Authors: Siying Wei, Yan Yue, Jiayu Yang, Jiayue Dan, Kangle He, Xiaofeng Xie
Categories: Research, Smartphone use diversity, Depressive symptoms, Older adults, Multilevel logistic regression, Propensity score matching
Source: BMC Public Health
Authors: Siying Wei, Yan Yue, Jiayu Yang, Jiayue Dan, Kangle He, Xiaofeng Xie
With the population aging and the popularization of intelligent technology, smartphones have emerged as a significant factor influencing the depressive symptoms of older adults through diversified functionalities, yet evidence remains limited. This study aims to determine the impact of the smartphone use diversity on depressive symptoms among older adults in China, and whether it varied by gender, residence, and education.
This study used data from the China Health and Retirement Longitudinal Study (CHARLS). We quantified smartphone use diversity via a composite index to measure smartphone usage capabilities among older adults. Multilevel logistic regression and propensity score matching were applied to analyze associations, with heterogeneity tests across gender, residence, and education. Additionally, latent class analysis (LCA) was conducted to identify distinct patterns of smartphone use and examine their associations with depressive symptoms.
Higher smartphone use diversity was significantly associated with reduced depressive symptoms (OR = 0.493, 95% CI: 0.376–0.646; P < 0.001) among older adults, even after adjusting for confounders. The protective effect was stronger in males (OR = 0.482, 95% CI: 0.333–0.697, P < 0.001), urban residents (OR = 0.450, 95% CI: 0.308–0.656, P < 0.001), and those with higher education (OR = 0.488, 95% CI: 0.314–0.760, P < 0.01). Latent class analysis further revealed two distinct usage “Limited Users” and “Multi-functional Users”. Compared with Limited Users, Multi-functional Users showed a significantly lower risk of depressive symptoms in both years (OR = 0.51 in 2018; OR = 0.66 in 2020).
Smartphone use diversity may mitigate depressive symptoms among older adults, particularly within specific subgroups. The identification of distinct usage patterns underscores the heterogeneity in digital engagement among older adults. These findings provide empirical support for technology-enabled mental health promotion interventions targeting older adults. Therefore, integrating digital literacy programs and diverse smartphone use initiatives into public health strategies aimed at enhancing mental well-being in aging populations is essential.
Population aging represents an objective trend in the development of modern economic and social structures. The “World Population Prospects 2022” report published by the United Nations Department of Economic and Social Affairs indicates that the proportion of the global population aged 65 and older increases from 10% in 2022 to 16% by 2050, corresponding to a population size of 1.5 billion [1]. This trend is associated with a significant increase in mental health risks among older adults [2]. Compared to younger individuals, older adults experience declines in physical function, increased risks of age-related diseases, and reduced social interaction after retirement, resulting in decreased participation in social activities [3]. Consequently, they are more likely to experience feelings of loneliness, even depression [4, 5]. A study on the prevalence of depression among older adults indicates that the overall rate of depression in the global elderly population continues to rise, reaching 35.1% [6]. Poor mental states are closely associated with declines in cognitive abilities, increased risks of dementia, and the onset of cardiovascular diseases among older adults [7]. These factors further impact quality of life in later years and lead to increased mortality rates [8, 9]. Therefore, the promotion of mental health among older adults is essential for the successful implementation of global healthy aging strategies.
The arrival of the digital and intelligent era brings profound changes to the modes of production, lifestyle, and work. The State of Mobile Internet Connectivity 2023, published by the Global System for Mobile Communications Association (GSMA), indicates that 54% of the global population (approximately 4.3 billion people) owns smartphones, with nearly 4 billion accessing mobile internet services through smartphones [10]. Driven by advancements in portability and user-centered design, smartphone adoption among older adults has grown substantially, positioning these devices as the dominant mode of internet access for this demographic [11]. With the development of mobile internet technology, smartphones increasingly evolve from basic communication tools into digital ecosystem platforms that integrate diversified functions, including social networking, entertainment, and health management. Social and entertainment applications, such as WeChat and Douyin, significantly enrich the social networks and recreational activities of older adults [12]. Emerging online health services and communication platforms, including health consultations and telemedicine, provide innovative technological pathways that support healthy aging [13].
In recent years, several empirical studies have explored the relationship between smartphone use and depressive symptoms among older adults [14–16]. Among these studies, most scholars have concluded that smartphone use effectively reduces the risk of depression, enhances quality of life, and facilitates social adaptation within this demographic [17]. However, existing research primarily focuses on the binary usage characteristics of smartphones and the effects of single function on depressive symptoms among older adults [18, 19]. This narrow focus neglects the holistic nature of smartphones as multifunctional interactive devices, limiting the comprehensive assessment of smartphone usage’s overall effects on depressive symptoms among older adults. To address this gap, our study leverages nationally representative longitudinal data from older adults in China to construct the indicator of “Smartphone Use Diversity” as the independent variable, with the objective of examining its association with depressive symptoms among older adults.
The data used in this study are derived from the China Health and Retirement Longitudinal Study (CHARLS). CHARLS is a comprehensive, interdisciplinary survey initiative conducted by the National Development Research Institute at Peking University. Its aim is to collect a high-quality, nationally representative sample of Chinese residents aged 45 and older to support scientific research on the elderly. The data can be accessed through the official website (http://charls.pku.edu.cn). The CHARLS database encompasses a wide range of survey content, including various dimensions such as socioeconomic status, health conditions, and lifestyle factors. Notably, the database systematically records older adults’ smartphone use diversity, including watching videos, playing games, reading news, and engaging in mobile payments. It also scientifically assesses the mental health status of older adults, focusing on key indicators such as depressive symptoms. Therefore, this study utilizes longitudinal data from the CHARLS database for 2018 and 2020, with the aim of in-depth exploring the impact of smartphone use diversity on depressive symptoms among older adults.
The dependent variable in this study is depressive symptoms, measured using the internationally recognized “10-item Center for Epidemiological Studies Depression Scale (CES-D)” within the CHARLS data. This scale has a high Cronbach’s alpha coefficient of 0.799 and a KMO value of 0.889 [20], demonstrating strong validity and reliability in measuring depressive symptoms among older adults in China [21, 22]. The scale consists of eight items that focus on positive emotions and two items that focus on negative emotions, as detailed in Table 1. Each item is rated on a four-point scale reflecting the frequency of the described “rarely or none,” “Some or a little,” “Occasionally or a moderate amount of the time,” and “most or all of the time.” The scoring for negatively oriented items ranges from 0, indicating “rarely or none,” to 3, indicating “most or all of the time,” while the scoring for positively oriented items is inverted. The values of the ten items are summed up to reflect mental health status, with total scores ranging from 0 to 30. Higher scores indicate poorer mental health and an increased risk of depression among older adults. Previous research indicates that a threshold of 12 was typically used to identify depression, and it has been demonstrated that this threshold effectively identifies clinically significant depression [23–25]. Therefore, the study established a dichotomous variable at the 12-point cut-off to determine whether respondents had depression (0 = No, 1 = Yes), considering those with a score greater than or equal to 12 as having depression.Table 1Indexes of depressive symptoms statusNoQuestion1I was bothered by things that don’t usually bother me2I had trouble keeping my mind on what I was doing3I felt depressed4I felt everything I did was an effort5I felt hopeful about the future6I felt fearful7My sleep was restless8I was happy9I felt lonely10I could not get “going”
In this study, smartphone use diversity serves as the core independent variable, reflecting the diversity of functional utilization of smartphones among older adults. The measurement of this variable was derived from eight core survey items pertaining to smartphone use in the CHARLS database. Drawing on existing literature that categorizes the functional dimensions of digital device usage [26–28], these items were consolidated into four distinct functional (1) social interaction, (2) entertainment, (3) information seeking, and (4) payment, Each dimension was measured using a set of binary items (0/1) composed of closely related specific behaviors. The specific classifications were detailed in Table 2.Table 2Definition and measurement of smartphone use diversity dimensionsDimensionSurvey Items in CHARLSSocial interactionChat; Use WeChat; Post WeChat MomentsEntertainmentWatch videos; Play gamesInformation seekingWatch newsPaymentUse mobile payments; Financial management
Based on this foundation, this study constructed the "Smartphone Use Diversity" indicator to measure the diversity of smartphone usage among older adults. The calculation formula was as 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ Smartphone\text{ }Use\text{ }Diversity_i=\frac{\sum_{s=1}^nk_s}n
In this context, *Smartphone Use Diversity*~*i*~ represents the smartphone use diversity among older adults. This metric is a continuous variable with values ranging from 0 to 1, where higher values indicate greater diversification in the individual’s smartphone usage. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {k_s} $$\end{document}indicates the score for the *s-th* smartphone usage feature, reflecting the individual’s ability to utilize that feature. Within each dimension, if the i-th individual exhibits one or more smartphone usage behaviors classified under that dimension (i.e., any item is answered “yes”), the dimension is recorded as “used” (assigned a value of 1); Otherwise, it is recorded as 0. N represents the number of dimensions in Smartphone Use Diversity, with *n* = 4 in this study. This method effectively captures the breadth of smartphone functionality usage among older adults, emphasizing the diversity of digital living rather than quantifying the intensity of individual usage. ### Covariables Considering the influence of additional factors on the depressive symptoms of older adults, this study selects control variables that encompass demographic and health-related factors [29–32]. The demographic factors include gender, age, area of residence, marital status, and education. Health-related factors include self-rated health status, chronic disease, alcohol consumption, and smoking status. The definitions and descriptions of these variables are presented in Table 3.Table 3Name and description of the variablesVariable nameVariable descriptionDependent variablesDepressive symptomsNo depression = 0, Depression = 1Independent variablesSmartphone use diversityThe degree of smartphone use diversityCovariatesGenderMale = 1, Female = 0Age65 ~ 74 = 0, ≥ 75 = 1Marital statusUnmarried = 1, Married = 2, Divorced = 3, Widowed = 4EducationPrimary school and below = 0, Junior school and above = 1Area of residenceUrban = 0, Rural = 1Self-rated healthVery poor = 1, Poor = 2, Fair = 3, Good = 4, Excellent = 5Chronic diseaseNo = 0, Yes = 1Alcohol consumptionNo = 0, Yes = 1Smoking statusNo = 0, Yes = 1 ### Statistical analysis This study utilizes Stata 18.0 and R 4.5.0 for data analysis. First, descriptive statistics present the demographic characteristics, health status, and distribution of depressive symptoms among participants. Second, to account for repeated measures within individuals over time, we employed a multilevel mixed-effects logistic regression model with random intercepts at the individual level. Fixed effects included time variable (year), primary exposure (smartphone use diversity) and covariates. Random intercepts were specified at the individual level to capture unobserved heterogeneity in baseline depression risk. Robust standard errors were estimated using a cluster-robust variance estimator at the individual level. Third, to further identify underlying patterns of smartphone use among older adults, latent class analysis (LCA) was conducted separately for 2018 and 2020. The optimal number of classes was determined using multiple fit indices (AIC, BIC, adjusted BIC, and entropy). The identified classes were subsequently examined in relation to depressive symptoms through logistic regression models, using the“Limited Users”group as the reference. This approach allows for a more nuanced understanding of the heterogeneity and evolving trajectories of smartphone engagement. Additionally, robustness was assessed by replacing the binary outcome variable with the continuous CES-D-10 score, and re-estimating associations using a multilevel linear regression model. Subgroup analyses further evaluated heterogeneity across gender, area of residence, and education level. Finally, to control for potential selection bias, propensity score matching (PSM) was employed. ## Results ### Descriptive statistics Table 4 presents detailed descriptive statistics for the study sample, stratified by depressive symptom status. The analysis includes longitudinal data from 2018 to 2020 waves of the CHARLS, comprising 8,657 unique participants (mean age 69.3 ± 7.2 years) with 14,183 person-observations. At baseline (2018), 31.2% of participants met the criteria for depressive symptoms (CES-D-10 score ≥ 12). The prevalence of depressive symptoms differed significantly across most sociodemographic and health-related factors. Older adults with depressive symptoms were more likely to be female, widowed, rural residents, and less educated (all *P* < 0.001). In addition, poorer self-rated health, presence of chronic conditions, and absence of alcohol consumption or smoking were all significantly associated with higher rates of depressive symptoms. No significant difference was observed across age groups (*P* = 0.841).Table 4Descriptive statistics of the main variableCharacteristicTotal(*n* = 14183)No Depression(*n* = 9756)Depression(*n* = 4427)*P*-valueAge group0.84165 ~ 7410,154 (71.6%)6,990 (71.6%)3,164 (71.5%)≥ 754,029 (28.4%)2,766 (28.4%)1,263 (28.5%)Gender< 0.001Male6,906 (48.7%)5,231 (53.6%)1,675 (37.8%)Female7,277 (51.3%)4,525 (46.4%)2,752 (62.2%)Marital status< 0.001Unmarried86 (0.6%)59 (0.6%)27 (0.6%)Married10,720 (75.6%)7,612 (78.0%)3,108 (70.2%)Divorced127 (0.9%)89 (0.9%)38 (0.9%)Widowed3,250 (22.9%)1,996 (20.5%)1,254 (28.3%)Area of residence< 0.001Urban5,576 (39.3%)4,241 (43.5%)1,335 (30.2%)Rural8,607 (60.7%)5,515 (56.5%)3,092 (69.8%)Education level< 0.001Primary school and below11,227 (79.2%)7,380 (75.6%)3,847 (86.9%)Junior school and above2,956 (20.8%)2,376 (24.4%)580 (13.1%)Self-rated health< 0.001Very poor1,095 (7.7%)388 (4.0%)707 (16.0%)Poor3,318 (23.4%)1,760 (18.0%)1,558 (35.2%)Fair6,899 (48.6%)5,134 (52.6%)1,765 (39.9%)Good1,550 (10.9%)1,310 (13.4%)240 (5.4%)Excellent1,321 (9.3%)1,164 (11.9%)157 (3.5%)Chronic disease< 0.001No1,766 (12.5%)1,476 (15.1%)290 (6.6%)Yes12,417 (87.5%)8,280 (84.9%)4,137 (93.4%)Alcohol consumption< 0.001No9,743 (68.7%)6,380 (65.4%)3,363 (76.0%)Yes4,440 (31.3%)3,376 (34.6%)1,064 (24.0%)Smoking status< 0.001No10,674 (75.3%)7,194 (73.7%)3,480 (78.6%)Yes3,509 (24.7%)2,562 (26.3%)947 (21.4%)Continuous variables presented as mean (SD) with Wilcoxon test; Categorical variables with Fisher/χ² test^1^*n* (%)^2^Fisher’s exact test Table 5 summarizes smartphone use among older adults across four functional dimensions. Overall smartphone use diversity remained low in the sample, with a mean diversity score of 0.08 (SD = 0.24) on a scale ranging from 0 to 1. Among specific functions, social interaction exhibited the highest utilization rate (mean = 0.10), followed by entertainment (mean = 0.09) and information seeking (mean = 0.09). In contrast, mobile payment had the lowest adoption rate (mean = 0.05). These findings highlight that although smartphone penetration has increased, older adults’ engagement remains concentrated in limited domains, with substantial room for functional diversification.Table 5Descriptive statistics of smartphone use among older adults (*N* = 14183)Min ValueMax ValueMeanStandard DeviationObsOverall diversity score010.080.2414,183Social interaction010.100.3014,183Entertainment010.090.2814,183Information seeking010.090.2814,183Payment010.050.2214,183 ### Benchmark regression analyses This study investigates the impact of the smartphone use diversity on depressive symptoms among older adults, with the basic regression results presented in Table 6. We employed a three-stage stepwise adjustment strategy to control for potential confounding factors. Specifically, Model 1 includes the core independent variable of smartphone use diversity. Model 2 incorporates demographic variables, including age, gender, marital status, area of residence, and education. Model 3 further adjusts for health status (self-rated health and chronic disease) and health behaviors (alcohol consumption and smoking status).Table 6Multilevel logistic regression analysis of smartphone use diversity and depressive symptoms among elderly peopleVariablesModel 1Model 2Model 3Primary exposureSmartphone use diversity, β (SE)−1.482***(−10.64)−0.827***(−5.69)−0.707***(−5.13)OR (95% CI)0.227(0.173 to 0.299)0.437(0.329 to 0.581)0.493(0.376 to 0.646)Time fixed effectsYear, Ref: 20182020, β (SE)0.271***(5.54)0.214***(4.34)0.242***(4.96)OR (95% CI)1.312(1.192 to 1.444)1.238(1.124 to 1.364)1.273(1.158 to 1.401)Demographic factorsAge, Ref: 65–74 years≥ 75 years, β (SE)−0.0801(−1.15)−0.125(−1.93)OR (95% CI)0.923(0.806 to 1.058)0.883(0.778 to 1.002)Gender, Ref: FemaleMale, β (SE)−0.823***(−12.00)−0.701***(−9.69)OR (95% CI)0.439(0.384 to 0.502)0.496(0.431 to 0.572)Socioeconomic factorsMarital status, Ref: UnmarriedMarried, β (SE)−0.335(−0.91)−0.287(−0.79)OR (95% CI)0.715(0.348 to 1.472)0.751(0.369 to 1.529)Divorced, β (SE)−0.176(−0.37)−0.0762(−0.16)OR (95% CI)0.838(0.327 to 2.146)0.927(0.372 to 2.309)Widowed, β (SE)0.0749(0.20)0.0986(0.27)OR (95% CI)1.078(0.518 to 2.242)1.104(0.537 to 2.269)Education, Ref: ≤Primary school≥Junior school, β (SE)−0.559***(−6.19)−0.461***(−5.62)OR (95% CI)0.572(0.479 to 0.683)0.630(0.537 to 0.740)Area of residence, Ref: UrbanRural, β (SE)0.741***(10.58)0.589***(9.20)OR (95% CI)2.098(1.829 to 2.407)1.801(1.589 to 2.042)Health-related factorsSelf-rated health, Ref: Very poorPoor, β (SE)−0.860***(−8.35)OR (95% CI)0.423(0.346 to 0.518)Fair, β (SE)−1.960***(−18.75)OR (95% CI)0.141(0.115 to 0.173)Good, β (SE)−2.680***(−19.54)OR (95% CI)0.069(0.052 to 0.090)Excellent, β (SE)−3.039***(−20.00)OR (95% CI)0.048(0.036 to 0.065)Chronic disease, Ref: NoYes, β (SE)0.641***(6.42)OR (95% CI)1.899(1.561 to 2.309)Alcohol consumption, Ref: NoYes, β (SE)−0.153*(−2.30)OR (95% CI)0.858(0.753 to 0.978)Smoking, Ref: NoYes, β (SE)0.202**(2.65)OR (95% CI)1.223(1.054 to 1.420)Model fit statisticsRandom intercept variance (SE)3.786***(13.69)3.441***(13.13)2.231***(11.39)ICC (95% CI)0.535(0.499–0.570)0.511(0.474–0.548)0.404(0.363–0.446)Log-likelihood−8358.052−8134.183−7600.787AIC16724.116290.3715237.57BIC16754.3416373.5215373.65AUC (95% CI)--0.952 (0.948–0.955)Observations (*N*)14,18314,18314,183(1) *OR* odds ratio, *CI* confidence interval, *Ref * reference(2) * *p* < 0.05, ** *p* < 0.01, *** *p* < 0.001 Multilevel logistic regression analyses revealed a robust inverse association between smartphone use diversity and depressive symptoms across sequential adjustments. Key covariate analysis indicates that the risk of depression for men is significantly lower than that for women (Model OR = 0.496, 95% CI: 0.431–0.572; *P* < 0.001). Additionally, older adults with an education level of junior school and above exhibit a significantly lower risk of depression compared to those with an education level of primary school and below (Model OR = 0.630, 95% CI: 0.537–0.740; *P* < 0.001). Furthermore, the risk of depression among rural older adults is significantly higher than that of their urban counterparts (Model OR = 1.801, 95% CI: 1.589–2.042; *P* < 0.001). Moreover, older adults in better health exhibit a significantly lower risk of depression compared to those in “very poor” health. The multilevel mixed-effects logistic regression models showed progressive improvement in model fit with the sequential inclusion of covariates. Model fit indices indicated improvement, with the AIC decreasing from 16,724.1 in Model 1 to 15,237.6 in Model 3, and the log-likelihood increasing from − 8,358.1 to −7,600.8. The variance of the random intercept at the individual level declined by 41.2% (from 3.786 to 2.231), while the intraclass correlation coefficient decreased from 0.535 (95% CI: 0.499–0.570) to 0.404 (95% CI: 0.363–0.446), suggesting that a larger proportion of individual-level heterogeneity was explained by the included covariates. In addition, the fully adjusted model (Model 3) yielded a high level of discriminative capacity, with an AUC of 0.952 (95% CI: 0.948–0.955), as illustrated in the receiver operating characteristic (ROC) curve (Fig. 1). Collectively, these results confirm that the applied multilevel modeling strategy is both appropriate and robust for addressing the structure of the data and for capturing variability in depressive symptoms among older adults. Besides, the forest plot (Fig. 2) visually presents the associations between various factors and depressive symptoms among older adults, based on the baseline regression results.Fig. 1Receiver operating characteristic (ROC) curve of the fully adjusted multilevel mixed-effects logistic regression modelFig. 2Factors associated with depression ### Latent class analysis results #### Model selection To explore the potential patterns of smartphone usage among older adults, this study conducted latent class analyses separately on the data from 2018 to 2020. As shown in Table 7, a comprehensive analysis of various model fit indices revealed that the data from both years supported the two-class model as the optimal solution. Specifically, the two-class models for 2018 and 2020 achieved the lowest values in the three key information AIC, BIC, and aBIC. Additionally, the entropy values reached 0.997 and 0.976, respectively, significantly surpassing those of other class models, indicating that this solution has good category differentiation accuracy. Therefore, the two-class model was ultimately selected for subsequent analysis.Table 7Model selection statistics of latent class analysisYearClasses*N*LogLikAICBICaBICEntropySmallest class201826645−1616.4243248.8483303.2613277.840.99753.8%36645−1616.4223254.8443329.6613294.710.471915.81%46645−1616.4213258.8423347.2633305.950.34872.35%56645−1616.4233248.8463303.2593277.840.39693.34%202027538−6523.85613063.7113119.1313093.710.97615.8%37538−6523.61313071.2313154.3613116.230.471915.81%47538−6522.76913071.5413161.6013120.290.38992.35%57538−6522.30213078.6013196.3713142.350.29032.35%Classes: Number of latent classes specified in the model; LogLik: Log-likelihood value of the model; higher (less negative) values indicate better model fit; AIC (Akaike Information Criterion): A model fit index; lower values suggest better fit, accounting for model complexity; BIC (Bayesian Information Criterion): A criterion for model selection; penalizes model complexity more strongly than AIC; Entropy: A measure of classification accuracy; values range from 0 to 1, with higher values indicating clearer class separation; aBIC, sample-size adjusted BIC #### The basic characteristics of the two smartphone usage patterns among older adults As shown in Table 8 and Fig. 3, the latent class analysis identified two distinct patterns of smartphone use in both 2018 and 2020. The first group, labeled Limited Users, exhibited near-zero probabilities of engaging in any of the four smartphone functions, indicating minimal functional participation. In contrast, the second group, labeled Multi-functional Users, demonstrated consistently high probabilities across all dimensions, with particularly high engagement in social interaction (2018: 0.98; 0.93) and information seeking (2018: 0.92; 0.82). These findings highlight a pronounced heterogeneity in digital behaviors, with one subgroup showing very limited use and the other displaying broad and diversified smartphone utilization.Table 8Conditional probabilities of smartphone use across latent classes among older adults (2018 and 2020)YearClassesSocial InteractionEntertainmentPaymentInformation seeking20181:Limited Users0.000.000.000.002:Multi-functional Users0.980.770.410.9220201:Limited Users0.010.010.000.012:Multi-functional Users0.930.800.510.82Values represent conditional probabilities of using smartphone functions within each latent class. Limited Users were characterized by minimal engagement across all dimensions, whereas Multi-functional Users showed consistently high probabilities of use across functionsFig. 3Patterns of Smartphone Use among Older Adults #### Results of the logistic regression analysis Based on the results of the logistic regression analysis, this study examined the association between different smartphone usage patterns and depressive symptoms among older adults (see Table 9). In both survey years, “Limited Users” was used as the reference group for the analysis. The results indicated that, compared to “Limited Users,” “Multi-functional Users” were significantly associated with a lower risk of depressive symptoms. This association remained robust even after adjusting for covariates such as age, gender, marital status, residence, education level, self-rated health, number of chronic diseases, alcohol consumption, and smoking status. This indicates that diversified smartphone usage patterns were an important protective factor for the mental health of older adults.Table 9Association between smartphone usage patterns and depressive symptoms (2018 & 2020)Variable2018OR (95% CI)2018 *p*-value2020OR (95% CI)2020 *p*-valueSmartphone usage patternLimited Users (Ref)1.001.00Multi-functional Users0.51 (0.33–0.79)0.0020.66 (0.56–0.78)< 0.001Model statisticsObservations6,6457,538Pseudo R²0.1180.117Wald χ²820.89< 0.001936.58< 0.001*OR * odds ratio, *CI* confidence interval All models adjusted for age, gender, marital status, residence, education, self-rated health, chronic conditions, drinking, and smoking. ### Robustness test To validate the robustness of the baseline regression results, the continuous CES-D-10 scale score is employed as an alternative outcome variable, and a multilevel linear regression model is constructed for robustness testing. The results indicate that smartphone use diversity remains significantly negatively correlated with depressive symptoms among older adults (β = −1.341, *p* < 0.001). Control variables, including gender, education, area of residence, and chronic disease, are also significantly associated with depressive symptoms, and the direction of these associations is highly consistent with the baseline regression results, further confirming the robustness of the baseline regression results. Detailed information is presented in Table 10.Table 10Robustness multilevel linear regression of depressive symptomsVariablesβ (SE)95% CISmartphone use diversity−1.341*** (0.198)(−1.730, −0.952)Year, Ref: 20182020, β (SE)0.753*** (0.084)(0.588, 0.917)Male−1.641*** (0.132)(−1.900, −1.382)Marital status, Ref: UnmarriedMarried−0.505 (0.717)(−1.910, 0.901)Divorced0.400 (0.877)(−1.319, 2.120)Widowed0.358 (0.727)(−1.067, 1.784)Education, Ref: ≤Primary school≥Junior school−1.029*** (0.137)(−1.297, −0.761)Area of residence, Ref: UrbanRural1.344*** (0.118)(1.112, 1.576)Self-rated health, Ref: Very poorPoor−2.398*** (0.230)(−2.848, −1.948)Fair−5.062*** (0.222)(−5.497, −4.626)Good−6.601*** (0.252)(−7.094, −6.107)Excellent−7.236*** (0.256)(−7.738, −6.734)Chronic disease, Ref: NoYes1.294*** (0.145)(1.009, 1.579)Alcohol consumption, Ref: NoYes−0.281** (0.116)(−0.509, −0.054)Smoking, Ref: NoYes0.394*** (0.131)(0.137, 0.651)Model fit statisticsRandom intercept variance12.95 (0.521)-Residual variance21.11 (0.471)-Log-likelihood−44713.49-AIC89464.99-BIC89608.63-Observations (N)14,183-*** *p* < 0.001, ** *p* < 0.01, * *p* < 0.05 ### Heterogeneity analysis This study analyzed the effects of smartphone use diversity on depressive symptoms among older adults with different genders, area of residence, and education. The results of the analysis were presented in Table 11. First, the analysis of gender heterogeneity indicates that smartphone use diversity has a similar impact on depressive symptoms among older adults of different genders, but the alleviation effect is significantly more pronounced among men (Male: OR = 0.482, 95% CI: 0.333–0.697, *P* < 0.001; Female: OR = 0.533, 95% CI: 0.359–0.793, *P* < 0.01). Second, the analysis of residential heterogeneity shows that smartphone use diversity is beneficial for older adults in both urban and rural areas, but the protective effect is significantly stronger for urban residents (Urban: OR = 0.450, 95% CI: 0.308–0.656, *P* < 0.001; Rural: OR = 0.647, 95% CI: 0.430–0.974, *P* < 0.05). Moreover, concerning educational level, the alleviation effect of smartphone use diversity on depressive symptoms is stronger among individuals with higher educational attainment (≥ junior high school) (OR = 0.488, 95% CI: 0.314–0.760, *P* < 0.01) compared to those with lower education (OR = 0.551, 95% CI: 0.384–0.790, *P* < 0.01). Figure 4 visually presents the heterogeneity in the impact of smartphone use diversity on depressive symptoms among older adults across different subgroups. The results indicate that smartphone use diversity significantly alleviates depressive symptoms among urban older adults, men, and individuals with higher educational levels.Fig. 4Results of heterogeneity analysisTable 11Results of heterogeneity analysisVariablesGenderArea of residenceEducationMaleFemaleRuralUrbanPrimaryschool and belowJunior school and aboveSmartphone use diversity, OR (95%CI)0.482*** (0.333–0.697)0.533** (0.359–0.793)0.647* (0.430–0.974)0.450*** (0.308–0.656)0.551** (0.384–0.790)0.488**(0.314–0.760)Control variablesYESYESYESYESYESYESN690672778607557611,2272956Pseudo R20.0850.0860.0760.1170.0820.119* *p* < 0.05, ** *p* < 0.01, *** *p* < 0.001 ### Propensity score matching (PSM) Estimation To mitigate potential confounding factors and diminish selection bias, we employed a 1 nearest neighbor propensity score matching approach with a caliper size set at 0.2. This method was implemented to ensure baseline balance between the treatment (smartphone users) and control (non-users) groups. As illustrated in Table 12; Fig. 5, the pre-match analysis revealed considerable imbalances across several covariates, with percentage biases reaching as high as 89.5% for education level and 61.1% for residence. After matching, the balance significantly improved, with all covariates registering biases below 10%, and the vast majority below 5%.Table 12Comparison of the difference of covariates before and after matching (full sample)VariablesMatchMeans%Bias%Reduct biasT - testTreatedControlt*P*>|t|GenderU0.580.4820.30-7.680.00M0.580.59−2.1089.70−0.600.55AgeU0.130.30−43.20-−14.820.00M0.130.131.1097.500.370.71Marital status (Married)U0.840.7424.40-8.720.00M0.840.85−1.7093.20−0.530.60Marital status (Divorced)U0.100.091.60-0.650.52M0.100.073.80−128.401.140.26Marital status (Widowed)U0.150.24−24.00-−8.560.00M0.150.140.8096.800.250.80ResidenceU0.350.64−61.10-−23.210.00M0.350.350.1099.800.040.97EducationU0.560.1689.50-38.640.00M0.560.550.1099.800.040.98Self-rated health (Poor)U0.170.24−17.70-−6.430.00M0.170.170.9094.900.280.78Self-rated health (Fair)U0.560.4816.20-6.140.00M0.560.57−1.3091.70−0.390.70Self-rated health (Good)U0.130.118.10-3.190.00M0.130.121.3083.700.360.72Self-rated health (Excellent)U0.090.090.30-0.120.90M0.090.091.50−360.600.420.67Chronic diseaseU0.900.886.90-2.560.01M0.900.90−1.5077.90−0.460.65Alcohol consumptionU0.440.3030.40-11.950.00M0.440.431.7094.500.460.65SmokingU0.240.25−1.20-−0.460.65M0.240.24−0.1088.200.040.97YearU0.840.5078.00-26.860.00M0.840.830.4099.500.140.89*U represents the difference between the treatment and control groups before pairing; M is the difference between the treatment and control groups after pairingFig. 5Standardized % bias across covariates The enhanced balance of post-matching is underscored by the significant reduction in mean bias from 28.2% pre-match to a mere 1.2% post-match, as shown in Table 13. Furthermore, the Pseudo R²value dropped from 0.246 to 0.001, highlighting the effective balancing of covariates between the groups. The matching procedure also led to a substantial decline in Rubin’s B from 143.8% to 6.5%, demonstrating the enhanced homogeneity of the matched samples.Table 13Overall fit indicators of propensity score modelSamplePseudo R2LR chi2*p* > chi2MeanBias(%)Rubin’s BRubin’s *R*Unmatched0.2462502.06< 0.00128.2143.8*0.89Matched0.0013.490.9991.26.51.28 The propensity score-matched multilevel logistic regression analysis revealed a robust protective association between smartphone use diversity and depressive symptoms(Table 14). After adjusting for covariates, each unit increase in usage diversity was associated with a 51.5% reduction in the odds of depression (OR = 0.485, 95% CI: 0.36–0.65, *p* < 0.001).Table 14Association between smartphone use and propensity Score-Matched multilevel logistic regression analysisVariablesOR95% CI*p*-valueSmartphone Use Diversity0.485[0.36, 0.65]< 0.001Age group, Ref: 65–74 years≥ 75 years0.878[0.73, 1.06]> 0.05Gender, Ref: FemaleMale0.421[0.35, 0.51]< 0.001Marital status, Ref: UnmarriedMarried3.197[0.17, 59.37]> 0.05Divorced1.127[0.04, 30.20]> 0.05Widowed4.976[0.27, 92.95]> 0.05Education, Ref: ≤Primary school≥Junior school0.654[0.54, 0.80]< 0.001Residence, Ref: UrbanRural1.930[1.64, 2.28]< 0.001Self-rated health, Ref: Very poorPoor0.337[0.24, 0.46]< 0.001Fair0.104[0.07, 0.14]< 0.001Good0.0388[0.02, 0.06]< 0.001Excellent0.0279[0.02, 0.05]< 0.001Chronic disease, Ref: NoYes1.669[1.22, 2.28]< 0.01Alcohol consumption, Ref: NoYes0.980[0.82, 1.17]> 0.05Smoking, Ref: NoYes1.220[0.99, 1.50]> 0.05Random intercept variance2.746[2.10, 3.59]< 0.001Observations (*N*)9057Log-likelihood−4799.9AIC9635.7BIC9763.7****p* < 0.001, ***p* < 0.01, **p* < 0.05*OR* Odds Ratio, *CI* Confidence Interval* ## Discussion This study finds that smartphone use diversity significantly reduces the risk of depression among older adults. This conclusion is supported by the use of a nationally representative sample from the CHARLS database, which mitigates the regional or segment-specific sample biases present in prior studies [33, 34]. Furthermore, to provide robust and multi-faceted evidence for understanding the relationship between smartphone use diversity and depressive symptoms among older adults, we employed a combination of statistical methods, including multilevel logistic regression, latent class analysis (LCA), robustness checks, and propensity score matching. Several studies have examined the association between smartphone use and depressive symptoms among older adults. Consistent with the findings of He et al. [35–37], this study confirms that smartphone use is associated with reduced depressive symptoms among older adults. However, unlike Ji et al. [38], which focused only on the binary distinction of whether smartphones are used, our study emphasizes the multidimensional nature of smartphone engagement. To capture this complexity, we introduces the concept of smartphone use diversity—a comprehensive indicator capturing an individual’s overall capacity to utilize smartphone functionalities across four key social interaction, entertainment, payment, and information seeking. The results demonstrate that smartphone use diversity significantly reduces the risk of depression among older adults, indicating that the ability to integrate digital technology serves as an important predictor of mental health and it can reflect its influence on the depressive symptoms of the elderly more comprehensively. Beyond this overall indicator, and in contrast to independent analyses of single functionalities such as playing games or reading news conducted by Fan et al. [34, 39], our study further employs Latent Class Analysis (LCA) to uncover distinct patterns of smartphone use. The analysis consistently reveals two profiles across both survey waves, identified as Limited Users and Multi-functional Users, highlighting the considerable heterogeneity in digital engagement among older adults. This finding is consistent with Hofer et al. [40], who argue that different patterns of digital technology usage serve distinct functions. Building on this perspective, the key to technological empowerment lies not merely in promoting device ownership, but in fostering active engagement with a wide range of smartphone functionalities. Accordingly, the mental health outcomes of older adults may depend less on the binary distinction of whether smartphones are used and more on the specific ways in which they are utilized. Building on this, the latent class analysis provides deeper insights into the usage patterns of smartphones among older adults and highlights the dynamic evolution of their digital engagement. First, from a longitudinal perspective, smartphone usage among older adults shows a trend of simultaneous “breadth expansion” and “depth deepening.” Specifically, “Limited Users” exhibit very low participation rates in digital behaviors such as social interaction, entertainment, payments, and information seeking. However, from nearly zero in 2018, their engagement shows a slight increase by 2020, indicating early signs of digital integration. This changing trend aligns with the progressive characteristics of technology adoption outlined in the theory of digital integration [41], this reflects that, despite their initially low acceptance rates, older adults are beginning to engage with and utilize technology against the backdrop of increasing digital technology prevalence [42]. Meanwhile, “Multi-functional Users” demonstrate deeper levels of digital integration while maintaining high participation rates across various functions. Notably, the probability of using payment functions significantly increases from 0.41 in 2018 to 0.51 in 2020, making it the functional dimension with the largest growth. This phenomenon is closely related to the comprehensive deepening of “cashless payments” in Chinese society between 2018 and 2020 [43]. The widespread adoption of payment functions often signifies a significant enhancement in the digital skills and trust levels of older adults, as this functionality involves both financial security and operational complexity [41, 44]. This aligns with the behavioral characteristics of late adopters in the innovation diffusion theory, who gradually follow suit after confirming the practicality and safety of the technology [45]. This change strongly illustrates that the evolution of the external social environment and technological ecosystem serves as a key driving force in advancing the digital integration of older adults to a higher level [46]. Furthermore, from a horizontal comparison across different dimensions, this study reveals the intrinsic hierarchical structure of smartphone usage behaviors among older adults. In both survey years, “Multi-functional Users” consistently display a preference the probability of using social interaction functions is highest, followed by information seeking and entertainment functions, while the probability of using payment functions is lower but shows significant growth. This structural model reveals the typical pathways for older adults’ acquisition of digital skills [47, 48]. They generally begin with low-risk, high-reward functions, such as instant messaging and video calls, which satisfy emotional and social connection needs. Gradually, they expand to information seeking and entertainment functions [49], ultimately attempting higher-risk, more complex functions related to financial security, such as mobile payments [44]. This hierarchical structure, progressing from easy to difficult and from enjoyable to practical, provides a scientific theoretical basis for designing and promoting elderly-oriented digital interventions. Ultimately, its robust association with depressive symptoms further corroborates the positive significance of diversified digital engagement. In both survey years, “Multi-functional Users” demonstrate a significantly lower risk of depression. Therefore, facilitating the transition of older adults from “limited use” to “multi-functional use,” and supporting the deepening of their digital engagement in line with technological and social developments, should be regarded as a vital component of policies aimed at promoting active and healthy aging. This study also finds that the impact of smartphone use diversity on depressive symptoms among older adults varies significantly by gender, area of residence, and education level. In terms of gender, the impact of smartphone use diversity on depressive symptoms is more pronounced among men. This may be related to the limited opportunities for many elderly women to engage with digital devices due to historical educational backgrounds or occupational experiences [50], their lower proficiency in smartphone use restricts their ability to fully utilize diverse functionalities, thereby diminishing the potential positive effects of smartphones on mental health [51]. Regarding area of residence, smartphone use diversity is significantly associated with reduced depressive symptoms among both urban and rural older adults, with a stronger effect observed in urban populations. Insufficient internet coverage in rural regions and the lower penetration rate of smartphones among older adults limit their ability to access the internet and utilize diverse smartphone functionalities [52, 53]. In contrast, older adults in urban areas benefit from well-developed internet infrastructure, allowing them to fully leverage various smartphone features and engage in diversified online social activities, which helps alleviate depressive symptoms. Finally, the impact of smartphone use diversity on depressive symptoms is significantly stronger among older adults with higher education levels compared to those with lower education levels. Education directly correlates with an individual’s ability to access and utilize digital technology; older adults with higher education typically demonstrate greater digital literacy [54, 55]. Their acceptance and learning ability of new technologies are higher, enabling them to quickly master and leverage diversified smartphone functionalities. This engagement fosters cognitive stimulation and a sense of achievement that alleviates depression. In contrast, older adults with lower education levels often experience difficulties with digital skills, resulting in apprehension toward the complex features of smartphones. Their usage typically remains confined to basic communication functions, thereby somewhat limiting the effectiveness of diverse functionalities in alleviating depressive symptoms. While this study identifies a significant association between greater diversity in smartphone use and lower levels of depressive symptoms among older adults, it is equally important to recognize the dual nature of digital technology. In the context of rapid technological development, the potential risk of smartphone addiction among older adults merits careful consideration. For example, Karas et al. [39], in a survey of 392 older adults, reported a strong association between smartphone addiction and adverse psychological outcomes, including depression and anxiety. This finding differs from the results of the current study, primarily due to differences in the research subjects. Karas et al. focus on older adults who actively use social media, while this study utilizes a large-scale national dataset from China that encompasses a broader demographic of older adults, enhancing its representativeness. Furthermore, prior studies have shown that the penetration rate of smartphones among older adults remains comparatively low relative to younger cohorts [56, 57]. For most older adults, smartphones primarily serve as tools for information seeking, social interaction, and entertainment, and large-scale dependency or addiction has not yet been observed. Nonetheless, these differences highlight that the benefits of digital technology must be considered alongside its potential risks. While diversified smartphone engagement may support psychological well-being, excessive or addictive use may conversely undermine it. This is especially critical for individuals with tendencies toward excessive use or addictive behaviors, for whom differentiated guidance on technology use and tailored mental intervention strategies should be implemented. Despite the rigorous analytical approaches employed, this study is subject to several limitations. First, the measurement of smartphone use diversity, while improved, remains constrained by the predetermined functional items in the CHARLS questionnaire. This framework does not fully capture emerging or specific applications, such as mobile health services, mobile commerce. Second, the primary limitation arises from the nature of the observational data. The short duration of the two-wave panel data limits the application of a fixed-effects model for causal inference, primarily due to restricted within-individual variation in smartphone usage patterns during the observed period. Furthermore, although a wide range of covariates is adjusted for, the potential for residual confounding by unmeasured factors (e.g., cognitive decline and specific life events) persists. ## Conclusions and implications This study, utilizing nationally representative data from China, establishes a significant negative association between smartphone use diversity and depressive symptoms among older adults. The results indicate that the mental health benefits associated with digital technologies go beyond mere access and are substantially influenced by the range of functions adopted and integrated into daily life. Latent class analysis further delineated two different usage profiles—Limited Users and Multi-functional Users—revealing pronounced heterogeneity in digital engagement patterns within the aging population. This classification underscores the critical need to address disparities in digital skills and functional usage. From a policy perspective, these findings suggest that initiatives aimed at digital inclusion should move beyond promoting basic device ownership to fostering competencies that support varied and purposeful digital engagement. Interventions designed to encourage diversified and context-sensitive smartphone use may not only help alleviate depressive symptoms but also contribute to narrowing the digital divide and advancing healthy aging in an evolving digital society.