A key unresolved controversy in severe COVID-19 pneumonitis in pregnancy is the optimum timing of delivery and whether delivery improves or worsens maternal outcomes. We aimed to assess clinical data on every intensive care unit (ICU) day for pregnant and postpartum women admitted to the ICU with COVID-19, with a particular focus on the days preceding and following delivery. In this multicentre, nationwide, prospective and retrospective cohort study, we evaluated all pregnant women who were admitted to an ICU in Israel with severe COVID-19 pneumonitis from the 13th week of gestation to the 1st week postpartum. We excluded pregnant patients in which the ICU admission was unrelated to severe COVID-19 pneumonitis. We assessed maternal and neonatal outcomes and longitudinal clinical and laboratory ICU data. The primary overall outcome was maternal outcome (worst of the following: no invasive positive pressure ventilation [IPPV], use of IPPV, use of extracorporeal membrane oxygenation [ECMO], or death). The primary longitudinal outcome was Sequential Organ Failure Assessment (SOFA) score, and the secondary longitudinal outcome was the novel PORCH (positive end-expiratory pressure [PEEP], oxygenation, respiratory support, chest x-ray, haemodynamic support) score. Patients were classified into four groups: no-delivery (pregnant at admission and no delivery during the ICU stay), postpartum (ICU admission ≥1 day after delivery), delivery-critical (pregnant at admission and receiving or at high risk of requiring IPPV at the time of delivery), or delivery-non-critical (pregnant at admission and not critically ill at the time of delivery). From Feb 1, 2020, to Jan 31, 2022, 84 patients were analysed: 34 patients in the no-delivery group, four in postpartum, 32 in delivery-critical, and 14 in delivery-non-critical. The delivery-critical and postpartum groups had worse outcomes than the other groups: 26 (81%) of 32 patients in the delivery-critical group and four (100%) of four patients in the postpartum group required IPPV; 12 (38%) and three (75%) patients required ECMO, and one (3%) and two (50%) patients died, respectively. The delivery-non-critical and no-delivery groups had far better outcomes than other groups: six (18%) of 34 patients and two (14%) of 14 patients required IPPV, respectively; no patients required ECMO or died. Oxygen saturation (SpO2), SpO2 to fraction of inspired oxygen (FiO2) ratio (S/F ratio), partial pressure of arterial oxygen to FiO2 ratio (P/F ratio), ROX index (S/F ratio divided by respiratory rate), and SOFA and PORCH scores were all highly predictive for adverse maternal outcome (p<0·0001). The delivery-critical group deteriorated on the day of delivery, continued to deteriorate throughout the ICU stay, and took longer to recover (ICU duration, Mantel-Cox p<0·0001), whereas the delivery-non-critical group improved rapidly following delivery. The day of delivery was a significant covariate for PORCH (p<0·0001) but not SOFA (p=0·09) scores. In patients who underwent delivery during their ICU stay, maternal outcome deteriorated following delivery among those defined as critical compared with non-critical patients, who improved following delivery. Interventional delivery should be considered for maternal indications before patients deteriorate and require mechanical ventilation. None.