Background and aim We aimed to compare different statistical methods for predicting clinical progression, among individuals with Huntington’s disease (HD).
Methods We compared the following methods: (1) Multiple Linear Regression, (2) Linear Mixed Models with different covariance structures (no correlation, autoregressive(AR)), and (3) Transition Models (Panel Models). We built models to predict the most commonly used clinical trial outcomes: (1) motor function (measured by the Total Motor Score on the UHDRS) and (2) daily function (measured by the Total Functional Capacity). Potential predictors considered included Cytosine-Adenine-Guanine repeat expansion length, age of onset, and years since diagnosis. We used genetic and longitudinal clinical data collected for the Cooperative Observational Research Trial (COHORT) study, led by the Huntington Study Group (United States, Canada, and Australia, 2006–2011). Data was randomly split into a training set (2/3) and testing set (1/3). We selected predictors for models using the Bayesian Information Criterion. We compared the predictive accuracy of different models using the Mean Squared Prediction Error.
Results Overall, models that took into account longitudinal functional scores for individuals (Mixed Models with AR covariate structure, and Transition Model) performed better than Multiple Linear Regression. However the Prediction Intervals were wide. Under 17% of participants had data for four or more visits, limiting our ability to use higher order AR structures for the Mixed and Transition Models.
Conclusions Mixed and Transition models improved prediction of individual disease progression in HD compared to Multiple Linear Regression, although current level of predictive accuracy is not high enough for clinical use.
- disease modelling
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.