Blood gas ratio may reveal hidden ICU death risk

Copied word for word from the study's registry entry, so the wording is the study team's rather than ours.

Inclusion criteria Adult patients (≥ 18 years old) at time of ICU admission. Rationale: Restrict to adult physiology and consent/ethics comparability. First ICU stay during hospital admission (index ICU admission). Rationale: Avoid within-hospital dependency and repeated measures bias. ICU admission captured in the MIMIC-II dataset between the database start and end dates. Rationale: Use all eligible records within the dataset timeframe. At least one arterial blood gas (ABG) measurement with both PaO₂ and PaCO₂ available during the ICU stay. Rationale: Primary exposure is PaO₂/PaCO₂ ratio; require both values to compute it. Available outcome data for ICU mortality (discharge status from ICU). Rationale: Necessary to ascertain primary endpoint. Available SpO₂ and mean arterial pressure (MAP) measurements in the same ICU stay (for confounder adjustment). Rationale: Study aims to evaluate independence from SpO₂ and MAP; require at least one measurement of each during the stay. (If either is missing, see handling below.) Exclusion criteria Age \< 18 years. Rationale: Exclude pediatric physiology. Missing or unusable arterial blood gas values (PaO₂ or PaCO₂ absent or non-numeric) for the ICU stay. Rationale: Cannot compute exposure. No recorded ICU outcome (missing ICU discharge status) or lost/invalid mortality data. Rationale: Cannot determine primary outcome. Extreme or physiologically implausible ABG values that suggest data error (e.g., PaO₂ ≤ 0 mmHg, PaCO₂ ≤ 0 mmHg, or PaO₂ \> 1000 mmHg or PaCO₂ \> 300 mmHg - thresholds may be tightened after exploratory data review). Rationale: Remove likely data-entry errors. Document thresholds and perform sensitivity analysis. Patients on extracorporeal life support (ECMO) during ABG measurement (if identifiable in MIMIC-II). Rationale: ECMO profoundly alters gas exchange and PaO₂/PaCO₂ physiology; consider exclusion or separate subgroup. Repeat ICU admissions within the same hospital stay beyond the index admission (if you choose to restrict to single index admission). Rationale: Prevent correlated observations - include only the first ICU stay per hospital admission. ICU stays with no recorded SpO₂ or MAP at any time (if you insist on requiring these for adjustment). Rationale: If primary models must adjust for SpO₂ and MAP, exclude records without those covariates or plan appropriate imputation. Do-not-resuscitate (DNR)/comfort-care only status recorded before ABG measurement (optional exclusion depending on study question). Rationale: Treatment limitation orders can confound mortality associations; you may choose instead to adjust or analyze separately.