This study is a secondary data based on the Korean Psychosocial Anxiety (KPA) survey, an epidemiological survey representing the South Korean population. KPA-survey was conducted from August to September 2015 by the Korea Institute for Health and Social Affairs. The study stratified the population, obtained from the Residence Registration Data (complete enumeration) conducted by the Ministry of Security and Public Administration as of June 2015, into 17 cities and provinces (i.e., Seoul, Busan, Daegu, Incheon, Gwangju, Daejeon, Ulsan, Sejong, Gyeonggi-do, Gangwon-do, Chungcheongbuk-do, Chungcheongnam-do, Jeollabuk-do, Jeollanam-do, Gyeongsangbuk-do, Gyeongsangnam-do, and Jeju-do). Afterward, this study extracted samples using the quota sampling method with considering the gender, age, and residential area ratios. This study selected 200 eups, myeons, and dongs from 3552 eups, myeons, and dongs in South Korea using the probabilities proportional to size (PPS) method. In order to secure randomness when extracting sampling sites, the PPS was applied after sorting the districts in the order of city, county, and district based on the administrative district code. After selecting 200 sampling sites, the 5^th household from the Community Service Center of each selected eup, myeon, or dong was chosen as the sampling household by visiting the selected sampling site, and finally, 7000 adults (19 years or older) were surveyed. This study conducted an in-person survey by having trained surveyors who received survey training visit the sample households using the computer assisted personal interview method. This study was approved by H University's Clinical Research Ethics Committee (No. 20180042). This study analyzed 1558 senior citizens (695 males and 863 females) who were 60 years or older and completed the KPA-survey.

Depression, an outcome variable, was measured using the short form of the Korean version CES-D (short form of CES-D)[16]. CES-D[15] is a standardized self-report depression test that is used most commonly in the world and can measure depression in healthy people easily. This test was developed by the National Institute of Mental Health for investigating the epidemiological status of depression in a community and has been widely used in many countries in various languages as a screening instrument for depression[16]. Many previous studies have proven the high internal consistency and validity of CES-D[17,18]. More recently, the scale of short form of CES-D, which has secured good reliability and validity, has also been developed and used for subjects who have difficulties in responding to the 20 items of the original CES-D (e.g., senior citizens, those with dyslexia, or illiterates)[19-23]. Short form of CES-D has been continuously used in epidemiological surveys because it reduces the response burden of the participants and it has the advantage that it can be easily used for the elderly who suffer from many physical symptoms and have difficulties in answering a long item due to a decrease in concentration[24]. The KPA-survey, the data source of this study, used standardized short form CES-D consisting of 11 items. This test classified depression into four levels based on the frequency of depression symptoms experienced by the subject over the past week. Moreover, the severity of symptoms was classified into four classes using a Likert scale (0 points = rare or 1 d or fewer per week; 1 point = occasional or 1 to 2 d per week; 2 points = often or 3 to 4 d per week; and 3 points = mostly or 5 d or more per week). A higher total score means more depression symptoms. This study defined the threshold of depression as 16 points based on the previous study[22] on the elderly in South Korea.

Explanatory variables were age (continuous variable), alcohol consumption (abstainers or normal drinkers, high-risk drinkers, and those with alcohol use disorder), self-esteem, suicidal ideation over the past year (yes or no), subjective social isolation (rarely, occasionally, often, or usually lonely), frequency of communication with other family members (a 10-point scale with a higher score infers more frequent communication), interaction with neighbors and friends [regular club activities (yes or no), perceived social support (a 10-point scale), trust neighbors (yes or no), and satisfied with help (support) from neighbors (yes or no)], satisfaction with the neighborhood environment (yes or no), satisfaction with the safety level of the neighborhood (yes or no), and satisfaction with the medical services in the region (yes or no)[25]. The self-esteem item was measured by using the Self-Esteem Scale (SES) comprising 10 items (a total score of 40 points)[26]: A low score (15 or less) indicates poor self-esteem. The Alcohol Use Disorders Identification Test (AUDIT)[27] developed by the World Health Organization for the preliminarily screening of high-risk drinkers comprises 10 items with a total score of 40 points: 0-15 points identify abstainers and normal drinkers, 16-19 points identify high-risk drinkers, and ≥ 20 points identify individuals with alcohol use disorder.

The general characteristics and depression prevalence of the subjects were presented/ present in percentages. The effects of depression were examined by Chi-square test. This study built a depression prediction model using logistic regression analysis to find out the effect of each variable on depression. This study selected variables using step-forward regression analysis. Moreover, this study presented an unadjusted crude model, which was not adjusted with confounding factors, and an adjusted model, which was adjusted with confounding factors.

The developed depression prediction model contained a nomogram to make it possible for clinicians to interpret the prediction results (prediction probability) easily. The nomogram was composed of 4 elements. The first was a point line. It was the line located at the top of the monogram to indicate the score corresponding to the risk range of a factor. In the case of a logistic nomogram, it ranged from 0 to 100 points. The second was a risk factor line. This line indicates the score range of a risk factor influencing the occurrence of an event. The number of risk factor lines was equal to the number of risk factors. The third was a probability line. The probability line was the sum of finally calculated nomogram scores and it was placed at the bottom of the nomogram to derive the probability (risk) of depression. The fourth is a total point line. It was constructed by calculating it based on a statistical model.

The sample size of this study was small (n = 1558). Therefore, when the prediction performance of the model is validated using a held-out validation method (i.e., a validation method that divides the dataset into a training dataset and a validation dataset at a 7:3 ratio), it poses a risk of overfitting. Consequently, it is more likely to decrease the reliability of prediction results. As a result, this study used 10-fold cross-validation to validate the prediction performance of the developed depression prediction model (nomogram). AUC, general accuracy, balanced accuracy, F1 score, sensitivity, specificity and calibration plot were presented. The calibration plot is a graph for visually confirming the degree of agreement (y, Calibration curve and x, predicted probability) between the predicted probability and the observed probability. Sensitivity indicates the ratio of true positives: The ratio of the developed model to predict the senior citizen living alone with depression as depression. Specificity indicates the ratio of true negatives: The ratio of the developed model to predict the senior citizen living alone without depression as no-depression accurately. R version 4.0.3 (Foundation for Statistical Computing, Vienna, Austria) was used for analyses and significance level was 0.05 in two-tailed test.

General characteristics of all subjects (1558 subjects) are presented in Table 1. The mean age was 67.9 ± 5.5 years old: 55.4% of them were women and 44.6% of them were men. The results of the AUDIT showed that most of the subjects were normal (52.4%), no suicidal urge over the past one year (91.0%), very rare subjective loneliness (44.1%), and homemaker (35.5%). The prevalence of depression among the senior citizens living alone in the community, measured by short form of CES-D was 7.7%.

The characteristics of the subjects according to the prevalence of depression are presented in Table 2. The results of chi-square test showed that depression was significantly (P < 0.05) affected by age, self-esteem, subjective frequency of communication with other family members, alcohol use disorder, the experience of suicidal urge over the past year, subjective trust satisfaction with neighbors, subjective satisfaction with help (support) from neighbors, subjective satisfaction of the safety level of the neighborhood, subjective satisfaction of the medical service of the region, regular club activities, and subjective loneliness.

Table 3 shows a model for predicting the depression of senior citizens living alone. The results of univariate logistic regression analysis (crude model) showed that all variables (age, self-esteem, alcohol use disorder, the experience of suicidal urge over the past year, subjective trust satisfaction with neighbors, subjective satisfaction of the living environment of the neighborhood, subjective satisfaction of the safety level of the neighborhood, subjective satisfaction of the medical service of the region, regular club activities, subjective loneliness, subjective frequency of communication with other family members, and subjective frequency of communication with neighbors/friends) were significantly (P < 0.05) related to the depression of senior citizens living alone. However, the results of the adjusted model revealed that the depression of senior citizens living alone was independently associated only with the experience of suicidal urge over the past year [odds ratio (OR): 3.57, 95% cumulative incidence (CI): 1.55-8.22], subjective satisfaction with help (support) from neighbors (OR: 0.29, 95%CI: 0.13-0.66), subjective loneliness (occasionally lonely: OR: 5.04, 95%CI: 1.00-4.65; often lonely: OR: 187.19, 95%CI: 23.17-1512.13; mostly lonely: OR: 758.12, 95%CI: 56.44-10183.32), age (OR: 1.17, 95%CI: 1.07-1.26), and self-esteem (OR: 0.81, 95%CI: 0.72-0.91).

A nomogram for predicting the depression of elderly living alone based on multiple risk factors is presented in Figure 1. Among the risk factors of depression of senior citizens living alone, subjective loneliness had the greatest influence, and the senior citizens living alone who responded that they felt lonely showed the highest risk of depression. For example, the nomogram for predicting depression predicted that the depression risk of senior citizens who had suicidal urge over the past year, were not satisfied with subjective help (support) with neighbors, thought that they were mostly lonely, had 15 points in Rosenburg SES scale (self-esteem), and were 80 years or older was 99.8% (Figure 2).

Open in New Tab   Full Size Figure   Download Figure

Figure 1 The area under the curve of the developed nomogram for predicting the depression of senior citizens living alone. AUC: Area under the curve.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 A nomogram for predicting the depression of elderly living alone.

The developed nomogram for predicting the depression of senior citizens living alone was validated by using AUC, general accuracy, balanced accuracy, F1 score, sensitivity, and specificity (Table 4). The results of 10-fold cross validation showed that the AUC of the nomogram was 0.96, the general accuracy of it was 0.95, the balanced accuracy of it was 0.80, the sensitivity of it was 0.98, the specificity of it was 0.62, and the F1 score of it was 0.97. The equation of the calibration plot was drawn along the ideal line; coefficient of determination (R squre) was 0.853 (Figure 3). The AUC of the developed nomogram for predicting the depression of senior citizens living alone is presented in Figure 4.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Calibration plot for nomogram model.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Application example of depression prediction nomogram for elderly living alone: Senior citizens who had suicidal urge over the past year, were not satisfied with subjective help with neighbors, thought that they were mostly lonely, had 15 points in the Self-Esteem Scale, and were 80 years or older.

Senile diseases are rapidly increasing globally due to the rapid aging of the population. Among these diseases, depression is an important and frequent psychiatric disorder in the senile stage, and it is predicted to be the second major factor in the global burden of disease.

Although the number of senior citizens living alone is increasing, only a few studies have identified factors related to the depression characteristics of senior citizens living alone by using epidemiological survey data that can represent a population group.

This study developed a nomogram that allows physicians to check the multiple risk factors of depression of senior citizens living alone using visual graphs and to calculate the prevalence probability of depression while considering the personal characteristics of a subject based on these results.

This study analyzed 1558 senior citizens (695 males and 863 females) who were 60 years or older. Depression, an outcome variable, was measured using the short form of the Korean version CES-D (short form of CES-D). This study built a depression prediction model using logistic regression analysis to find out the effect of each variable on depression. The developed depression prediction model contained a nomogram to make it possible for clinicians to interpret the prediction results easily.

In this study, the significant predictors of depression of the senior citizens living alone were the experiences of suicidal urge over the past year, dissatisfaction with help (support) from neighbors, subjective loneliness, age, and low self-esteem.

The results of this study implied that it is necessary to develop a support system customized for subjects to strengthen the relation network for preventing depression in senior citizens living alone so that they can receive actual support from acquaintances such as neighbors rather than the frequency of physical contact.

It is needed to establish an improved mental health policy that identifies high depression risk groups among senior citizens living alone based on multiple risk factors and continuously manages them.