Arterial-to-end-tidal carbon dioxide gradient as a risk marker for recorded 90-day mortality in intracranial neurosurgery: a retrospective INSPIRE cohort study of patients with early blood gas monitoring

STUDY OBJECTIVE: To evaluate the first-hour arterial-to-end-tidal carbon dioxide gradient as a marker of recorded 90-day mortality, distinguishing association, component interpretation and incremental prediction. DESIGN: Retrospective cohort study. SETTING: INSPIRE v1.4.2, covering surgery from 2011 to 2020. PATIENTS: Adults undergoing their first eligible intracranial operation with arterial carbon dioxide tension (PaCO₂) recorded within 60 min of anesthesia start. End-tidal carbon dioxide (ETCO₂) was paired within ±5 min. MEASUREMENTS: The outcome was database-recorded all-cause death within 90 days. Firth regression evaluated the gradient without conditioning on either component. The revised principal association model included clinical covariates and contemporaneous mean arterial pressure, heart rate and minute ventilation. Component and internally validated prediction models were secondary. MAIN RESULTS: Among 1907 patients, 95 deaths were recorded. The clinical gradient odds ratio (OR) per 5 mmHg was 1.38 (95% confidence interval [CI], 1.09-1.73). The revised principal association model included 1830 patients and 92 events (OR, 1.34; 95% CI, 1.06-1.69). Conditional component estimates did not establish predominance of either component. Adding the gradient to clinical variables, contemporaneous physiology and ETCO₂ changed optimism-corrected area under the curve from 0.832 to 0.833, without improving the Brier score. CONCLUSIONS: A larger first-hour gradient was associated with recorded mortality in patients selected for early paired monitoring. Within the evaluated models and metrics, incremental prediction was limited. These findings do not establish a causal mechanism, treatment target or externally validated prediction tool.