Pathway Is Not Diagnosis: Identifiability Limits of Cardiorespiratory Pattern Monitoring in Preterm Infants
- 1SAIT Polytechnic
Abstract
This paper argues that a monitored cardiorespiratory signal carries a fixed amount of information about the cause of a preterm infant's apnea–bradycardia–desaturation events, and that this information is measured by the distance between the signal laws of the candidate etiologies. The author generalizes two propositions from his neonatal monograph—that etiologies producing the same pathway vector cannot be separated, and that context can change posterior odds—into a graded account. Total-variation distance quantifies how separable two etiologies are; no processing of the signal can increase it, so the error floor is fixed; and when laws coincide exactly, the finest resolvable diagnosis is a class of causes, with any split inside that class set by the prior. Imperfect reference labels shrink measured separation in a computable direction, and selective verification biases sensitivity and specificity. Likelihood ratios are the transportable output, sufficient for every prior and unchanged between units with the same signal laws, whereas posteriors and fold increases over average risk depend on prevalence. Context, the infant's own baseline, additional measurement channels, and clinically indicated interventions are the means by which further information enters. Six neonatal mechanisms of confounding are analyzed: mixing over reserve, stable between-infant differences, the limits of effort-only respiratory measurement, artifact, correlated events that overstate evidence, and the distinction between mixed and uncertain causes, which requires longitudinal data. Detection delay scales inversely with the divergence between pre- and post-change laws. The paper proposes a two-layer report—Layer 1 giving the pathway and its route, Layer 2 giving likelihood ratios over signal-equivalence classes with the prior stated—and a reporting checklist. Heart-rate characteristics monitoring is discussed as an example of a class-level index that can be useful without being diagnostic. All numerical results come from a synthetic seven-etiology model; no clinical data are used, and the paper does not claim that pattern monitoring is useless or that any existing product overclaims.
Related
- Route, Not Score: A Decision-Theoretic Basis for Pathway-Labelled Early Warning in Neonatal Cardiorespiratory Monitoring
- Embedded Cardiorespiratory Monitoring Using a 77 GHz FMCW Radar: On-Chip Multi-Timescale Tiny 1D CNN and Whole-Node Power Evaluation
- Acute Cardiorespiratory and Metabolic Responses to Different Endurance Exercise Modalities in Sedentary College Students Based on Physiological and Perceptual Load Monitoring
- Improved cardiorespiratory fitness is associated with functional changes in hippocampus during episodic memory pattern completion.
- Development and validation of an integrated wearable system for continuous cardiorespiratory monitoring using photoplethysmography and optical respiratory detection
BibTeX
@misc{doucette2026pathway,
title = {{Pathway Is Not Diagnosis: Identifiability Limits of Cardiorespiratory Pattern Monitoring in Preterm Infants}},
author = {Doug Doucette},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23406730},
url = {https://zenodo.org/records/23406730},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Pathway Is Not Diagnosis: Identifiability Limits of Cardiorespiratory Pattern Monitoring in Preterm Infants
AU - Doug Doucette
PY - 2026
DA - 2026/10/16
PB - Zenodo
DO - 10.5281/zenodo.23406730
UR - https://zenodo.org/records/23406730
AB - This paper argues that a monitored cardiorespiratory signal carries a fixed amount of information about the cause of a preterm infant's apnea–bradycardia–desaturation events, and that this information is measured by the distance between the signal laws of the candidate etiologies. The author generalizes two propositions from his neonatal monograph—that etiologies producing the same pathway vector cannot be separated, and that context can change posterior odds—into a graded account. Total-variation distance quantifies how separable two etiologies are; no processing of the signal can increase it, so the error floor is fixed; and when laws coincide exactly, the finest resolvable diagnosis is a class of causes, with any split inside that class set by the prior. Imperfect reference labels shrink measured separation in a computable direction, and selective verification biases sensitivity and specificity. Likelihood ratios are the transportable output, sufficient for every prior and unchanged between units with the same signal laws, whereas posteriors and fold increases over average risk depend on prevalence. Context, the infant's own baseline, additional measurement channels, and clinically indicated interventions are the means by which further information enters. Six neonatal mechanisms of confounding are analyzed: mixing over reserve, stable between-infant differences, the limits of effort-only respiratory measurement, artifact, correlated events that overstate evidence, and the distinction between mixed and uncertain causes, which requires longitudinal data. Detection delay scales inversely with the divergence between pre- and post-change laws. The paper proposes a two-layer report—Layer 1 giving the pathway and its route, Layer 2 giving likelihood ratios over signal-equivalence classes with the prior stated—and a reporting checklist. Heart-rate characteristics monitoring is discussed as an example of a class-level index that can be useful without being diagnostic. All numerical results come from a synthetic seven-etiology model; no clinical data are used, and the paper does not claim that pattern monitoring is useless or that any existing product overclaims.
ER -