Question & Answer

Answer (Analysis Blueprint & Expected Signals)

TL;DR

Fit a multivariable model for 6MWD (preferably percent-predicted 6MWD) measured within 90 days of RRT/EOS. Add a pre-specified block of device parameters (capture threshold, lower-rate limit, rate-response settings, R-wave, impedance) on top of clinical covariates. Use nested-model tests, partial R², and permutation importance to evaluate independent predictive contribution. Hypothesis: higher chronic capture threshold and suboptimal rate-response are associated with shorter 6MWD even after adjustment.

1) Outcome Choice

• Primary: %-predicted 6MWD (robust to body size); compute from a single pre-chosen reference equation across all patients.
• Secondary: Absolute 6MWD (meters) for clinical interpretability (MCID ≈ 20–30 m in many cardiopulmonary cohorts).

2) Predictors

3) Model Specification

1. Base model (clinical): %-pred 6MWD ~ clinical covariates.
2. Full model (clinical + device):

   %pred_6MWD ~ Acute_CT + Chronic_CT + LRL + RR_slope + RR_threshold + Rwave + Impedance + (clinical covariates)

   Use standardized predictors; model nonlinearity with restricted cubic splines for Chronic_CT and LRL.

3. Penalization: Ridge/LASSO or elastic net to reduce overfitting; report shrunk coefficients.
4. Multicollinearity: Check VIF; if high between acute & chronic CT, prefer chronic CT (long-term interface) and place acute CT in sensitivity analysis.
5. Interactions (pre-specified): Chronic_CT × LRL; Chronic_CT × RR_slope (biologic plausibility: higher thresholds + muted rate response may limit walk capacity).
6. Missing data: Multiple imputation by chained equations with outcome included in imputation model.

4) Evidence of Independent Prediction

• Incremental fit: ΔAdjusted R² and likelihood-ratio test comparing Full vs Base.
• Partial R²: For the device block and each device variable.
• Permutation importance: Loss in model performance when shuffling each device predictor.
• Calibration & internal validation: Bootstrapping (≥1,000 resamples) for optimism-corrected performance.

5) Expected Directions (Pre-specified)

• Chronic capture threshold: higher ⇒ lower %-pred 6MWD (worse interface, higher output, potential non-capture episodes).
• LRL (within physiologic range): modestly higher LRL may increase %-pred 6MWD up to a point; very low LRL may blunt HR reserve early in the walk.
• RR slope/threshold: more sensitive/appropriate slope ⇒ longer 6MWD; too low sensitivity can under-drive rate response.
• R-wave amplitude: very low values may signal poor contact that co-occurs with higher thresholds; expect a positive association with 6MWD.
• Impedance: extremes (very low or very high) may correlate with poorer performance; test for nonlinearity.

6) Sample Size & Power (Planning)

• Aim for ≥10–20 participants per effective degree of freedom after penalization; with ~12–16 df (including splines), target ≥150–250 patients.
• Alternatively, use prediction-model sample size formulas (e.g., Riley et al.) to set minimum N for desired shrinkage and small optimism in R².

7) Sensitivity Analyses

• Outcome as absolute 6MWD (meters) and as z-score vs reference equation.
• Exclude recent med titration (<14 days) or intercurrent illness.
• Stratify by documented non‑capture burden near test, if available from EGM/telemetry.
• Device-vendor fixed effects if multiple platforms are included.

8) Reporting Template

Base (clinical) adj. R²: ____; RMSE: ____ m
Full (clinical + device) adj. R²: ____ (Δ = ____); LRT p = ____
Partial R² (device block): ____
Key coefficients (standardized):
  Chronic_CT: β = ____ (95% CI ____ to ____), p = ____
  LRL (spline): overall p = ____; effect at ____ bpm: ____
  RR_slope: β = ____ (95% CI ____ to ____), p = ____
  R-wave: β = ____ (95% CI ____ to ____), p = ____
  Impedance (spline): overall p = ____
Calibration slope: ____; optimism-corrected R²: ____

9) Caveats

• 6MWD is influenced by motivation and corridor characteristics; standardize protocol.
• Rate‑response algorithms differ by vendor—harmonize parameters to comparable constructs (slope/sensitivity).
• Acute CT may reflect immediate post-interrogation physiology; prioritize chronic CT for “steady-state” interface.

This blueprint supports a rigorous test of independent prediction of 6MWD at RRT/EOS by device parameters, beyond clinical status.