Modified automated detection of the first ventilatory threshold using V-slope regression: validation against manual assessment in healthy adults

Background

The first ventilatory threshold (VT1) is a key marker of aerobic fitness, yet its determination typically relies on expert visual assessment, which is time-consuming and subject to interobserver variability. Automated solutions may offer a more objective and time-efficient alternative. This study aimed to test the accuracy of the recently proposed modified V-slope method, which has not been formally evaluated yet.

Methods

In this retrospective study, VT1 was assessed in 271 healthy adults who completed cycle ergometer cardiopulmonary exercise testing (CPET). VT1 was determined using a modified V-slope algorithm (Sarthy et al), which identifies the point of maximum perpendicular distance between a global linear regression and a quadratic fit of the VCO2–VO2 curve, and was compared to a manual V-slope assessment by two independent observers. Thirteen participants were excluded due to requirements of the automated method or due to linear or abnormal VCO2–VO2 curves. Agreement between methods was evaluated using Bland–Altman analysis, equivalence testing (TOST), Deming regression, and intraclass correlation coefficients (ICC). Interobserver reliability for manual VT1 determination was assessed using ICC.

Results

Oxygen uptake at VT1 was slightly higher with the automated method compared to manual assessment, both in absolute terms (1154 (IQR 944, 1406) vs. 1100 (IQR 936, 1372) ml/min, p < 0.001) and relative to body weight (15.3 (IQR 12.5, 17.8) vs. 14.9 (IQR 12.2, 17.9) ml/kg/min, p = 0.018). Similar minor differences were observed for VO2 at VT1 relative to predicted VO2max (58% (IQR 52, 65) vs. 57% (IQR 50, 64), p < 0.001), heart rate at VT1 (117 ± 18 vs. 113 ± 17 bpm, p < 0.001), and load at VT1 (80 (IQR 60, 100) vs. 80 (IQR 60, 100) Watts, p < 0.001).

Bland–Altman analysis revealed a mean bias of 36 ml/min with 95% limits of agreement from −190 to 262 ml/min, indicating a small systematic difference but substantial individual variability (Figure 1). Equivalence within ±60 ml/min (p < 0.001) was confirmed, although only 45.7% of subjects fell within this range. Deming regression revealed a slope of 0.97 (95% CI 0.93 to 1.02) and an intercept of 0.14 ml/min (95% CI −50.08 to 50.04) (Figure 2), while the ICC was 0.95 (95% CI 0.93 to 0.96), demonstrating close agreement between the automated and manual VT1 measurements.

Inter-rater reliability for manual VT1 determination in a subset of 30 subjects was also high (ICC = 0.94, 95% CI 0.79 to 0.98), indicating consistent measurements between both observers.

Conclusions

The modified automated V-slope method provides VT1 values with minimal systematic bias to manual assessment but with some within-patient variability. This automated approach is reliable, reproducible, and suitable for initial large-scale CPET analyses, though manual verification remains advisable.Figure 1Figure 2