Association Between Drinking Water Sources and Osteoarthritis Incidence in Chinese Elderly Population: National Population-Based Cohort Study

Background The elderly is a vulnerable group susceptible to toxic environmental factors; however, the association between unsafe water and osteoarthritis (OA) incident among this population is poorly investigated. This study aimed to examine the effects of drinking water sources on OA risk in the Chinese elderly population. Methods Data were obtained from the China Longitudinal Healthy Longevity Survey to perform a 16-year longitudinal study. All participants aged ≥65 years at baseline were potentially eligible. Information on drinking water sources was recorded via a structured questionnaire. The water source was classified as well, surface, spring, or tap. OA was selected as the health outcome. Cox proportional hazard models, adjusted socio-demographics, lifestyle, and chronic disease were used to estimate the hazard ratio (HR) for OA. Subgroup analyses were performed to assess the potential interactive effect. Results A total of 12,543 participants were included. With restriction to the longest follow-up time, the mean survival time of follow-up was 14.56 years. Among the cohort members, 1,585 members suffered from OA. In the full model, drinking from well water and spring water was associated with a higher risk of OA compared with drinking from tap water (HR: 1.37, 95% CI: 1.22–1.54; HR: 1.34, 95% CI: 1.03–1.74). Except for age, the effects of modifications from socioeconomic status, lifestyles, and health conditions were non-significant (p for interaction <0.05). After multiple imputations for missing data and excluding deaths in the first year of follow-up and participants with a history of the disease and using the Shared-frailty Cox model, sensitivity analysis indicated a robust association between the drinking water source and OA incidence. Conclusion Drinking tap water was associated with a low risk of OA among older adults in China. The use of clean water sources as a marker of decreased OA and arthritis risks must be revisited in low- and middle-income countries.


INTRODUCTION
Osteoarthritis (OA) is a chronic pathological joint disease characterized by changes in cartilages and hyperplasia around bones (1). Patients with OA often experience knee joint injury, low activity, poor self-efficacy, severe fatigue, social ability limitation, and great functional disability (2). According to a Global Burden of Disease report, hip and knee OA ranked the 11th highest contributor to global disability and 38th highest in disability-adjusted life years (3). In China, the prevalence rates of mild, moderate, and severe knee OA among 17,459 middle and older people are 1.5, 3.3, and 3.9%, respectively (4), and total and per 100,000 population number of years lived with disability for knee OA are 4,149,628 and 968, respectively (4).
Osteoarthritis pathogenesis is unclear and linked with age, obesity, inflammation, trauma, and genetic and environmental factors. Consumable water in China contains a dangerous concentration of arsenic, fluorine, and sulfates (5), and an excess intake of fluoride and nitrate is associated with an increased risk of OA (6)(7)(8). In China, chemical and phosphate fertilizer plants discharge waste gas with fluorine into natural water sources (9,10). Except for Shanghai and Hainan, the fluorine level for drinking water in most parts of China is over the limit (9,10). An epidemiological study suggested a connection between OA prevalence and water districts contaminated with perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS) (6). A research using a rat OA model also revealed that (18) F-fluoride is potentially related to early OA changes (11). These empirical results indicate that drinking unprocessed water or polluted water increases the risk of OA possibly through a biological mechanism that fluorides or nitrides exert an adverse influence on the apatite of bone with secondary inflammation and oxidative stress in osteoporosis and osteosclerosis (12).
The global elderly population is at an increased risk of OA (13), and this cohort (aged ≥60 years) is rapidly growing in China (14). Therefore, the association between water contamination and OA incidence in this age group must be examined in the Chinese context. Water source indirectly reflects the quality of drinking water for residents. However, only a few studies were conducted on the association between drinking water sources and OA incidence in older people, especially in low-and middleincome countries.
Considering that OA might be affected by different water sources, this study evaluated the long-term effects of drinking water sources on OA risk among the Chinese elderly population. aged 80 years and above and performed multistage, stratified cluster sampling. Participants aged 65-79 years were additionally recruited since 2002. All the participants were interviewed on determinants of healthy longevity for the elderly, such as socioeconomic characteristics, lifestyle, social interaction, residential environment, cognitive function, and physical/mental condition. This study was approved by the Research Ethics Committee of Peking University (IRB00001052-13074), and written consent was obtained from all participants or their legal representatives.

Study Design and Subjects
To date, six waves of surveys have been performed. Individuals still living were followed up in each wave. If a participant has already died when being surveyed, then his/her relatives and/or family members were interviewed for the cause of his/her death, health condition, and utilization of health service before death. The questionnaires were divided into two sections, namely, the "survivors questionnaire" (for participants alive) and the "deceased questionnaire" for the relatives/family members (15). The current study extracted the dataset from 2002,2005,2008,2011,2014, and 2018 waves. Participants aged 65 years and above were included.
In the first wave (2002), 16,064 individuals were selected. Among which, 12,543 participants with complete information on drinking water sources and OA history at baseline and follow-up surveys were eligible. A total of 599 participants with missing data or ambiguous answers, 2,710 participants with OA at baseline, and 212 participants leaving the place of residence were excluded. A fixed cohort was employed in this study.

Arthritis Assessment
During the succeeding surveys in the cohort, OA diagnosis was self-reported ("Have you suffered from osteoarthritis?"). The outcome was defined as OA that occurred in the followup period for participants without a history of OA at baseline. All interviewees were followed up from baseline until arthritis incidence, dropout, or latest interview.

Assessment of Water Sources
Water sources were identified by one item: "What source of water source did you mainly drink from at present [well, surface water (river, lake, pond, and pool), spring, and tap water]?"

Covariates
Several potential confounders were adjusted in the analyses. Sociodemographic data included age, sex, residential place (rural/urban), and education level (illiterate/primary school/high school or above). Lifestyle factors comprised smoking status (never/previously/currently), drinking status (never/previously/currently), dietary diversity scores (low/medium/high), and social activity (yes/no). Health status was composed of body weight (kg), chronic diseases (hypertension, diabetes, heart disease, stroke, pneumonia, and tuberculosis; yes/no), cognitive function [as measured by Mini-Mental State Examination (MMSE) score], and depression status (measured by five items). Detailed definitions of covariates are described in the Appendix methods in Supplementary Material.

Statistical Analysis
Continuous variables were described as mean ± SD and analyzed by Chi-square test, and categorical variables were described as frequency and proportion (%) and tested by ANOVA or Kruskal-Wallis test. Multivariable Cox regression was used to assess the association between water sources and OA incidence. Drinking well water was set as the reference group. A set of Cox regression models were built: Model 1-adjusting age and gender; Model 2-further adjusting residence, education, income, region, and weight; Model 3-further adjusting social activity and physical activity based on model 2; Model 4further adjusting drinking alcohol, smoking, and dietary scores; and Model 5-further adjusting chronic disease, depressive symptoms, and cognitive impairment.
Subgroup analyses were conducted to test multiplicative interactions between water sources and covariates by adding a cross-product term in the full multivariable Cox regression model, namely, sex, age, residence, chronic disease, drinking, smoking, or exercise.
Three sensitivity analyses were performed to address the potential influence of dropout. First, deaths in 2005 (second wave) were excluded to ensure that the participants were pre-mortality. Second, individuals with chronic diseases (e.g., hypertension, diabetes, heart attack, stroke, pneumonia, and tuberculosis) were excluded to minimize the hysteretic effects of chronic diseases on the outcome. Third, multiple imputations based on five replications and Monte Carlo simulations were performed to account for missing data because the covariates were disproportional and randomly missing. The Cox model with shared frailty was used to treat the cluster, and participants with equal value were assumed to have shared frailty across different groups in controlled provinces. The effect within the province group was estimated separately to identify the significance of regional heterogeneity.
All analyses were performed using Stata 16.1 (Stata Corporation, College Station, TX, USA). Statistical significance was considered when the two-side p was <0.05 (two-sided).

Baseline Characteristics of Participants
This study included 12,543 eligible participants at baseline. The regional sample distribution is displayed in Figure 1. The average age and body weights were 86.71 years and 49.13 kg, respectively. A total of 5,489 (43.8%) men and 7,054 (42.6%) women were enrolled. After the restriction to the longest follow-up, the mean survival time of follow-up was 14.56 years. In the cohort, 1,585 suffered from OA. In terms of drinking water sources, 4,372 people drank from a well, 171 drank from the surface water, 400 drank from the spring water, and 7,600 drank from tap water. More than half of participants were illiterate (7,679, 61.2%), living in rural areas (6,707, 53.5%), never drank alcohol (8,368, 66.9%), never smoked (8,125, 65.0%), and had low leisure time in physical activity (7,099, 56.8%). Most of the participants

Subgroup Analysis Between Drinking Water Sources and OA Incidence
A series of subgroup analyses was conducted by using predefined factors (sex, age, living area, chronic disease, smoking, drinking, and physical activity; Table 3). Compared with tap water, drinking mainly from well and spring water showed a positive association with OA incidence that was generally persistent across subgroups. A significant interaction between drinking water sources and age (p < 0.001) was also observed. In terms of drinking mainly from well, a significantly large effect size was observed among female participants, those aged below 80 years, living in rural areas, had no chronic disease, never smoked and drunk, and had no physical activity. In terms of drinking main from spring water, a significantly large effect size was observed among women, those living in rural areas, had chronic diseases or smoking/drinking habits, and participated in physical activity.

Sensitivity Analysis
After imputing missing values in the sensitivity analyses of Cox regressions, drinking from the well and spring water was associated with a high risk of OA (HR for well water:  Table 4). Furthermore, drinking well water might be associated with a relatively strong risk of OA in southern (HR: 1.24, 95% CI: 1.01-1.52) and western regions (HR: 1.49, 95% CI: 1.06-2.11), but the p for interaction was not supported for regional heterogeneity (p > 0.05; Table 5).

DISCUSSION
In this national longitudinal study that includes 12,543 Chinese adults aged 65 years and above, the association between drinking water sources and OA incidence was examined. The primary finding was that drinking tap water was associated with a lower risk of OA compared with drinking well and spring water. For drinking tap water, the HRs of OA were 30% higher for those drinking well water and spring water, regardless of other risk factors. Effect modifications from subgroups were  non-significant except for age, indicating that the associations were robust.
To the authors' knowledge, this study was the first to evaluate the association between drinking water sources and OA. Related works have reported about drinking water and arthritis. Acute diarrheal pathogens are often a high-risk issue for residents' unclean water supplies and gastrointestinal disease. Various potential sequelae follow enteric infections, such as diabetes, myocarditis, and reactive arthritis (16). However, the burden of pathogens for health conditions is often temporary. A followup survey involving 9,500 residents in Finnish measured the complaints or symptoms of participants after extensive backflow that caused household water contamination (17). The outbreak's long-term consequences were minor in terms of the self-reported symptoms or complaints (17). Therefore, the long-term effect of unprocessed drinking water on health change should be considered in chemicals and minerals. One retrospective cohort study reported that participants who lived in communities where the drinking water from private well water with a high nitrate N content had a great risk of intestinal difficulties and OA (8).
Fluoride-and arsenic-contaminated drinking water are positively associated with skeletal fluorosis worldwide, especially in rural areas (18). An increase in groundwater fluoride concentration is a potential risk factor for arthritis in Asian countries (19)(20)(21). Approximately 2.58% of the area in Jinghui irrigation district, China groundwater arsenic level exceeded the limit of 10 µg/L for drinking water (22). Although tap water is the best water source, issues about water delivery should be further examined. For example, the WHO Guidelines for Drinking-water Quality reported that cadmium contamination of drinking water might occur due to the presence of cadmium as an impurity in the zinc of galvanized pipes or cadmium-containing solders in fittings, water heaters, water coolers, and taps (23). Cadmium can increase bone resorption, affect the activity of osteoclasts and calcium absorption, and impair kidney function, thus leading to osteoporosis, OA, and rheumarthritis (24). The present findings were consistent with the abovementioned evidence. In general, tap water was the most beneficial for decreasing OA risk. Except for age adjustment, the interactive effect of drinking water sources and OA was robust. Age is a determining risk factor associated with OA (25,26). The estimated prevalence of OA was 25.03% for lumbar, 21.51% for knee, 20.46% for cervical, and 8.99% for hand in middle-aged and older Chinese adults according to the current meta-analysis (26). Owing to the projected increase in life expectancy, the population of middleaged and elderly OA patients will also continue to grow. The meta-analysis also revealed a trend of increasing prevalence with age before 70 years old and a slight decrease in the oldest ages (26); this finding was comparable with the present results. This work suggests the need for implementing indoor clean water sources and the development of cost-effective interventions to address the adverse health effects of unprocessed drinking water and reduce the increasing prevalence of OA from the pollutant water use of chemicals and minerals in this vulnerable segment of the population in rural areas.
The underlying biological mechanisms of promoting OA incidence due to long-term intake of unprocessed drinking water remain unclear. Some studies on molecular biology attempted to illustrate the association. Arsenic induces senescence in human articular chondrocytes by increasing the activity of βgalactosidase (SA-β-Gal) and protein expression of p16, p53, and p21 (27). In a rat experiment, increased cartilage senescence and abrasion were found in the rats treated by long-term drinking water with arsenic (0.05 and 0.5 ppm) for 36 weeks compared with those in age-matched control rats (27). The phosphorylation of p38 and JNK and the induction of GATA4-NF-κB signaling and senescence-associated secretory phenotype (SASP) were also enhanced in the rat cartilages (27). This signaling pathway has been used as the most specific explanation for the pathological mechanism of OA (28,29). For fluoride, in ex vivo models and animal studies linked PFOS and PFOA to inflammation (30-32), oxidative stress (33,34), and disturbance of calcium homeostasis (35). In particular, PFOA increases inflammation by inducing proinflammatory cytokines (32). By binding to peroxisome proliferator-activated receptor (PPAR)-PPAR-α and γ-PFOS, and PFOA trigger changes in bone metabolism that are related to the progression and onset of OA symptoms (6).
This study shows several strengths. First, this work is the first attempt to analyze the effect of drinking water sources on the OA incidence of the Chinese elderly. The findings provide novel evidence for improving the availability of water, sanitation, and hygiene and the prevention of OA in the elderly. Second, this study was based on CLHLS, a large-scale population-based cohort study. Sufficient statistical power was obtained to infer the casual associations of drinking water sources with OA incidence. Finally, three sensitive analyses were conducted to verify the accuracy and credibility of the association, and the results were consistent with those from the analysis of the overall sample.
This study has several limitations. First, sufficient and valid information on the diagnosed OA cases was not available from the dataset. Recall bias could be generated due to retrospective longitudinal design and self-report data. Second, detailed food data were not accessible, and the proxy variable of the dietary pattern cannot be adopted in the baseline questionnaire. Therefore, the association between dietary patterns and drinking water sources cannot be explored. Third, the proportion of participants drinking spring water and surface water was relatively small. A large sample size should be employed to improve the power of statistical inference. Finally, the doseresponse relationship of chemical substances and OA (e.g., fluorides or nitrate) cannot be assessed. Further evidence from environmental medicine perspective must be obtained to clarify the underlying mechanism between drinking water sources and OA. The conclusion of this research was proposed carefully.

CONCLUSION
Drinking well and spring water was associated with a high risk of OA among the population of age >65 years old in China. Therefore, the association between unprocessed drinking water with a multilevel exposure and OA incidence must be re-examined in low-and middle-income countries. Further longitudinal studies are also needed to confirm the relationship as causal. This work paves the way for an evidence-based intervention for the policymakers to ease the burden of OA related to water pollution for the elderly in such a globally aging population.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://sites.duke. edu/centerforaging/programs/chinese-longitudinal-healthylongevity-survey-clhls/.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the Institutional Review Board, Duke University (Pro00062871), and the Biomedical Ethics Committee, Peking University (IRB00001052-13074). The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
YZ and TW contributed to the conception, analysis, interpretation of data, drafted the report, and received the final version for publication. YL, YX, SS, JY, XC, XW, and XL contributed to analyzing and interpreting the data, commented on the report, revised the manuscript, and approved the final version for submission. All authors have read and agreed to the published version of the manuscript.