Personalised Medicine in the Early Identification of Preclinical Cognitive Impairment. Development of a Predictive Risk Model.

The "Comprehensive Plan for Alzheimer's and other Dementias" shows that more than 50% of cases of cognitive impairment (CI) in population-based studies are undetected. The figure is particularly striking in the case of mild dementias, of which up to 90% are undiagnosed. The aim is to use the combined power of the integration of clinical, molecular, proteomic, genomic, care, social, environmental and behavioural data in patients, using advanced artificial intelligence techniques for data processing and analysis, in order to generate predictive models for the preclinical detection of CI in the population aged 55-70 years.

Multicentre, non-interventional, convergent mixed methods observational study, with a prospective observational design part and a qualitative design part. Sample recruited randomly among users of the public health system in the participating geographical locations. Data will be collected in 6 regions (Andalucia, Castilla-Mancha, Catalonia, Valencia, Madrid and the Basque Country) and their rural and urban Primary Care (PC) networks.

Non-institutionalised subjects, aged between 55 and 70 years, assigned to PC centres in the territories included in the study, with a "living history" (recorded in the last 12 months) and without an established diagnosis of CI.

A descriptive analysis of the characteristics of the population will be carried out using frequencies and percentages or measures of central tendency and dispersion, with their 95% confidence intervals. Baseline socio-demographic and clinical characteristics will be compared in order to study the homogeneity of the sample. For the comparison of qualitative variables, the Chi-square test or Fisher's exact test will be used and for the comparison of quantitative variables, the t-test or Wilcoxon test will be used. Logistic regression models are proposed to analyse health outcome factors associated with mild cognitive impairment. All models will include repeated measures for each individual. All models will adjust for different risk factors, and for those factors that may change over time, the interaction between time and that factor will be studied.

Initially, multivariate linear latent models will be used for the predictive model of cognitive impairment risk. The integration of data from multiple sources of information will be done using multivariate probabilistic models, in order to find a representation of the patient in a feature space influenced by all data sources (visits).