Observed and expected hospital deaths, and standardised mortality ratio for infants admitted to intensive care units in 1993 and 1994, adjusted for risk using gestation alone
| Neonatal intensive care unit | Observed deaths | Expected deaths | Total infants | p Value | Standardised mortality ratio (observed/ expected deaths) | 95% CI |
| Australian | 216 | 215.83 | 1628 | 0.991 | 1.00 | — |
| All Scottish | 205 | 121.58 | 923 | < 0.001 | 1.69 | 1.52 to 1.85 |
| Scottish tertiary | 172 | 101.12 | 775 | < 0.001 | 1.70 | 1.52 to 1.88 |
| Scottish non-tertiary | 33 | 20.46 | 148 | 0.006 | 1.61 | 1.21 to 2.02 |
A logistic regression model predicting hospital mortality was fitted using gestation as the explanatory variable in infants admitted to Australian neonatal intensive care units only. The model was then used to calculate the number of deaths expected in all Scottish neonatal intensive care units, and in Scottish tertiary and non-tertiary neonatal intensive care units respectively, using the Australian neonatal intensive care units as a reference. As in table 3, significantly more deaths were observed in Scottish neonatal intensive care units than in Australian neonatal intensive care units. Hence early treatment bias—that is, inappropriate early treatment leading to inappropriately increased initial illness severity in Australia—is unlikely to explain the excess mortality observed in table 3.