Subgroup | Mortality | Extubation from mechanical ventilation | Hospital discharge | ||||||

Tocilizumab n (%) | No tocilizumab n (%) | Adjusted HR* (<1.0 favours tocilizumab) | Tocilizumab n (%) | No tocilizumab n (%) | Adjusted HR* (>1.0 favours tocilizumab) | Tocilizumab n (%) | No tocilizumab n (%) | Adjusted HR* (>1.0 favours tocilizumab) | |

Overall | 86 (46.7) | 2399 (60.0) | 1.21 (0.98–1.50) | 84 (45.7) | 1672 (41.8%) | 0.71 (0.57–0.88) | 50 (27.2) | 812 (20.3) | 0.66 (0.50–0.88) |

Subgroups | |||||||||

Time from start of mechanical ventilation to tocilizumab | |||||||||

≤2 days | 41 (43.2) | 2399 (60.0) | 1.08 (0.79–1.47) | 54 (56.8) | 1672 (41.8) | 1.24 (0.95–1.63) | 35 (36.8) | 812 (20.3) | 1.19 (0.85–1.66) |

3–10 days | 45 (50.6) | 2399 (60.0) | 1.59 (1.18–2.15) | 30 (33.7) | 1672 (41.8) | 0.46 (0.32–0.66) | 15 (16.9) | 812 (20.3) | 0.36 (0.22–0.61) |

Age group | |||||||||

<60 years | 35 (38.0) | 640 (44.3) | 1.52 (1.10–2.11) | 46 (50.0) | 732 (50.7) | 0.57 (0.42–0.76) | 28 (30.4) | 430 (29.8) | 0.52 (0.36–0.77) |

≥60 years | 51 (55.4) | 1759 (68.8) | 1.08 (0.81–1.43) | 38 (41.3) | 940 (36.8) | 0.84 (0.60–1.16) | 22 (23.9) | 382 (14.9) | 0.79 (0.51–1.22) |

Corticosteroid use | |||||||||

Yes | 76 (52.8) | 2332 (61.6) | 1.43 (1.14–1.79) | 56 (38.9) | 1519 (40.1) | 0.61 (0.47–0.80) | 27 (18.8%) | 706 (18.6%) | 0.49 (0.33–0.72) |

No | 10 (25.0) | 67 (31.5) | 0.48 (0.25–0.92) | 28 (70.0) | 153 (71.8%) | 0.62 (0.42–0.92) | 23 (57.5%) | 106 (49.8%) | 0.60 (0.38–0.94) |

Sensitivity analysis | |||||||||

1:1 Propensity score match† | 86 (46.7) | 70 (38.0) | 1.70 (1.21–2.38) | 84 (45.7) | 100 (54.4) | 0.78 (0.58–1.05) | 50 (27.2) | 59 (32.0) | 0.72 (0.49–1.08) |

*Adjusted for study site (Southern California vs Northern California), index month, sex, age, race/ethnicity, body mass index and time-dependent variables of receipt of remdesivir, convalescent plasma or corticosteroids. Weighted using inverse probability weighting to estimate average treatment effect for the treated. Cox model used maximum likelihood estimation method. Subgroup model of age <60 years adjusted for all factors above and for heart rate, systolic blood pressure, diabetes and Charlson Comorbidity Index as absolute standardised difference >0.2 after weighting. Subgroup model of age ≥60 years adjusted for all factors above and for diabetes and Charlson Comorbidity Index as absolute standardised difference >0.2 after weighting. Subgroup model of corticosteroid use adjusted for all factors above and for systolic blood pressure and Charlson comorbidity index as absolute standardised difference >0.2 after weighting. Subgroup model of no corticosteroid use adjusted for all factors above and for heart rate and Charlson Comorbidity Index as absolute standardised difference >0.2 after weighting.

†N=184:184. Propensity score match with optimal method, calliper=0.25, exact match on site. In addition to the variables adjusted above, 1:1 match model also adjusted for temperature since the propensity score match yielded a >0.2 absolute standardised difference.