Interpreting peripheral oxygen saturation variability in critical illness: A directional framework adjusted for hypoxia severity

Peripheral oxygen saturation ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> ) exhibits a complex pattern of fluctuations during hypoxia, which can be quantified using entropy measures. <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> entropy analysis provides insights into dynamic physiological regulation by non-invasively reflecting the body's capacity to adapt to internal or external physiological challenges. However, the interpretation of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> entropy alone is limited without contextualisation and the degree of physiological challenge encountered (e.g. the severity of hypoxia). This proof-of-concept retrospective study analysed continuous 1 Hz <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> recordings extracted from MIMIC-III dataset's Intensive Care Unit ICU patients with sepsis (n = 164), chronic obstructive pulmonary disease (COPD) (n = 58), acute liver failure (ALF) (n = 59), or cirrhosis (n = 169). Sample entropy was computed directly from raw 20-min <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> signals and normalised to mean <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> using directional parenclitic deviation (δ), derived from a healthy hypoxia-exposure reference dataset. Cox-regression models assessed 30-day ICU mortality. In sepsis, δ was significantly higher in non-survivors (hazard ratio (HR) = 2.20, P < 0.0001) and independently predicted 30-day mortality (HR = 1.79, P < 0.0001). δ was not predictive in the COPD, ALF and cirrhosis cohorts. Unlike other patient groups, the cirrhosis group demonstrated unexpected mean negative δ values, suggesting aberrant regulatory engagement, potentially related to the pathophysiology of hepatopulmonary syndrome. These findings demonstrate that δ provides physiological contexts to entropy-based <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:semantics><mml:msub><mml:mi>S</mml:mi> <mml:mrow><mml:mi>p</mml:mi> <mml:msub><mml:mi>O</mml:mi> <mml:mn>2</mml:mn></mml:msub> </mml:mrow> </mml:msub> <mml:annotation>${S_{{\mathrm{p}}{{\mathrm{O}}_2}}}$</mml:annotation></mml:semantics> </mml:math> analysis. By linking variability to the severity of hypoxia, this framework enables a more interpretable and a potentially clinically applicable biomarker of systemic regulation in critical illnesses. Future validation across diverse cohorts could support its potential to aid in personalised care within intensive care settings.