[o1–12–03]: Modeling and Prediction of the Natural History of Neurodegeneration From Longitudinal Trial Data
Year: 2017
Type: article
Abstract: Disease progression modeling in Alzheimer's disease(AD) aims at 1) quantifying the relationship between different biomarkers, and 2) staging patients based on individual observations. However, clinical datasets generally lack of a well-defined temporal reference, since the onset of the pathology varies across individuals. In this study we propose data-driven probabilistic regression and time reparameterization for estimating and predicting the natural history of the disease from longitudinal trial data, accounting for different biomarkers, missing data and unknown disease staging. We collected longitudinal ADNI data for 154 individuals with baseline values of CSF Abeta amyloid<192 pg/ml (Table 1). The probabilistic regression was applied to normal individuals, mild cognitive impairment subjects converted to AD (MCI conv), and AD patients having at least one measurement for each of the following biomarkers: 1)hippocampal, ventricular, entorhinal, and whole brain volumes, 2)average normalized FDG uptake, 3)average normalized AV45 uptake, and 4)ADAS13, RAVLT, FAQ. The testing set was composed by the remaining subjects with at least a missing biomarker, as well as by the subgroup of non converted MCI (Table 2). The estimated progression spans across roughly 20 years, and at the initial stages shows high-levels of AV45, followed by abnormal ventricles and FDG uptake. Lately we observe increasing abnormality of the volumetric measures, and finally a steady worsening of neuropsychological scores (Fig.1). Healthy individuals are associated with the early stages of the pathology in both training and testing data, while MCI and AD are characterized by late stages (p<1e-4 for the group-wise comparison)(Fig.2). The temporal cut-off based on the 10th quantile of the staging in the training AD population(t=3.6 years) leads to a discrimination accuracy of 0.84 for MCI converters vs stable in the testing data. We proposed a data-driven statistical model of the natural history of the disease based on short-term clinical data, enabling novel applications beyond the state-of-art, such as the probabilistic prediction of disease staging in unseen patients. Furthermore, the model naturally accounts for missing data, and provides uncertainty quantification of the biomarker evolutions. The model provided remarkable modeling and predictive performance when tested on a large clinical cohort, and thus represents a promising instrument for the analysis of clinical trials data of disease modifying drugs. Modelled biomarker progression in Abeta amyloid positive individuals in the training group (solid/dashed lines: mean/sd). NL: normal individuals, MCI: mild cognitive impairment converted to AD, AD: Alzheimer's patients. The estimated progression spans across roughly 20 years, and at the initial stages shows high-levels of AV45, followed by abnormal ventricles and FDG uptake. Lately we observe increasing abnormality of the volumetric measures, and finally a steady worsening of neuropsychological scores. Posterior prediction for the individual time shift in the testing data. Healthy individuals are generally displaced at the early stages of the pathology, while the predictions for MCI and AD patients are associated with respectively intermediate and late progression stages. NL: normal individuals, MCI: mild cognitive impairment, AD: Alzheimer's patients. Healthy individuals are associated with the early stages of the pathology in both training and testing data, while MCI and AD are characterized by late stages (p<1e-4 for the group-wise comparison). The temporal cut-off based on the 10th quantile of the staging in the training AD population (t=3.6 years) leads to a discrimination accuracy of 0.84 for MCI converters vs stable in the testing data.
Source: Alzheimer s & Dementia
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4252574091