Article Text
Abstract
Background TRACK-HD and PADDINGTON are both longitudinal studies whose aims included assessing the potential of various biomarkers as outcomes for clinical trials of new Huntington’s disease (HD) therapies. One such biomarker is caudate atrophy. This potential is dependent upon the distribution of change in patients, with the distribution of change in healthy controls being indicative of what a successful treatment might achieve.
Aims Appropriate statistical models may be fitted to repeated measures to predict the distribution of changes that would be seen over time intervals not utilised in the index study. TRACK-HD involved yearly follow up over 36 months whilst PADDINGTON assessed changes over shorter intervals. Our aim was to compare the distributions of observed 6, 9 and 15 month caudate atrophy seen in PADDINGTON with that predicted from TRACK-HD.
Methods/techniques Linear mixed models were fitted separately to data from the early-HD and control subjects in TRACK-HD to predict the mean and standard deviation of caudate atrophy rates over 6, 9 and 15 months. These predictions (and their 95% confidence intervals) were compared with the observed means and standard deviations from PADDINGTON and their confidence intervals. Predicted and observed effect sizes were also compared.
Results/outcome As assessed using confidence intervals, predictions from TRACK-HD and observed results in PADDINGTON were consistent with one another. Confidence intervals for predictions made from TRACK-HD were narrower than those for the analogous observed estimates from PADDINGTON. The predictions did not exhibit implausible patterns of behaviour over time that sometimes occurred with the observed PADDINGTON data, which were likely the result of chance variability.
Conclusions Appropriate statistical models,when used with repeated measures of caudate atrophy, have great utility in predicting the distribution of atrophy rates time over intervals shorter than those used in the index study.
- caudate atrophy
- predictive modelling
- effect sizes