Perfusion indices cannot be reproduced from publicly available perioperative waveform data: a reproducibility and usability assessment of VitalDB

Abstract Background. The peripheral perfusion index (PI) is a monitor-derived plethysmographic measure reported to predict postoperative acute kidney injury (AKI). VitalDB, the largest public perioperative waveform database, originates from the same institution as the original PI–AKI study; we asked whether it is reproducible from the public release. Methods. Using the public VitalDB release (6,388 cases), we attempted four sequential reproduction steps: identification of a monitor-derived PI/PVI numeric channel; derivation of PI from the raw plethysmographic waveform with assessment of cross-case scale comparability; construction of a within-case relative PI (rel-PI); and computation of pulse pressure variation (PPV). The outcome was KDIGO AKI. Associations were evaluated in the waveform core set (n = 3,518) with threshold and multivariable analyses; explicit hypotension (mean arterial pressure < 65 mmHg) served as a positive control. Results. No monitor-derived PI/PVI numeric channel exists in the public release. Waveform-derived PI ranged 6.4–115.9% across cases (86% of 3,507 cases above the physiological upper bound ≈ 25%): the display scale is auto-ranged and not comparable across cases. The relative index was measurable and reproducible, and was not attributable to noise or automatic gain artefact (split-half reliability r = 0.967), but was unrelated to AKI. The primary exposure (a persistent ≥ 2-min episode with a ≥ 50% fall in the relative perfusion signal at normal arterial pressure) showed no association (crude OR 0.94, 95% CI 0.76–1.17; adjusted OR 0.98, p = 0.86); a nominally significant crude association at ≥ 30 min (OR 1.40, p = 0.026) was attenuated after adjustment (OR 1.31, p = 0.148); no association was seen without explicit hypotension (OR 0.84, 95% CI 0.62–1.14). In an underpowered pilot (241 cases), PPV showed no association (OR 1.17, 95% CI 0.52–2.63). The positive control remained robust (adjusted OR 1.62, 95% CI 1.21–2.16, p = 0.0011). Conclusions. A monitor-derived perfusion index cannot be reproduced from public perioperative waveform data: the monitor channel is absent and the released waveform scale is not comparable across cases. The within-case surrogate is measurable and reproducible but a different construct and does not reproduce the reported association. These constraints reflect open waveform archive design, not VitalDB alone.