A prediction model of dementia conversion for mild cognitive impairment by combining plasma pTau181 and structural imaging features

Abstract

Aims

The early stages of Alzheimer's disease (AD) are no longer insurmountable. Therefore, identifying at-risk individuals is of great importance for precise treatment. We developed a model to predict cognitive deterioration in patients with mild cognitive impairment (MCI).

Methods

Based on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, we constructed models in a derivation cohort of 761 participants with MCI (138 of whom developed dementia at the 36th month) and verified them in a validation cohort of 353 cognitively normal controls (54 developed MCI and 19 developed dementia at the 36th month). In addition, 1303 participants with available AD cerebrospinal fluid core biomarkers were selected to clarify the ability of the model to predict AD core features. We assessed 32 parameters as candidate predictors, including clinical information, blood biomarkers, and structural imaging features, and used multivariable logistic regression analysis to develop our prediction model.

Results

Six independent variables of MCI deterioration were identified: apolipoprotein E ε4 allele status, lower Mini-Mental State Examination scores, higher levels of plasma pTau181, smaller volumes of the left hippocampus and right amygdala, and a thinner right inferior temporal cortex. We established an easy-to-use risk heat map and risk score based on these risk factors. The area under the curve (AUC) for both internal and external validations was close to 0.850. Furthermore, the AUC was above 0.800 in identifying participants with high brain amyloid-β loads. Calibration plots demonstrated good agreement between the predicted probability and actual observations in the internal and external validations.

Conclusion

We developed and validated an accurate prediction model for dementia conversion in patients with MCI. Simultaneously, the model predicts AD-specific pathological changes. We hope that this model will contribute to more precise clinical treatment and better healthcare resource allocation.