A Comparison of Multi-Attribute Approaches for Assessment of Mechanical Ventilation in ICU Patients Using MIMIC-IV Data
Author
Rosager, Signe Kirstine
Term
4. term
Publication year
2026
Abstract
This thesis compares two multi-attribute approaches to assessing the burden of invasive mechanical ventilation in the ICU: time-weighted average mechanical power (TWA-MP) and a Beacon-derived time-weighted average total penalty (TWA-TP). Adult patients in MIMIC-IV v3.1 who started invasive ventilation within 24 hours of their first ICU admission and remained ventilated for at least 48 hours were analyzed using data from 24–48 hours after initiation. Ventilator and laboratory measurements were used to compute TWA-MP and six physiologically grounded penalties (oxygenation, oxygen toxicity, barotrauma, acid-base, and respiratory muscle stress and atrophy for support modes), which were summarized as TWA-TP. Associations with ICU and in-hospital mortality were evaluated using logistic regression adjusted for age and sex, and discriminative performance was assessed with ROC-AUC. The final cohort comprised 4271 patients (mean age 60.4 years; 59.5% male); ICU and in-hospital mortality were 25.0% and 30.5%, respectively. Higher TWA-MP and TWA-TP were both significantly associated with mortality, with TWA-MP showing stronger adjusted associations than TWA-TP. However, discrimination was limited for both metrics (AUC around 0.54–0.57), indicating that they are not sufficient for clinical decision-making on their own. Overall, the findings suggest that both metrics capture aspects of ventilatory burden linked to mortality, while highlighting the need for further validation and integration with other clinical variables.
Dette arbejde sammenlignede to multi-attributmetoder til at vurdere belastningen ved invasiv mekanisk ventilation på intensiv: tidsvægtet gennemsnitlig mekanisk power (TWA-MP) og en Beacon-afledt tidsvægtet gennemsnitlig total penalty (TWA-TP). Voksne patienter i MIMIC-IV v3.1, der fik påbegyndt invasiv ventilation inden for 24 timer efter første ICU-indlæggelse og var ventileret i mindst 48 timer, blev analyseret med data fra 24–48 timer efter opstart. Ventilator- og laboratorieparametre blev anvendt til at beregne TWA-MP og seks fysiologisk forankrede penalties (oksygenering, ilttoksicitet, barotraume, syre-base samt respirationsmuskelstress og -atrofi ved støtteventilation), som blev summeret til TWA-TP. Associationer med ICU- og hospitalsmortalitet blev vurderet med logistisk regression justeret for alder og køn, og diskrimination blev vurderet med ROC-AUC. I alt indgik 4271 patienter (gennemsnitsalder 60,4 år; 59,5 % mænd); ICU- og hospitalsmortalitet var henholdsvis 25,0 % og 30,5 %. Højere TWA-MP og TWA-TP var begge signifikant associeret med dødelighed, og TWA-MP viste stærkere justerede associationer end TWA-TP. Den diskriminative evne var dog begrænset for begge mål (AUC omkring 0,54–0,57), hvilket indebærer, at metrikkerne alene ikke kan understøtte klinisk anvendelse. Samlet tyder resultaterne på, at begge metrikker afspejler aspekter af ventilatorisk belastning, som er forbundet med mortalitet, men at yderligere validering og kombination med andre kliniske faktorer er nødvendig.
[This apstract has been generated with the help of AI directly from the project full text]