A Review of Early Detection of Cognitive

Frailty

ABSTRACT

This article reviewed the literature on Physical Frailty to find out how to detect
Physical Frailty. The methods used to detect Cognitive Frailty in the Medical
Electronic Health Record were searched by reading the Medical Electronic Health
Record literature. The literature on TCSA mobile phone app was reviewed to find out
the methods for detecting Cognitive impairment. Finally, the literature on Machine
Learning models was reviewed to find out the types of models that are beneficial for
disease prediction.

1. INTRODUCTION

Nowadays, cognitive frailty is considered a geriatric syndrome, usually involving

over 65 years old people, representing a state of vulnerability with increased risk of
poor health outcomes, including falls, incident disability, hospitalization mortality.
Common signs and symptoms of cognitive frailty are weight loss, fatigue, muscle
weakness and reduced physical and mental performances. Several population-based
studies estimated the prevalence rate of cognitive frailty to be in the range of 1.0% to
12.0%, whereas in clinical settings, the figure was much higher at 10.7% to 40%.
Cognitive frailty can be influenced by a number of risk factors, including vascular,
lifestyle, physical activity, smoking status, and psychosocial factors as well as
potential effects of a poor nutritional status. The prevalence of robust, cognitive pre-
frailty and cognitive frailty among older adults in Malaysia was 60.4%, 37.4%, and
2.2%, respectively[1].

1/9
2. Physical Frailty

Physical frailty was defined based on the frailty phenotype including weight loss,
low grip strength, exhaustion, slow gait speed, and low physical activity[2]. In
paper[3], Fried et al. evaluated the prevalence and characteristics of frailty in older
adults and its association with falls, disability, hospitalization, and mortality. Baseline
evaluations included interviews, physical function assessments, and self-reported
health information. Lewis proposed the frailty phenotype model which was based on a
cycle of factors including disease, chronic undernutrition, sarcopenia, and reduced
resting metabolic rate[4]. The frailty phenotype has been adapted and operationalized
in various study designs and settings, including low and middle-income countries
(LMIC). Data were collected using handheld tablet devices and analyzed using
descriptive statistics and chi-squared values.
Marques used the smartphone application to measure temporal variables during
the Single Sit-to-Stand Test with older adults showed excellent reliability, high
accuracy, and strong relationships with the reference criterion (a digital video camera)
[5]. The results showed excellent reliability, high accuracy, and strong relationships
between devices in measuring stand-up time and total time. The mobile app provides
real-time results and eliminates the need for other materials like chronometers. It is a
valuable tool for assessing the sit-to-stand performance of older adults in clinical or
controlled settings.
Yee XS et al. also did sit-to-stand tests which examined the relationship between
sit-to-stand tests and measured of functional fitness and sarcopenia diagnosis in older
adults living in the community[6]. Sit-to-stand tests are commonly used to evaluate
lower body strength and are simple to administer. The study collected data on various
physical performance measures, including grip strength, timed-up-and-go, habitual
gait speed, six-minute walk test, and short physical performance battery. The results
showed that there was no significant difference in sit-to-stand performance between
men and women or across different age groups. Multiple linear regression analysis
revealed that strength, mobility, endurance, and balance were all associated with sit-
to-stand performance. The study highlights the importance of considering muscle
mass in addition to physical performance measures when screening for sarcopenia.
The conclusion of the research is that sit-to-stand tests (STS) in older adults better
represent overall physical performance rather than just muscle strength. The use of
STS tests identified a different subset of the population with possible sarcopenia
compared to grip strength tests. The lack of association between STS performance
and muscle mass results in a lower prevalence of confirmed sarcopenia compared to
grip strength, but may better reflect changes in muscle quality. Further research is
needed to evaluate muscle quality in the community using advanced technology.
Jung et al. classified the risk of cognitive impairment in elderly individuals using
sequential gait characteristics and machine learning techniques[7]. The participants
were classified into three groups based on their MMSE scores: cognitively normal,

2/9
low-risk CI, and high-risk CI. The study proposed a long short-term memory (LSTM)
network-based classifier for classification. Gait characteristics, such as gait speed,
were analyzed and significant differences were found between usual-paced and fast-
paced walking in all groups. The LSTM network-based classifier showed high
classification performance, with an F-score of 0.975 for the validation sets. The input
features with the highest predictive power included initial and terminal double-limb
support, stance, and stride times.

3. Medical Electronic Health Record

Electronic Health Record (EHR) systems have become widely used in healthcare
settings, providing valuable data for research and improving patient care. Deep
learning techniques in machine learning can automatically learn optimal features from
EHR data, eliminating the need for manual feature engineering. These techniques
have been applied to tasks such as concept extraction, temporal event extraction,
relation extraction, outcome prediction, and computational phenotyping.
The authors in [8] created models based on convolutional neural networks (CNN)
are effective in extracting features from medical data and performing intelligent
recognition for disease risk assessment. The study explored multi-modal medical data
feature representation learning, including analysis of medical text data and medical
image data. The use of CNN in text analysis allowed for the extraction of text features
and performing tasks such as text classification and sentiment analysis. The
experiments demonstrated the feasibility and versatility of the proposed models for
disease risk assessment. Additionally, the multimodal data fusion method for disease
risk assessment performed better than the text-based method in terms of training time
and stability.
Bertini et al. proposed two predictive models for frailty in the elderly using
multidimensional socioclinical databases[9]. The data warehouse construction process
involved loading, cleaning, and integrating data from 12 socioclinical databases. The
data set included continuous, categorical, and Boolean variables related to clinical and
socioeconomic aspects and health service resources utilization. The data set was
refined by integrating it with other knowledge bases such as ICD-9-CM taxonomy,
drug databases, and exemption code databases. Logistic regression was used to
construct the predictive models, and odds ratios and confidence intervals were
computed to evaluate the association between variables and the outcome. The models
were validated externally using a local database of frail subjects identified through the
Brief Self-Sufficiency Index.
Idris et al. used a machine learning model to classify elderly patients into
different levels of cognitive frailty using parameters from blood samples[10]. The
dataset used in the study is obtained from the Malaysian Elders Longitudinal Research
(MELoR) study, which includes socio-demographic and medical data from
participants aged 55 years and above. The dataset was pre-processed by cleaning the

3/9
data, removing inconsistent scaling and mismatched data types, and standardizing and
normalizing the values. The classification algorithms used in the study were Logistic
Regression, Linear Discriminant Analysis, k-Nearest Neighbor, Classification &
Regression Tree, Gaussian Naive Bayes, Support Vector Machines, and Random
Forest Classifier. The performance of the models was evaluated in terms of accuracy,
precision, recall, and f1-score using holdout method.

4. Cognitive Impairment

Cognitive impairment(CI), a progressive neurocognitive disorder, is commonly

experienced in old age and is characterized as declines in brain functions such as
memory, language, and problem-solving. The spectrum of both frailty and CI can
greatly impact the quality of life[11].
Nie et al. introduced a novel two-step touchscreen-based cognition assessment tool
called the Thoven Cognitive Self-Assessment (TCSA)[12]. The TCSA was designed
to quickly and comprehensively assess cognitive function in older adults, particularly
in the early stages of cognitive impairment. It can be used as a screening tool for early
cognitive impairment and has the ability to differentiate between healthy individuals
and those at risk for dementia. It consisted of a primary screening section and a
secondary screening section, with a total score of 11 points for the primary section
and 30 points for the secondary section. Participants needed to complete tasks related
to memory, attention, calculation, language, and orientation. Receiver operating
characteristic analysis indicated that a score below 7.5 for the TCSA primary task and
a score below 22.5 for the TCSA secondary task were indicators of MCI. The TCSA
demonstrated good test-retest reliability and concurrent validity when compared to the
Montreal Cognitive Assessment (C-MoCA) test. The TCSA showed significant
differences in scores between the control group, mild cognitive impairment (MCI)
group, and dementia group.

4. Machine Learning in Detection of Disease

The existing work on disease diagnosis approaches was categorized into rule-based
methods, support vector machine methods, neural network methods, Bayesian
methods, and deep learning methods[13]. The existing work on disease diagnosis
approaches was categorized into rule-based methods, support vector machine
methods, neural network methods, Bayesian methods, and deep learning methods.
Rule-based methods involved extracting detailed medical information and resolving
co-reference relationships in patient records. Machine learning methods, such as
support vector machines, Bayesian methods, and decision trees, were commonly used

4/9
for constructing medical databases from electronic health records. Deep learning
methods, including convolutional neural networks and recurrent neural networks,
were also used for disease diagnosis.
Li et al. proposed a brain age prediction model using a similarity metric
convolutional neural network for resting-state functional MRI data[14]. The model
included convolution, batch normalization, and pooling steps to learn features from
two groups of rs-fMRI images and calculate the similarity between them. It was
optimized using a loss function and the average value of the three groups of sample
labels with the greatest similarity is taken as the final predicted brain age. It achieved
an absolute mean error of 5.337 and a correlation coefficient of 0.6279 on the
Southwest University Longitudinal Imaging Dataset. The dataset was used in the
experiments includes rs-fMRI images from 494 healthy subjects, and the model was
trained using contrastive loss as the loss function. The experimental results showed
that the proposed method outperforms some deep learning methods in terms of
prediction accuracy.
The research conducted by Yang et al. focused on the use of an improved back-
propagation neural network for classifying and predicting Tibetan medical syndromes,
specifically Chronic Atrophic Gastritis (CAG)[15]. The study first used a clustering
algorithm to objectively classify the syndrome type of CAG, reducing subjective
judgment in diagnosis and treatment. Then, a classification and prediction model
using the improved neural network was constructed based on clinical data, achieving
high accuracy in determining the syndrome type. The experiment achieved an
accuracy of 99.09%, which was significantly better than other classification methods.
The study also compared the proposed method with traditional machine learning
algorithms and demonstrates its superior performance in terms of accuracy and time
consumption. Overall, the improved neural network model showed promise in
providing scientific decision support for Tibetan medicine diagnosis.
The paper explored the relationship between cognitive and physical functions of the
human body and proposed the use of machine learning techniques to predict frailty
based on cognitive assessments[16]. The dataset used in the study consisted of 104
participants from a senior housing community, with various sociodemographic,
cognitive, physical, and mental variables collected. The paper discussed techniques to
navigate limited and imbalanced datasets commonly found in clinical datasets. The
study showed that oversampling is an effective method to increase the size and
balance of the dataset. The paper also investigated the importance of different features
in predicting physical function, with age being identified as an important feature. The
study compared classification and regression models and found that the binary
classification model achieves high accuracy in predicting physical functionality from
cognitive features. It demonstrated that adhering to clinically determined thresholds
yields better results than arbitrary cutoffs in creating balanced datasets. The results
suggested that the TUG test was more highly related to cognitive functioning than the
SPPB test. The proposed data preprocessing technique using SMOTE oversampling
improved the performance of the models and overcame the imbalance in the dataset.
Delmastro et al. demonstrated the efficacy of physical activity on cognitive
5/9
performance in frail older adults while maintaining low stress levels[17]. The authors
used a real dataset collected in a semi-controlled environment to evaluate a binary
stress detection system based on different classification algorithms, which showed
promising results compared to solutions used in completely controlled settings. They
provided an overview of related works on physical activity and cognitive performance
in older adults, as well as stress monitoring and detection systems. The research
proposed a novel Decision Support System (DSS) with a mobile pervasive
architecture for online stress monitoring, which provided personalized feedback
during therapy and can be used in both clinical settings and remote monitoring
systems. The methods section described the analysis of physiological data collected
during training sessions, including data preprocessing, feature extraction, and
statistical analysis. The research also evaluated the performance of machine learning
techniques for stress detection, including dataset preparation, feature selection, and
classification algorithms. The proposed system can accommodate multiple predictive
models and can be extended to include additional wearable and smartphone sensors.
The research report written by Alsaedi et al. focused on using the Extended Cox
Proportional Hazard Model to analyze and predict the conversion from mild cognitive
impairment (MCI) to Alzheimer's Disease (AD)[18]. The Mini-Mental State
Examination (MMSE) was a widely used cognitive test, but its predictability can be
influenced by various factors such as age, education, language, and cultural
background. The report proposed combining MMSE with other predictors and
demographic data to improve its accuracy in predicting AD conversion. The
Anderson-Gill (AG) model and frailty model were used in the analysis. Recurrent
events, such as multiple occurrences of the same type of event, were considered in the
models. The report also discussed the significance of different covariates and their
interactions with MMSE. The research used a dataset divided into training and testing
sets and examines the factors that impact MMSE as a predictor for AD. The
predictability of the Mini-Mental State Examination (MMSE) in identifying the
conversion from mild cognitive impairment to Alzheimer's Disease was improved
when combined with other predictors.

5. CONCLUSION

This article reviewed the literature on Physical Frailty to find out how to detect
Physical Frailty. The use of sit-to-stand tests (STS) in older adults was valuable for
assessing physical performance and identifying sarcopenia. Gait characteristics, such

as gait speed, can be analyzed using machine learning techniques （such as LSTM）
to classify the risk of cognitive impairment in elderly individuals. Electronic Health
Record (EHR) systems provided valuable data for research and improving patient
care, and deep learning techniques can automatically learn optimal features from EHR

6/9
data. The Thoven Cognitive Self-Assessment (TCSA) was a touchscreen-based tool
that can assess cognitive function in older adults and differentiate between healthy
individuals and those at risk for dementia. Machine learning techniques can be used to
predict the conversion from mild cognitive impairment (MCI) to Alzheimer's Disease
(AD) by combining cognitive tests with other predictors and demographic data.
Predictive models for frailty in the elderly can be constructed using multidimensional
socioclinical databases, and logistic regression can be used to evaluate the association
between variables and the outcome. Convolutional neural networks (CNN) were
effective in extracting features from medical data and performing intelligent
recognition for disease risk assessment.

REFERENCES

[1]. Malek Rivan NF, Shahar S, Rajab NF, Singh DKA, Din NC, Hazlina M, et al.
Cognitive frailty among Malaysian older adults: baseline findings from the LRGS
TUA cohort study. Clin Interv Aging. (2019) 14:1343–52. doi:
10.2147/CIA.S211027.
[2]. Liu, W., Puts, M., Jiang, F. et al. Physical frailty and its associated factors among
elderly nursing home residents in China. BMC Geriatr 20, 294 (2020).
https://doi.org/10.1186/s12877-020-01695-5
[3]. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, Seeman
T, Tracy R, Kop WJ, Burke G, et al. Frailty in older adults: evidence for a
phenotype. J Gerontol A Biol Sci Med Sci. 2001;56(3):M146–56.
[4]. Lewis, E., Coles, S., Howorth, K. et al. The prevalence and characteristics of
frailty by frailty phenotype in rural Tanzania. BMC Geriatr 18, 283 (2018).
https://doi.org/10.1186/s12877-018-0967-0
[5]. Marques DL, Neiva HP, Pires IM, Zdravevski E, Mihajlov M, Garcia NM, Ruiz-
Cárdenas JD, Marinho DA, Marques MC. An Experimental Study on the Validity
and Reliability of a Smartphone Application to Acquire Temporal Variables
during the Single Sit-to-Stand Test with Older Adults. Sensors (Basel). 2021 Mar
15;21(6):2050. doi: 10.3390/s21062050. PMID: 33803927; PMCID:
PMC8000467.
[6]. Yee XS, Ng YS, Allen JC, Latib A, Tay EL, Abu Bakar HM, Ho CYJ, Koh WCC,
Kwek HHT, Tay L. Performance on sit-to-stand tests in relation to measures of
functional fitness and sarcopenia diagnosis in community-dwelling older adults.
Eur Rev Aging Phys Act. 2021 Jan 8;18(1):1. doi: 10.1186/s11556-020-00255-5.
PMID: 33419399; PMCID: PMC7791746.
[7]. Jung D, Kim J, Kim M, Won CW, Mun KR. Classifying the Risk of Cognitive
Impairment Using Sequential Gait Characteristics and Long Short-Term Memory
Networks. IEEE J Biomed Health Inform. 2021 Oct;25(10):4029-4040. doi:

7/9
 10.1109/JBHI.2021.3073372. Epub 2021 Oct 5. PMID: 33857005.
[8]. W. Liu et al., "Research on Medical Data Feature Extraction and Intelligent
Recognition Technology Based on Convolutional Neural Network," in IEEE
Access, vol. 7, pp. 150157-150167, 2019, doi: 10.1109/ACCESS.2019.2943927.
[9]. F. Bertini, G. Bergami, D. Montesi, G. Veronese, G. Marchesini and P. Pandolfi,
"Predicting Frailty Condition in Elderly Using Multidimensional Socioclinical
Databases," in Proceedings of the IEEE, vol. 106, no. 4, pp. 723-737, April 2018,
doi: 10.1109/JPROC.2018.2791463.
[10]. S. Idris and N. Badruddin, "Classification of Cognitive Frailty in Elderly
People from Blood Samples using Machine Learning," 2021 IEEE EMBS
International Conference on Biomedical and Health Informatics (BHI), Athens,
Greece, 2021, pp. 1-4, doi: 10.1109/BHI50953.2021.9508514.
[11]. Damluji AA, Ijaz N, Chung SE, Xue QL, Hasan RK, Batchelor WB, Orkaby
AR, Kochar A, Nanna MG, Roth DL, Walston JD, Resar JR, Gerstenblith G.
Hierarchical Development of Physical Frailty and Cognitive Impairment and
Their Association With Incident Cardiovascular Disease. JACC Adv. 2023
May;2(3):100318. doi: 10.1016/j.jacadv.2023.100318. Epub 2023 May 10.
PMID: 37538136; PMCID: PMC10399211.
[12]. Nie J, Yang Y, Gao Y, Jiang W, Aidina A, Sun F, Prieto LR, Yu J, Ju K, Song
L, Li X. Newly self-administered two-step tool for screening cognitive function
in an ageing Chinese population: an exploratory cross-sectional study. Gen
Psychiatr. 2023 Feb 3;36(1):e100837. doi: 10.1136/gpsych-2022-100837. PMID:
36760346; PMCID: PMC9900047.
[13]. J. Latif, C. Xiao, S. Tu, S. U. Rehman, A. Imran and A. Bilal,
"Implementation and Use of Disease Diagnosis Systems for Electronic Medical
Records Based on Machine Learning: A Complete Review," in IEEE Access, vol.
8, pp. 150489-150513, 2020, doi: 10.1109/ACCESS.2020.3016782.
[14]. Li H, Satterthwaite TD, Fan Y. BRAIN AGE PREDICTION BASED ON
RESTING-STATE FUNCTIONAL CONNECTIVITY PATTERNS USING
CONVOLUTIONAL NEURAL NETWORKS. Proc IEEE Int Symp Biomed
Imaging. 2018 Apr;2018:101-104. doi: 10.1109/ISBI.2018.8363532. Epub 2018
May 24. PMID: 30079125; PMCID: PMC6074039.
[15]. S. Yang, X. Zhu, L. Zhang, L. Wang and X. Wang, "Classification and
Prediction of Tibetan Medical Syndrome Based on the Improved BP Neural
Network," in IEEE Access, vol. 8, pp. 31114-31125, 2020, doi:
10.1109/ACCESS.2020.2973304.
[16]. S. Kumar, C. Du, S. Graham and T. Nguyen, "Using Machine Learning to
Predict Frailty from Cognitive Assessments," 2021 43rd Annual International
Conference of the IEEE Engineering in Medicine & Biology Society (EMBC),
Mexico, 2021, pp. 1648-1652, doi: 10.1109/EMBC46164.2021.9630386.
[17]. F. Delmastro, F. D. Martino and C. Dolciotti, "Cognitive Training and Stress
Detection in MCI Frail Older People Through Wearable Sensors and Machine
Learning," in IEEE Access, vol. 8, pp. 65573-65590, 2020, doi:
10.1109/ACCESS.2020.2985301..
8/9
[18]. A. Alsaedi, I. Abdel-Qader, N. Mohammad and A. C. Fong, "Extended cox
proportional hazard model to analyze and predict conversion from mild cognitive
impairment to alzheimer's disease," 2018 IEEE 8th Annual Computing and
Communication Workshop and Conference (CCWC), Las Vegas, NV, USA,
2018, pp. 131-136, doi: 10.1109/CCWC.2018.8301669.

9/9