The study sample was obtained from the (CHARLS), a representative national cohort with middle-aged and older adults in China, as described previously (Zhao et al., 2014). At baseline, 17,708 participants from 450 urban and rural areas of 28 provinces were recruited in 2011, with follow-up evaluations in 2013, 2015, and 2018. During each survey, information on age, sex, marital status, educational level, family income, residence, social activities, retirement, cigarette smoking, alcohol consumption, sleep duration, number of chronic diseases, and body mass index (BMI) was collected from structured questionnaires and physical examinations by trained interviewers and physicians, respectively. Adults ≤ 60 years of age (N = 10,255), and those with missing depressive symptom assessments (N = 454) and demographic data (N = 50) were excluded. In addition, 1,646 or 1,361 participants without sufficient evaluation data on the frailty index or phenotypic frailty were excluded, respectively. Thus, using the baseline data in 2011, 5,303 or 5,117 participants underwent a cross-sectional analysis between frailty, as defined by a frailty index or phenotype, respectively, and depressive symptoms, using the baseline data in 2011. In the cohort study (2011–2018), we further excluded those participants with baseline frailty or depressive symptoms and those lost to follow-up. Therefore, 3,157 or 3,082 participants remained in the cohort analysis with baseline frailty, as defined by a frailty index or phenotype, respectively, and the incidence of depressive symptoms, while 2,086 or 1,491 participants were included for the cohort analysis between baseline depressive symptoms and incidence of frailty assessed by a frailty index or phenotype, respectively. Additionally, with the limited information, such as physical examination indicators, for the definition of phenotypic frailty in 2018, the association between baseline depressive symptoms and the incidence of phenotypic frailty was only evaluated during follow-up from 2011 to 2015. The flow chart of the selection process for participants is shown in Supplementary Figure 1.

This study was approved by the Peking University Ethical Committee. Informed consent was obtained from all participants.

The 10-item Center for Epidemiological Studies Depression Scale (CES-D) was used to evaluate depressive symptoms (Andresen et al., 1994). As reported elsewhere, the CES-D includes 10 items with four answers for each item, as follows: “rarely (<1 day/week);” “some days (1–2 days/week);” “occasionally (3–4 days/week);” “most (5–7 days/week).” For the negative items, the score was assigned 0, 1, 2, and 3 points for “rarely,” “some days,” “occasionally,” and “most,” respectively; whereas for the positive items, the score was defined as 3, 2, 1, and 0 points for “rarely,” “some days,” “occasionally,” and “most,” respectively. Then, the total score was summed for the 10 items, ranging from 0 to 30. In agreement with a prior study verified the validity of CES-D in CHARLS (Chen and Mui, 2014), the study participants were classified as depressed with a CES-D score ≥ 12 and non-depressed with a CES-D score < 12.

Frailty was assessed by both phenotype and a frailty index. As reported elsewhere, phenotypic frailty was determined by the physical frailty phenotype (PEP) scale, which includes weakness, slowness, exhaustion, inactivity and weight loss (Wu et al., 2017). Weak was defined as a maximum handgrip strength for either hand less than the 20th percentile for the sex- and BMI- adjusted weighted population distribution. Slow was the average time of repeated walking tests over a 2.5-m course that exceeded the 80th percentile for the sex- and height- adjusted weighted population distribution. Exhaustion occurred when subjects felt that anything they did was an effort or they could not get going. Inactivity was defined as subjects who walked continuously for < 10 min in a typical week. Weight loss was defined as a self-rated loss of ≥5 kg in the previous year or a BMI ≤ 18.5 kg/m². Using the above information, phenotypic frailty was categorized into three levels, as follows: robust (meeting none of the five domains); and prefrail and frail (meeting any one or more criteria).

The modified procedure of the China Kadoorie Biobank was used to evaluate the frailty index (Fan et al., 2020), and included 20 deficits, including chronic diseases (i.e., hypertension, heart disease, stroke, emphysema or bronchitis, tuberculosis, asthma, peptic ulcer, gallstone diseases, rheumatoid arthritis, fracture, neurasthenia, diabetes, cancer, and chronic kidney diseases), symptoms and signs (i.e., sleep disturbances, body pain or discomfort, unintentional weight loss, feeling sad or depressed, and poor health status), and physical measurements (i.e., BMI). Each deficit was dichotomized or mapped from 0 (the healthiest status) to 1 (the unhealthiest status). The frailty index was calculated as a ratio of the number of deficits for each participant to the total number of deficits, with a range from 0 to 1. Two subgroups were created, as follows: robust (≤0.10); and pre-frailty and frailty (>0.10). Because the number of frail people over 60 aged in the CHARLS is limited, we had to combine the prefrailty and frailty conditions.

The distributions of the study sample were determined by frailty status. Means ± standard deviations (SDs) or numbers (percentages) were calculated to perform continuous or categorical variables, respectively. Student’s t-test, the Mann-Whitney U-test and the chi-square test were used to compare the distribution of covariates in univariate analyses. Logistic regression models were used to determine the bidirectional associations between frailty and depressive symptoms with a cross-sectional design. Cox proportional hazard regression models were used to determine the relationship between frailty and the incidence of depression, as well as the relationship between depression and the incidence of frailty. For example, in the association between frailty index and the incidence of depressive symptoms, if individuals without depressive symptoms at baseline (2011), but were evaluated to be with depressive symptoms in 2013, then the time interval in Cox proportional hazards model was considered as 2 years. Similarly, if participant without depressive symptoms in 2011 and 2013, but were assessed to be with depressive symptoms in 2015, then the time interval was defined as 4 years; if individuals without depressive symptoms in 2011, 2013, and 2015, but were evaluated to be with depressive symptoms in 2018, then the time interval was considered as 7 years. Additionally, to ensure the reliability of the results, we further calculated the per SD change value after zero-mean normalization of the continuous independent variable. The standardization process was calculated according to the following formula:

where z_ij represents the standardized independent variable, x_ij represents the original independent variable, x_i represents the mean value of the independent variable, and s_i represents the SD of the independent variable.

Moreover, subgroup analyses based on sex and residence were both performed to verify such associations, either in a cross-sectional or prospective study. In addition, a series of sensitivity analyses were further conducted. Firstly, because the definition of frailty might involve depressive symptoms-related factor (“Overall in the last month, how much of a problem did you have with feeling sad, low, or depressed?”), we excluded the associated factors and performed sensitivity analyses. Secondly, since the age limit established for old age is 65 years, we conducted the sensitivity analyses among people over the age of 65, as well. Thirdly, the survival time is unlikely to be precisely observed in the queuing setting, and thus at best it is known only to fall within an interval between two consecutive surveys, that is interval-censoring. Briefly, it treats the right-truncated observation value as a special interval with an infinite right boundary and the exact event time as a zero-length interval. The PROC ICLIFETEST (Guo et al., 2014) provides a non-parametric statistical method for estimating survival functions and a statistical test of interval-censored data. Therefore, we used this procedure to investigate the association between baseline frail status and incidence of depressive symptoms, as well as the association between baseline depressive symptoms and the incidence of frailty.

In the present study, the crude model was used to evaluate the odd ratio (OR) or hazard ratio (HR) and the 95% confidence interval (CI) without any adjustment. Model 1 adjusted for age and sex, and model 2 additionally adjusted for education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration. All statistical analyses were conducted using SAS (version 9.4; SAS Institute Inc., Cary, NC, United States). All P-values were two-tailed, and a P < 0.05 was considered statistically significant.

Table 1 shows the basic characteristics of the study population. As calculated by the frailty index, greater than one-half of 5,303 participants (53.84%) were pre-frail or frail, primarily affecting the majority of women and retirees. These subjects tended to have a higher BMI, lower educational level and family income, shorter sleep duration, and several chronic diseases. In addition, the prevalence of depressive symptoms among pre-frail and frail subjects was significantly greater than the robust subjects. Indeed, 65.88% of subjects had pre-frailty and frailty when determined based on phenotype. Specifically, the pre-frail and frail subjects were more likely to be older, females, no alcohol consumption, countrymen, and retirees. Furthermore, the pre-frail and frail subjects tended to have several chronic diseases, but a lower BMI, educational and income levels, and shorter sleep duration. The prevalence of depressive symptoms in pre-frail and frail subjects was significantly greater than the robust subjects.

Figure 1 presents the association of frailty with depressive symptoms in the cross-sectional study. Pre-frail and frail participants had higher risks of depression before and after adjustments for confounders when compared to the robust participants. As shown in Supplementary Table 1, the crude ORs (95% CIs) for depressive symptoms among pre-frail and frail participants, as defined by the frailty index and phenotype, were 3.25 (2.87–3.69) and 10.56 (8.73–12.89), respectively. The ORs (95% CIs) for depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 2.03 (1.91–2.17), and 2.58 (2.41–2.77), respectively. The ORs (95% CIs) for depressive symptoms among pre-frail and frail participants, as defined by the frailty index and phenotype were 3.19 (2.82–3.62) and 10.86 (8.95–13.29) after adjustment for age and sex, respectively. The ORs (95% CIs) for depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 2.01 (1.88–2.14), and 2.76 (2.56–2.97), respectively. The full-adjusted ORs (95% CIs) for depressive symptoms among pre-frail and frail people defined by index and phenotype were 3.05 (2.68–3.49) and 9.78 (8.02–12.03), respectively. The ORs (95% CIs) for depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 1.95 (1.82–2.08), and 2.61 (2.42–2.82), respectively.

When assigning depressive symptoms as an independent variable, as shown in Supplementary Table 2, depressed patients had higher risks of pre-frailty and frailty, with crude ORs (95% CIs) of 3.25 (2.87–3.69) for the frailty index and 10.56 (8.73–12.89) for phenotypic frailty. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation of the depressive symptoms score were 1.90 (1.79–2.02) and 3.78 (3.46–4.16), respectively. After adjusting for age and sex, depressed patients had higher risks of pre-frailty and frailty, with ORs (95% CIs) of 3.19 (2.81–3.62) for the frailty index; and 10.88 (8.97–13.31) for phenotypic frailty. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation of the depressive symptoms score were 1.89 (1.77–2.01) and 3.90 (3.55–4.29), respectively. Moreover, depressed patients had higher risks of pre-frailty and frailty, with ORs (95% CIs) of 3.07 (2.69–3.50) for the frailty index and 9.95 (8.15–12.24) for phenotypic frailty, after adjusting for age, sex, education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation of the depressive symptoms scores were 1.89 (1.77–2.02) and 3.97 (3.59–4.40), respectively.

Because the definition of frailty might involve depressive symptom-related factors, sensitivity analysis was performed after excluding the related factors, as shown in Supplementary Table 3. The results were similar to that of the entire population. Pre-frail and frail participants had higher risks of depressive symptoms before and after adjustments for confounders, when compared to the robust participants. In the crude model, pre-frail and frail participants were at higher risk of depressive symptoms when assessed by the frailty index and phenotype [ORs (95% CIs), 3.46 (3.00–4.00) and 1.69 (1.51–1.91), respectively], compared to robust subjects. In addition, the ORs (95% CIs) for depressive symptoms per each incremental increase in the standard deviation for the frailty index and phenotypic frailty scores were 1.95 (1.83–2.08) and 1.32 (1.25–1.40), respectively. After adjusting for age and sex, pre-frail and frail participants were at higher risk of depressive symptoms when assessed by the frailty index and phenotype [ORs (95% CIs), 3.32 (2.87–3.84) and 1.73 (1.53–1.95), respectively], respectively, compared to robust subjects. In addition to the ORs (95% CIs) for depressive symptoms per each incremental increase in the standard deviation for the frailty index and phenotypic frailty scores were 1.93 (1.81–2.06), and 1.35 (1.27–1.43). In the full-adjusted model, pre-frail and frail participants were at higher risk of depressive symptoms when assessed by an index and phenotype [ORs (95% CIs), 3.12 (2.68–3.63) and 1.59 (1.40–1.82), respectively], compared to robust subjects. The ORs (95% CIs) for depressive symptoms per each incremental increase in the standard deviation for the frailty index and phenotypic frailty scores were 1.88 (1.76–2.01) and 1.28 (1.20–1.37), respectively.

When assigning depressive symptoms as an independent variable, as shown in Supplementary Table 4, depressed participants had a higher risk of pre-frailty and frailty before and after adjustments for confounding factors, including age, sex, education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration, when compared to the non-depressed participants. In the crude model, the ORs (95% CIs) for pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 3.46 (3.00–4.00) and 1.69 (1.51–1.91), respectively. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation for the depressive symptoms scores were 1.98 (1.84–2.11) and 1.34 (1.27–1.42), respectively. After adjusting for age and sex, the ORs (95% CIs) for pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 3.31 (2.87–3.84) and 1.73 (1.53–1.95), respectively. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation for the depressive symptoms scores were 1.93 (1.80–2.07) and 1.36 (1.28–1.44), respectively. The full-adjusted ORs (95% CIs) for pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients were 3.15 (2.71–3.67) and 1.59 (1.40–1.82), respectively. The ORs (95% CIs) for pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation for the depressive symptoms scores were 1.92 (1.79–2.08) and 1.31 (1.23–1.40), respectively.

Consistently shown in Supplementary Figure 2, after adjusting for age, sex, education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration, pre-frail and frail participants (>65 years) were at higher risk of depressive symptoms when assessed by an index and phenotype [ORs (95% CIs), 2.78 (2.35–3.30) and 11.03 (8.31–14.94), respectively], compared to robust subjects. In contrast, the full-adjusted ORs (95% CIs) for pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients were 2.79 (2.36–3.31) and 11.23 (8.45–15.22), respectively.

The incidence of depressive symptoms was 56 per 1,000-person years. The association between frailty and the incidence of depressive symptoms is shown in Table 2. Compared with the robust participants, prefrail or frail participants had higher risks of depressive symptoms before and after adjusting for confounders, including age, sex, education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration. The crude HRs (95% CIs) for depressive symptoms among prefrail and frail participants, as defined by the frailty index and phenotype, were 1.39 (1.23–1.58) and 1.38 (1.21–1.56), respectively. The HRs (95% CIs) for the incidence of depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 1.24 (1.17–1.31), and 1.19 (1.12–1.26), respectively. After adjusting for age and sex, the HRs (95% CIs) for depressive symptoms among prefrail and frail participants, as defined by the frailty index and phenotype, were 1.38 (1.22–1.57) and 1.38 (1.21–1.57), respectively. The HRs (95% CIs) for the incidence of depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 1.23 (1.16–1.30), and 1.19 (1.12–1.27), respectively. The full-adjusted HRs (95% CIs) for depressive symptoms among prefrail and frail people, as defined by an index and phenotype, were 1.38 (1.22–1.57), and 1.30 (1.14–1.48), respectively. The HRs (95% CIs) for the incidence of depressive symptoms for each incremental increase in the standard deviation of the frailty index and phenotypic frailty scores were 1.23 (1.16–1.30), and 1.15 (1.08–1.30), respectively.

When setting depressive symptoms as an independent variable, the incidences of frailty defined by phenotype and index were 45 per 1,000 person years and 94 per 1,000-person year. As shown in Table 3, depressed participants had a higher risk of pre-frailty and frailty when assessed by the frailty index, although the risk was not statistically significant. The crude HRs (95% CIs) for the incidence of pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 1.17 (0.98–1.39) and 1.47 (0.95–2.28), respectively. The HRs (95% CIs) for the incidence of pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation of the depressive symptoms scores, were 1.06 (0.996–1.12) and 1.33 (1.17–1.53), respectively. With adjustment for age and sex, the adjusted HRs (95% CIs) for the incidence of pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 1.15 (0.96–1.37), and 0.98 (0.63–1.52), respectively. The HRs (95% CIs) for the incidence of pre-frailty and frailty, as assessed by the frailty index and phenotype, for each incremental increase in the standard deviation of the depressive symptoms scores, were 1.05 (0.99–1.11) and 1.14 (0.995–1.31), respectively. Moreover, the full-adjusted HRs (95% CIs) for the incidence of pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 1.18 (0.98–1.41) and 0.96 (0.61–1.50), respectively. The HRs (95% CIs) for the incidence of pre-frailty and frailty, as assessed by the frailty index and phenotype for each incremental increase in the standard deviation of the depressive symptoms scores were 1.06 (0.99–1.13) and 1.12 (0.96–1.29), respectively.

In agreement with the results from the entire study population, the sensitivity analysis showed that pre-frail or frail participants had a higher risk of depressive symptoms before and after adjustments for confounding factors, including age, sex, education level, smoking status, alcohol consumption, marital status, place of residence, income, participation in social activities, number of chronic diseases, retirement status, and sleep duration, when compared with the robust participants (Supplementary Table 5). The crude HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype, were 1.58 (1.39–1.81) and 1.17 (1.03–1.34), respectively. The depressive symptom HRs (95% CIs) for each incremental increase in standard deviation of the frailty index and phenotypic frailty scores were 1.22 (1.15–1.30) and 1.09 (1.03–1.16), respectively. After adjusting for age and sex, the HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype were 1.55 (1.36–1.77) and 1.19 (1.04–1.35), respectively. The depressive symptom HRs (95% CIs) for each incremental increase in standard deviation of the frailty index and phenotypic frailty scores were 1.21 (1.15–1.29) and 1.10 (1.03–1.17), respectively. The multivariate-adjusted HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype, were 1.55 (1.35–1.77) and 1.16 (1.02–1.32), respectively. The depressive symptoms HRs (95% CIs) for each incremental increase in standard deviation of the frailty index and phenotypic frailty scores were 1.22 (1.15–1.29) and 1.09 (1.02–1.16), respectively. When depressive symptoms were set as an independent variable (Supplementary Table 6), no significant relationship of existed between depressive symptoms and the incidence of frailty.

Using the ICLIFETEST procedure, the results were consistent with that of the main analysis. As shown in Supplementary Table 7, the crude HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype, were 1.44 (1.27–1.63) and 1.41 (1.24–1.60), respectively. With adjustment for age and sex, the HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype were 1.43 (1.26–1.62) and 1.42 (1.24–1.61), respectively. Moreover, the multivariate-adjusted HRs (95% CIs) for depressive symptoms among pre-frail and frail participants, as assessed by the frailty index and phenotype, were 1.44 (1.27–1.63) and 1.35 (1.19–1.54), respectively. However, considering the depressive symptoms as an independent variable, no significant relationship of existed between depressive symptoms and the incidence of frailty (Supplementary Table 8).

Moreover, for participants aged over 65, we got a similar result. As shown in Supplementary Figure 3, the multivariate-adjusted HRs (95% CIs) for depressive symptoms among pre-frail and frail participants (age over 65), as assessed by the frailty index and phenotype, were 1.25 (1.05–1.49) and 1.27 (1.06–1.53), respectively. Whereas considering depressive symptoms as an independent variable, the full-adjusted HRs (95% CIs) for the incidence of pre-frailty and frailty, as defined by the frailty index and phenotype among depressed patients, were 1.22 (1.00–1.49) and 0.92 (0.43–1.97), respectively.

Tables 4, 5 present the subgroup analysis of the association between frailty and incidence of depressive symptoms, according to sex and place of residence. No significant interaction effects of frailty were detected with sex, or the place of residence. Pre-frailty and frailty participants, as assessed by the frailty index, had a higher risk of depressive symptoms [HRs (95% CIs), 1.27 (1.05–1.53) for males; 1.48 (1.24–1.76) for females; 1.53 (1.21–1.94) for urban residents; and 1.33 (1.14–1.55) for rural residents]. The HRs (95% CIs) for the incidence of depression for each incremental increase in the standard deviation for frailty index scores were 1.28 (1.15–1.42) for males, 1.28 (1.16–1.42) for females, 1.37 (1.20–1.56) for urban residents, and 1.25 (1.14–1.36) for rural residents. Pre-frail and frail participants assessed by phenotype had a higher risk of depressive symptoms [HRs (95% CIs), 1.45 (1.19–1.75) for males; 1.25 (1.05–1.50) for females; 1.35 (1.07–1.72) for urban residents; and 1.28 (1.10–1.50) for rural residents). The HRs (95% CIs) for the incidence of depressive symptoms for each incremental increase in the standard deviation for phenotypic frailty scores were 1.26 (1.13–1.41) for males, 1.16 (1.05–1.29) for females, 1.18 (1.02–1.36) for urban residents, and 1.18 (1.08–1.29) for rural residents. The results of subgroup analysis between depressive symptoms and the incidence of frailty were consistent with the entire population (data not shown). Thus, no significant interaction effects of depressive symptoms were detected with sex, or the place of residence.