Objective Cerebral oedema is a potentially devastating complication of diabetic ketoacidosis (DKA). The relationship between osmolar changes, acid-base changes and development of cerebral oedema during therapy is unclear. Design Retrospective cohort study on 53 children with severe DKA (mean pH at presentation 6.92 +/- 0.08). Cerebral oedema was diagnosed using neurological status, response to osmotherapy, and neuroimaging, and classified as: early (occurring <= 1 h after presentation, n=15), late (1-48 h, n=17) or absent (controls, n=21). The temporal profiles for various osmolar and acid-base profiles were examined using a random coefficients fractional polynomial mixed model, adjusted for known risk factors. Results The three groups could not be differentiated by demographic, osmolar or acid-base variables at presentation. All osmolar and acid-base variables showed non-linear temporal trajectories. Children who developed late onset oedema showed dramatically different temporal profiles for effective osmolality and glucose-corrected serum sodium (both p<0.001). Glucose-corrected sodium provided better qualitative discrimination, in that it typically fell in children who developed late oedema and rose in controls. The maximum between-group difference for both variables approximated the median time of clinical cerebral oedema onset. Blood glucose and acid-base temporal profiles did not differ between the groups. Late onset oedema patients received more fluid in the first 4 h, but this did not influence the osmolar or glucose-corrected sodium trajectories in a predictable fashion. Conclusions Glucose-corrected serum sodium may prove a useful early warning for the development of cerebral oedema in DKA.