STRENGTHS AND LIMITATIONS OF THIS STUDY

User-friendly interfaces in smartphones and smartwatches can help to screen major depressive disorder (MDD) at the early stage in late life.

A strength of this study is longitudinal data collection over 2 years from the same participants using wearable devices compared with the typical short collection periods in the previous research.

By recruiting a subsample from a large national cohort study founded over 20 years ago in South Korea, this research can discover a developing pattern of late-life MDD from the very early stages.

The limitations of this study are the increased likelihood of missing data due to updates in the software and random malfunctions of a wearable device.

Introduction

Major depressive disorder (MDD) is a major global contributor to mortality. Despite the wide array of cultural backgrounds and methods measured in previous MDD studies, the lifetime prevalence of MDD ranges from 12.8% to 16.2%.1 Although the estimated prevalence of MDD is high, the number of individuals seeking treatment is only a few. For example, according to a recent study analysing the WHO survey data from 21 countries,2 56.7% of those who experienced MDD based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders Fourth Edition during the last 12 months recognised their need for treatment, but only 16.5% received minimally adequate treatment. South Korea was not included in that study, but South Korea ranks the highest in the prevalence of depression among Organisation for Economic Co-operation and Development countries (37% in 2020) and deaths by suicide increased by 46% from 2000 to 2019.3

The course and nature of MDD have changed over the past few decades. Originally, MDD was considered an acute and self-limiting disease, while it is now categorised as a chronic, life-long illness.4 This phenomenon has arisen due to an increase in life expectancy, which also extends the average exposure duration to life stressors.5 Furthermore, due to sociological and demographic changes, the rates of elderly depression will increase substantially in developed countries.5 With the increased risks of medical comorbidities, neurological disorders, social isolation, or bereavement, the older population has a greater risk of MDD compared with other populations.6–10 This results in lower remission but higher relapse rates for the older population compared with younger individuals.9 MDD in the elderly is often referred to as ‘late-life depression’ or ‘geriatric depression’ and is highly prevalent and life-threatening.11 Elderly people with MDD have a greater risk of developing all-cause and cardiovascular mortality, sleep disturbances, dementia, fragility and other chronic illnesses.6 12–15 However, the diagnosis of MDD is often underestimated due to their low response rate,11 and stigmas about mental health problems significantly hinder seeking medical help.16

Since the traditional diagnostic method for MDD relies on subjective and retrospective reports collected in a cross-sectional design, there is a growing interest in collecting more objective data through smartphones/wearable devices.17–22 This type of data makes it feasible to monitor their current mood and fluctuations as well.17 18 However, small convenient samples have been a major limitation of previous research.23 To overcome this limitation, a recent study recruited 334 participants aged 18–60 years for 12 weeks.24 They collected the data including Patient Health Questionnaire-9 (PHQ-9), daily mood changes, phone usage logs, sleep, step count and heart rate through mobiles and wearable devices. Their classification results analysed all the data by feature selection, and machine learning proposed high accuracy rates (75%–80%) and high recall rates (84%–97%) between submodels of the prediction.24 Although most of the previous research has targeted patients with MDD only (eg,21 24), we anticipate the data gathered from smartphones and smartwatches would be feasible as well for self-screening MDD among the general population. Marzano et al suggested that the passive data collected through the sensors on a smartphone and a wearable device can be used in research ‘to understand complex psychological processes over long periods of time’.25 By using smartphones and smartwatches, we expect the elderly, who have never visited a mental health clinic, can easily screen for depression from home.

The elderly are less adept in using technology, so prior studies that apply wearable devices to the elderly population are extremely limited, making this a challenging research area because there is a gap in the knowledge available. Previous research using wearable devices for tracking depressive symptoms has mainly recruited young population (eg,26 27). Recently, there has been a growing interest in applying Internet of Things and digital sensors for health monitoring among elderly people,28 29 with a focus on addressing their vulnerability in both physical and psychological health. However, this endeavour requires a significant budget and delicate procedures for ethical reasons such as participants’ privacy, particularly in developing new devices and installing multiple antennas for secure transmission. Although there are several traditional screening instruments for depression diagnoses in elderly people,30 the implementation of the traditional screening method is expensive and limited due to barriers to medical access and high concerns of stigma. In this situation, using wearable devices for screening late-life depression is promising as an adjunct to traditional clinical assessments.31 Monitoring real-time symptoms with smartphones and smartwatches is a great strength in addition to lower costs and time. A recent scoping review demonstrates a notable surge in academic publications related to this topic over the past 5 years.32 However, the scoping review also highlights small sample sizes as a consistent limitation observed across these publications.32 There is a significant difference in physical activity between people with and without late-life depression,33 and despite small sample sizes, previous studies have supported the idea that data collection through smartphones can be a reliable tool for monitoring the physical activity of elderly people.34 During the COVID-19 pandemic, elderly people in Korea tended to use smartphones more frequently to communicate with their families by transcending physical boundaries. Smart devices are usually designed to be intuitive, which facilitates universal usage regardless of participants’ ages. Elderly people tend to be concerned about the burden they pose to their families, so this research can increase their self-sufficiency by monitoring health conditions.

Purpose of this research

This study is designed to construct longitudinal data using wearable devices and to develop a prediction algorithm for depression in late life. Based on the developed algorithm from this study, we will suggest public policies to promote elderly access to mental health services and community care. We hope this approach will facilitate the early diagnosis or prevention of MDD in later life.

Methods and analysisRecruitment

This research has recruited participants from a larger cohort study, Korean Genome and Epidemiology Study—Cardiovascular Disease Association Study (KoGES_CAVAS).35 Since 2005, KoGES_CAVAS recruited a total of approximately 28 500 people, men and women, across provinces in South Korea, aged 40 years or older, to understand genetic and environmental causes of non-communicable chronic diseases among the Korean population. Subsequently, KoGES_CAVAS has collected the follow-up data from 17 575 people until 2017 and continues their data collection. To analyse the association between the participants’ clinical, functional and inflammatory characteristics and MDD in late life, we recruited a subsample from this large cohort study in 2020. We restricted potential participants to those residing in Wonju City because Wonju has a unique strength as a complex community that includes both urban and rural populations in Korea. From the list of 3620 KoGES_CAVAS participants in Wonju, 1896 people changed or did not report their cell phone numbers. After excluding 1039 cases out of 1724 elderly participants who had cell phone numbers based on the exclusion criteria below, 685 people were finally recruited. This study was conducted from 1 July 2020 to 31 December 2023.

Eligibility

Potential participants were excluded if they satisfied any of the following criteria: (1) younger than 55 years old in Korean age, (2) unable to stay for 1.5 hours to complete the baseline interview, (3) using an outdated mobile device which is not able to connect to Wi-Fi (eg, 2G phone), (4) non-compliant with the research procedure, (5) immobile, (6) having cognitive impairments that impeded their understanding of the study, (7) previously diagnosed with an alcohol or substance use disorder or (8) not wanting to join due to other reasons (eg, health problems and transportation issues).

Patient and public involvement

No patient involved.

Overview of protocol

Figure 1 illustrates the recruitment process. In total, 685 participants visited campus for the face-to-face baseline interviews. A trained researcher spoke with each participant for approximately 1–1.5 hours in a semistructured one-on-one interview. We also developed a new smartphone application for app-based longitudinal data collection, including daily, weekly and monthly self-report surveys and passive sensing data. We had hoped to install this smartphone application on participants’ mobiles at the time of their baseline interviews, but we realised our participants were elderly people who used a variety of Android OS versions with which the application could not cope. Because of this, we chose not to install the application during the baseline interviews. Instead, soon after the baseline interviews (within approximately 3 months after each participant completed the baseline interview), trained research staff members made home visits and installed the application for those who agreed to participate in follow-up data collection. Beyond self-reporting, the application is connected to the Samsung Health platform and automatically retrieves daily step counts, heart rates and sleep data collected from smart sensors. When participants open our application to complete self-report app surveys, the collected Samsung Health data are sent to our database. Initially, our application was not designed to collect passive sensing social network data (eg, call, text and mobile app usage). However, COVID-19 shifted a great deal of social connection from in-person meetings to the virtual space. This was especially true for our participants, whose age placed them at high risk from the virus. This shift to virtual connection made the collection of passive sensing social network data critical to the development of our prediction algorithm for late-life depression. After obtaining Institutional Review Board (IRB) approval for our revised research plan, we upgraded our mobile application to include social network data collection. Approximately 12 months after the baseline interview, participants were invited to participate in the first follow-up interview. When participants visited the campus for the first follow-up interview, the upgraded version of our research application was installed on their smartphones to collect not only app surveys and Samsung Health data but also social network data over the following 12 months. Approximately 24 months after the baseline interview, the second follow-up interview is conducted using the same procedure. Key measures and their collected time points are suggested in figure 2. All data to be collected in the research project are presented in the Supplementary Table.

Figure 1

Process of recruitment and data collection. KoGES_CAVAS, Korean Genome and Epidemiology Study—Cardiovascular Disease Association Study.

Figure 2

Schema of the study protocol. GAD-7, Generalized Anxiety Disorder-7; HAM-D, Hamilton Rating Scale for Depression; MINI, Mini-International Neuropsychiatric Interview; PHQ-9, Patient Health Questionnaire-9.

Data sourcesBaseline and annual follow-up face-to-face interviews

Baseline interviews included demographic characteristics, economic situations, physical health, psychological well-being, social activities, stress, anxiety and personal or family history of clinical depression. For depressive symptoms, three measures were recorded: PHQ-9, Hamilton Rating Scale for Depression (HAM-D) and Mini-International Neuropsychiatric Interview (MINI). Additional information such as personal histories of clinical depression, mental illness, family histories of clinical depression, participants’ early trauma and their concurrent anxiety disorders were also collected. Annual responses on stress, loneliness and perceived social support were collected as well. One year later, 327 participants revisited the campus and completed the first follow-up interview following the same procedure and questionnaire. After an additional year, the same participants are invited back, and the second follow-up interview is performed following the same procedure and questionnaire.

App-based self-report surveys

Participants are asked to respond to daily, weekly and monthly surveys. The daily survey comprises one question about participants’ daily mood. The weekly survey conducted every Sunday to Monday asks three questions about stress exposure and major stressors. During the last week of each month, a monthly survey is opened to ask participants to complete a PHQ-9. Participants can only submit one response for each survey. All app surveys are open from 7:00 pm–11:30 pm.

Samsung Health data

Samsung Health records daily step counts via smartphone and smartwatch sensors. In addition to step counts, Samsung Health also collects heart rate and sleep data through the Galaxy-series smartwatches. Our smartphone survey application developed exclusively for research purposes is designed to retrieve daily step counts, heart rate and sleep data saved in the Samsung Health application to our database server when a participant opens our research app to complete their self-report daily, weekly or monthly surveys.

Social network data

During the first follow-up survey, participants reinstalled the smartphone application which has the additional function of collecting social network data. The social network data are composed of the total amount of daily calls, text messages and usage of mobile applications. Social network data are collected via passive sensing data to indicate daily usage of social networks throughout the year, from the first to the second follow-up interviews.

Measures

Detailed information about the measures used in the baseline and annual follow-up surveys is listed below.

Mini-International Neuropsychiatric Interview

MINI is a structured test36 that identifies clinical depression. This test is administered by a trained clinician towards each participant at baseline and annual follow-up face-to-face interviews.

Hamilton Rating Scale for Depression

HAM-D is a widely used instrument for depression with a high internal consistency and reliability.37 In our research, a trained interviewer scores 17 items of HAM-D scale38 for each participant at baseline and annual interviews.

Patient Health Questionnaire-9

PHQ-9 is a brief self-report 9-item measure for screening depressive symptoms.39 In our research, PHQ-9 is measured not only at the baseline and annual interviews but also at the monthly surveys on smartphone application.

Early Trauma Inventory—Short Form

The original measure of Early Trauma Inventory—Short Form (ETI-SF) was developed as a 56-item clinician-administered instrument.40 However, ETI-SF was adapted as a self-report measure for research convenience later.41 This scale covers physical, emotional and sexual abuse and general traumas. In our research, since this scale focuses on childhood experiences, participants responded to this scale only at the baseline interview.

Generalized Anxiety Disorder-7

Generalized Anxiety Disorder-7 is a brief 7-item measure to assess generalised anxiety disorder.42 In our research, this instrument is used at the baseline and annual follow-up interviews. Participants report how many days they have experienced symptoms described in each item during the last 2 weeks.

Questions from the Korean National Health and Nutrition Examination Survey—Short Form

For stress, participants respond to a set of 10 questionnaires about perceived stress during the last month from the Korean National Health and Nutrition Examination Survey—Short Form.43 It consists of one item choosing a category for primary stressor during the last month and nine items covering burn-out, depression and anger. The original scale was developed with 21 items, but this 9-item Short Form is more preferred for easy assessment.44

UCLA loneliness scale

The University of California, Los Angeles (UCLA), loneliness scale is a self-report 20-item measure about loneliness.45 Participants score this measure at the baseline and annual interviews. The scale of each item ranges from 1, never, to 4, often, and the sum score will be created.

Multidimensional Scale of Perceived Social Support

Multidimensional Scale of Perceived Social Support is a self-report 12-item questionnaire of perceived social support.46 It covers social support from significant other, family and friends. Participants report it at the baseline and annual interviews. The scale of each item ranges from 1, never, to 5, always, and the sum score will be created.

Sense of Acceptance in Community Activities

Sense of Acceptance in Community Activities is a self-report 8-item questionnaire of participants’ psychological integration.47 This measure was missing at the baseline, but it is added since the first follow-up interview. Participants score each item from 1, strongly agree, to 4, strongly disagree, and the sum score will be created.

Lubben Social Network Scale-18

Lubben Social Network Scale-18 is developed as a self-report 18-item measure about the social network of elderly people.48 49 It covers the size, closeness and frequencies of contacts with family, friends and neighbours. However, since there have been restrictions on face-to-face gathering during the COVID-19, we add two questions for the size of contacts and two questions for the frequency of contacts in each network (family, friends and neighbours) by dividing the contacts into face-to-face gathering and digital communication through cell phone. 12 items (4 items×3 network areas) are added in total. Participants answer this self-report 24-item measure for their social network at the baseline and annual follow-up interviews.

History of mental health

Three questions are asked to the participants if they have ever been diagnosed with major depression, if they have ever been diagnosed with other mental illness, and if any family member has ever been diagnosed with major depression. For those who report ‘yes’ to their own mental health history of major depression, additional nine items are asked including the age, medical treatment after diagnosis and the recent status. For those who report ‘yes’ to other mental illnesses, additional one item is asked about the type of mental illness. For those who report ‘yes’ to the history of major depression among their family members, additional one item is asked about the relationship(s) between the participant and the corresponding family member(s).

Social activities

This is a set of nine items about participants’ social activities during the last year. Out of the nine items, eight items ask the frequency of each type of social activity, and one item asks major obstacles to joining social activities.

Physical health

This is a set of 15 items regarding participants’ physical health. It consists of six items about medical history (eg, high blood pressure, hyperlipidaemia and diabetes), one item about smoking, three items about drinking, one item about physical activity in normal daily lives, three items about exercise and one item about total hour of sleeping.

Demographics

This is a set of 22 items regarding participants’ demographic characteristics. It includes a participant’s age, gender, education, work status, family structure, periods of residence, caregiver burdens in family relationships, household income and household expenditures.

Analysis plan

To develop a prediction algorithm, we will use machine learning. Machine learning has proven useful in providing objective and easily accessible diagnostic tools for depression.50 Additionally, this methodology is a good supplement to traditional methods (ie, clinical interviews and self-report questionnaires), especially when a patient tends not to report their symptoms accurately or if MDD needs to be differentiated from other diagnoses (eg, anxiety) with similar symptoms.50 In this project, we will analyse features and extract key features. We will test several models using logistic regression, support vector machine, decision tree, recurrent neural network and so on. The model fit of these models will be evaluated by Receiver Operating Characteristic curve (ROC curve) and Area under the ROC curve (AUC) or F-scores, and the best-fitted model will be selected. We anticipate it can help find a unique pattern for screening of MDD among the elderly population. If the collected data seem to have too many features, principal component analysis will be used to reduce the dimensions of the data. The ultimate goal of our analyses is to validate which parameters are significantly associated with the occurrence of depressive symptoms, while simultaneously developing a prediction algorithm for detecting MDD in older adults. In addition, an integrated clustering method for natural language processing will be also performed to analyse the Korean text data regarding major stressors of older adults reported in weekly app surveys for 2 years. At least 2–3 experts will manually clean and review the data as a way to conduct data preprocessing for hypernym detection. Python will be a primary tool used throughout our statistical analyses.

Ethics and dissemination

The protocol of this study has been reviewed and approved by the Yonsei University Mirae Campus IRB in South Korea (1041849–2 02 111 SB-180-06). All participants were informed of this research project and voluntarily provided written consent before their participation began. We obtain the written consent of the participants each year. Participants can withdraw from this research at any time if the person is unwilling to continue. All printed versions of participant data are stored in a locked cabinet in a secured room. Electronic records are all encrypted and stored on a password-protected hard drive. Following the data security policy at Yonsei University, all data will be destroyed 3 years after the study’s completion. The collected data will be anonymised in the data cleaning process so that it is impossible to identify a specific participant from the cleaned data. The university’s IRB approval is renewed annually.

Given that this research is conducted in natural settings, potential risks are expected to be minimal. There will be no direct treatment or intervention in this research, but a list of local psychiatrists’ clinics will be given to participants with PHQ-9 scores ≥15 (indicating moderately severe/severe depression) as an option they can use if they feel a need for medical support.

Research findings will be presented at academic conferences in medical, engineering, social science and other relevant fields. As convergence research, this research team has been actively building up academic collaboration with other convergence research teams funded by the National Research Foundation of Korea. We will also publish our major findings in academic journals. Published results will be disseminated to both traditional media organisations and social media so that the findings of this research project can be used to improve the mental health of elderly people.

Discussion

This research is designed to explore the possibility of using wearable devices to screen for depressive symptoms among older adults. Based on collaboration with professionals working in a variety of fields, we have developed a smartphone application for research purposes. We are also collaborating with professionals at the Yonsei University Mirae Campus and Yonsei University Wonju College of Medicine to interpret a large longitudinal dataset collected via passive sensors and traditional methods (eg, self-report or observation) to understand MDD development in late life. By recruiting a subsample from a large national medical cohort study, the collected data would enable us to link late-life MDD to past medical information. Another unique component is that this research targets the general population. We hope to explore the possibility of using wearable devices for early screening of MDD among the elderly population. Even the data from the participants without MDD during the data collection period will be used to get insights to determine what factors are helpful in preventing MDD among the older population. Extensive data including but not limited to self-report, observational and passive sensing data through annual face-to-face interviews and smartphone app-based data collection will be collected over 2 years. These data reveal development patterns of late-life MDD and help to investigate significant risk factors as well as protective factors related to late-life MDD. Given that the older population has been often disregarded in the previous research using wearable devices, the collected data of this research project are anticipated to have essential benefits to psychiatrists and other professionals in relevant fields to increase medical access and healthcare services.

Of course, there are several limitations to this research. First, due to technical restrictions, this research includes only participants who use Android devices. Second, the change within smartphone OS and application is everchanging, which makes data collection through wearable devices challenging and unstable. To resolve these issues, we are prioritising communication with relevant companies and updating our application constantly. However, even after an update is completed, the challenge of continuous updates is not resolved. This increases our burden to contact all participants again whenever they need to upgrade the application. On one hand, we will perform missing data imputation methods or maximum likelihood techniques in the data analyses in order to handle inevitable missing data. On the other hand, we operate 24-hour phone numbers to which our participants can leave their calls, text messages or voicemails whenever they need support. This is mainly because we anticipate the elderly population will require additional technological support (ie, unexpected malfunctions in smartphones and smartwatches). Responses in an untimely manner can result in the failure of successful implementation of our app-based data collection system, which will increase the likelihood of missing data. Another challenge we have is that this cohort grew up witnessing a high level of public stigma about mental illness and mental health. Therefore, our participants generally have a defence mechanism about the possibility of experiencing MDD. This tendency increases the likelihood of missing data. On the other side, due to this cultural aspect, we believe our research using wearable devices for elderly people’s mental health is a timely project because the collected real-time passive sensing data through smartphones and smartwatches will supplement self-report scores.

In summary, not only will we ask participants to self-report stress and MDD tests, but we will also collect more objective passive sensing data regarding depressive symptoms. The tests administered by researchers at annual face-to-face interviews will shed light on the mysterious high suicide attempts compared with low treatment rates among the elderly population. Our goal is that the big data collected in this research project will provide basic information to reveal the mechanisms surrounding late-life MDD and give insights on how to support our community to become more caring of mental health of the elderly population.

Ethics statementsPatient consent for publication

Consent obtained directly from patient(s).