Rigas F. Soldatos1*†
Dimitrios Kasselimis2,3†
Christina Parpoula2
Eleni Konidari4Vassilis Dimitriou1Everina Katirtzoglou1
Dimitris Kiosses5
Konstantinos Tsibanis6Anastasia Konsta7
Theofanis Vorvolakos8
Panagiotis Alexopoulos9,10
Antonios Politis1- 1First Department of Psychiatry, National and Kapodistrian University of Athens Medical School, Eginition Hospital, Athens, Greece
- 2Department of Psychology, Panteion University of Social and Political Sciences, Athens, Greece
- 3Neuropsychology and Language Disorders Unit, First Neurology Department, Eginition Hospital, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece
- 4Mental Health Services, School of Health Sciences, University of Patras, Patras, Greece
- 5Weill Cornell Institute of Geriatric Psychiatry, Weill Medical College of Cornell University, White Plains, NY, United States
- 6E-learning Services, National and Kapodistrian University of Athens, Athens, Greece
- 7Department of Psychiatry, School of Health Sciences, University of Thessaloniki, Thessaloníki, Greece
- 8Department of Psychiatry, School of Health Sciences, University General Hospital of Alexandroupolis, Democritus University of Thrace, Alexandroupolis, Greece
- 9Mental Health Services, University General Hospital of Patras, Patras, Greece
- 10Department of Medicine, School of Health Sciences, University of Patras, Patras, Greece
Introduction: Older individuals who suffer from mental disorders may encounter accessibility difficulties related to factors such as remoteness and socioeconomic status. The present analysis provides empirical evidence from the INTegRated InterveNtion of pSychogerIatric Care (INTRINSIC) and shows that this network could aid towards the incorporation of tele-psychiatry and tele-neuropsychology into primary healthcare. We propose that such integration, situated within comprehensive health digitalization initiatives, represents a scalable approach to expanding mental health access.
Methods: 1,143 individuals from 2022 to 2025, from 11 different sites of INTRINSIC were recruited. Data collection was facilitated via the HEllenic Remote MEntal health Services for old-age (HERMES) Digital Platform, including demographic information, Mini-Cog scores, as well as information based on the Old Age Behavioral Risk Factor Surveillance System (OLA-BRFSS). A machine learning (ML) model was developed, trained, and evaluated using nested cross-validation. The classification analysis outcome was the Mini-Cog scores and eighty-three known risk factors were analyzed. Features were selected using Elastic Net regularization. A Random Forest classifier was then trained on the selected feature, and was employed to classify individuals into two Mini-Cog cognitive performance groups.
Results: The ML algorithm employed in this study revealed eight features to be positively associated with low Mini-Cog scores, including subjective complaints of cognitive problems, retirement, polypharmacy, and history of falls. Five variables demonstrated a positive association with higher Mini-Cog scores, including prior diagnosis of an anxiety disorder, insomnia, and physical exercise. The model achieved a ROC-AUC of 0.76 (Figure 3 and Table 4), with a BAC of 0.65.
Discussion: The present paper presents the first large-scale study on INTRINSIC, including multiple sites and integrating psychiatric, cognitive, medical, as well as sociodemographic variables in state-of-the-art ML models. Our results add to the existing literature on the complex interrelationships of factors affecting cognitive status in older individuals. We propose that INSTRINSIC may function as a benchmark for integrating psychiatric and neuropsychological services within primary healthcare settings, thereby addressing disparities in access to care and diagnostic equity.
1 Introduction
Adults affected by age-related brain disorders are expected to increase significantly in the next decades. Mental disorders that affect older people (Skoog, 2011; Mendez and Cummings, 2003) pose detrimental effects on overall quality of life (Banerjee et al., 2006), adding to the increased prevalence of depression among older individuals (Abdoli et al., 2022). The interplay between mood disorders and pathological cognitive decline in dementia and Mild Cognitive Impairment (MCI) has been well-established (Bennett and Thomas, 2014; Rubin, 2018; Petersen and Negash, 2008), with genetic polymorphisms being connected to the observed comorbidity between Alzheimer’s Disease and Major Depressive Disorder (Brzezińska et al., 2020). Emerging evidence has revealed a complex framework of interaction between cognition, mood, and health status (Alexopoulos, 2005; Taylor, 2014). It has been argued that healthy aging can also result in cognitive slowing and difficulties in emotional regulation, which may lead to changes and challenges in everyday life (Woodard et al., 2012; La Corte et al., 2016; Pettigrew and Martin, 2014). There are other neurological and/or psychiatric disorders that can cause subtle cognitive deficits, such as cerebral microbleeds (Rosenbaum Halevi et al., 2019), leukoencephalopathy (Boyle et al., 2016; Van Den Berg et al., 2018), multiple sclerosis (Katsari et al., 2020), as well as anxiety disorders and/or mood disorders (Ma, 2020). Identifying “normal aging” may be one of the greatest challenges in modern clinical practice (Pertl et al., 2017; Poirier et al., 2021). Contemporary literature shows that any cognitive or psychological changes demonstrated by older individuals should be examined in a multifactorial framework, integrating biological, behavioral and psychosocial variables (Harada et al., 2013; Deary et al., 2009).
There are three main barriers of a multifactorial framework: geographical disparities of psychogeriatric services, reduced accessibility and resource allocation. Recent systematic reviews confirmed that the prevalence of cognitive decline in rural areas is higher than the prevalence in urban ones (Xue et al., 2018; Mollalo et al., 2025). The obstacles to healthcare access are widespread in old populations, with the consequence that old people with mental disorders and cognitive decline do not benefit from specialized mental and cognitive health services. The need to increase accessibility and methods of optimization has been suggested, which could eventually lead to equity (Levesque et al., 2013; Wang, 2012). In Greece, older patients with mental disorders such as depression and cognitive decline face great difficulties accessing specialized services and mental and cognitive health services are lacking in remote and rural areas. In terms of resources, there is an imbalance worldwide in mental health care resources between urban and rural areas. Greece has been examined in the past in a large-scale comparative study on European healthcare systems, and was found to have the lowest public health expenditure (as a percentage of the total health expenditure) and the highest out-of-pocket payments, among European countries (Wendt, 2009). It should be also noted that neuropsychology is a quite new discipline in Greece (Liozidou et al., 2023) and has not yet been incorporated into the national health system.
Taking into account that remoteness and insularity undermine access to older adults mental and cognitive health services, tele-health practices should be considered as a facilitator of remote assessment and treatment. Several attempts towards this goal have been made. Research shows that telepsychogeriatrics is of comparable efficacy compared to face-to-face treatment (Hagi et al., 2023) and is of high acceptability among clinicians and beneficiaries (Abuyadek et al., 2024; Sharma and Devan, 2023). Several researchers argue in favor of remote cognitive assessment via tele-neuropsychology, given the accessibility difficulties discussed above (Hewitt et al., 2022; Sperling et al., 2024). Data seem promising, showing efficacy of teleneuropsychology when compared to traditional in-person cognitive assessment (Alva et al., 2025; Monteiro et al., 2025), even with older healthy adults and patients with dementia (Hunter et al., 2021). Telehealth, including telepsychogeriatrics and teleneuropsychology, offers a solution to the healthcare accessibility problem. The cost of comprehensive neuropsychological and psychiatric services to all patients would be unbearable (Alzheimer’s disease facts and figures, 2024; Sweet et al., 2002). We thus find ourselves between Scylla—the detrimental effects of mental disorders and cognitive decline on people in low resource areas—and Charybdis—the monetary drainage of the healthcare system.
Community old age mental and cognitive health services in remote low resources areas may prove crucial in the delivery of health care, providing support to people with mental health problems closer to their homes. Remote healthcare, when incorporated into primary care, can achieve increased access to mental and cognitive health services in low-resource settings (Garg et al., 2022), in a broader framework of health digitalization (Majcherek et al., 2024).
A digital transformation and modernisation can help redesign and reorganise core community mental and cognitive health services moving towards a new place-based, integrated multidisciplinary team aligned with primary care networks. The aim of the study is to present detailed data from the INTegRated InterveNtion of pSychogerIatric Care (INTRINSIC) project. The INTRINSIC project was developed in order to address the mental and cognitive imbalance of healthcare needs of older adults residing in low-resources areas of Greece, through integrated digitally enabled mental and cognitive care services, easily accessible by all.
2 Methods
2.1 Sample cohort
The present analysis was developed, tested and validated using cross-sectional data from the INTRINSIC cohort (Politis et al., 2023; Aggeletaki et al., 2024). The diagnostic and treatment protocols adhered to the principles of the sixth revision of the Declaration of Helsinki and received approval from the Bioethics and Research Ethics Committee of the Eginition Hospital of the University of the National and Kapodistrian University of Athens (approval no. 1,036/31/12/2021, ΑΔΑ 6ΘΞ146Ψ8Ν2-1ΗΙ). Written informed consent was obtained from all beneficiaries or their legally authorized representatives prior to being enrolled in INTRINSIC. All analyses included 1,143 individuals enrolled in the study from 2022 to 2025. Participant characteristics, including demographics and descriptive statistics, are presented in Table 1.
Table 1. Participant characteristics for the INTRINSIC cohort.
2.2 INTRINSIC study
The recruitment process and study design of INTRINSIC are detailed in previous publications (Politis et al., 2023; Aggeletaki et al., 2024). In brief, an ecosystem of innovative and easy to use information systems has been developed and actively supported, which functions as a single Digital Platform. This technological pillar of the project aims at the electronic interconnection of benefiting populations in rural areas with specialized scientists in psychogeriatric with the important mediation of doctors in remote Health Centers (Figure 1).
Figure 1. Map of participating healthcare units: primary healthcare units (
), tertiary coordination centers (
), and the home care healthcare unit in Tripoli (
).
The HEllenic Remote MEntal health Services for old-age (HERMES) Digital Platform1 integrates innovative Remote Mental Health Services and supports interoperability with other external applications and services. More specifically, it integrates: (a) an innovative System of Telehealth and a Surveillance of Mental and Cognitive Health Risk Factors Survey, (b) a sophisticated e-training system for the members of the project offering self-paced learning courses and webinars in state of the art psychogeriatric topics, a set of sensory health assessments, the Modified Problem Adaptation Therapy (M-PATH) for old-age patients with depression and their carers and (c) an open to the public information and awareness raising materials about mental and cognitive health of old individuals.
As mentioned above, in addition to demographic and medication data, cognitive, behavioral, physical and mental health risk factors are also collected through HERMES.
The Old Age Behavioral Risk Factor Surveillance System (OLA-BRFSS) is a premier system of mental health-related digital survey that collects data from remote Health centers, about Greek old age residents regarding their mental and cognitive health and use of health services. The OLA-BRFSS was established in 2021 as part of the Integrated Psychogeriatric Care Project funded by Hellenic Ministry of Health and designed and implemented in 2022 by the 1st Department of Psychiatry in 8 remote health centers in Greece. By collecting behavioral health risk data at the 11 remote health and community centers, OLA-BRFSS will become a powerful tool for targeting and building integrated mental health promotion activities. The OLA-BRFSS digital survey includes a 20-item questionnaire of clinical significance in older adults: polypharmacy, loneliness, emotional support, hearing loss, falls, weight loss, insomnia, cognitive complaints, traumatic/stressful events, depressive symptoms, anxiety symptoms and cognitive impairment. Psychopathological and cognitive data is collected through the Patient Health Questionnaire-2 (PHQ-2; Kroenke et al., 2003), the Generalized Anxiety Disorder-2 (GAD-2; Sapra et al., 2020) and the Mini-Cog (Fage et al., 2021).
Mini-Cog scores were dichotomized using a threshold of ≥3 (Kroenke et al., 2003; Sapra et al., 2020), for statistical analysis purposes, facilitating the use of advanced classification techniques. Participant characteristics according to Mini-Cog Binary Class stratification are described in Table 2.
Table 2. Participant characteristics for the INTRINSIC cohort stratified by Mini-Cog binary class.
2.2.1 Study variables and outcome
Eighty-eight known risk factors collected through HERMES at initial assessment were analyzed (Supplementary Table S1). These included medical conditions such as hypertension, obesity, infections, heart disease, dyslipidemia, stroke, and Parkinson’s disease. Psychiatric and neuropsychological conditions included a diagnosis of depression, psychosis, anxiety disorder, bipolar disorder, MCI and dementia. Medication history encompassed the use of antidiabetics, antiarrhythmic medications, antihypertensives, thyroid medications, benzodiazepines, lithium, antidepressants, and antipsychotics. PHQ-2 and GAD-2 scores were used in binary form for the Machine Learning (ML) analysis with a cut-off of ≥3. Finally, all items from the previously described 20-item questionnaire documenting clinically significant factors in older adults were included. Further to sample statistics, the classification analysis outcome were the Mini-Cog scores.
2.2.2 Data preprocessing and ML pipeline
Previously, Artificial Intelligence (AI) methods have been applied in healthcare with the aim of supporting decision-making by improving the effectiveness of services through the use of individualized data (Khosravi et al., 2024; Hobensack et al., 2023). The machine learning (ML) model was developed, trained, and evaluated on the INTRINSIC cohort using nested cross-validation (CV), a robust approach that supports both hyperparameter tuning and model selection on a dataset using single-patient cohorts (Cearns et al., 2019). In this setup, the inner CV performs hyperparameter selection, while the outer CV provides an unbiased estimate of model accuracy with CV-based hyperparameter tuning. Because hyperparameter search only accesses a subset of the data defined by the outer CV folds, the risk of overfitting is minimized, yielding a less biased estimate of the tuned ML model’s performance on the dataset (Cawley, 2010).
The pipeline began with data imputation using Multivariate Imputation by Chained Equations (MICE; Azur et al., 2011) implemented via an iterative imputation procedure, which reconstructs missing values by iteratively estimating each incomplete variable from the others until convergence (Zhang, 2016). Imputation models were fitted using Bayesian ridge regression with five imputation iterations and posterior sampling enabled to account for uncertainty. A fixed random seed ensured reproducibility. To prevent information leakage, imputation was performed exclusively within training data for each cross-validation fold. Continuous variables were then standardized to zero mean and unit variance to improve predictive performance (Bishop, 1995). To reduce dimensionality and address multicollinearity, features were selected using Elastic Net regularization, which combines LASSO (L1) and Ridge (L2) penalties to retain important variables while shrinking correlated coefficients (Zou and Hastie, 2005). Models were fitted across a range of L1 mixing parameters (0.7, 0.9, and 1.0) and 200 penalty strengths, using 10-fold cross-validation to select optimal regularisation parameters. To determine the optimal number of predictors, candidate feature set sizes (5, 10, 15, 20, 30, and 50 predictors) were evaluated. For each candidate size, a Random Forest classifier was trained using stratified five-fold cross-validation, and mean area under the receiver operating characteristic curve (ROC AUC) was computed. The feature set size yielding the highest mean ROC AUC was selected for final modelling. A Random Forest (RF) classifier was then trained on the selected features. RF builds an ensemble of decision trees using bagging (bootstrap aggregation), enhancing accuracy, preventing overfitting, enabling automatic feature selection, and generating robust ML models (Zou and Hastie, 2005). It can detect anomalies, identify important features, uncover patterns, and provide insightful graphics. In this study, RF was employed to classify individuals into two Mini-Cog cognitive performance groups. The outer loop consisted of repeated stratified cross-validation to estimate generalisation performance. The inner loop employed stratified 10-fold cross-validation for hyperparameter optimisation using randomised search. Hyperparameters included tree depth, number of trees, node size constraints, and feature subsampling strategies. Class imbalance was addressed using inverse-frequency class weighting. All preprocessing steps were refitted within each training fold. Table 3 summarizes the analytical steps of the Mini-Cog group classification procedure.
Table 3. Summary of the Mini-Cog group classification procedure.
The entire process was implemented in a reproducible pipeline (Figure 2). The pipeline was specifically structured to prevent information leakage between training and testing phases. Hyperparameter tuning was performed for RF within in the inner CV loop using 10-fold CV repeated 10 times (Malakouti et al., 2023), optimizing parameters such as regularization strength (Elastic Net; Supplementary Table S2) and number of trees (RF). Model performance was assessed using the Area Under the Receiver Operating Characteristic Curve (ROC-AUC), a measure that quantifies the model’s ability to distinguish between classes across classification thresholds, with values near 1 indicating excellent discriminative power (Cook, 2007). Model performance was further assessed using accuracy, balanced accuracy, F1 score, sensitivity, specificity, predictive values, likelihood ratios, and diagnostic odds ratios. Ninety-five per cent confidence intervals for key metrics (ROC AUC, accuracy, balanced accuracy, and F1 score) were estimated using non-parametric bootstrap resampling (2,000 iterations) of pooled out-of-fold predictions. All random seeds, hyperparameters, and feature selection decisions were recorded to ensure full reproducibility.
Figure 2. RF model building and Mini-Cog group generation. The dataset was imputed, standardized, and reduced to the selected features using elastic net following hyperparameter tuning. Using the selected hyperparameter set, 10 trials of a 10-fold nested CV were performed to generate Mini-Cog grouping and associated metric values. These values were then used to produce RF plots and metric (ROC–AUC) plots.
A similar rationale for the aforementioned approaches to developing and evaluating ML models for dementia prediction can be found in Nori et al. (2019) and Twait et al. (2023), and references therein. For a recent review of algorithms using electronic health record (EHR) data to identify patients with Alzheimer’s disease and related dementias (ADRD) and to advance their use in research and clinical care, see Walling et al. (2023).
All analyses were performed using Python 3.11 using scikit-learn, scipy, statsmodels, pandas, and matplotlib libraries. The full code used in the analyses is available from the corresponding author upon reasonable request.
3 Results
3.1 Sample descriptive characteristics
The total number of individuals included in the analyses was 1,143, of whom 359 (32%) were male and average age was 72.26 years (SD = 10.02). Regarding physical illnesses, 720 (65%) participants had a diagnosis of hypertension, 602 (54%) had dyslipidemia and 337 (30%) had heart disease. With respect to psychiatric and cognitive disorders, 81 (7%) had a prior diagnosis of dementia, 424 (38%) had a diagnosis of anxiety, and 389 (35%) had a diagnosis of depression. Mean PHQ-2 was 1.59, with 288 individuals (26%) scoring above the cut-off of 3. The mean GAD-2 score was 1.34, while 226 individuals (20%) scoring above the cut-off of 3. Finally, 216 (16%) participants had a Mini-Cog below the cut-off score, indicating possible cognitive decline. See Table 1 for details.
Statistical comparisons between the two Mini-Cog groups (see Table 2), revealed significant differences. Individuals in Group 2 (Mini-Cog <3) were older, had a lower level of education, were more likely to have a diagnosis of dementia and take related medication, were more likely to take diuretics, were more likely to be on more than five medications (polypharmacy; Pazan and Wehling, 2021), and subjective cognitive complaints were more frequently reported. Finally, traumatic and stressful events were more often reported by individuals whose Mini-Cog performance was above the cut-off score.
3.2 Binary Mini-Cog classification analysis
To perform an in-depth, multivariate classification analysis, a Random Forest (RF) classifier was applied (Breiman, 1996). RF is a supervised learning classification algorithm that consists of multiple classifiers, called decision trees, built independently. The decision about which class each participant belongs to is based on the majority of the results of these classifiers (i.e., majority voting). RF is particularly suitable for large datasets due to its ability to handle numerous input variables, manage missing data, and identify variable associations. Additionally, it provides data insights through variable importance rankings (Shaik and Srinivasan, 2019) and has been successfully applied in psychiatry, psychology and the medicine (Sarica et al., 2017; Bracher-Smith et al., 2020; Dimitriadis and Liparas, 2018; Del Casale et al., 2023; Kaur and Sharma, 2019; Lima and De Castro, 2019; Yarkoni and Westfall, 2017; Wallace et al., 2023; Speiser et al., 2019; Prinzi et al., 2024). The ML pipeline used is illustrated in Figure 2.
Fifteen variables out of eighty-three were selected by the Elastic Net (Supplementary Table S3), following a process of hyperparameter optimization, with positive or negative weights assigned to each. Positive weights favored Group 2 (Mini-Cog <3), whereas negative weights favored Group 1 (Mini-Cog ≥3). This set of variables was then used by the RF algorithm and they were ranked according to their relative importance from most important to least important (Elastic Net weighting), as follows: (1) use of antidementia medication (+), (2) prior diagnosis of dementia (+), (3) subjective complaint of recent memory (+), (4) subjective complaint of mental clouding (+), (5) being retired from work (+), (6) prior diagnosis of an anxiety disorder (−), (7) documented polypharmacy (+), (8) hearing problems (+), (9) the use of diuretic medication (+), (10) experiencing stressful events (−), (11) complaints of insomnia symptoms (−), (12) ability to provide personal means of transport (−), (13) a history of falls (+), (14) ability to exercise regularly (−), and (15) living in the same household with their children (+).
The model achieved a ROC-AUC of 0.76 (Figure 3 and Table 4), with a balanced accuracy (BAC) of 0.65. A negative predictive value (NPV) of 0.87 indicates strong performance in identifying Group 1 cases, whereas a positive predictive value (PPV) of 0.40 reflects limited precision in confirming Group 2 classification. All performance metrics are included in Table 4. Learning curve diagram, variability of selected features and confusion matrix can be found in Supplementary Figures S1–S3.
Figure 3. ROC curve.
Table 4. Outer loop cross validation metrics.
4 Discussion
The INTRINSIC study has thus far enabled access to a substantial community population that underwent neurocognitive screening with the aim to identify individuals exhibiting cognitive decline. A multitude of factors were systematically collected at this stage, to investigate possible associations with neurodegenerative disorders. Cross-sectional analysis of correlations between these factors and cognitive functioning can aid towards organizing timely therapeutic interventions, also revealing interactions with demographic and clinical features. Central to this effort is the integrated electronic system utilized, providing a digital hub for communication, information, data management and patient monitoring. Through this framework clinicians are able to identify low prevalence conditions such as Parkinson’s disease, and explore the fine grain of clinical conditions, including late and early depressive disorder.
The current project can provide clinically useful information based on descriptive statistics alone. For example, there is a clear sex bias, i.e., there were much more women visiting the sites of INTRINSIC, compared to men. Even though this has been reported in the literature and could be attributed to several factors (Bertakis et al., 2000; Barsky et al., 2001), the identification of such a sex-dependent imbalance in healthcare utilization, can serve as a criterion for planning group-focused awareness-raising activities for the community. Similarly, differences between sites with regard to mean age of beneficiaries or proportions of actually cognitively impaired individuals visiting the clinical setting (in contrast to those who had subjective complaints without any objective signs of impairment), could be also helpful in that direction. Moreover, we have found a significant proportion of individuals who had an indication of cognitive decline, despite lack of prior dementia diagnosis. In a recent meta-analysis, Lang and colleagues have shown that the prevalence of undiagnosed dementia is quite high, related to several factors including socioeconomic status, sex, educational level, and age (Lang et al., 2017). In our sample, a considerable portion of participants had lower education. Thus, the indications of undetected dementia could be attributed to less years of formal schooling, among other factors. This has been already documented in the literature (Savva and Arthur, 2015; Amjad et al., 2018) and poses a significant impediment to diagnostic equity. That is where projects like INTRINSIC could aid in amending such inequalities, by raising awareness via educational activities and detecting patients with dementia in a primary care framework.
The ML algorithm employed in this study revealed some interesting factors that are associated with the Mini-Cog scores. Eight features were found to be positively associated with low Mini-Cog scores, namely, a subjective complaint of recent memory impairment, a subjective complaint of mental clouding, being retired from work, polypharmacy, hearing impairment, diuretic medication use, history of falls, and residing with adult children to live in the same household. All of these have previously been recognized to participate in the trajectory of dementia (Mendonça et al., 2016; Basagni and Navarrete, 2025; Hallam et al., 2021; Dufouil et al., 2014; Leelakanok and RR, 2019; Powell and Reed, 2024; Lad et al., 2024; Liu et al., 2022; Tully et al., 2016; Fernando et al., 2017; Desai et al., 2020). The convergence of machine learning-derived results with existing evidence base strengthens the validity of these associations and underscores the multifactorial nature of cognitive decline.
Five variables demonstrated a positive association with higher Mini-Cog scores. A prior diagnosis of an anxiety disorder, experiencing stressful events, self-reported symptoms of insomnia, access to personal means of transport and being able to exercise physically regularly. The protective effects of maintained mobility and regular physical activity against cognitive decline have been well-established in previous research (Alty et al., 2020; Ahlskog et al., 2011). Interestingly, previous studies have found that clinical anxiety, stress and insomnia are positively correlated to dementia (Santabárbara et al., 2019; Gulpers et al., 2016; de Almondes et al., 2016; Hung et al., 2018). It could be argued, however, that unlike previous investigations that primarily examined clinical cohorts, our sample stemmed from community individuals, representing a different demographic profile, potentially affecting these variables. The results of this study align with previous research demonstrating negative associations between anxiety, stressful life events, and Alzheimer’s dementia incidence, though findings in this area remain mixed (Franks et al., 2021; Santabárbara et al., 2019). These findings suggest that the relationship between these features and the progression of cognitive function may be more context-dependent than previously understood, requiring further investigation of the underlying mechanisms. Integrating these findings in clinical settings would require further exploring the application of the ML algorithm in a variety of populations to test replicability and generalizability of these results. This process includes several caveats, such as access to care, data noise, missing values and indication bias (Khosravi et al., 2024).
In the context of primary care, efficient and reliable predictive models for cognitive decline are particularly valuable because clinicians often face constraints in time, resources, and patient burden. The INTRINSIC project, which assesses a wide range of cognitive health–related risk factors—including social and emotional support, lifestyle behaviors, sensory impairments, mental health history, and direct cognitive screening, among others—provides a rich multidimensional dataset that can inform early dementia detection. Within this framework, efficient study design is crucial given the high cost of data collection. As a future direction, we aim to leverage this rich, multidimensional dataset to develop a database-driven design selection scheme that integrates metaheuristics and data mining to identify the most influential variables for MiniCog prediction and early dementia detection. This approach would enable optimal design retrieval from such a large observational dataset, enhancing model precision, interpretability, and clinical relevance. Building on the Design of Experiments (DOE) paradigm (Pumplün et al., 2005), we plan to identify the most informative observations and features, initially using rule mining techniques to uncover hidden relationships, followed by the application of a genetic algorithm to construct an optimal supersaturated design that isolates the vital few variables with run-size economy (Parpoula et al., 2014). This alternative approach to variable selection (Parpoula et al., 2014), applied to a multidimensional database of observations, enables the utilisation of only a small fraction of the available runs, making statistical analysis of large databases computationally feasible and cost-effective. From a primary care perspective, the ultimate goal is to identify a parsimonious set of predictors that maintain predictive accuracy while minimizing the number of assessments required. In this regard, the planned future approach of database-driven design selection aligns closely with clinical priorities. Moreover, this strategy could support both the identification of important individual features and the exploration of meaningful feature combinations, paving the way for robust, interpretable, and clinically actionable models for cognitive health screening. Such models could inform clinical support systems focused on predictive classification and early intervention.
There are a few limitations of this study. First, due to the nature of the data collection pipeline, there are key data missing for some participants (as shown in Table 1), which are impossible to retrieve because of anonymization. Related to that, we should also mention possible measurement errors due to multiple assessors in various sites. In addition, all individuals included were of Greek descent, this leads to exclusion of minority populations. Sites are distributed throughout the mainland and the Greek islands, with recruitment numbers varying depending on various factors, such as the size of the local population and enrollment strategy. Second, there is a lack of comprehensive neuropsychological assessment and full, structured psychiatric evaluation. However, we argue that, at least to some degree, this is considered an inevitable drawback, when attempting to establish a low-resource, multi-site, primary care network. In this sense, a low-cost, broad and quick evaluation can help us identify the specific individuals in need of a thorough psychiatric and/or neuropsychological assessment, thus protecting resources, which are already low in such sites. Third, there is a sample bias, regarding the manner in which beneficiaries visited the sites. Nevertheless, with advancing time, and with more data accumulated, subsamples will be more representative of the populations the sites correspond to. Regarding the ML methodology, (Bleeker et al., 2003; Ramspek et al., 2021) as found in prior studies, the use of a multivariate model in a cross-sectional context could allow for the identification of a potential signature of the factors that correlate to cognitive functioning in this patient cohort (Zhou et al., 2023; Li et al., 2023). As indicated by the relatively low PPV value, further exploration of the generalizability of this model could allow for adaptation and potential application in the clinical context. An external validation sample would provide robust evidence of replicability of results, testing the algorithm’s reproducibility. In this case, the nature of the data collected used impedes efforts to identify a compatible external sample. Albeit the large number of the sample allows for a nested CV method to provide an indication of the model’s external performance (Bleeker et al., 2003). Once a predictive algorithm is found to successfully replicate in an external sample, a cohort with different patient characteristics can be utilized to test transportability (Ramspek et al., 2021).
In conclusion, this paper presents for the first time a large-scale study on INTRINSIC, including multiple sites and integrating psychiatric, cognitive, medical, as well as sociodemographic variables in state-of-the-art ML models. Our findings provide insights into the complex interrelationships of factors affecting cognitive status in old individuals. We further argue that INTRINSIC can serve as a benchmark for the incorporation of psychiatric and neuropsychological care of old people, especially those suffering from dementia or MCI, in the broader framework of primary healthcare, in an effort to remediate the issues of access and diagnostic inequality discussed in this paper.
Data availability statement
The datasets presented in this article are not readily available because the original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding authors. Requests to access the datasets should be directed to YXBvbGl0aXNAbWVkLnVvYS5ncg==.
Ethics statement
The studies involving humans were approved by Bioethics and Research Ethics Committee of the Eginition Hospital of the University of the National and Kapodistrian University of Athens (approval no. 1,036/31/12/2021, ΑΔΑ 6ΘΞ146Ψ8Ν2-1ΗΙ). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
RS: Writing – original draft. DKa: Writing – original draft. CP: Writing – original draft. EKl: Writing – review & editing. VD: Writing – review & editing. EKv: Writing – review & editing. DKl: Writing – review & editing. KT: Writing – review & editing. AK: Writing – review & editing. TV: Writing – review & editing. PA: Writing – review & editing. AP: Writing – original draft, Writing – review & editing.
Funding
The author(s) declared that financial support was received for this work and/or its publication. INTRINSIC is funded and supported by the Greek Ministry of Health and Social Solidarity.
Acknowledgments
The authors extend their profound appreciation to the head physicians of the primary healthcare centers of Andros, Erymanthia, Halandirtsa, Karditsa, Katouna, Loutraki, Soufli, Tinos, and Xanthi. General Hospital of Siros and the Institute of Healthy Ageing “Mentor” and the Municipality of Tripoli for their invaluable support in the implementation of INTRINSIC. Rigas Soldatos, Dimitris Kasselimis and Antonios Politis are Topic Editors for Frontiers.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors RS, DK and AP declared that they were an editorial board member of Frontiers at the time of submission. This had no impact on the peer review process and the final decision.
Generative AI statement
The author(s) declared that Generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2025.1696407/full#supplementary-material
Footnotes
References
Abdoli, N., Salari, N., Darvishi, N., Jafarpour, S., Solaymani, M., Mohammadi, M., et al. (2022). The global prevalence of major depressive disorder (MDD) among the elderly: a systematic review and meta-analysis. Neurosci. Biobehav. Rev. 132, 1067–1073. doi: 10.1016/j.neubiorev.2021.10.041,
Abuyadek, R. M., Hammouda, E. A., Elrewany, E., Elmalawany, D. H., Ashmawy, R., Zeina, S., et al. (2024). Acceptability of tele-mental health services among users: a systematic review and Meta-analysis. BMC Public Health 24, 1–18. doi: 10.1186/s12889-024-18436-7,
Aggeletaki, E., Stamos, V., Konidari, E., Efkarpidis, A., Petrou, A., Savvopoulou, K., et al. (2024). Telehealth memory clinics in primary healthcare: real-world experiences from low-resource settings in Greece. Front. Dement. 3:1477242. doi: 10.3389/frdem.2024.1477242,
Ahlskog, J. E., Geda, Y. E., Graff-Radford, N. R., and Petersen, R. C. (2011). Physical exercise as a preventive or disease-modifying treatment of dementia and brain aging. Mayo Clin. Proc. 86, 876–884. doi: 10.4065/mcp.2011.0252,
Alexopoulos, G. S. (2005). Depression in the elderly. Lancet 365, 1961–1970. doi: 10.1016/S0140-6736(05)66665-2,
Alty, J., Farrow, M., and Lawler, K. (2020). Exercise and dementia prevention. Pract. Neurol. 20, 234–240. doi: 10.1136/practneurol-2019-002335,
Alva, J. I., Brewster, R. C., Mahmood, Z., Harrell, K. M., Kaiser, N. C., Riesthuis, P., et al. (2025). Are tele-neuropsychology and in-person assessment scores meaningfully different? A systematic review and meta-analysis. Clin. Neuropsychol. 39, 1037–1072. doi: 10.1080/13854046.2025.2493343,
Alzheimer’s disease facts and figures (2024). 2024 Alzheimer's disease facts and figures. Alzheimers Dement. 20, 3708–3821. doi: 10.1002/alz.13809,
Amjad, H., Roth, D. L., Sheehan, O. C., Lyketsos, C. G., Wolff, J. L., and Samus, Q. M. (2018). Underdiagnosis of dementia: an observational study of patterns in diagnosis and awareness in US older adults. J. Gen. Intern. Med. 33, 1131–1138. doi: 10.1007/s11606-018-4377-y,
Azur, M. J., Stuart, E. A., Frangakis, C., and Leaf, P. J. (2011). Multiple imputation by chained equations: what is it and how does it work? Int. J. Methods Psychiatr. Res. 20, 40–49. doi: 10.1002/mpr.329,
Banerjee, S., Smith, S. C., Lamping, D. L., Harwood, R. H., Foley, B., Smith, P., et al. (2006). Quality of life in dementia: more than just cognition. An analysis of associations with quality of life in dementia. J. Neurol. Neurosurg. Psychiatry 77, 146–148. doi: 10.1136/jnnp.2005.072983,
Barsky, A. J., Peekna, H. M., and Borus, J. F. (2001). Somatic symptom reporting in women and men. J. Gen. Intern. Med. 16, 266–275. doi: 10.1046/j.1525-1497.2001.016004266.x,
Basagni, B., and Navarrete, E. (2025). Suggestions for the decision making in subjective cognitive complaints. Aging Clin. Exp. Res. 37, 1–5. doi: 10.1007/s40520-024-02875-8,
Bennett, S., and Thomas, A. J. (2014). Depression and dementia: cause, consequence or coincidence? Maturitas 79, 184–190. doi: 10.1016/j.maturitas.2014.05.009,
Bertakis, K. D., Azari, R., Helms, L. J., Callahan, E. J., and Robbins, J. A. (2000). Gender differences in the utilization of health care services. J. Fam. Pract. 49, 147–147,
Bleeker, S. E., Moll, H. A., Steyerberg, E. W., Donders, A. R. T., Derksen-Lubsen, G., Grobbee, D. E., et al. (2003). External validation is necessary in prediction research:: a clinical example. J. Clin. Epidemiol. 56, 826–832. doi: 10.1016/j.ijmedinf.2022.104978
Boyle, P. A., Yu, L., Fleischman, D. A., Leurgans, S., Yang, J., Wilson, R. S., et al. (2016). White matter hyperintensities, incident mild cognitive impairment, and cognitive decline in old age. Ann. Clin. Transl. Neurol. 3, 791–800. doi: 10.1002/acn3.343,
Bracher-Smith, M., Crawford, K., and Escott-Price, V. (2020). Machine learning for genetic prediction of psychiatric disorders: a systematic review. Mol. Psychiatry 26, 70–79. doi: 10.1038/s41380-020-0825-2,
Brzezińska, A., Bourke, J., Rivera-Hernández, R., Tsolaki, M., Woźniak, J., and Kaźmierski, J. (2020). Depression in dementia or dementia in depression? Systematic review of studies and hypotheses. Curr. Alzheimer Res. 17, 16–28. doi: 10.2174/1567205017666200217104114,
Cawley, G. (2010). On over-fitting in model selection and subsequent selection bias in performance evaluation. J. Mach. Learn. Res. 11, 2079–2107. doi: 10.1007/978-981-13-2354-6_27
Cearns, M., Opel, N., Clark, S., Kaehler, C., Thalamuthu, A., Heindel, W., et al. (2019). Predicting rehospitalization within 2 years of initial patient admission for a major depressive episode: a multimodal machine learning approach. Transl. Psychiatry 9. doi: 10.1038/s41398-019-0615-2,
Cook, N. R. (2007). Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 115, 928–935. doi: 10.1161/CIRCULATIONAHA.106.672402,
de Almondes, K. M., Costa, M. V., Malloy-Diniz, L. F., and Diniz, B. S. (2016). Insomnia and risk of dementia in older adults: systematic review and meta-analysis. J. Psychiatr. Res. 77, 109–115. doi: 10.1016/j.jpsychires.2016.02.021,
Deary, I. J., Corley, J., Gow, A. J., Harris, S. E., Houlihan, L. M., Marioni, R. E., et al. (2009). Age-associated cognitive decline. Br. Med. Bull. 92, 135–152. doi: 10.1093/bmb/ldp033,
Del Casale, A., Sarli, G., Bargagna, P., Polidori, L., Alcibiade, A., Zoppi, T., et al. (2023). Machine learning and pharmacogenomics at the time of precision psychiatry. Curr. Neuropharmacol. 21, 2395–2408. doi: 10.2174/1570159X21666230808170123,
Desai, R., John, A., Stott, J., and Charlesworth, G. (2020). Living alone and risk of dementia: a systematic review and meta-analysis. Ageing Res. Rev. 62:101122. doi: 10.1016/j.arr.2020.101122,
Dimitriadis, S. I., and Liparas, D. (2018). How random is the random forest? Random forest algorithm on the service of structural imaging biomarkers for Alzheimer’s disease: from Alzheimer’s disease neuroimaging initiative (ADNI) database. Neural Regen. Res. 13, 962–970. doi: 10.4103/1673-5374.233433,
Dufouil, C., Pereira, E., Chêne, G., Glymour, M. M., Alpérovitch, A., Saubusse, E., et al. (2014). Older age at retirement is associated with decreased risk of dementia. Eur. J. Epidemiol. 29, 353–361. doi: 10.1007/s10654-014-9906-3,
Fage, B. A., Chan, C. C. H., Gill, S. S., Noel-Storr, A. H., Herrmann, N., Smailagic, N., et al. (2021). Mini-cog for the detection of dementia within a community setting. Cochrane Database Syst. Rev. 2021:CD010860. doi: 10.1093/ckj/sfaa188
Fernando, E., Fraser, M., Hendriksen, J., Kim, C. H., and Muir-Hunter, S. W. (2017). Risk factors associated with falls in older adults with dementia: a systematic review. Physiother. Can. 69, 161–170. doi: 10.3138/ptc.2016-14,
Franks, K. H., Bransby, L., Saling, M. M., and Pase, M. P. (2021). Association of stress with risk of dementia and mild cognitive impairment: a systematic review and meta-analysis. J Alzheimer's Dis 82, 1573–1590. doi: 10.3233/JAD-210094,
Garg, A., Agrawal, R., Velleman, R., Rane, A., Costa, S., Gupta, D., et al. (2022). Integrating assisted tele-psychiatry into primary healthcare in Goa, India: a feasibility study. Glob. Mental Health 9, 26–36. doi: 10.1017/gmh.2021.47,
Gulpers, B., Ramakers, I., Hamel, R., Köhler, S., Oude Voshaar, R., and Verhey, F. (2016). Anxiety as a predictor for cognitive decline and dementia: a systematic review and Meta-analysis. Am. J. Geriatr. Psychiatry 24, 823–842. doi: 10.1016/j.jagp.2016.05.015,
Hagi, K., Kurokawa, S., Takamiya, A., Fujikawa, M., Kinoshita, S., Iizuka, M., et al. (2023). Telepsychiatry versus face-to-face treatment: systematic review and meta-analysis of randomised controlled trials. Br. J. Psychiatry 223, 407–414. doi: 10.1192/bjp.2023.86,
Hallam, B., Rees, J., Petersen, I., Cooper, C., Avgerinou, C., and Walters, K. (2021). How are people with mild cognitive impairment or subjective memory complaints managed in primary care? A systematic review. Fam. Pract. 38, 669–683. doi: 10.1093/fampra/cmab014,
Harada, C. N., Natelson Love, M. C., and Triebel, K. L. (2013). Normal cognitive aging. Clin. Geriatr. Med. 29, 737–752. doi: 10.1016/j.cger.2013.07.002,
Hewitt, K. C., Block, C., Bellone, J. A., Dawson, E. L., Garcia, P., Gerstenecker, A., et al. (2022). Diverse experiences and approaches to tele neuropsychology: commentary and reflections over the past year of COVID-19. Clin. Neuropsychol. 36, 790–805. doi: 10.1080/13854046.2022.2027022,
Hobensack, M., Song, J., Scharp, D., Bowles, K. H., and Topaz, M. (2023). Machine learning applied to electronic health record data in home healthcare: a scoping review. Int. J. Med. Inform. 170:104978. doi: 10.19139/soic.v2i2.75,
Hung, C. M., Li, Y. C., Chen, H. J., Lu, K., Liang, C. L., Liliang, P. C., et al. (2018). Risk of dementia in patients with primary insomnia: a nationwide population-based case-control study. BMC Psychiatry 18, 1–7. doi: 10.1186/s12888-018-1623-0,
Hunter, M. B., Jenkins, N., Dolan, C., Pullen, H., Ritchie, C., and Muniz-Terrera, G. (2021). Reliability of telephone and videoconference methods of cognitive assessment in older adults with and without dementia. J Alzheimer's Dis 81, 1625–1647. doi: 10.3233/JAD-210088,
Woodard, J. L., Nielson, K. A., Sugarman, M. A., Smith, J. C., Seidenberg, M., Durgerian, S., et al. (2012). Lifestyle and genetic contributions to cognitive decline and hippocampal structure and function in healthy aging. Curr. Alzheimer Res. 9, 436–446. doi: 10.2174/156720512800492477,
Katsari, M., Kasselimis, D. S., Giogkaraki, E., Breza, M., Evangelopoulos, M. E., Anagnostouli, M., et al. (2020). A longitudinal study of cognitive function in multiple sclerosis: is decline inevitable? J. Neurol. 267, 1464–1475. doi: 10.1007/s00415-020-09720-8,
Kaur, P., and Sharma, M. (2019). Diagnosis of human psychological disorders using supervised learning and nature-inspired computing techniques: a Meta-analysis. J. Med. Syst. 43, 1–30. doi: 10.1007/s10916-019-1341-2,
Khosravi, M., Zare, Z., Mojtabaeian, S. M., and Izadi, R. (2024). Artificial intelligence and decision-making in healthcare: a thematic analysis of a systematic review of reviews. Health Serv. Res. Manag. Epidemiol. 11:4863. doi: 10.1016/s0895-4356(03)00207-5
Kroenke, K., Spitzer, R. L., and Williams, J. B. W. (2003). The patient health questionnaire-2: validity of a two-item depression screener. Med. Care 41, 1284–1292. doi: 10.1097/01.MLR.0000093487.78664.3C,
La Corte, V., Sperduti, M., Malherbe, C., Vialatte, F., Lion, S., Gallarda, T., et al. (2016). Cognitive decline and reorganization of functional connectivity in healthy aging: the pivotal role of the salience network in the prediction of age and cognitive performances. Front. Aging Neurosci. 8:27616991. doi: 10.3389/fnagi.2016.00204,
Lad, M., Sedley, W., and Griffiths, T. D. (2024). Sensory loss and risk of dementia. Neuroscientist 30, 247–259. doi: 10.1177/10738584221126090,
Lang, L., Clifford, A., Wei, L., Zhang, D., Leung, D., Augustine, G., et al. (2017). Prevalence and determinants of undetected dementia in the community: a systematic literature review and a meta-analysis. BMJ Open 7:e011146. doi: 10.1136/bmjopen-2016-011146,
Leelakanok, N., and RR, D. C. (2019). Association between polypharmacy and dementia – a systematic review and metaanalysis. Aging Ment. Health 23, 932–941. doi: 10.1080/13607863.2018.1468411,
Levesque, J. F., Harris, M. F., and Russell, G. (2013). Patient-centred access to health care: conceptualising access at the interface of health systems and populations. Int. J. Equity Health 12, 18–19. doi: 10.1186/1475-9276-12-18,
Li, E., Ai, F., and Liang, C. (2023). A machine learning model to predict the risk of depression in US adults with obstructive sleep apnea hypopnea syndrome: a cross-sectional study. Front. Public Health 11:1348803. doi: 10.1186/s12911-023-02244-x
Lima, A. C. E. S., and De Castro, L. N. (2019). TECLA: a temperament and psychological type prediction framework from twitter data. PLoS One 14:e0212844. doi: 10.1371/journal.pone.0212844,
Liozidou, A., Traikapi, A., Stanitsa, E., Kontaxopoulou, D., Fragkiadaki, S., Beratis, I., et al. (2023). Neuropsychology in Greece: results from a survey of practicing professionals. Appl. Neuropsychol. Adult 30, 340–351. doi: 10.1080/23279095.2021.1944145,
Liu, C. J., Chang, P. S., Griffith, C. F., Hanley, S. I., and Lu, Y. (2022). The nexus of sensory loss, cognitive impairment, and functional decline in older adults: a scoping review. Gerontologist 62, e457–e467. doi: 10.1093/geront/gnab082,
Ma, L. (2020). Depression, anxiety, and apathy in mild cognitive impairment: current perspectives. Front. Aging Neurosci. 12:12. doi: 10.3389/fnagi.2020.00009,
Majcherek, D., Hegerty, S. W., Kowalski, A. M., Lewandowska, M. S., and Dikova, D. (2024). Opportunities for healthcare digitalization in Europe: comparative analysis of inequalities in access to medical services. Health Policy 139:4950. doi: 10.1016/j.healthpol.2023.104950,
Malakouti, S. M., Menhaj, M. B., and Suratgar, A. A. (2023). The usage of 10-fold cross-validation and grid search to enhance ML methods performance in solar farm power generation prediction. Clean Eng. Technol. 15:100664. doi: 10.1016/j.clet.2023.100664,
Mendonça, M. D., Alves, L., and Bugalho, P. (2016). From subjective cognitive complaints to dementia. Am. J. Alzheimers Dis. Other Dement. 31, 105–114. doi: 10.1177/1533317515592331,
Mollalo,, Kramer, M., Cutty, M., and Hoseini, B. (2025). Systematic review and meta-analysis of rural-urban disparities in Alzheimer’s disease dementia prevalence. J. Prev. Alzheimers Dis. 12:100305. doi: 10.1016/j.tjpad.2025.100305,
Monteiro, S. S., Geraldo, A., Pinto, J. O., Santos, I. M., DeFilippis, N., and Ferreira-Santos, F. (2025). Neuropsychological assessment by video teleconference in adults: a systematic review. Appl. Neuropsychol. Adult, 1–18. doi: 10.1080/23279095.2025.2455538,
Nori, V. S., Hane, C. A., Crown, W. H., Au, R., Burke, W. J., Sanghavi, D. M., et al. (2019). Machine learning models to predict onset of dementia: a label learning approach. Alzheimers Dement. 5, 918–925. doi: 10.3390/geriatrics6010033
Parpoula, C., Drosou, K., Koukouvinos, C., and Mylona, K. (2014). A new variable selection approach inspired by supersaturated designs given a large-dimensional dataset. J. Data Sci. 12, 35–52. doi: 10.1177/23333928241234863
Parpoula, C., Koukouvinos, C., Simos, D. E., and Stylianou, S. (2014). Supersaturated plans for variable selection in large databases. Stat. Optim. Inform. Comp. 2, 161–175. doi: 10.19139/soic.v2i2.75
Pazan, F., and Wehling, M. (2021). Polypharmacy in older adults: a narrative review of definitions, epidemiology and consequences. Eur. Geriatr. Med. 12, 443–452. doi: 10.1007/s41999-021-00479-3,
Pertl, M. T., Benke, T., Zamarian, L., and Delazer, M. (2017). Effects of healthy aging and mild cognitive impairment on a real-life decision-making task. J Alzheimer's Dis 58, 1077–1087. doi: 10.3233/JAD-170119,
Petersen, R. C., and Negash, S. (2008). Mild cognitive impairment: an overview. CNS Spectr. 13, 45–53. doi: 10.1017/S1092852900016151,
Pettigrew, C., and Martin, R. C. (2014). Cognitive declines in healthy aging: evidence from multiple aspects of interference resolution. Psychol. Aging 29, 187–204. doi: 10.1037/a0036085,
Poirier, G., Ohayon, A., Juranville, A., Mourey, F., and Gaveau, J. (2021). Deterioration, compensation and motor control processes in healthy aging, mild cognitive impairment and Alzheimer’s disease. Geriatrics 6:33. doi: 10.3389/fpubh.2023.1348803
Politis, A., Vorvolakos, T., Kontogianni, E., Alexaki, M., Georgiou, E. Z., Aggeletaki, E., et al. (2023). Old-age mental telehealth services at primary healthcare centers in low- resource areas in Greece: design, iterative development and single-site pilot study findings. BMC Health Serv. Res. 23, 1–11. doi: 10.1016/j.ijnurstu.2023.104562
Powell, D. S., and Reed, N. S. (2024). Sensory loss rehabilitation among people with dementia: a low-risk strategy to enhance quality of life. Lancet Healthy Longev. 5:100640. doi: 10.1016/j.lanhl.2024.100640,
Prinzi, F., Currieri, T., Gaglio, S., and Vitabile, S. (2024). Shallow and deep learning classifiers in medical image analysis. Eur. Radiol. Exp. 8, 1–13. doi: 10.1186/s41747-024-00428-2,
Pumplün, C., Weihs, C., and Preusser, A. (2005). “Experimental Design for Variable Selection in data bases” in Classification — the Ubiquitous Challenge. Studies in Classification, Data Analysis, and Knowledge Organization, eds. C. Weihs and W. Gaul 192–199. doi: 10.1007/3-540-28084-7_20
Ramspek, C. L., Jager, K. J., Dekker, F. W., Zoccali, C., and M, V. D. I. (2021). External validation of prognostic models: what, why, how, when and where? Clin. Kidney J. 14, 49–58. doi: 10.5555/1756006.1859921
Rosenbaum Halevi, D., Bursaw, A. W., Karamchandani, R. R., Alderman, S. E., Breier, J. I., Vahidy, F. S., et al. (2019). Cognitive deficits in acute mild ischemic stroke and TIA and effects of rt-PA. Ann. Clin. Transl. Neurol. 6, 466–474. doi: 10.1002/acn3.719,
Rubin, R. (2018). Exploring the relationship between depression and dementia. J. Am. Med. Assoc. 320, 961–962. doi: 10.1001/jama.2018.11154,
Santabárbara, J., Lipnicki, D. M., Villagrasa, B., Lobo, E., and Lopez-Anton, R. (2019). Anxiety and risk of dementia: systematic review and meta-analysis of prospective cohort studies. Maturitas 119, 14–20. doi: 10.1016/j.maturitas.2018.10.014,
Santabárbara, J., Lopez-Anton, R., de la Cámara, C., Lobo, E., Gracia-García, P., Villagrasa, B., et al. (2019). Clinically significant anxiety as a risk factor for dementia in the elderly community. Acta Psychiatr. Scand. 139, 6–14. doi: 10.1111/acps.12966,
Sapra, A., Bhandari, P., Sharma, S., Chanpura, T., and Lopp, L. (2020). Using generalized anxiety Disorder-2 (GAD-2) and GAD-7 in a primary care setting. Cureus 12:8224. doi: 10.7759/cureus.8224,
Sarica, A., Cerasa, A., and Quattrone, A. (2017). Random forest algorithm for the classification of neuroimaging data in Alzheimer’s disease: a systematic review. Front. Aging Neurosci. 9:284242. doi: 10.3389/fnagi.2017.00329,
Savva, G. M., and Arthur, A. (2015). Who has undiagnosed dementia? A cross-sectional analysis of participants of the aging, demographics and memory study. Age Ageing 44, 642–647. doi: 10.1093/ageing/afv020,
Shaik, A. B., and Srinivasan, S. (2019). A brief survey on random forest ensembles in classification model. Lect. Notes Netw. Syst. 56, 253–260. doi: 10.1002/14651858.CD010860.pub3
Sharma, G., and Devan, K. (2023). The effectiveness of telepsychiatry: thematic review. BJPsych Bull. 47, 82–89. doi: 10.1192/bjb.2021.115,
Skoog, I. (2011). Psychiatric disorders in the elderly. Can. J. Psychiatr. 56, 387–397. doi: 10.1177/070674371105600702,
Speiser, J. L., Miller, M. E., Tooze, J., and Ip, E. (2019). A comparison of random forest variable selection methods for classification prediction modeling. Expert Syst. Appl. 134, 93–101. doi: 10.1016/j.eswa.2019.05.028,
Sperling, S. A., Acheson, S. K., Fox-Fuller, J., Colvin, M. K., Harder, L., Cullum, C. M., et al. (2024). Tele-neuropsychology: from science to policy to practice. Arch. Clin. Neuropsychol. 39, 227–248. doi: 10.1093/arclin/acad066,
Sweet, J. J., Peck, E. A., Abramowitz, C., and Etzweiler, S. (2002). National Academy of neuropsychology/division 40 of the American Psychological Association practice survey of clinical neuropsychology in the United States, part I: practitioner and practice characteristics, professional activities, and time requirements. Clin. Neuropsychol. 16, 109–127. doi: 10.1076/clin.16.2.109.13237,
Taylor, W. D. (2014). Clinical practice. Depression in the elderly. N. Engl. J. Med. 371, 1228–1236. doi: 10.1056/NEJMcp1402180,
Tully, P. J., Hanon, O., Cosh, S., and Tzourio, C. (2016). Diuretic antihypertensive drugs and incident dementia risk: a systematic review, meta-analysis and meta-regression of prospective studies. J. Hypertens. 34, 1027–1035. doi: 10.1097/HJH.0000000000000868,
Twait, E. L., Andaur Navarro, C. L., Gudnason, V., Hu, Y. H., Launer, L. J., and Geerlings, M. I. (2023). Dementia prediction in the general population using clinically accessible variables: a proof-of-concept study using machine learning. The AGES-Reykjavik study. BMC Med. Inform. Decis. Mak. 23, 1–12. doi: 10.1186/s12913-023-09583-5
Van Den Berg, E., Geerlings, M. I., Biessels, G. J., Nederkoorn, P. J., and Kloppenborg, R. P. (2018). White matter Hyperintensities and cognition in mild cognitive impairment and Alzheimer’s disease: a domain-specific Meta-analysis. J Alzheimer's Dis 63, 515–527. doi: 10.3233/JAD-170573,
Wallace, M. L., Mentch, L., Wheeler, B. J., Tapia, A. L., Richards, M., Zhou, S., et al. (2023). Use and misuse of random forest variable importance metrics in medicine: demonstrations through incident stroke prediction. BMC Med. Res. Methodol. 1, 23, 1–12. doi: 10.1186/s12874-023-01965-x,
Walling, A. M., Pevnick, J., Bennett, A. V., Vydiswaran, V. G. V., and Ritchie, C. S. (2023). Dementia and electronic health record phenotypes: a scoping review of available phenotypes and opportunities for future research. J. Am. Med. Inform. Assoc. 30, 1333–1348. doi: 10.1093/jamia/ocad086,
Wang, F. (2012). Measurement, optimization, and impact of health care accessibility: a methodological review. Ann. Assoc. Am. Geogr. 102, 1104–1112. doi: 10.1080/00045608.2012.657146,
Wendt, C. (2009). Mapping European healthcare systems: a comparative analysis of financing, service provision and access to healthcare. J. Eur. Soc. Policy 19, 432–445. doi: 10.1177/0958928709344247
Xue, J., Li, J., Liang, J., and Chen, S. (2018). The prevalence of mild cognitive impairment in China: a systematic review. Aging Dis. 9, 706–715. doi: 10.14336/AD.2017.0928,
Yarkoni, T., and Westfall, J. (2017). Choosing prediction over explanation in psychology: lessons from machine learning. Perspect. Psychol. Sci. 12, 1100–1122. doi: 10.1177/1745691617693393,
Zhang, Z. (2016). Multiple imputation with multivariate imputation by chained equation (MICE) package. Ann. Transl. Med. 4, 30–30. doi: 10.3978/j.issn.2305-5839.2015.12.63,
Zhou, Y., Han, W., Yao, X., Xue, J. J., Li, Z., and Li, Y. (2023). Developing a machine learning model for detecting depression, anxiety, and apathy in older adults with mild cognitive impairment using speech and facial expressions: a cross-sectional observational study. Int. J. Nurs. Stud. 146:104562. doi: 10.1016/j.trci.2019.10.006
Keywords: dementia, low resource areas, machine learning, remote areas, telehealth, Telepsychiatry, telepsychology
Citation: Soldatos RF, Kasselimis D, Parpoula C, Konidari E, Dimitriou V, Katirtzoglou E, Kiosses D, Tsibanis K, Konsta A, Vorvolakos T, Alexopoulos P and Politis A (2026) INTegRated InterveNtion of pSychogerIatric Care: real-world application and implementation of an advanced integrated telehealth system incorporating machine learning. Front. Psychol. 16:1696407. doi: 10.3389/fpsyg.2025.1696407
Edited by:
Fabrizio Stasolla, Giustino Fortunato University, ItalyReviewed by:
Prathap Rudra, M.S. Ramaiah University of Applied Sciences, IndiaAn Huang, Huzhou University, China
Copyright © 2026 Soldatos, Kasselimis, Parpoula, Konidari, Dimitriou, Katirtzoglou, Kiosses, Tsibanis, Konsta, Vorvolakos, Alexopoulos and Politis. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Rigas F. Soldatos, cnNvbGRhdG9zQGxpdmUuY29t
†These authors have contributed equally to this work