Designing a Predictive Model for Successful Aging among the Elderly Using Machine Learning Techniques

doi:10.21203/rs.3.rs-1944869/v1

Download PDF

Research Article

Designing a Predictive Model for Successful Aging among the Elderly Using Machine Learning Techniques

https://doi.org/10.21203/rs.3.rs-1944869/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: the aging phenomenon has an increasing trend worldwide which caused the emergence of the SA¹ concept. It is believed that the SA prediction model can increase the QoL² in the elderly by decreasing physical and mental problems and enhancing their social participation. Most previous studies noted that physical and mental disorders affected the QoL in the elderly but didn't pay much attention to the social factors in this respect. Our study aimed to build a prediction model for SA based on the physical, mental, and social factors by considering all factors affecting SA.

Materials and methods: in this descriptive, applied, and retrospective study, the data of 975 related to SA and non-SA of the elderly were investigated. We used the Chi-square test at P<0.05 to determine the best factors affecting the SA. The AB³, J-48, RF⁴, ANN⁵, SVM⁶, BLR⁷, and NB⁸ algorithms were used for building the prediction models. To get the best model predicting the SA, we compared them using the sensitivity, specificity, accuracy, F-measure, and AUC.

Results: The Chi-square test showed that 28 variables had a meaningful relationship with SA. The results of comparing the ML⁹ model's performance showed that the RF with sensitivity=0.91, specificity= 0.98, accuracy= 0.95, F-test=0.9, and AUC-test= 0.884 is the best model for predicting the SA.

Conclusion: using prediction models can increase the QoL in the elderly and consequently reduce the economic cost for people and societies. The RF can be considered an optimal model for predicting SA in the elderly.

successfulaging

data mining

prediction model

quality of life

elderly

Aging is a concept that is increasingly widespread worldwide(1). So far, this concept and the precise amount of age at which the elderly period begins have not been well defined(2, 3), and there is a difference among nations in this respect. The precise amount of the agreed age rate of the elderly has not been introduced by pundits, and they have different viewpoints on the ages beginning the elderly. In some studies, the ranges of age associated with the elderly are 60, 65, or even 80 years old to higher, in which the phenomenon named fall occurs(4–6). This phenomenon deteriorates individuals' physical and cognitive health, increases the degree of dependence, and declines the aged group's quality of life and self-confidence(7, 8). In recent years, the population's life expectancy has increased due to the decrease in birth rate, improvements in training people, medical and health conditions, technology status, and nutrition betterment(9, 10). It is estimated that the number of elderly around 80 years old will increase from 143 million in 2019 to 426 million by 2050(11). Also, the life expectancy will increase to 82 years old in 2100(12). Due to the increased elderly population and life expectancy and attention to support the elderly from various physical, mental, and social factors, the SA concept is widespread worldwide by humans leading to a better quality of life among the elderly(13). This trend has raised much attention to understanding this concept and improving the quality of life among the elderly(14).

Some theories that emerged from the 1960s to 1970s described the SA concept. The disengagement theory, introduced by Cumming & Henry, characterized SA as the increased participation of the elderly in social activities(15). The activity theory discussed engaging the elderly in community relationships was represented by Havighurst. Atchley has represented the continuity theory describing the ability of the elderly to do personal and social activities in the latest duration of their life, similar to early(16). A successful aging pattern and tracing its peculiarities is a qualitative approach to the aging period(17). Rowe and Kahn have introduced that the SA model as a comprehensive approach pertains to people's lifestyles in three dimensions: 1- the low possibility of physical disease and disease-associated conditions, 2- high mental and physical performance capacity, and 3- active participation in social life(18). They believed that genes do not influence biological aging, and lifestyle has also potentially affected aging(19). This idea also has been scrutinized and confirmed by the world's great gerontologists(20–22). Therefore, environmental conditions and focus on the people's lifestyle can be considered predictive factors in determining the SA, leading to a better quality of life in the elderly (18). Using a model for predicting SA in the early stages of the elderly's life or even in middle-aged people leads to higher physical performance, lower physical and mental disorders, and better social situations increasing the success in these people's last years of life (23). Several studies have shown that the predictive models have a potential role in different medical conditions such as prognosis, diagnosis, and treatment of physical, cognitive, and mental disorders (24–26). The studies have focused on the predictive models in the SA and QoL in the elderly based on physical and cognitive factors. They have paid less attention to the social factors affecting the SA prediction(27, 28). Social activities for the elderly are crucial and should be considered in SA(29). The predictive models' ML can forecast SA betterment more efficiently(30). This study aims to build the SA model using various ML techniques to better predict the SA in people by considering their physical, mental, and social factors.

This study was an applied and descriptive method and performed in two steps as follows:

1. Data description: In this study, we used the 975 records for investigating the most important factors affecting the SA and building the prediction model, which included 515 and 460 records associated with females and males, respectively, in the database gathered from January 2019 to January 2021 in three elderly centers of Hamdeli, Mehrjooyan, and Hasti in Ahvaz city.

Predictor Features

Socio-demographic variables

this category consists of age, gender (male and female), literacy level, marital status, occupation type, source of income, monthly income, and insurance situation.

Chronic diseases

This class of variables is divided into cancer, cardiovascular accident (CVA), diabetes, high blood pressure, liver, renal, eye, bone, muscle, and other functional diseases, depression, and convalescences.

Behavioral and psychosocial factors

By considering the factors existed in this research, a SA person has a pleasure level of personal independence (the grade ranging from 90 to 100 according to the Barthel index), best life satisfaction (the grade ranging from 20 to 35 from the Diener Life Satisfaction Scale which was ranged from 5 to 35) with a pleasant quality of life (the grade higher than 70 in the SF36 questionnaire which was ranged from the 0 to 100). The determinant of these factors can be described below

Quality of life: This factor can evaluate the quality of life and health. Ware and Sherbourne designed it in 1992. It included 36 questionnaires and eight contexts: physical and social ability, physical and cognitive active participation, psychological health, assessment vivacity, physical pain, and general health status. Moreover, 36-SF consists of two general measurements of physical, mental, and social health known as the total physical and mental component score. The subject in these contexts is ranged from 0 to 100. The higher amounts mean that the quality of life for the elderly is better. The reliability and validity of this questionnaire have been investigated and justified among the Iranian samples(31, 32).

Individual independence

independence in people is calculated through the Barthel Index. It poses ten questions and computes the level of physical health in people. This variable specifies each person's ability in various domains of daily functions on one scale ranging from 0 to 100. The 0 to 20 pertained to very dependent people, 20 to 60 were entirely dependent, 61 to 90 were medium dependent, 99 to 91 were little dependent, and the grade of 100 was associated with the fully independent person(33).

Satisfaction with Life scale (SWLS)

This criterion was introduced by Diener et al. It included five items assessing the cognitive parts of health and well-being. Each condition consists of seven options and ranges from 1 to 7, from highly disagree to agree strongly, in which a higher score shows better life satisfaction. The validity of this questionnaire was justified by Bayani et al. in 2007(34).

Lifestyle

determining the lifestyle depends on the whole grade obtained. It can be obtained by specifying a grade ranging from 42 to 98 associated with the unpleasant lifestyle, 99 to 155 for medium lifestyle, and 156 to 211 pertained to a pleasant lifestyle. It evaluates exercise, healthy eating, stress management, social and interpersonal relationships, physical activity, and recreation(35).

2. Data analysis: After gaining the essential factors affecting the SA and preparing our dataset, we performed the data analysis. This step includes preprocessing the dataset, selecting and implementing the ML models, and assessing to get the best predictive model affecting the SA. In preprocessing phase, first, we resolved inconsistencies in our database when integrating the different databases in various elderly centers. Second, we cleaned our data. In this respect, we identified and smoothed any noises in our databases by determining the quantiles of our data point distributions. For the data points that had the missing values in the output class, we removed the records having that missing class. Also, to fill the missing values in the features predicting the SA, we imputed the missing values through regression methods to predict the missing values by other existing values in attributes. The trimmed mean was used for numerical data to replace the missing numerical data with the most negligible bias. Third, because of inequity in the number of samples pertained to output classes, we used the Synthetic Oversampling Technique to balance the number and to prevent form biases when evaluating the algorithms' performance. Forth, to better analyze the dataset and reduce the data dimensions, we used the Feature Selection (FS) method. FS is selecting the best variables and decreasing the dataset dimensions. This process has potential benefits such as removing irrelevant data, precluding form overfitting of the algorithm, increasing the ML training speed, reducing the data redundancy, and higher learning accuracy(36–38). In this study, we used the independence test of Chi-square (χ²) to determine the relationship between each factor affecting the SA prediction and SA. The variables with the relationship at P < 0.05 were considered statistically meaningful in this regard. Other factors that didn't meet this statistical level were removed from the study.

To perform the ML process to get the most common model determining the success among adults based on the best factors affecting the success obtained from statistical analysis, we used the Weka V 3.9 software. Seven ML algorithms were selected and implemented in this respect because the most commonly used with higher performance than other algorithms in recent studies.

• RF: the RF decision tree is a well-known machine learning algorithm for creating decision models. This algorithm consists of many subtrees for classifying the dataset samples. The RF algorithm trees are built using the Classification and Regression (CART) technique with no pruning. In this algorithm, the splitting process occurs through the random selection of the variables. This feature gave the RF a high capability for sample classification, especially in large dataset dimensions. This algorithm is suitable for the high-dimensional dataset with many variables and uses the voting process to classify the samples with high accuracy. In other words, this algorithm performance is shared in its most sub-trees with the increased classification capability. In large datasets, this algorithm, through distributing the samples in subtrees with different dataset attributes, can pose a high accuracy even with noisy data and prevent the overfitting of the algorithm. Generally, the RF advantages can be noted as fast and accurate in the training process with resistance to noisy data as its flexibility(39–42).

• Ada-boost: the ada-boost or adaptive-boost in one type of boosting algorithm from ensemble category using the weak classifiers to classify the cases simultaneously and identify and remove the errors in the classified cases in each classifier sequentially. Some advantages of this algorithm can be enumerated as the generalizability with high accuracy, efficient calculation ability, flexibility for various tasks with sophisticated data, easily adaptable, and the capability to integrate with other algorithms(43, 44).

• J-48: this algorithm is the newer model but similar to the ID3 algorithm. Its generic name is C4.5, but in Weka software is named J-48. This algorithm uses the info gain to split the tree using the variable with the highest entropy. Therefore, this algorithm's splitting process is not random, contrary to the RF. The tree is built based on features capability in classifying samples with the highest entropy recognized during the training process. This algorithm has more pruning features than others, which allows the J-48 to have less capability to be overfitted than other ML models. Generally, some beneficial algorithm features are: using the confidence factor to set the tree size and preclude from overfitting process, the capability of working with continuous variables contrary to other decision tree algorithms, extracting rules with the maximum discrimination between the output classes, and the ability to work with missing values(45–47).

• ANN: the ANN tries to imitate human behavior, and its structure consists of neurons and weights between them to transfer messages around the network like humans. Generally, the ANN includes three layers, including input layers which reciters the input data such as signals, images, or any other data type. The hidden layer is responsible for calculation in ANN, receiving data from input layers, and considering the calculation nodes that process the input data. The results of the computation process that occurred in the hidden layers will transfer to the output layers. The last layer is the output layer, and in this layer, the users can see the output of the ANN, which is represented by this layer. So far, the ANN has various applications for solving highly complex computational problems in medical conditions(48–51).

• SVM: The SVM is one of the classification and regression algorithms using the hyperplane concept for performing the classification task. This algorithm uses the mapping process to convert the low dimensional data points to the higher dimensions for classifying the cases, which can be used through the kernel functions. Different kernel types are used for various types of datasets in terms of the complexity of data so that the SVM algorithms can be categorized as linear and non-linear types(52, 53).

• NB: The NB algorithm is a statistical classification based on the Bayesian theory using probabilistic hypothetic concepts. It determines the probability of each sample belonging to different data classes using Eq. 1. The new samples will be classified in one class having the highest probability. In this algorithm, each variable's occurrence is independent of determining the dependent variable(54).

Equation 1:\(\text{P}\left(\text{A}/\text{B}\right)=\frac{\text{P}\left(\text{B}/\text{A}\right)}{\text{P}\left(\text{B}\right)}\left(\text{P}\left(\text{A}\right)\right)\)

• BLR: This algorithm (Eq. 2) uses the statistical method to predict the probability of occurrence of each state of the output class through the input variables. In contrast with the NB, the occurrence of each independent variable is not separate from the other. Therefore, these variables affect each other in predicting the output class and have a mixed correlation. The BLR is adaptable to the dataset with two values of output class, for example, having a disease or not.

Equation 2: \(\text{P}=\frac{{\text{e}}^{\text{B}0+\text{B}1\text{X}1+\text{B}2\text{X}2+\text{B}3\text{X}3+\dots .}}{1+{\text{e}}^{\text{B}0+\text{B}1\text{X}1+\text{B}2\text{X}2+\text{B}3\text{X}3+\dots .}}\)

To get the best model to determine the success among the elderly, we evaluated the performance of selected ML algorithms using the confusion matrix (Table 1) and sensitivity (Eq. 3), specificity (Eq. 4), accuracy (Eq. 5), F-Measure (Eq. 6), and AUC (area under the ROC (Receiver Operator Characteristics) curve) obtained from the confusion matrix. In Table 1, the True Positive (TP) and True Negative (TN) are the successful and unsuccessful cases correctly classified by the decision models. The False Negative (FN) and False Positive (FP) are the successful and unsuccessful adults incorrectly categorized by the algorithms. In this study, 70% and 20%, and 10% of the samples were assigned for training, testing, and validation of the algorithms. The ten-fold cross-validation was considered to measure the errors while evaluating the performance criteria.

Table 1

Confusion matrix
		Predicted cases
		+	-
Real cases	+	TP	FP
Real cases		FN	TN

Equation 3:\(\text{S}\text{e}\text{n}\text{s}\text{i}\text{t}\text{i}\text{v}\text{i}\text{t}\text{y}=\frac{\text{T}\text{P}}{\text{T}\text{P}+\text{F}\text{N}}\)

Equation 4: \(\text{S}\text{p}\text{e}\text{c}\text{i}\text{f}\text{i}\text{c}\text{i}\text{t}\text{y}=\frac{\text{T}\text{N}}{\text{T}\text{N}+\text{F}\text{P}}\)

Equation 5: \(\text{A}\text{c}\text{c}\text{u}\text{r}\text{a}\text{c}\text{y}=\frac{\text{T}\text{P}+\text{T}\text{N}}{\text{T}\text{P}+\text{F}\text{N}+\text{F}\text{P}+\text{T}\text{N}!}\)

Equation 6:\(\text{F}-\text{M}\text{e}\text{a}\text{s}\text{u}\text{r}\text{e}=2\text{*}\left(\frac{\text{S}\text{e}\text{n}\text{s}\text{i}\text{t}\text{i}\text{v}\text{i}\text{t}\text{y}\text{*}\text{S}\text{p}\text{e}\text{c}\text{i}\text{f}\text{i}\text{c}\text{i}\text{t}\text{y}}{\text{S}\text{e}\text{n}\text{s}\text{i}\text{t}\text{i}\text{v}\text{i}\text{t}\text{y}+\text{S}\text{p}\text{e}\text{c}\text{i}\text{f}\text{i}\text{c}\text{i}\text{t}\text{y}}\right)\)

After removing the records with missing values in the output class, 15 records (1%) belonged to successful and unsuccessful cases were excluded from the study. Therefore, after using the SMOTE method, 1465 records were used for analysis; 746 and 719 cases were associated with unsuccessful and successful cases, respectively. The results of using the independence test of Chi-square are shown in Table 2.

Table 2

The results of the Chi-square test of all variables
Variables	Variables encode	Frequency	χ²	P_-Value
Age	60–70 (1) 70–80 (2) > 80 (3)	60–70 (964) 70–80 (371) > 80 (130)	3.943	0.0139
Sex	Female (1) Male (2)	Female (758) Male (707)	20.441	< 0.001
Educational level	No literacy (1) Elementary (2) Diploma (3) Academic (4)	No literacy (861) Elementary (338) Diploma (172) Academic (94)	6.113	0.106
Marital situation	Married (1) Single (2) Divorced (3) Widowed (4)	Married (1014) Single (39) Divorced (23) Widowed (389)	2.019	0.568
Occupation type	No job (1) Housekeeper (2) Retired (3) Self-employment (4)	No job (240) Housekeeper (605) Retired (402) Self-employment (218)	3.090	0.378
Income level	Under poverty line (1) On poverty line (2)	Under poverty line (1134) On poverty line (331)	15.546	< 0.001
Family support	Have (0) Haven't (1)	Have (675) Haven't (790)	14.121	0.01
Insurance situation	Have (1) Haven't (2)	Have (1293) Haven't (172)	0.005	0.946
High blood pressure	Have (1) Haven't (2)	Have (1124) Haven't (341)	82.943	< 0.001
Cardiovascular diseases	Have (1) Haven't (2)	Have (472) Haven't (993)	30.832	< 0.001
Bone diseases	Have (1) Haven't (2)	Have (790) Haven't (675)	9.067	0.003
Renal diseases	Have (1) Haven't (2)	Have (245) Haven't (1220)	0.444	0.505
Liver diseases	Have (1) Haven't (2)	Have (143) Haven't (1322)	4.331	0.037
Muscle diseases	Have (1) Haven't (2)	Have (887) Haven't (578)	24.539	< 0.001
Depression	Have (1) Haven't (2)	Have (449) Haven't (1016)	5.864	0.015
Convalescences	Have (1) Haven't (2)	Have (78) Haven't (1387)	2.446	0.118
Eye disease	Have (1) Haven't (2)	Have (286) Haven't (1179)	0.197	0.657
Diabetes	Have (1) Haven't (2)	Have (738) Haven't (727)	58.542	< 0.001
Cancer	Have (1) Haven't (2)	Have (136) Haven't (1329)	12.647	< 0.001
Other diseases	Have (1) Haven't (2)	Have (68) Haven't (1397)	0.828	0.661
Activity daily living	Have (1) Haven't (2)	Have (876) Haven't (589)	10.225	< 0.001
Sporting exercises	Have (1) Haven't (2)	Have (453) Haven't (1012)	8.889	0.03
Exercise duration	Haven’t (1), < 0.5 hour (2), > 0.5 hour (3)	Haven’t (778), < 0.5 hour (354), > 0.5 hour (333)	14.334	0.021
Exercise type	Laborious exercise (1), soft exercise (2), none (3)	Laborious exercise (89), soft exercise (598), none (778)	13.336	0.08
Sexual condition	Healthy (1) Unhealthy (2)	Healthy (525) Unhealthy (940)	7.775	0.01
Tension control capability	Low (11) High (12)	Low (989) High (476)	1.614	0.02
Social relationships with other persons	Weak (9) Strong (10)	Weak (418) Strong (1047)	1.379	0.001
Assessment of the nutritional situation	Fair (1) Good (2)	Fair (923) Good (542)	16.621	0.001
Assessment of malnutrition situation	Have (1) Haven't (0)	Have (620) Haven't (845)	14.228	0.001
Life satisfaction	Pleasant (1) Unpleasant (2)	Pleasant (1051) Unpleasant (414)	41.028	< 0.001
The general explanation of lifestyle	Undesirable (1) Medium (2) Desirable (3)	Undesirable (244) Medium (1203) Desirable (18)	13.188	0.001
General health	Have (1) Haven't (2)	Have (924) Haven't (541)	10.453	< 0.001
Assessment of body pain	Have (1) Haven't (2)	Have (674) Haven't (791)	7.42	0.11
Physical dysfunction	Have (1) Haven't (2)	Have (370) Haven't (1095)	13.212	0.01
Fatigue	Have (1) Haven't (2)	Have (855) Haven't (610)	8.260	0.162
Mental dysfunction	Have (1) Haven't (2)	Have (616) Haven't (819)	32.365	< 0.001
Social dysfunction	Have (1) Haven't (2)	Have (935) Haven't (530)	18.552	< 0.001
Quality of life	Low (3) High (4)	Low (994) High (471)	13.768	< 0.001
Physical activity	Low (1) High (2)	Low (1092) High (373)	3.655	0.056
Performing debarment activities when occurring disease	Have (7) Haven't (8)	Have (949) Haven't (516)	4.096	0.027

Based on the information provided in Table 2, the variables of age, sex, income level, family support, high blood pressure, cardiovascular diseases, bone diseases, liver diseases, muscle diseases, depression, diabetes, cancer, activity in daily living, sporting exercises, exercise duration, sexual condition, tension control capability, social relationship with other persons, assessment of the nutritional situation, assessment of malnutrition situation, life satisfaction, the general explanation of lifestyle, general health, physical, mental, and social dysfunctions, quality of life, and performing debarment activities when occurring diseases with independence test of chi-square at P < 0.05 gained the meaningful relationship with SA as the output class statistically. The variables of educational level, marital situation, occupation type, insurance situation, renal diseases, convalescences, eye diseases, other diseases, exercise type, assessment of body pain, fatigue, and physical activity with the independence chi-square at P > 0.05 were not recognized as the essential factors statistically; therefore, they were removed from the study. The results of classifying the samples using the confusion matrix are shown in Table 3.

Table 3

The results of the confusion matrix of selected models
Algorithm	Hyperparameter	TP	FN	FP	TN
Ada-boost	Number of decimal places= "2"; Number of iterations= "15"; Classifier= "Rep-tree"; Weight threshold= "100"; Use resampling= "true";	637	109	90	629
ANN	Learning rate= "0.3"; Hidden layer= "15"; Validation threshold= "20"; Normalize attributes and numeric class= "true"; Momentum= “0.2”; Training time= "500";	606	140	209	510
SVM	Kernel type= “RBF”; C= "15"; Gamma= “1.0”; RBF gamma= '0.1'; Regression precision= ‘0.1’;	585	161	108	611
J-48	Confidence factor= “0.2”; Number of folds= "3"; Minimum number of objects per leafs= "2"; Collapse tree= "true"; Binary splitting =" true";	620	126	142	577
RF	Ensemble decision tree type= "Decision Stump"; Maximum depth= "8"; Number of iterations= "20"; Number of execution slots= "2"; Number of randomly selected features= "2";	679	67	12	707
BLR	Ridge value in log-likelihood= “1.0E-8”; Gradient descent= “true”; Maximum number of iterations= "10"; Number of decimal places= "2";	543	203	238	481
NB	Estimator type= “simple”; Number of decimal places= "2"; Use AD tree= "true"; Search algorithm= "K2 algorithm";	506	240	283	436

Based on Table 3, the RF algorithm with TP = 679, FN = 67, FP = 12, and TN = 707 has categorized the positive and negative cases with higher performance than other ML models. In contrast, the NB algorithm with TP = 506, FN = 240, FP = 283, and TN = 436 had the lowest performance in this regard. The results of sensitivity, specificity, and accuracy of selected ML models have been illustrated in Fig. 1.

Based on Fig. 1, the RF model with sensitivity = 0.91, specificity = 0.98, and accuracy = 0.95 obtained the best performance of other ML models in classifying successful and unsuccessful cases. Also, the adaboost with sensitivity = 0.85, specificity = 0.87, and accuracy = 0.86, and J-48 with sensitivity = 0.83, specificity = 0.8, and accuracy = 0.82 with sensitivity, specificity, and accuracy more than 0.8 gained favorable performance in this regard. The NB algorithm with sensitivity = 0.68, specificity = 0.61, and accuracy = 0.61 obtained the lowest performance in classifying the successful and unsuccessful cases than others. The results of the F-measure in train, test, and validation modes of all ML models are shown in Fig. 2.

Based on Fig. 2, the RF model with F-_train=0.98, F_{− validation}=0.95, and F_− test=0.9 had a higher performance than other ML models in all states of the train, validation, and test modes. Generally, the results of assessing the F-measure of all ML models showed that the F-measure rate in test mode had not dropped much. Therefore, all the algorithms have not been overfitted by the data of training samples. Also, the Ada-boost with F_− train= 0.89, F_{− validation}= 0.87, and F_− test= 0.81 and SVM with F_− train=0.85, F_{− validation}=0.82, and F_− test=0.81 obtained favorable performance during training, test, and validation modes of algorithms. The ROC curve of all selected ML models is shown in Fig. 3 (the vertical and horizontal vertices are ranged from 0 to 1 and represent sensitivity and 1-specificity, respectively).

Based on Fig. 3, The ROC curve for the RF model was closest to the specificity vertices. The RF model with AUC-train = 0.955, AUC-validation = 0.931, and AUC-test = 0.884 obtained the best ability to classify successful and unsuccessful cases. The NB model' ROC curve was closest to the 1-specificity vertices than other ML models, and with AUC-train = 0.67, AUC-validation = 0.634, and AUC-test = 0.555 has gotten the lowest performance in classifying the cases. Generally, in this study, the RF model with sensitivity = 0.91, specificity = 0.98, and accuracy = 0.95, F-train = 0.98, F-validation = 0.95, and F-test = 0.9, AUC-train = 0.955, AUC-validation = 0.931, and AUC-test = 0.884 obtained the best performance than other ML models in classifying the successful and unsuccessful cases associated with the elderly, therefore, was considered as the best model for predicting the SA among the elderly. In Fig. 4, we depicted the part of the RF model, and some of the extracted rules from this model are described and interpreted.

Based on the information existed in Fig. 4, we extracted three important rules with more classified instances found in leaf nodes.

IF (hypertension = 2) && (life satisfaction = 1) Then SA = 1;

IF (hypertension = 1) && (life satisfaction = 1) && (diabetes = 2) Then SA = 1;

IF (hypertension = 1) && (life satisfaction = 2) Then SA = 0;

In rule1, in the condition that one elderly haven't blood pressure and has a pleasant life satisfaction, the RF model predicts that the elderly is successful, and 210 instances confirm this rule. 41 instances classified this rule as unsuccessful, but most categorized it as successful (precision = 0.83). In rule 2, if one elderly has hypertension, their life satisfaction is pleasant, and he hasn't diabetes, the RF will predict that person is a successful elderly. Among 270 cases in the leaf, 217 of them confirmed this rule (precision = 0.803). Rule 3 represents that in the situation that one older adult has high blood pressure and unpleasant life, satisfaction will be predicted as the unsuccessful person by the RF model. For this rule, among the 360 classified cases, 283 were classified correctly by the model, and the precision of the prediction was 0.786. The relative importance (RI) of factors for predicting the SA obtained by the RF model are shown in Table 4 based on the average impurity decrease and the number of nodes using that attribute in the model.

Table 4

The most important features obtained from the RF model
Variable	Number of nodes used in RF model	RI
Lifestyle pleasure	1456	0.4
Quality of life	2618	0.32
Assessment of the nutritional situation	2819	0.3
Diabetes	1300	0.29
CVA	1573	0.28
Life satisfaction	1337	0.28
Sex	2141	0.28
Tension	2970	0.27
Age	2611	0.27
High blood pressure	1125	0.27
Performing debarment activities when occurring disease	2341	0.27
Bone diseases	3020	0.26
Muscle diseases	2126	0.26
Interpersonal relationships with others	2714	0.25
Depression	2274	0.24
Insurance situation	1702	0.22
Liver diseases	1536	0.22
Cancer	904	0.17

Based on Table 4, the RI of all factors associated with the SA prediction ranged from 0.17 to 0.4. The three variables of lifestyle pleasure (RI = 0.4), quality of life (RI = 0.32), and assessment of the nutritional situation (RI = 0.3) were considered the best factors affecting the SA among the selected variables. The three variables of insurance situation (RI = 0.22), liver diseases (RI = 0.22), and cancer (RI = 0.17) obtained the lowest relative importance score by the RF model; therefore, they were recognized as the least important variables affecting the SA prediction by the model.

In this study, we aimed to build a predictive model for SA by using ML models. We first reviewed and extracted the articles associated with factors affecting the SA. Then the most critical factors were obtained based on the expert opinions, and the factors were in agreement by more than 60% of the experts included in this study. Second, to investigate the relationship of each factor with SA and get the essential factors statistically, we used the independence test of Chi-square at P < 0.05. In the third step, after preprocessing the data and gaining the best factors affecting the SA, we selected the appropriate ML models to implement the model predicting the SA. To this end, we selected the AB, J-48, RF, ANN, SVM, BLR, and NB algorithms. Finally, the best predictive model for SA is obtained based on comparing and evaluating all the ML models ' performance and selecting the best one with the highest performance through various performance criteria such as sensitivity, specificity, accuracy, F-measure, and AUC in all training, testing, and validation states.

So far, few studies have been conducted concerning SA prediction models, especially in the ML domain. Most studies have focused on the elderly's quality of life (QoL) and investigated the physical, mental, and social factors separately. Cai et al. built a prediction model for SA based on the physical factors using machine learning approaches. Their studies showed that the deep learning model with AUC = 90%, specificity = 93.1%, and accuracy = 83.9% gained the best performance to predict the SA. Also, the variables of age, sitting and standing for 30 seconds, arm curl, and reaction time were recognized as essential factors affecting the SA by the model(30). In our study, the mental and social factors affecting the SA prediction are considered in addition to the physical factors. Based on these factors, the RF model with sensitivity = 0.91, specificity = 0.98, and accuracy = 0.95 obtained the best performance predicting the SA, and variables of lifestyle pleasure, quality of life, and assessment of the nutritional situation were extracted as the best factors by the RF model. Li et al. estimated the QoL using physical factors, laboratory measurements, demographic factors, and healthy behaviors based on machine learning algorithms. They found that age, walking speed, sleep duration, handgrip strength, body mass index, blood pressure, and sitting and standing duration are the most critical factors affecting the QoL(55). In the present study, the QoL is considered one of the dimensions affecting the SA, and the social and mental factors are considered in addition to the introduced factors in that study. Kyoung-Sae Na conducted a study to predict cognitive impairment two years later in the elderly. They showed that the Gradient boosting machine (GDM) algorithm with sensitivity = 0.96, specificity = 0.825, and AUC = 0.921 gained the best ability for cognitive impairment prediction, and age was considered the best factor obtained from GDM(56). In the current study, age with (RI = 0.27) is considered the most factor for predicting SA.

Moreover, in our study, we spotted the social and physical factors and mental factors to build the prediction model for increasing the QoL in the elderly. Byeon constructed the predictive model for social participation of the elderly using the ANN and Quest algorithms. The ANN with AUC = 0.718 and Quest with 0.754 with 10-fold cross-validation were relatively efficient in classifying the elderly as a social activity. The subjective health situation with (Normalized importance) NI = 100% and age with NI = 89% were reported as the essential factors in this regard(57). In the current study, the factors affecting SA were noticed more with the inclusion of diseases that may be existed in the elderly and constrained physical and social participation. Moreover, more ML models, including ensemble and non-ensemble, are utilized to predict the SA among the elderly.

So far, the predictive model for predicting the SA using ML models has not been introduced. So, the novelty of our study was using the physical, mental, and social factors based on Rowe and Kahn's theory to build the model for predicting the SA based on ML models. Our model can predict the success or unsuccess of the people in the earlier periods of the elderly; therefore, earlier detection of SA in the elderly can assist them by introducing the solution for people and physical and mental care providers to advance the lifestyle and quality of life in this aged group people. The mental and physical disorders will be detected in the earlier lifespan of the elderly, and the probability of SA in these people will be increased.

The limitations of our study were first using the qualitative variables in the database as the factors affecting the SA. Using the qualitative variables may somewhat decrease the performance of our ML models. But, considering the physical, mental, and social factors and many variables, the qualitative variables might be beneficial. Second, we were forced to fill by regression method and the feature with missing values and also deleted the records having missing values in the class field; therefore, this may affect the performance and generalizability of our ML models. Third, when integrating three databases, we may miss some attribute values, which may cause the model's performance to be reduced. Fourth, due to the imbalance in the number of output classes pertained to the SA and non-SA cases, we used the (Synthetic Minority Oversample Technique) SMOTE to troubleshoot this problem, and this may affect our model performance by increasing the number of samples without the high quality of data existed in their features. In the future, to get a more optimal SA predictive model for better generalization, we suggest training ML models with more data from more nursing centers and filling missing data in features by acquiring the actual data instead of imputing methods. Also, we recommend training and testing the data through more algorithms to get better performance in models predicting the SA.

Using the predictive model for SA based on ML models can lead to better QoL by decreasing mental and physical morbidity and increasing the physical and social participation of the elderly. This study used AB, J-48, RF, ANN, SVM, BLR, and NB algorithms to develop prediction models for SA. The research shows that the variables of lifestyle pressure and QoL are the best factors affecting SA, among others. We conclude that the RF is the best model for SA based on evaluating various ML models. Therefore, this model can assist the gerontologist in increasing the speed of assessing the situation of SA in elderlies and introduce the best solution for increasing the QoL by considering the physical, mental, and social factors using the best knowledge extracted from the data.

SA = Successful aging
QoL = Quality of life
AB= Adaptive-Boost
RF= Random Forest
ANN= Artificial Neural Network
SVM= Support Vector Machine
BLR= Binary Logistics Regression
NB= Naïve Bayes
ML= Machine Learning

Ethics approval and consent to participate

This article was extracted from the research project approved by the research committee affiliated with the Iran University of Medical Sciences (IUMS) with the ethical code of IR.IUMS.REC.1401.150. All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all subjects and/or their legal guardian(s).

Consent for publication

Not applicable

Availability of data and materials

The datasets generated and/or analyzed during the current study are not publicly available to the privacy concerns of research committee but are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

The Iran University of Medical Sciences financially supported this article.

Authors' contributions

Study concept and design: M. A., and R. N.; analysis and interpretation of data: R. N., and S. N.; drafting of the manuscript: M. A.; critical revision of the manuscript for important intellectual content: M. A., R. N., and S. N.; statistical analysis: R. N.

Acknowledgements

We thank the people and specialists who assisted us in all steps of this study, especially, in completing the questionnaire.

Machado AH, Casarin VA, De Mirinda RL, Franco MB, Dos Santos AV, Wbatuba BBR. O comportamento de consumo de vestuário de um grupo de terceira idade: Estudo de caso em um Município do Estado do Rio Grande do Sul. Revista Espacious. 2016;37(5):19.
Coto M, Lizano F, Mora S, Fuentes J, editors. Social Media and Elderly People: Research Trends. Social Computing and Social Media Applications and Analytics; 2017 2017//; Cham: Springer International Publishing; 2017.
Pisano M, Ceresoli M, Cimbanassi S, Gurusamy K, Coccolini F, Borzellino G, et al. 2017 WSES and SICG guidelines on acute calcolous cholecystitis in elderly population. World Journal of Emergency Surgery. 2019;14(1):1–16.
Wiggins T, Markar SR, Mackenzie H, Jamel S, Askari A, Faiz O, et al. Evolution in the management of acute cholecystitis in the elderly: population-based cohort study. Surgical Endoscopy. 2018;32(10):4078–86.
Yu Z, Kong D, Peng J, Wang Z, Chen Y. Association of malnutrition with all-cause mortality in the elderly population: A 6-year cohort study. Nutrition, Metabolism and Cardiovascular Diseases. 2021;31(1):52–9.
Prasad GL, Anmol N, Menon GR. Outcome of traumatic brain injury in the elderly population: a tertiary center experience in a developing country. World neurosurgery. 2018;111:e228-e34.
Sotoudeh GR, Mohammadi R, Mosallanezhad Z, Viitasara E, Soares JJF. The prevalence, circumstances and consequences of unintentional falls among elderly Iranians: A population study. Archives of Gerontology and Geriatrics. 2018;79:123–30.
Mentis M, Gouva M, Antoniadou E, Mpourdoulis K, Kesoudidou I, Marneras C, et al. Psychological and Physical Problems in Elderly People with Problems of Falls. European Psychiatry. 2017;41(S1):S659-S60.
Chalise H. Provincial situation of elderly population in Nepal. Aging Sci Res. 2020;1(1):9–11.
Afshar PF, Asgari P, Shiri M, Bahramnezhad F. A review of the Iran's elderly status according to the census records. Galen Medical Journal. 2016;5(1):1–6.
Pilleron S, Soto-Perez‐de‐Celis E, Vignat J, Ferlay J, Soerjomataram I, Bray F, et al. Estimated global cancer incidence in the oldest adults in 2018 and projections to 2050. International journal of cancer. 2021;148(3):601–8.
Purnomo E, Komalasari R, Manurung EI, editors. Elderly Patients Quality of Life at Outpatient Department Hospital X, Banten. The 1ST ICGH international conference on global health; 2016; Jakarta. Universitas Indonesia: Faculty of Public Health Universitas Indonesia; July, 2017.
Estebsari F, Dastoorpoor M, Khalifehkandi ZR, Nouri A, Mostafaei D, Hosseini M, et al. The concept of successful aging: a review article. Current aging science. 2020;13(1):4–10.
Zanjari N, Sharifian Sani M, Chavoshi MH, Rafiey H, Mohammadi Shahboulaghi F. Successful aging as a multidimensional concept: An integrative review. Med J Islam Repub Iran. 2017;31:100-.
Asiamah N. Social engagement and physical activity: Commentary on why the activity and disengagement theories of ageing may both be valid. Cogent Medicine. 2017;4(1):1289664.
Kooij DTAM, Zacher H, Wang M, Heckhausen J. Successful aging at work: A process model to guide future research and practice. Industrial and Organizational Psychology. 2020;13(3):345–65.
Emlet CA, Harris L, Furlotte C, Brennan DJ, Pierpaoli CM. ‘I'm happy in my life now, I'm a positive person’: approaches to successful ageing in older adults living with HIV in Ontario, Canada. Ageing Society. 2017;37(10):2128–51.
Rowe JW, Kahn RL. Successful Aging1. The Gerontologist. 1997;37(4):433–40.
Rowe JW, Kahn RLJAirrt. Successful aging and disease prevention. Advances in renal replacement therapy. 2000;7(1):70–7.
Crowther MR, Parker MW, Achenbaum WA, Larimore WL, Koenig HG. Rowe and Kahn's model of successful aging revisited: Positive spirituality—The forgotten factor. The gerontologist. 2002;42(5):613–20.
Vaillant GE, Mukamal K. Successful aging. American Journal of Psychiatry. 2001;158(6):839–47.
Depp CA, Jeste DV. Definitions and predictors of successful aging: a comprehensive review of larger quantitative studies. The American Journal of Geriatric Psychiatry. 2006;14(1):6–20.
Torregrosa-Ruiz M, Gutiérrez M, Alberola S, Tomás JM. A successful aging model based on personal resources, self-care, and life satisfaction. The Journal of Psychology. 2021;155(7):606–23.
Pedras S, Carvalho R, Pereira MG. A predictive model of anxiety and depression symptoms after a lower limb amputation. Disability health journal. 2018;11(1):79–85.
Zeng F, Deng G, Cui Y, Zhang Y, Dai M, Chen L, et al. A predictive model for the severity of COVID-19 in elderly patients. Aging Sci Res. 2020;12(21):20982.
DuMontier C, Clough-Gorr KM, Silliman RA, Stuck AE, Moser A. Health‐related quality of life in a predictive model for mortality in older breast cancer survivors. Journal of the American Geriatrics Society. 2018;66(6):1115–22.
Chodosh J, Reuben DB, Albert MS, Seeman TE. Predicting cognitive impairment in high-functioning community‐dwelling older persons: MacArthur Studies of Successful Aging. Journal of the American Geriatrics Society. 2002;50(6):1051–60.
Byeon H. Development of a physical impairment prediction model for Korean elderly people using synthetic minority over-sampling technique and XGBoost. International Journal of Advanced Computer Science Applications. 2021;12(1):36–41.
Romo RD, Wallhagen MI, Yourman L, Yeung CC, Eng C, Micco G, et al. Perceptions of Successful Aging Among Diverse Elders With Late-Life Disability. The Gerontologist. 2013;53(6):939–49.
Cai T, Long J, Kuang J, You F, Zou T, Wu L. Applying machine learning methods to develop a successful aging maintenance prediction model based on physical fitness tests. Geriatrics Gerontology International. 2020;20(6):637–42.
Asghari Moghaddam M, Faghehi S. Validity of the sf-36 health survey questionnaire in two iranian samples. Clinical Psychology Personality. 2003;1(1):1–10.
Montazeri A, Goshtasebi A, Vahdaninia M, Gandek B. The Short Form Health Survey (SF-36): translation and validation study of the Iranian version. Quality of life research. 2005;14(3):875–82.
Tagharrobi Z, Sharifi K, Sooky Z. Psychometric evaluation of Shah version of modified Barthel index in elderly people residing in Kashan Golabchi nursing home. KAUMS Journal. 2011;15(3):213–24. [In Persian].
Bayani AA, Koocheky AM, Goodarzi H. The reliability and validity of the satisfaction with life scale. Developmental Psychology. 2007;3(11):259–65.
Bandari R, Mohammadi Shahboulaghi F, Montazeri A. Development and psychometric evaluation of the healthy lifestyle questionnaire for elderly (heal). Health Quality of Life Outcomes. 2020;18(1):1–9.
Chandrashekar G, Sahin F. A survey on feature selection methods. Computers & Electrical Engineering. 2014;40(1):16–28.
Miao J, Niu L. A Survey on Feature Selection. Procedia Computer Science. 2016;91:919–26.
Li J, Cheng K, Wang S, Morstatter F, Trevino RP, Tang J, et al. Feature selection: A data perspective. ACM computing surveys. 2017;50(6):1–45.
Subasi A, Alickovic E, Kevric J. Diagnosis of chronic kidney disease by using random forest. CMBEBIH 2017: Springer; 2017. p. 589 – 94.
Speiser JL, Miller ME, Tooze J, Ip E. A comparison of random forest variable selection methods for classification prediction modeling. Expert systems with applications. 2019;134:93–101.
Belgiu M, Drăguţ L. Random forest in remote sensing: A review of applications and future directions. ISPRS journal of photogrammetry remote sensing. 2016;114:24–31.
Tang F, Ishwaran H. Random forest missing data algorithms. Statistical Analysis Data Mining: The ASA Data Science Journal. 2017;10(6):363–77.
Chengsheng T, Huacheng L, Bing X, editors. AdaBoost typical Algorithm and its application research. MATEC Web of Conferences; 2017; Poland. Cracow University of Technology: EDP Sciences; 2017.
Schapire RE. Explaining AdaBoost. In: Schölkopf B, Luo Z, Vovk V, editors. Empirical Inference: Festschrift in Honor of Vladimir N Vapnik. Berlin, Heidelberg: Springer Berlin Heidelberg; 2013. p. 37–52.
Priya MB, Juliet PL, Tamilselvi P. Performance analysis of liver disease prediction using machine learning algorithms. International Research Journal of Engineering Technology. 2018;5(1):206–11.
Iliyas MMK, Shaikh MIS, Student M. Prediction of Heart Disease Using Decision Tree. Allana Inst of Management Sciences, Pune. 2019;9:1–5.
Abdar M, Kalhori SRN, Sutikno T, Subroto IMI, Arji G. Data mining on the heart disease with the use of different algorithms. International Journal of Electrical Computer Engineering. 2015;5(6).
Shanmuganathan S. Artificial Neural Network Modelling: An Introduction. In: Shanmuganathan S, Samarasinghe S, editors. Artificial Neural Network Modelling. Cham: Springer International Publishing; 2016. p. 1–14.
Wu Y-c, Feng J-w. Development and Application of Artificial Neural Network. Wireless Personal Communications. 2018;102(2):1645–56.
Kukreja H, Bharath N, Siddesh C, Kuldeep S. An introduction to artificial neural network. Int J Adv Res Innov Ideas Educ. 2016;1:27–30.
El_Jerjawi NS, Abu-Naser SS. Diabetes Prediction Using Artificial Neural Network. International Journal of Advanced Science and Technology. 2018;121:54–64.
Suthaharan S. Support Vector Machine. In: Suthaharan S, editor. Machine Learning Models and Algorithms for Big Data Classification: Thinking with Examples for Effective Learning. Boston, MA: Springer US; 2016. p. 207–35.
Pisner DA, Schnyer DM. Chapter 6 - Support vector machine. In: Mechelli A, Vieira S, editors. Machine Learning. United States: Academic Press; 2020. p. 101–21.
Zhang Z. Naïve Bayes classification in R. Annals of translational medicine. 2016;4(12):1–5.
Lee S-H, Choi I, Ahn W-Y, Shin E, Cho S-I, Kim S, et al. Estimating quality of life with biomarkers among older Korean adults: a machine-learning approach. Archives of gerontology and geriatrics. 2020;87:103966.
Adnan K, Akbar R. An analytical study of information extraction from unstructured and multidimensional big data. J Big Data. 2019;6(1):91.
Byeon H. Developing a model to predict the social activity participation of the senior citizens living in South Korea by combining artificial neural network and quest algorithm. International Journal of Engineering Technology. 2019;8(1.4):214–21.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Designing a Predictive Model for Successful Aging among the Elderly Using Machine Learning Techniques

Status:

Version 1

Abstract

Figures

Introduction

Methods

Predictor Features

Results

Discussion

Limitation And Suggestion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1