed, (2) + KL grade, (3) + KAMnorm (peak & impulse). Compare adjusted R² and partial R² of KAM terms.
- Secondary: Replace KAM with foot progression angle, step width to test mediation; include alignment × KAM interaction.
- Known‑groups: Compare pain across KAM tertiles (ANOVA, trend test) controlling for KL.
- Responsiveness (sub‑study): ΔWOMAC vs ΔKAM: expect slope β ≈ 0.15–0.25 pain points per %BW·Ht reduction; compute SRM, MCID mapping.
- Reliability: Test–retest (1 week) ICC ≥0.85 for KAMnorm measures in stable participants.
6) Benchmarks to claim success (pre‑specified)
| Property | Threshold |
|---|
| Criterion validity | KAM terms add ≥0.08 partial R² to pain beyond KL and covariates |
| Model performance | Adjusted R² ≥ 0.35 for final model |
| Known‑groups | Mean pain differs ≥0.5 SD across KAM tertiles (p<0.01) |
| Responsiveness | Correlation ΔKAM ↔ Δpain |r| ≥ 0.30; SRM ≥ 0.5 in responders |
| Reliability | ICC ≥ 0.85; MDC95 small enough to detect 10% KAM change |
7) Sample size (sketch)
- Detect partial R² = 0.08 for KAM with ~10 predictors, α=0.05, power=0.90 → N≈200–220.
- Sub‑study to detect r=0.30 for ΔKAM–Δpain, α=0.05, power=0.80 → n≈85; inflate to 120 for attrition.
8) Expected findings & interpretation
- KAM outperforms KL for explaining pain variability; structure (KL) remains important but weaker for symptoms.
- Alignment matters: Varus alignment strengthens KAM→pain association (significant interaction).
- Interventions (valgus brace / medial wedge) lower KAM by 5–15% and yield small‑to‑moderate pain improvements within 2 weeks, proportional to KAM reduction.
- Clinical translation: Use IMU‑KAM to identify high‑load gait mechanics and tailor unloading strategies (brace, gait retraining to increase toe‑out/step width, weight loss).
9) Practical cut‑points (to validate)
- Low load: KAMnorm < 3.0 %BW·Ht
- Moderate load: 3.0–3.9 %BW·Ht
- High load: ≥ 4.0 %BW·Ht — consider unloading brace and gait retraining
Notes — KAM estimates from IMUs should be validated against 3D motion capture + force plates in a subset (target RMSE ≤ 10–15% for peak KAMnorm). KAM is most relevant to medial knee OA; pain can also reflect inflammatory and central factors, so multicomponent care remains essential.