Background

The continuous progress in medical technology and advancements in living standards have enabled an increasing number of people to live to an advanced age. However, with old age comes a multitude of health issues, and simply living longer does not guarantee a state of good health. According to the World Health Organization (WHO) report, the share of the older population that is above 60 years of age is expected to double between 2015 and 2050, among which nearly 80 percent are from low and middle-income countries [1]. The persistence of communicable diseases, along with an increase in the prevalence of non-communicable diseases, has led to a reduction in the health and well-being of older people. To benefit from the changes in demographics, every nation faces significant challenges in ensuring that its healthcare and social systems are well prepared.

Health and well-being of the elderly

The absence of sickness is not the only indicator of good health. Health encompasses the aspects of physical, psychological, and social well-being [2]. The researchers given evidence that many older persons categorized as healthy based on the medical tests had vulnerabilities exposed to lifestyle-related or psychological problems, which affected their risks of dying or being handicapped within five years [3]. On the other side, some persons with chronic conditions possessed several qualities that kept them healthy. Overweight older persons with otherwise good physical and mental health, for example, had the lowest chance of dying or becoming handicapped. So, there are different kinds of illnesses faced by older adults. Physical vulnerabilities such as vision problems, loss of hearing, functional disability, chronic obstructive pulmonary diseases, cerebrovascular, cardiovascular diseases, and cancer are common illnesses experienced by old age people all over the world. Apart from these, obesity is another health concern that is affecting an increasing number of seniors around the country. Age-related increases in obesity peak in the 60 s and 70 s increasing the risk of diabetes, arthritis, cardiovascular disease, and several cancer types. According to the Longitudinal Ageing Study of India (LASI), 75% of the elderly suffer from chronic diseases and 40% have a disability due to various factors [4].

The mental health of older is equally significant as their physical health. With the increase in life expectancy, mental illness has significantly influenced the older person’s quality of life. Cognitive deficits, dementia, depression, and anxiety disorders are some of them. It has been pointed out by many researchers that middle- and low-income countries invest a very lesser amount for mental illness and most of it is neglected. At the same time, mental health disease awareness is minimal. Many individuals assume that memory issues and emotions of depression or pessimism are inevitable as they become older, so they delay or avoid getting help. As reported in the World Alzheimer Report 2021, 75% of people with dementia globally are undiagnosed [5]. Many elderly people, particularly those residing in long-term care institutions, are affected by depressive disorders and symptoms. As a result, the demand for mental health treatment in old age is becoming more widespread and urgent.

Machine learning in geriatric research

Machine Learning (ML) has revolutionized the technology sector with its day-to-day applications. Many developed countries are now adopting this technology to enhance their healthcare service [6]. This raises the question of whether ML will be a powerful tool for enhancing gerontological research. Developing an accurate and quick diagnosis is one of the most difficult aspects of care for geriatric patients. Such individuals bring complicated medical histories and clinical circumstances to healthcare settings, necessitating a focus on how to enhance patient care outcomes for this group. A statistical foundation underpins ML. This should be self-evident, given that ML requires data, which must be characterized using a statistical framework. It allows the user to submit an enormous quantity of data to a computer algorithm, which the computer may then evaluate and make data-driven suggestions and judgments based only on the supplied data [7]. This approach is promising since it readily discovers trends and patterns in a large dataset, is easy to handle multi-dimensional and multi-variety data, has wide applicability in practically every discipline, and is constantly improved.

Need of the study

Numerous systematic literature reviews have delved into the utilization of machine learning within the realm of geriatric research. Choudhury and colleagues conducted a comprehensive examination of the application of machine learning in geriatric clinical care, and they observed the absence of standardized metrics for evaluating machine learning models, as well as the pressing need for tailored data governance in the healthcare domain [8]. Another literature review explored the integration of machine learning and artificial intelligence in the context of geriatric mental health [9]. This review revealed that dementia stands as the most extensively researched mental health concern within this field, with inconsistent information availability for other mental health issues. Olender et al. undertook a systematic review of studies that used machine learning techniques to predict clinical outcomes in older populations [10], while Leghissa et al. scrutinized research papers focused on the detection, classification, and prediction of frailty in older adults using machine learning methods [11]. The summary of the objective and outcome of these studies have been described in Table 1. In the latter two decades of life, geriatric patients suffer from a variety of ailments, including chronic illnesses, frailty, cognitive decline, and functional dependency. These individuals require high-quality clinical treatment since their issues may lead to hospitalization. As a result, effective solutions for enhancing geriatric clinical care are required. Numerous studies from all over the world have used ML to identify older people at high risk for dementia, predict weakness, risk of falls, pneumonia, delirium, and acute kidney disease, and provide them care. The scope of this study encompasses a comprehensive examination of aging-related concerns, encompassing various facets such as brain age prediction, biological age prediction, chronic diseases, mental health issues, and cognitive disorders. Unlike previous literature, which often concentrated on singular aspects of aging, our research presents a holistic overview, shedding light on the multifaceted aspects of the aging process from creating successful ageing index to using genome data to understand biological ageing, from risk prediction, classification of around all geriatric diseases to multimorbidity, the current study has a broader coverage of literatures in ageing field. The present review mainly focuses on studies that have applied different ML algorithms for accessing the health and well-being and diseases in the elderly population.

Table 1 An overview of reviewed studies in the field of geriatrics and machine learning
Full size table

In the present study, we sought to explore two research questions:

  1. 1.

    What is the current state of research on the application of machine learning in addressing aging-related issues?

  2. 2.

    How has machine learning been applied to study geriatric diseases, the type of population, methods, and datasets used?

So, the objective of the study is to understand the application of ML in solving ageing-related issues by studying the available literature and looking into the more refined measures and methodologies that will show a much better picture of the issue.

Methods

Literature search strategy

The systematic review followed the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) [14]. In the current study, older adults are defined as individuals aged 45 and above. The study focuses on the healthy and successful attainment of ageing, with the illness older adults face during this process. PubMed and Scopus databases were used to search original articles, as the authors have access to these databases. The search was conducted focusing on the following three major domains:

  • Machine learning

  • Geriatric population

  • Diseases that occur in old age

For machine learning, we used the keywords Machine Learning, Unsupervised Machine Learning, Supervised Machine Learning. The second set of keywords is Geriatrics, Aged, Older Population for older adults. For the diseases, Diabetes Mellitus, Hypertension, Cancer, Cardiovascular Diseases, Heart Diseases, Lung Diseases, Chronic Obstructive Pulmonary Disease, Alzheimer’s Disease, Parkinson's Disease, Dementia, Mental Health Disorders were used as keywords with OR operator in both PubMed and Scopus for searching papers. Then we combined these three sets of keywords with AND operator to get results for all three domains at once. In the PubMed database, we used the available MeSH (Medical Subject Headings) terms related to our search. Only title, abstract, and keyword sections were selected for the search to get comprehensive and only important literature in the field. The literature search was done on 7th April 2023 and 10th October 2023. The search strategy is clearly explained in the Table 2.

Table 2 Search strategy
Full size table

The inclusion criteria of the study were:

  • Peer-reviewed papers with available full-text

  • Study includes older population and must have focused on any geriatric disease

  • Implementation of machine learning algorithm

Primarily, the research articles were excluded using the filter option of the database, based on the language (only English language articles were selected). We excluded papers focused on only the younger population, or have applied different methodology other than machine learning, and the papers whose full-text version is not available on the web are also excluded from our study. However, we included the studies covering both younger and older populations; and some studies have used deep learning methods, as this method is the advanced application of machine learning methods and is often considered as a subfield of it [15], the studies using only deep learning methods were decided to be included. Both the reviewers (A.D. & P.D.) carefully screened the titles and then abstracts of all the papers and discarded the papers that did not come under the scope of our research and violated our inclusion criteria. The selection of articles was initially done by one reviewer (A.D.) and then checked by another reviewer (P.D.). Figure 1 illustrates the PRISMA flowchart showing the number of articles screened, included, and excluded in each step.

Fig. 1
figure 1

PRISMA Flowchart

Full size image

Data collection process and critical analysis:

Both the reviewers (A.D. & P.D.) studied all the selected articles to collect useful information. The research articles were thoroughly read by one author (A.D.) and the objectives, type of data used and the machine learning methods used were summarised in an Excel sheet, which was further examined by another author (P.D.) to check the correctness.

Quality assessments of studies

The 70 studies were reviewed for quality by two authors, AD and PD, independently, and arrived at one score with mutual agreement using the Joanna Briggs Institute's (JBI) critical appraisal tool [16]. This tool is commonly employed for evaluating the methodological quality of studies, specifically analytical cross-sectional studies. It comprises eight questions that address issues related to internal validity and the risk of bias, including aspects such as confounding, selection, and the clarity of study sample reporting. A high risk of bias was determined if positive responses were 49% or lower, a moderate risk of bias if the measure fell between 50 and 69%, and a low risk of bias if positive responses exceeded 70%.

Results

The use of machine learning in geriatrics and ageing research has been on a significant rise since 2019, which is shown in Fig. 2. In 2021, the number of published papers was the highest compared to any other year.

Fig. 2
figure 2

Research papers using ML over the years (from PubMed and Scopus databases)

Full size image

Initially, we found 249 research articles from the PubMed database and 1415 research articles from the Scopus database, after using the filters according to our inclusion criteria, the number was cut down to 120 peer-reviewed research articles. However, out of these, only 100 papers had free access. After screening the titles of the articles, and screening the abstracts, a total of 70 articles were selected for the critical analysis. The score of the quality assessment of the studies conducted using the Joanna Briggs Institute's (JBI) critical appraisal tool which is available in Table- S2 of the supplementary file. Among the 70 studies, a specific number, only 3 studies showed a moderate risk of bias, while the remaining 67 studies demonstrated a low risk of bias. Importantly, none of the studies included in our review were identified as having a high risk of bias. A summary of the machine learning applications in different geriatric diseases and their objective, datasets, and findings are included in Table 3. Following are the findings from the table, grouped by disease type.

Table 3 Summary of studies
Full size table

Biological and brain age prediction

Biological age refers to an individual's age as determined by their physical and physiological health, as opposed to their chronological age brain age on the other hand, is a measurement of the age of an individual's brain based on its physical condition and function. The concept of biological age and brain age is useful in understanding an individual's overall health and risk of age-related diseases. By using neural networks and supervised ML algorithms on clinical datasets, biological age prediction [15] and brain age prediction [12] were conducted by two different studies. Ageing-related problems [13], measurement of healthy ageing [17], and the association between the health status of older adults and environmental and social factors [18,19,20] were studied extensively.

Neurodegenerative disorders

There are many studies focused on neurodegenerative disorder, which includes Alzheimer’s disease, Parkinson’s disease, Dementia, and any type of cognitive disorders. Different types of deep neural networks were applied and compared with Cox proportional hazard models to predict the neurodegenerative disorders using population-based datasets [66], and another comparative analysis was done to understand the association between ageing process and these diseases [58]. Among all kinds of neurodegenerative diseases, Alzheimer’s disease was focused a lot. Some studies used biomarker data for example white matter hyperintensities [62], and particulate matter [70] to explore their role in causing Alzheimer’s disease. On the other hand, interestingly many studies have used social media (Twitter) data, to capture the sentiment of a large number of populations regarding the stigma of the disease [67]. Insurance claim data [57] is also used for the study. Most of the studies used regularised regression models, logistic regression, and deep learning models for capturing the risk. Among neurodegenerative disorders, another part of the disease is dementia [54, 61] and cognitive dysfunctions. Machine learning was used to diagnose and predict cognitive dysfunction mostly using population-based data [55, 56, 60, 72, 75], mostly using regression models from supervised ML, another type of studies have used biomarker variables [65], digital device features [59, 69], and hospital records [68] to analyse the risk factors of cognitive dysfunction. Similarly, for dementia, most of the studies used population-based surveys [63, 71, 73, 74] and clinical datasets [64] using classification and deep learning methods of ML. Mostly, logistic regression and random forest regression performed better than the other models applied.

Non-communicable diseases

Among NCDs, diabetes, hypertension, chronic kidney diseases, cancer, and cardiovascular diseases were explored. Diabetes mellitus is very common among the older population and machine learning has been extensively applied for the detection, prediction, and identification of risk factors of the disease. Studies have developed predictive models [34, 35] based on supervised ML (logistic regression, XG Boost, decision tree, etc.) and some studies identified associated risk factors [32, 33] using clustering algorithms (like principal component analysis), logistic regression classifiers, and other supervised ML algorithms. Predicative models for hypertension were developed by using population-based datasets [39], with the association of risk factors like high waist circumference [38], cognitive impairment [37], and sleep & pulmonary measures [36] discussed. LASSO (Least Absolute Shrinkage and Selection Operator) and Ridge regression were popularly used for finding the association of the risk factors and feature section for model building. Chronic kidney disease detection [42], predicting stone-free status [40], and identifying distinct types of kidney transplants [41] were the focus areas. Patient records from the hospital were the only source for this kind of study. To detect chronic kidney disease, image processing, and deep learning algorithms were used, for predicting stone-free status, supervised ML algorithms (like logistic regression, random forest, and XG Boost regressor) were used, and lastly for identifying the kidney transplants clustering approach was used on the organ sharing data of patients. We found two studies based on colorectal cancer [22] which evaluated its risk using a random forest model on primary care health records. Another study predicted the quality of colorectal cancer surgery [24] with the 30-day mortality data from a hospital record and applied supervised ML algorithms. Many studies applied cross-sectional health survey data [25, 26, 28] and administrative databases [27, 29, 31] for predicting the risk of cardiovascular diseases. These studies used deep learning algorithms, cluster analysis, and ensemble ML to identify risk factors.

Mental health conditions

Except physical health, mental health has equal importance for the overall body condition of older adults. Many types of mental health-related issues among the older population, for example, predicting depression from smartphone data using supervised ML models [46], identifying patients with depressive symptoms using random forest decision models on primary care visits [44], suicide prediction model [43], analysing the effect of environmental factors on mental health [45, 49], quantifying the psychotherapy content and its effect [47], and studying loneliness using social media data Twitter [48] and sentiment analysis were addressed by different studies. Most of the studies applied logistic regression, random forest, and deep learning models. The studies identifying environmental factors have used image data of surroundings along with neural networks and deep learning algorithms.

Furthermore, other diseases like oral health-related diseases, bone diseases, and multimorbidity were also covered. From our advanced search, we only found one study in the field of oral health, which predicted tooth loss among older adults using a population-based dataset and supervised ML classification algorithms [84]. One study on osteoporosis [21], which used CT (Computed Tomography) scan image data to develop various supervised ML models to develop prediction models was also found. Classification and prediction models for multimorbidity were developed using deep learning models, by comparing them with other ML algorithms [50]. For developing a score for an overall health condition, a set of multimorbidity was taken into account and analyzed with the help of logistic regression, and a gradient-boosted tree classifier [53].

Type of methods and datasets used in the included studies

We've provided an overview of model performance metrics in the final column of Table 3. These metrics include the Area Under the Curve—Receiver Operating Characteristic (AUC-ROC), accuracy, specificity, sensitivity, precision, and the F1 score. Based on these performance scores, the Random Forest algorithm demonstrated the highest performance, with Extreme Gradient Boosting and Neural Networks following as the second and third-best performers, respectively. To clarify, the AUC-ROC measures the ability of a model to distinguish between different classes, accuracy represents the proportion of correctly classified instances, specificity measures the ability to correctly identify negative cases, sensitivity assesses the ability to correctly identify positive cases, precision quantifies the accuracy of positive predictions, and the F1 score combines precision and sensitivity to provide a balanced evaluation of the model's performance.

Figure 3 (a) & (b) shows the application of different types of ML methods and types of datasets used respectively. 67% of the papers included in our study have used supervised ML models, in which Random Forest, Logistic Regression, and the Decision Tree were the top three mostly used algorithms. After supervised ML, deep learning was the second most used algorithm, and lastly unsupervised ML. In unsupervised ML, the clustering technique was used often. Fig. 3 (b) shows the different types of datasets used. Nearly half of the studies (45%) used population-based surveys, while electronic health records and hospital-based data were the second highest used dataset. In the population-based datasets, two data sources English Longitudinal Study of Ageing (ELSA) and China Health and Retirement Longitudinal Study (CHARLS) were used more often. Then MRI/ CT images were the second highest used dataset. Many kinds of digitally traced data were also used, which includes internet databases like Kaggle, mobile application-based datasets, and data retrieved from social media APIs (Application Programming Interface) like Twitter.

Fig. 3
figure 3

a Different machine learning methods used for geriatric diseases*. b Type of dataset used in the included studies

* The total number of papers here is more than 70 because many papers have used more than one algorithm and are counted more than one time to make this chart

Full size image

The application of ML methods can be explained in three parts that is classification, risk prediction, and disease detection & diagnosis. The highest number of studies have predicted the risk factors or have built risk prediction models for different diseases. Under risk prediction, health survey datasets from population-based cross-sectional studies or biobank studies have been used to build the models. For the same, survival models like the Cox proportional hazard ratio model penalized regressions and supervised ML (decision tree, random forest, support vector machine, and logistic regression) were mostly used. Another type of data is nationally represented administrative population-based datasets, which have mostly focused on the sociodemographic and behavioral risk factors of the diseases. These studies have used supervised ML and deep learning methods to get the desired result. Except for these, some studies used hospital records, primary care visit records, and insurance data. The second category is disease detection and classification. Disease detection refers to the identification of the presence of a disease or condition usually through screening and testing. While, on the other hand, disease diagnosis is the process of identifying the specific disease or condition that is causing a person’s symptoms [85]. Generally, population-based datasets, biomarker data, and image-based data such as MRI, and retinal image data were used to detect or diagnose the disease in the studies. For image-based data, hierarchical clustering is used to detect the disease. In the case of other datasets, a mix of deep learning and supervised machine learning was used. The third category is disease classification. In the context of machine learning, disease classification refers to the process of developing algorithms or models that can automatically classify or categorize diseases based on labeled data separating diseases and not diseases or the severity of the illness. Ensemble ML algorithms, deep neural networks, random forests, and decision trees were the methods used for the classification of diseases using the health survey datasets containing phenotype data and national health survey datasets [86].

Discussion

The ML methods have been largely used to study different types of illness and in different kinds of datasets. The use of machine learning in the development of new treatments and interventions leads to the development of new drugs that can slow down the disease progression process [87]. The exponential increase in the use of machine learning in geriatric care is evidenced by the growing number of research studies in this field. There has been a significant number of published studies identified in the last decade and it’s growing day by day.

Machine learning is a relatively new technology in the field of geriatrics, and it has the potential to revolutionize the way we diagnose, treat, and manage geriatric diseases. In comparison to traditional methods, machine learning has several advantages that make it a promising tool for geriatric care. One of the most important advantages of ML is its ability to analyse a large amount of unstructured data like image data CT scans, MRI, tumor images, etc. It can find patterns and detect those morbidities quickly and more accurately [88,89,90]. While traditional methods rely on manual data analysis which is time-consuming and prone to human error, ML algorithms with their ability to learn and adapt, can be trained on large datasets and improve their accuracy over time [91]. In contrast, traditional methods rely on fixed rules and protocols that may not be able to adapt to the changing needs of geriatric patients. Machine learning algorithms can analyse patient data to identify those at high risk of developing a disease. This early identification can enable clinicians to intervene early and provide appropriate treatment to prevent or slow down the progression of the disease [92]. A paper by Ali et al. systematically reviewed 180 research articles according to the application of artificial intelligence in healthcare benefits, challenges, methodologies, and functionalities, which concluded that this novel method continues to outperform humans in terms of accuracy, efficiency, and fast execution of clinical processes [93]. Another systematic literature review by Battineni et al. suggested that in real-time clinical practice, there is no universally accepted approach for determining the optimal method, as each machine learning technique comes with its own set of strengths and limitations, however, Support Vector Machines (SVM) and Logistic Regression (LR) are two common machine learning methods that are used in most of the studies [94]. In another review article, the majority of the examined studies emphasized that the use of only machine learning methods or combining it with other intelligent techniques is popularly used to prevent emergencies [95]. This approach holds a significant promise for uncovering substantial patterns in both structured and unstructured datasets. The widespread adoption of these techniques generates curiosity about their global evolution and which countries utilize them most extensively. According to Tran et al., the trend of usage of machine learning and artificial intelligence in research is highest in the United States, followed by China and Italy [96]. Previous systematic literature studies have highlighted mostly clinical aspects of geriatrics and have only focused on chronic diseases [8, 10] and some have focused only on mental health disorders [9] or frailty [11]. The current study covers the total breadth of ageing from diseases to mental health problems and also the successful ageing, brain age, and biological age prediction. The study has also included literature which have used population-based surveys to build a successful ageing index by exploring new methods and datasets. Many kinds of digitally traced data [30, 46,47,48, 67, 69] are used in studies that have a future scope of application for improving geriatric research.

However, it is important to note that machine learning is not without its limitations. ML algorithms require large amounts of high-quality data to be effective, and there may be issues with data quality or bias that can impact the accuracy of the algorithms. Additionally, machine learning algorithms are not always transparent in their decision-making [97], which can make it difficult for clinicians to understand how the algorithms arrived at a particular diagnosis or treatment recommendation. So, in most cases, it may be suitable to use both the conventional methods and the ML methods side by side to get better results [98]. To summarize, machine learning offers several benefits in geriatric care such as its ability to rapidly and accurately analyse large amounts of data, learn and improve over time, and enhance the precision of diagnoses and treatment recommendations. However, it is crucial to acknowledge the potential limitations of machine learning and take necessary measures to ensure that the algorithms are fair, transparent, and validated before their implementation in clinical settings. The future scope of ML in geriatrics is vast and promising. With the aging population on the rise globally, there is a growing need for innovative technologies that can enhance the quality of care for the elderly. However, it is crucial to continue research and development to ensure that the algorithms are fair, transparent, and validated before their widespread implementation in clinical practice.

Conclusion

The current review found a wide variety of research papers analyzing different diseases using various machine learning algorithms in different kinds of datasets. Disease diagnostic criteria, risk prediction models, and factors were also highlighted and the application of machine learning in the field of geriatrics and care is well explored, but still, there is scope for future development. There is a need to validate that constructed machine learning models in large-scale datasets generalize the results across all age groups, gender, ethnicity, and other crucial factors. Also, there is a huge scope in using internet-based data and digital datasets from personalized applications in digital devices like smartphones and wearable technologies to provide customized patient-centric care for older populations.