Exploring the association between Frailty Index and low back pain in middle-aged and older Chinese adults: a cross-sectional analysis of data from the China Health and Retirement Longitudinal Study (CHARLS)


  • Data were analysed from the China Health and Retirement Longitudinal Study, a nationally representative survey.

  • The large sample size enhances the statistical strength and generalisability of the results to the middle-aged and older Chinese population.

  • The reliance on self-reported low back pain data may introduce bias.

  • The study’s cross-sectional design limits our ability to establish causality; further longitudinal studies are necessary.


Low back pain (LBP) is the most frequent musculoskeletal condition, representing 36.8% of cases in 2017.1 By 2020, 619 million people globally suffered from LBP, with projections increasing to 843 million by 2050, largely due to an ageing population.2 Among G20 nations, China ranks second in disability-adjusted life years for significant musculoskeletal disorders, excluding the European Union.3 LBP often presents as severe discomfort or neuropathic pain that extends to the lower limbs. It is typically labelled as ‘nonspecific’ when its precise cause is undetermined.4 Recent studies underline the challenges in diagnosing and managing LBP, noting that the use of multiple risk indicators can improve diagnostic precision.5 6 Patients’ expectations significantly influence treatment outcomes for LBP, highlighting the psychological and social dimensions of managing this condition.7 This underscores the complex interaction between patient perceptions and clinical outcomes, adding to LBP’s overall burden. With a lifetime prevalence of approximately 80%, LBP poses a major health issue. Spinal disorders are the fourth most common reason for medical consultations in the USA.8 Although the precise source of pain is often unclear in most cases, the condition typically resolves within a short period through self-care methods.4 9 Given the inherent risk of chronicity or disability associated with LBP, it is important to acknowledge that the prognosis for narrowly defined acute LBP, including cases with sciatica, is more challenging than often perceived, especially regarding return-to-work outcomes. This situation highlights the necessity for comprehensive approaches that address both primary and secondary prevention of recurrent and chronic LBP.10

Individual frailty can significantly contribute to a variety of negative consequences, including increased risk of disability, hospitalisation, premature mortality and impaired resilience to stressors.11 The degree of frailty, as measured by the Frailty Index (FI)—a ratio of current to projected health deficits—may be a reliable predictor for tailoring treatments. A larger FI suggests greater health deficiencies, which corresponds with increased frailty.12 Thus, regular assessment of frailty using the FI is crucial for preventing or delaying LBP and its associated negative effects.13 However, studies explicitly connecting the FI to the occurrence of LBP remains sparse, emphasising an area suitable for additional exploration.14

Despite LBP’s well-documented global impact and high prevalence among the elderly, research into the relationship between frailty and LBP in China is limited. Our study fills a significant gap by looking into the relationship between FI and LBP in middle-aged and older Chinese people, notably those aged 45 and up. This research intends to determine whether there is a correlation between a greater FI and a higher prevalence of LBP using data from the China Health and Retirement Longitudinal Study (CHARLS). Our approach not only relies on the existing body of research but also brings forth new perspectives on the relationship between ageing and health in a large demographic group. This serves as a crucial basis for future interventions and policies specifically designed for this important yet understudied population.


Data source and participants

This research made use of data from CHARLS, a longitudinal survey that specifically focused on persons aged 45 and older in China. CHARLS seeks to gather extensive data in order to comprehend the socioeconomic effects and results of the ageing process. The study comprises comprehensive inquiries into the elder population’s economic status, encompassing both physical and mental health, demography, and social networks. The response rate for CHARLS’s annual interviews has regularly exceeded 80%.15 The analyses were based on publicly available data that have been approved by relevant review boards. The CHARLS survey received approval from the Biomedical Ethics Committee of Peking University (IRB00001052-11015), and all participants were obligated to provide informed consent by signing.16–18 The CHARLS data is accessible to the public. The reporting of our study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.19

Pain questionnaires are a customary element of CHARLS. However, the FI can only be computed using the datasets from the years 2011, 2013 and 2015.15 Therefore, our research relied on cross-sectional data derived from these particular waves of the survey. We selected individuals aged 45 years or older who had comprehensive data on LBP and the FI from the CHARLS 2011, 2013 and 2015 datasets.

Data collection

Data was methodically gathered through organised household interviews carried out by skilled staff. These interviews included demographic data, including age, gender, residential address, marital status, educational level and medical history, with a particular emphasis on chronic illnesses. The demographic information is essential for modifying our analysis to consider possible confounding variables that may impact the association between LBP and FI.

Additionally, comprehensive data were collected on health behaviours, such as smoking status, alcohol consumption and Body Mass Index (BMI). Incorporating these behaviours into our study provides a more comprehensive picture of the direct and indirect factors that contribute to both LBP and frailty. The health behaviour data, particularly BMI, were assessed using established techniques, whereas information on smoking and drinking habits was obtained through self-reporting. The CHARLS data collection procedures adhered to the ethical norms established by the Peking University Research Ethics Committee, guaranteeing the protection of participants’ privacy and autonomy throughout the research process.

Calculation of the FI

A person’s FI was determined by dividing the sum of all measured deficits by the number of impairments related to age. This methodology takes advantage of the extensive data available in the CHARLS dataset and is in line with accepted research standards.12 In this sense, a ‘health deficit’ is any age-related physical or mental disability that raises the probability of unfavourable consequences.

In the index, you may see a variety of impairments, including those related to activities of daily living and instrumental activities of daily living, which cover 11 different tasks like personal hygiene, dressing and money management. Additionally, it takes into consideration restrictions in physical function (nine items), chronic diseases (nine items), markers of psychological health (five items) and subjective assessments such as self-rated health.

In order to create this index, we used a binary coding system where ‘0’ meant there was no deficit and ‘1’ meant there was. In cases where a variable included an intermediate response option, such as ‘sometimes’ or ‘possibly’, a value of ‘0.5’ was assigned to reflect partial deficits. Using this method, a more complex picture of a person’s health can be created. Online supplemental table S1 in the supplemental materials lists all of the components of the FI.

Supplemental material

Indexes containing 30–40 variables are effective in predicting unfavourable health outcomes, according to previous research.20 21 For this reason, we validated and relied on our results by included only individuals with 30 or more real health deficiencies in the FI calculation.

Definition of LBP

The participants were originally queried about the frequency of their physical pain, to which they were instructed to respond with a simple ‘yes’ or ‘no’. Individuals who expressed a favourable response were subsequently asked to indicate the precise site of their discomfort from a predetermined inventory of anatomical regions, encompassing choices such as the cranium, shoulders, arms, wrists, fingers, chest, stomach, back, lower back, buttocks, legs, knees, ankles, toes, neck and other areas. Participants who reported frequent physical discomfort and specified that the pain was located in the lower back were categorised as having LBP.22–24


The assessment of covariates included sociodemographic characteristics, lifestyle factors and chronic diseases. The collected sociodemographic data included age, gender, place of residence (classified as rural or urban), educational attainment (categorised as illiterate, junior high school or below, high school or above) and marital status (options included married, never married, separated, divorced or widowed). The BMI was determined by measuring the weight and height of individuals (kg/m²) and categorised into three categories based on the guidelines set by the WHO: underweight/normal (<25), overweight (25 to 30) and obese (≥30).25

The lifestyle factors that were evaluated were smoking and alcohol intake. The smoking status was ascertained by considering the respondents’ self-reported smoking history, which was categorised as either never smoked or ever smoked. The latter group comprised those who reported any past smoking experience, regardless of whether it was frequent or sporadic. The categorisation of alcohol intake was based on self-reported history, distinguishing between individuals who have never consumed alcohol and those who have consumed alcohol in the past, regardless of the frequency or quantity.

Chronic diseases were classified according to individuals’ self-reported medical history. Hypertension was identified by self-reported diagnosis or medication use, as well as by clinical criteria: a systolic blood pressure of ≥140 mm Hg or a diastolic blood pressure of ≥90 mm Hg, based on the average of three consecutive measurements obtained by proficient medical staff throughout the interview. Diabetes was ascertained through self-reported medical diagnosis, administration of insulin injections or diabetic tablets, or clinical measurements indicating an HbA1c level of ≥6.5%, or a fasting glucose level of ≥126 mg/dL. Dyslipidaemia was determined by a self-reported diagnosis or medication use or by having a total cholesterol level >240 mg/dL, HDL-C level <40 mg/dL, LDL-C level ≥140 mg/dL or a triglyceride level ≥150 mg/dL, as measured from blood samples collected during the health assessment.

Statistical analysis

Participant characteristics were summarised using descriptive statistics. The normal distribution of continuous data was assessed, and the mean values were reported together with their SD. Categorical variables were expressed as percentages (%). The FI was treated both as a continuous variable and categorised into quartiles based on the distribution within our sample: Q1 (≤0.15), Q2 (0.16–0.18), Q3 (0.19–0.22) and Q4 (≥0.23). Group differences were evaluated using one-way analysis of variance for continuous data that followed a normal distribution and the chi-square test for categorical variables. Logistic regression models were applied to estimate ORs and 95% CIs for the association between FI and LBP. In the continuous model, the FI was multiplied by a factor of 10 to make it easier to understand the ORs. The study was conducted in a sequential manner, starting with an unadjusted model, followed by Model 1 that accounted for sociodemographic data and finally, Model 2 that additionally accounted for lifestyle aspects. The upper limit for BMI values was set at 40 in order to reduce the impact of extreme values that deviate significantly from the norm. The Model 3 was comprehensively modified, incorporating all factors discussed before, as well as hypertension, diabetes and dyslipidaemia.

To investigate possible non-linear relationships, we employed restricted cubic spline regression, with knots placed at the 5th, 35th, 65th and 95th percentiles of the FI. The inflection point was identified by the utilisation of piecewise regression analysis, with the application of the likelihood ratio test and bootstrap resampling approach to validate significant points of effect.

To evaluate the consistency of the association between FI and LBP across different strata, subgroup analyses under Model 3 considered variables like gender, age categories and chronic diseases. The clinical relevance and potential for interaction effects informed the subgroup selection process. Interactions were investigated using the likelihood ratio test, and subgroup heterogeneity was tested using multivariate logistic regression.

Sensitivity analysis was conducted on participants with a broader range of health deficits (more than 10) to test the robustness of our findings. Since the study sample depended on the available data, no power calculation was performed a priori.

All analyses were conducted using the R statistical package V.4.1.2 (http://www.R-project.org) and Free Statistics software V.1.9.26 A p value of below than 0.05 was considered statistically significant, including adjustments for multiple comparisons when necessary.

Patient and public involvement



Participant selection

Among the 24 958 individuals who participated in the data collection, 2231 were below the age of 45 and were deliberately omitted from the study in order to concentrate on the middle-aged and older population, where LBP is more prevalent and often more impactful. We removed participants with incomplete data on LBP (n=886) to ensure the precision of our outcome measures. In order to ensure the reliability of our study’s focus, we removed individuals (n=9986) who had fewer than 30 health deficit items. This is because a lower number of deficits would not yield a dependable FI. In addition, we removed data for participants (n=5480) who had missing covariate data. These covariates included important sociodemographic and health behaviour information that was crucial for the modified models.

A total of 6375 participants from the CHARLS database, spanning the years 2011 to 2015, were selected as subjects for this cross-sectional analysis. The procedure for selecting and excluding participants is outlined in online supplemental figure S1.

Participant characteristics

The table 1 displays the characteristics of the study subjects, categorised into quartiles based on the FI. Out of all the participants, 1733 cases (27.2%) reported LBP. The average age was 60.6±9.5 years, and 3781 (59.3%) were female. The observed associations suggest that factors such as older age, female gender, rural residency, unmarried status or living alone, lower educational levels, being overweight or obese, current or past smoking, never drinking, and the presence of hypertension, diabetes, or dyslipidaemia were associated with higher levels of frailty (table 1).

Table 1

Characteristics of participants

Relationship between FI and LBP

Univariate analysis demonstrated that variables including age, gender, place of residence, level of education, BMI, smoking and hypertension exhibited associations with the prevalence of LBP. The specific ORs and CIs for these associations are detailed in online supplemental table S2. Across all models, the multivariate logistic regression analysis revealed a significant association between FI and LBP. The study found a strong and consistent correlation between FI and the likelihood of reporting LBP. As FI increased, the odds of reporting LBP also increased. This association was observed in all three models: Model 1 (OR=1.51, 95% CI: 1.40 to 1.63, p<0.001), Model 2 (OR=1.66, 95% CI: 1.53 to 1.80, p<0.001) and Model 3 (OR=1.67, 95% CI: 1.54 to 1.81, p<0.001).

When the FI was divided into four equal groups, persons in the highest quartile (Q4, FI ≥0.23) had considerably greater chances of experiencing LBP compared with those in the second group (Q2, FI 0.16–0.18), with adjusted ORs ratios revealing a significantly increased risk (OR=2.90, 95% CI: 2.45 to 3.42, p<0.001). The ORs for the first quartile (Q1, FI ≤0.15) and third quartiles (Q3, FI 0.19–0.22) were adjusted and found to have increased odds of LBP compared with the Q2. The OR for Q1 was 1.21, while the OR for Q3 was 1.45, as shown in table 2.

Table 2

Multivariate logistic regression analyses of FI and LBP

Therefore, there was a U-shaped curve in the relationship between FI and LBP (non-linear, p<0.001) (figure 1). A threshold effect analysis indicated an inflection point at an FI of 0.176. Below a certain threshold, there was a minor decrease in the likelihood of experiencing LBP as FI increased. However, above this threshold, there was a substantial positive link between FI and LBP. This suggests that when frailty develops beyond this point, the ORs of experiencing LBP also increase (table 3).

Figure 1
Figure 1

Dose–response association between Frailty Index and low back pain. Solid and dashed lines represent the predicted value and 95% CIs, respectively. Orange bars represent the distribution of the entire cohort and red bars indicate the distribution of low back pain. Adjustments were made for age, sex, residence, marital status, education level, BMI, smoking, drinking, hypertension, diabetes and dyslipidaemia. The figure displays 99% of the data. BMI, Body Mass Index.

Table 3

Threshold effect analysis between FI and LBP

Stratified analysis

We analysed stratified data to see how age, gender and chronic diseases may influence the connection between FI and LBP. The analyses conducted did not show any significant interaction effects across the subgroups, as seen by the interaction p values displayed in figure 2.

Figure 2
Figure 2

Stratified analyses of the association between Frailty Index and low back pain. The p value for interaction represents the likelihood of interaction between the variable and Frailty Index. Ref, reference.

Notably, persons under 60 years of age showed a more distinct U-shaped connection between FI and LBP. Among the younger participants, those with the lowest FI (Q1, ≤0.15) had a greater likelihood of experiencing LBP compared with those in the reference group (Q2, FI 0.16–0.18), with an adjusted OR of 1.43 (95% CI: 1.14 to 1.79). This indicates that those with extremely low or extremely high levels of frailty are more likely to experience lower back pain in this particular age group. The modifications made to the ORs took into consideration the potential confounding factors that were identified in the multivariate models.

Sensitivity analysis

In a sensitivity analysis, we re-evaluated the connection between FI and LBP by including a wider range of individuals from the research group. This analysis incorporated 21 827 individuals. We excluded persons who were under 45 years of age, those who had missing data on LBP and those who had less than 10 health deficit items. This was done to provide a more comprehensive sample while preserving the accuracy and reliability of the data.

After accounting for all possible confounding factors, the ORs for LBP were computed for each quartile of FI, compared with the second quartile (Q2, FI 0.13–0.15). The adjusted ORs for the lowest quartile (Q1, FI <0.13), the third quartile (Q3, FI 0.16–0.19) and the highest quartile (Q4, FI ≥0.2) were 1.04 (95% CI: 0.86 to 1.28, p=0.665), 1.24 (95% CI: 1.06 to 1.44, p=0.007) and 3 (95% CI: 2.6 to 3.47, p<0.001), respectively. The ‘adjusted’ ORs take into consideration the potential confounding factors that were detected in the multivariate analysis, as outlined in online supplemental table S3.

A threshold was identified at a FI of 0.139 through a study of inflection points. When the FI value was below a certain threshold, there was no significant association with the risks of LBP (p>0.05). However, when the FI value was equal to or more than 0.139, there was a significant increase in the odds of LBP (online supplemental table S4). This indicates a shift in the correlation between FI and LBP at this juncture.

Furthermore, a restricted cubic spline model, with 0.139 as the reference point, showed that within this larger sample, the relationship between FI and LBP indeed follows a U-shaped curve (online supplemental figure S2).


The findings of this study suggest that there is a U-shaped relationship between FI and prevalence of LBP in middle-aged and older Chinese individuals. Both very low and very high levels of frailty are associated with an increased likelihood of LBP. The study identified an inflection point at a FI of 0.176, beyond which the likelihood of experiencing LBP significantly increases. The intricate relationship between FI and LBP was supported by a multivariate logistic regression analysis. This analysis demonstrated a positive correlation between increasing levels of FI and LBP, which remained strong even after accounting for other potential confounding factors such as age, gender and BMI. Stratified analyses were conducted to investigate the potential influence of age and gender on the link between FI-LBP, but no significant interaction effects were found in these subgroups. However, a more distinct U-shaped pattern was observed, specifically among individuals below the age of 60. This pattern suggests that both the lowest and greatest degrees of frailty substantially increase the likelihood of experiencing lower back pain in this younger population. Conducting sensitivity analysis on a larger group of participants in the study confirmed the U-shaped correlation, with a revised turning point at a FI of 0.139. Below this threshold, FI was not significantly associated with LBP; however, odds of LBP significantly increased when FI reached or exceeded this point. The results remained comparable across different quartiles of FI after accounting for potential confounding factors, demonstrating the strength and reliability of the findings across a broader range of participants.

Furthermore, the prevalence of LBP among rural dwellers was notably higher than that among urban residents, potentially reflecting the disparity in access to medical care. The disparity between rural and urban areas indicates that LBP might have a significant negative influence on everyday activities and emotional well-being, functioning as a hidden threat in rural communities. Hence, enhancing healthcare facilities and delivering focused education on LBP in these regions could effectively reduce its consequences. The results of this study indicate that health interventions should focus on individuals who are not only frail but also those who appear to be healthier but are still at risk. This provides a basis for future research to investigate preventive and treatment measures that are specifically designed for levels of frailty.

Prior studies have gradually clarified the intricate connection between FI and LBP. Research constantly shows that frail individuals frequently experience limitations in their ability to move around in different environments and experience decreases in muscular strength and mass. These factors hinder their capacity to carry out basic everyday tasks and participate in physical activities.27 This decline is directly associated with a higher likelihood of acquiring LBP and associated impairments. Furthermore, existing evidence indicates a connection between frailty and certain orthopaedic diseases that are known to cause LBP.28 29 Our study extends these findings by uncovering that not only high but also low FI levels are associated with an elevated prevalence of LBP. This unexpected result may be attributed to the ‘J-shaped’ or ‘U-shaped’ risk curve often observed in epidemiological studies, where both very low and very high exposures (or risk factors) lead to increased outcomes. When considering frailty, persons with a very low FI may not have noticeable physical limitations yet, but they may have other underlying diseases or lifestyle variables, such as sedentary behaviour, that are not accounted for by typical frailty measures. These characteristics make them more vulnerable to experiencing LBP. A recent discovery indicates that the intensity of LBP can directly impact the occurrence of pre-frail and frail disorders in older persons. This reveals a clear connection between the degree of pain and indicators of frailty.30 This observation prompts a broader consideration of frailty beyond the conventional scope, suggesting that even those perceived as ‘less frail’ could be at risk and should be considered in preventive and management strategies for LBP.

The relationship between pain and frailty is closely linked, with the involvement of pro-inflammatory cytokines in their pathophysiology highlighting this connection. Rocha et al’s research emphasises that there is a strong correlation between higher degrees of frailty and increased pain intensity in older women who are having LBP.31 This association is exacerbated by the systemic inflammatory response, wherein elevated levels of TNF-α, interleukin-6 (IL-6) and C reactive protein enhance the severity of LBP.32–36 These cytokines, including IL-1 and TNF, are known to trigger nociceptors, promoting pain hypersensitivity and neural activity that lowers the threshold for pain perception.36–38 Inflammation not only intensifies pain but also plays a role in muscle breakdown, resulting in sarcopenia and exacerbating the disability associated with LBP.33 37 39 The term ‘inflammation’ is used to describe the chronic low-grade inflammation that is commonly seen in ageing persons. This inflammation can make older people more susceptible to persistent LBP.40 The reciprocal relationship between inflammation and LBP indicates that heightened levels of inflammatory mediators can arise from and also contribute to heightened pain intensity, hence, impacting the everyday activities and movement of individuals experiencing LBP.38 41 42

Our findings suggest that future research should focus on conducting longitudinal studies to gain a deeper understanding of the causal connections between FI and LBP in middle-aged and older populations. These studies could specifically target the phases of frailty progression to identify critical intervention points that could delay or prevent the onset of LBP. Furthermore, it is possible to conduct intervention studies to evaluate the efficacy of focused techniques designed to address frailty, such as strength training, nutritional supplementation and comprehensive lifestyle management programmes. These interventions attempt to enhance physical resilience and decrease the likelihood of experiencing LBP. Evaluating the efficacy of these interventions could provide actionable insights for healthcare providers to mitigate the impacts of frailty on LBP prevalence. Furthermore, the inclusion of additional components such as genetic indicators of frailty, comprehensive evaluations of muscle quality rather than solely focusing on bulk, and psychological elements could augment our comprehension of the intricate interplays that contribute to low back pain in frail individuals. These variables could also assist in defining more individualised methods for managing and preventing LBP.

Although our findings provide vital insights for public health, it is crucial to recognise the limitations of this study. The result of LBP is reliant on data provided by patients themselves, which could potentially be influenced by their memory recall bias. Nevertheless, the data obtained from CHARLS were gathered by researchers who underwent professional training to reduce this bias, so offering a more dependable foundation for our discoveries. In order to improve the validity of future studies, it is recommended to incorporate objective measurements, such as clinical assessments.43 44 Furthermore, although cognitive function is a factor in determining the FI, we encountered a constraint due to the disparity between the cognitive evaluation instrument used in CHARLS and the Mini-Mental State Examination (MMSE) commonly applied in standardised FI development protocols. This discrepancy made it challenging to incorporate cognitive function directly into our FI calculation. Nevertheless, our FI encompasses 35 items, which adheres to research suggesting that an index with 30–40 variables is sufficiently accurate for predicting adverse outcomes, thus, ensuring the reliability of our frailty measure. Future research should strive to use standardised cognitive evaluations that are compatible with the MMSE in order to provide a more thorough evaluation of frailty. Furthermore, the cross-sectional design of our study unavoidably restricts the ability to establish causality. Subsequent investigations can enhance this study by using a longitudinal design, which would offer a stronger foundation to analyse the temporal connection and causality between FI and LBP. Such prospective studies would not only strengthen the evidence for causality but also enhance our understanding of the dynamics underlying the association between FI and LBP. Also, our study did not incorporate comprehensive variables such as physical activity levels, dietary habits and direct measurements of muscle strength due to the unavailability of such data in the CHARLS database. In future study, it is crucial to include these significant covariates in order to conduct a more thorough analysis that effectively accounts for a wider array of possible confounding factors. Adding this element would augment our comprehension of the complex interconnections among lifestyle factors, muscle strength and low back pain. Further, it is important to acknowledge that our data set did not include prescription drugs for pain due to constraints in the CHARLS database. However, it is worth noting that the Frailty Index we used encompasses a wide range of chronic diseases, which may somewhat compensate for this limitation. Ultimately, blood biomarkers were gathered during Waves 1 (2011) and Waves 3 (2015) in order to align with Health and Retirement Study and other ageing surveys, which take blood samples every alternate wave.45 The cross-sectional data we compiled present a challenge in including inflammatory and related biochemical markers. To address this constraint, future research should strive to reliably integrate these biomarkers. By including inflammatory biomarkers into future research, we can gain a more comprehensive understanding of the molecular mechanisms that connect frailty and LBP. This could help identify specific targets for therapeutic interventions and open up new possibilities for treatment and preventative methods.


In this cross-sectional analysis, we identified a U-shaped relationship between the FI and LBP among middle-aged and older Chinese individuals. Both low and high FI scores were linked to a higher occurrence of LBP, with a notable inflection point. Although these connections are persuasive, the fact that our study is cross-sectional prevents us from establishing causality. Consequently, our study emphasises the necessity of future longitudinal research to investigate the time-related interactions between FI and LBP, which could possibly inform the creation of focused interventions for the prevention and treatment of LBP in the context of ageing. Additionally, a thorough examination could evaluate the effects of measures to reduce frailty on LBP.

Data availability statement

Data are available in a public, open access repository. The CHARLS datasets generated and analysed during the current study are available in the website of the CHARLS home page at http://charls.pku.edu.cn/en. The CHARLS data are deidentified. Respondents are identified by a unique ID number.

Ethics statements

Patient consent for publication

Ethics approval

This study involves human participants. The CHARLS survey project was approved by the Biomedical Ethics Committee of Peking University (IRB00001052-11015). All participants gave signed informed consent at the time of participation.

This post was originally published on https://bmjopen.bmj.com