Objectives: To describe and classify individual trajectories of 15-day changes in delirium severity.

Methods: A longitudinal hospital-based study was carried out with 230 medical inpatients aged 65 and over admitted to St Mary's Hospital in Montreal, Canada, between 1996 and 1999, diagnosed with delirium at enrollment, and who had at least four measurements of delirium severity during the next 15 days. Delirium severity was assessed using the Delirium Index (DI). To classify patients' individual trajectories, we applied a new method that relies on principal factor analysis and cluster analysis. We used multiple linear regression to investigate if clusters were associated with DI scores measured at an 8-week follow-up. Multivariable Cox's proportional hazards regression was used to assess whether the clusters were associated with survival over the next 12 months.

Results: Individual patterns were classified into five clusters: Steady (n = 89, 38.9%), Fluctuating (n = 36, 15.7%), Worsening (n = 15, 6.6%), Fast Improve-ment (n = 26, 11.3%), and Slow Improvement (n = 63, 27.5%). The Fast Improvement cluster had much lower prevalence of dementia (38.5% vs. 55.6% to 77.8% in other clusters, p = 0.003). Subjects whose 2-week patterns were classified as Fast or Slow Improvement had a significantly lower DI at 8 weeks than those in the Steady or Fluctuating clusters. The Worsening cluster had the largest percentage of deaths. The Fast Improvement and Worsening clusters initially had a high risk of death in the first 2 weeks (adjusted relative risks of approximately 3 and 6, respectively) but that risk decreased rapidly thereafter.

Conclusion: Two-week trajectories of delirium severity were associated with short-term mortality and delirium severity at 8-week follow-up.