MRI and cerebrospinal fluid biomarkers for predicting progression to Alzheimer's disease in patients with mild cognitive impairment: a diagnostic accuracy study

Abstract

Objectives: To assess the incremental value of MRI and cerebrospinal fluid (CSF) analysis after a short memory test for predicting progression to Alzheimer's disease from a pragmatic clinical perspective.

Design: Diagnostic accuracy study in a multicentre prospective cohort study.

Setting: Alzheimer Disease Neuroimaging Initiative participants with complete data on neuropsychological assessment, MRI of the brain and CSF analysis.

Participants: Patients with mild cognitive impairment (MCI; n=181) were included. Mean follow-up was 38.9 months (range 5.5-75.9).

Main outcome measures: Diagnostic accuracy of individual instruments and incremental value of entorhinal cortex volume on MRI and p-τ/Aβ ration in CSF after administration of Rey's Auditory Verbal Learning Memory Test are calculated and expressed as the 'Net Reclassification Improvement' (NRI), which is the change in the percentage of individuals that are correctly diagnosed as Alzheimer or non-Alzheimer case.

Results: Tested in isolation, a short memory test, MRI and CSF all substantially contribute to the differentiation of those MCI patients who remain stable during follow-up from those who progress to develop Alzheimer's disease. The memory test, MRI and CSF improved the diagnostic classification by 21% (95% CI 15.1 to 26.9), 22.1% (95% CI 16.1 to 28.1) and 18.8% (95% CI 13.1 to 24.5), respectively. After administration of a short memory test, however, the NRI of MRI is +1.1% (95% CI 0.1 to 3.9) and of CSF is -2.2% (95% CI -5.6 to -0.6).

Conclusions: After administration of a brief test of memory, MRI or CSF do not substantially affect diagnostic accuracy for predicting progression to Alzheimer's disease in patients with MCI. The NRI is an intuitive and easy to interpret measure for evaluation of potential added value of new diagnostic instruments in daily clinical practice.

Keywords: Alzheimer; Mild cognitive impairment; Net reclassification improvement; diagnostic accuracy.

Figure 1

Receiver-operator characteristic-curves (ROC; upper panel) and Net Reclassification Improvements (lower panel) of Rey's Auditory Verbal Learning memory test (RAVLT), entorhinal cortex volume on MRI and p-τ/Aβ ration in cerebrospinal fluid (CSF) in participants with mild cognitive impairment (MCI). The upper panel depicts the ROC curves for the RAVLT test of memory (MEM), entorhinal cortex volumetry (MRI) and the p-τ/Aβ ratio in CSF contrasting MCI participants without progression to dementia on follow up (N=100) and those progression to Alzheimer's disease (N=81). The corresponding area under curves are provided in table 2. The lower panel shows the net reclassification improvement for the same comparisons as a result of performing a single test (left) or on the right, in the grey area, for the MRI and CSF examination, after having incorporated the result of memory testing.

Figure 2

Reclassification and Net Reclassification Improvement (NRI) of participants as no progression to Alzheimer's disease (AD) or progression to AD after a basic memory test (Rey's Auditory Verbal Learning memory test) followed by MRI (A) or cerebrospinal fluid (B) Illustration of the effects of memory testing and subsequent MRI on diagnostic classification of a group of 181 participants belonging either to a group of stable MCI patients (blue, N=100) or patients with MCI who progressed to AD during follow-up (red, N=81). For reasons of typographical clarity not all possible changes of diagnostic category are delineated.