Machine Learning Approach for Improved Longitudinal Prediction of Progression from Mild Cognitive Impairment to Alzheimer's Disease

Abstract

Mild cognitive impairment (MCI) is cognitive decline that can indicate future risk of Alzheimer's disease (AD). We developed and validated a machine learning algorithm (MLA), based on a gradient-boosted tree ensemble method, to analyze phenotypic data for individuals 55-88 years old (n = 493) diagnosed with MCI. Data were analyzed within multiple prediction windows and averaged to predict progression to AD within 24-48 months. The MLA outperformed the mini-mental state examination (MMSE) and three comparison models at all prediction windows on most metrics. Exceptions include sensitivity at 18 months (MLA and MMSE each achieved 0.600); and sensitivity at 30 and 42 months (MMSE marginally better). For all prediction windows, the MLA achieved AUROC ≥ 0.857 and NPV ≥ 0.800. With averaged data for the 24-48-month lookahead timeframe, the MLA outperformed MMSE on all metrics. This study demonstrates that machine learning may provide a more accurate risk assessment than the standard of care. This may facilitate care coordination, decrease healthcare expenditures, and maintain quality of life for patients at risk of progressing from MCI to AD.

Keywords: Alzheimer’s disease; disease progression; machine learning; mild cognitive impairment.

Figure 1

Diagram of inputs and outputs for the machine learning algorithm (MLA). Abbreviations: Alzheimer’s Disease Assessment Scale (ADAS), Functional Activities Questionnaire (FAQ), mini-mental state examination (MMSE).

Figure 2

(A) Attrition chart for the dataset used for modeling. We selected individuals receiving a mild cognitive impairment (MCI) diagnosis at their baseline visit in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study. Selection criteria included data availability to confirm MCI stability or progression to Alzheimer’s disease (AD). The filtered dataset was split 80/20 into training and test sets, respectively. (B) Examples of subjects’ diagnostic statuses/tags over the course of the study. The vertical red line is at month 12, the first prediction timepoint/window investigated, before which all subjects were filtered to have MCI (not AD).

Figure 3

The receiver operating characteristic (ROC) curves for AD class vs. non-AD class as given by the MLA and the MMSE classifier showing the performance of the MLA and the MMSE over time: (A) 12-month prediction window; (B) 24-month prediction window; (C) 48-month prediction window.

Figure 4

The AUROC (A), Sensitivity (B), and Specificity (C) vs. Prediction Window for the MLA and MMSE classifier across all 7 prediction windows showing the performance of the MLA and the MMSE over time.

Figure 5

The ROC curves for AD class vs. non-AD class as given by the averaged MLA for the 24–48 months lookahead timeframe, and the MMSE classifier at 48 months.