Aims Jon Dorling 24/9/07 This thesis set out to test whether it is possible to predict neurodevelopmental outcome at two years of age using data collected in the first 12 hours of life or during the entire admission of a premature newborn infant. Outcomes were tested separately (severe disability in survivors against other survivors) or in combination (death or severe disability against survival without severe disability) Methods The hypotheses were tested in three cohorts; the East Anglian Very Low Birthweight Database, the Trent Neonatal Survey 'ABC study' and the United Kingdom Trial of Oscillation (UKOS). After exploration of the cohort data quality, each was used in tum to develop a model for predicting outcome using data in the first 12 hours of life. The UKOS dataset was also used to test prediction using data available from the entire admission. Univariate analysis was used to determine which variables were associated with the outcome of interest. Logistic regression was then used to develop the models and ROC curve analysis performed to test the predictive ability of the models. Results Each of the three models for predicting the combined outcome of death or severe disability appeared to predict reasonably well with areas under the curve of 0.808, 0.793 and 0.798. Further testing however showed that these models were good at predicting death but that they were very poor at predicting disability. Data from the entire admission were successfully used to develop a model that predicted disability adequately in survivors (Az= 0.842) suggesting the importance of this time-period. Conclusions Further study is needed to determine whether additional data from before 12 hours of age would enable outcome prediction. Prediction of disability amongst survivors appears possible using data from the whole admission, and after testing in other cohorts, offers a number of future epidemiological uses. Supplied by The British Library - 'The world's knowledge'