Validation of a 5‐Year Prognostic Model for Parkinson's Disease

Abstract Background A simple prognostic model was previously developed to predict the probability of recently‐diagnosed patients reaching negative outcomes (postural instability, dementia or death) in a 5‐year period. Objectives To validate this model in an independent cohort and establish utility at later time points. Methods Validation was performed using data collected in an incident cohort at baseline, 2 and 4 years. Predicted negative outcome probabilities were compared to actual 5‐year outcomes. Results The model, based on age, MDS‐UPDRS axial score and 60‐second animal fluency, predicted poor 5‐year outcome when applied at baseline, (area under the curve (AUC) 0.80), 2 years (AUC 0.82) and 4 years (AUC 0.71). Power calculations showed that selecting a subgroup with prognostic score >0.5 reduced the sample size required for a disease‐modifying trial. Conclusions This 5‐year prognostic model has good accuracy when employed up to 4 years from diagnosis and may help stratification for disease‐modifying trials.