Exposure to high altitude is associated with an elevated risk of hip fracture: a retrospective cohort study using data from the CHARLS

Abstract

Background

Previous investigations have suggested that high-altitude residency may increase the risk of bone loss; however, evidence on the impact of high-altitude exposure on the risk of hip fracture is lacking. Notably, hip fractures are the most serious complication of osteoporosis.

Methods

We designed and conducted a retrospective cohort analysis using the China Health and Longitudinal Study data from 2011 to 2018. Participants were divided into two groups based on the altitude of their residence, with a threshold of 1500 m. Propensity score matching was used to balance covariates, including socioeconomic indicators and medical histories. Univariate Cox regression analysis was performed to evaluate the effect size of high-altitude exposure on the risk of hip fractures. To perform subgroup analysis, the participants were stratified based on sex, body mass index, smoking, and drinking.

Results

After matching, 399 participants in the high-altitude group and 3192 in the low-altitude group were enrolled. Cox regression showed that exposure to high altitude increased the risk of hip fracture (hazard ratio [HR]: 1.65, 95% confidence interval [CI]: 1.06–2.57, p = 0.03). Subgroup analysis showed that women (HR: 1.996, 95% CI: 1.142–3.491, p = 0.025) and participants with overweight (HR: 2.703, 95% CI: 1.186–6.165, p = 0.034) were vulnerable to the impact of high-altitude exposure compared with men and participants with normal weight.

Conclusion

Exposure to high altitudes increases the risk of hip fractures, and prevention interventions should receive greater attention in these districts.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12877-025-06492-6.

Introduction

Osteoporosis is an endocrine disorder caused by an imbalance in bone metabolism, leading to decreased bone density. This disorder poses a major threat to well-being and places a great burden on individuals and the healthcare system. Approximately 438,000 all-cause deaths and 16.6 million all-cause disability-adjusted life-years owing to low bone mineral density (BMD) have been reported [1]. BMD loss considerably increases the risk of hip fracture, the most lethal complication of osteoporosis [2], and up to 30% of patients die within 1 year of sustaining a hip fracture [3–5].

Exposure to simulated or real-world hypoxic environments can lead to decreased bone mass and skeletal deterioration [6]. An investigation of mountaineers found that short-term exposure to extreme altitudes can cause a 5% decrease in BMD [7]. Zuo et al. performed an analysis based on a Chinese multi-ethnic cohort (CMEC). They concluded that people living in highland areas had lower bone density than those living in plain areas, with a mean difference of 0.373 standard deviation (SD) [8]. In vitro studies using mouse or rat models also suggest that exposure to hypobaric hypoxia leads to a decrease in BMD, cortical bone thickness, and the number and thickness of the trabeculae [9, 10].

Because highlanders have an elevated risk of osteoporosis, they may also be at higher risk of hip fractures. Considering the high mortality rate associated with hip fractures, evaluating the long-term impact of residing in high-altitude environments on the risk of hip fractures is crucial for informing decision-making regarding potential interventions and preventive measures. However, studies regarding the relationship between high-altitude exposure and hip fractures are lacking, especially data analyses based on large-scale, nationwide, and longitudinal studies. Therefore, we evaluated the causal relationship using the China Health and Longitudinal Study (CHARLS) cohort.

Materials and methods

Study population and dataset

The CHARLS is a prospective national cohort study that enrolled 17,708 participants in 2011, and three waves of follow-up were conducted in 2013, 2015, and 2018. Participants were randomly selected using a probability-proportional-to-size (PPS) technique and a four-stage random sample method. The workgroup selected 150 counties in 28 provinces. Administrative villages in rural areas and neighborhoods in urban areas were the primary sampling units (PSUs). Three PSUs within each county-level unit were selected using PPS sampling. Detailed information on the methodology and cohort profile has been reported previously [11].

We collected data on participants enrolled in the baseline investigation who attended all three follow-up investigations and aged above 60 years old. Those with a history of cancer were excluded because the progression and treatment of cancer affect multiple organs and systems throughout the body; this aspect could have introduced substantial bias into our study. During the data cleaning process, we identified participants with implausible data regarding body weight and height, specifically heights less than 100 cm or weights less than 20 kg. These values were considered erroneous owing to likely data entry mistakes; therefore, participants with an abnormal body mass index (BMI) (BMI < 10 or BMI > 60) in the baseline investigation were excluded.

Altitude data acquisition

The participants’ place of residence was determined by their community ID. The latitude and longitude were acquired through amap api (https://restapi.amap.comv3/geocode), and the local altitude was acquired based on the latitude and longitude using geodata (version 0.5-8) [12] packages and the raster package (version 3.6–23) [13]. Participants were stratified into a low-altitude or high-altitude group based on a criterion of 1500 m, which was used in previous studies [14].

Variable collection

The primary outcome was hip fractures reported by the participants, which was defined by their answer to the question, “Have you fractured your hip since the last interview?” Participants choosing “yes” were considered to have experienced a hip fracture, and the time point the investigation occurred was recorded as the time the event happened.

Individual income, marital status, medical history, sex, BMI, smoking status, alcohol consumption, and age were selected as covariates for propensity score matching (PSM). Individual income was acquired from harmonized CHARLS data and evenly divided into five groups. Educational status was retrieved from the answer to the question: “What is the highest level of education completed?” and re-coded into preschool, primary, secondary, and higher education. Marital status was retrieved from the answer to the question: “What is your marital status?” and re-coded as unmarried, married, widowed, divorced (or long-term separation). History of stroke, cardiovascular dysfunction, falls, hip fractures, chronic lung diseases, and arthritis were retrieved from the answer to the question: “Have you been diagnosed with any of the following by a doctor?” The presence of specific diseases was determined based on whether the participant selected “Yes” to the following conditions: (1) “Chronic lung diseases, such as chronic bronchitis and emphysema (excluding tumors or cancer),” (2) “Heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems,” (3) “Stroke,” and (4) “Arthritis or rheumatism.” BMI was calculated using the following formula: BMI (kg/m^[2]) = body weight (kg)/body height (m)².

Propensity score matching

Individual income, marital status, medical history, sex, BMI, smoking status, alcohol consumption, and age were selected for PSM. A general linear model was used to calculate the distance, and the nearest method was used for matching with a matching ratio of 1:8 (high altitude: low altitude). After matching, a balance test was performed to evaluate the imbalance between the two groups in the matched data. The MatchIt package [15] was used for matching, and the Cobalt package [16] was used for the balanced test and plotting.

Statistical analysis

Categorical variables are shown as counts (percentages, %), and continuous variables are shown as mean ± SD. To compare the differences in baseline data between the two groups, chi-square and t-tests were used for categorical and continuous variables, respectively. Kaplan–Meier survival analysis and Cox regression were used to compare the differences between the two groups regarding fall and hip fracture risks.

Subgroup analysis was performed, and the participants were divided into subgroups according to sex, smoking history, and overweight status. The effects of high-altitude exposure on fall and hip fracture risks in different subgroups were evaluated using Cox regression analysis.

To test the robustness of our results, we performed different sensitivity analyses as follows: (1) We randomly sampled 80% and 90% of all participants, respectively, and replicated the previously described analysis workflow. (2) Using optimal matching as the PSM method and a matching ratio of 1:8, we replicated the previously described analysis workflow. (3) Cox regression was performed with and without adjustments for covariates without matching the participants.

All analyses were performed using R (version 4.3.1). The survival analysis was performed using the survival package (version 3.5-5) [17] and visualized using the Survminer package (version 0.4.9) [18]. The comparison of baseline characteristics and the generation of tables was performed using the gtsummary package (version 1.7.2) [19].

Results

Baseline characteristics before and after matching

A total of 5669 participants were enrolled in the baseline investigation, 399 of whom lived in high-altitude areas and 5270 in low-altitude areas. The participants allocated to the two groups were statistically different in most covariates, except for sex, age, previous hip fractures, menopause status, smocking status and history of stroke (Table S1).

After matching, 399 participants in the high-altitude group and 3192 in the low-altitude group were enrolled for the subsequent analysis (Table 1). The overall age range of the participants was 60 to 95 years, with a mean age of 68.22 and a standard deviation of 6.77 All covariates were statistically indifferent except for BMI and alcohol usage.

Table 1.

Baseline characteristics of matched data

Characteristic	Low altitude N = 3192	High altitude1 N = 399	SMD2	P-value
Sex			0.01	0.92
Male	1,591 (49.84%)	200 (50.13%)
Female	1,601 (50.16%)	199 (49.87%)
Age	68.23 ± 6.83	68.14 ± 6.36	0.01	0.79
BMI3	21.80 ± 3.40	21.44 ± 3.84	0.10	0.011
Education			0.03	0.72
Preschool education	2,130 (66.73%)	274 (68.67%)
Primary education	725 (22.71%)	84 (21.05%)
Secondary education	314 (9.84%)	37 (9.27%)
Higher education	23 (0.72%)	4 (1.00%)
Marital status			−0.05	0.14
Unmarried	30 (0.94%)	1 (0.25%)
Married	2,362 (74.00%)	296 (74.19%)
Widowed	748 (23.43%)	90 (22.56%)
Divorced or separated	52 (1.63%)	12 (3.01%)
Fall history			0.06	0.25
Yes	606 (19.31%)	86 (21.77%)
No	2,532 (80.69%)	309 (78.23%)
Hip fracture history			−0.02	0.73
Yes	47 (1.47%)	5 (1.25%)
No	3,145 (98.53%)	394 (98.75%)
Menopause			−0.01	0.88
Yes	1,461 (91.3%)	181 (91.0%)
No	140 (8.7%)	18 (9.0%)
Smoking			−0.02	0.76
Yes	1,314 (41.17%)	161 (40.35%)
No	1,878 (58.83%)	238 (59.65%)
Alcohol use			−0.02	0.045
Drink more than once a month	710 (22.24%)	79 (19.80%)
Drink but less than once a month	197 (6.17%)	37 (9.27%)
None of these	2,285 (71.59%)	283 (70.93%)
Chronic lung disease history			−0.06	0.25
Yes	348 (10.90%)	36 (9.02%)
No	2,844 (89.10%)	363 (90.98%)
Heart disease history			−0.05	0.39
Yes	229 (7.17%)	24 (6.02%)
No	2,963 (92.83%)	375 (93.98%)
Stroke			0.00	0.97
Yes	73 (2.29%)	9 (2.26%)
No	3,119 (97.71%)	390 (97.74%)
Arthritis or rheumatism history			0.08	0.12
Yes	1,358 (42.54%)	186 (46.62%)
No	1,834 (57.46%)	213 (53.38%)
Personal income			0.07	0.55
Lowest quintile	863 (27.04%)	109 (27.32%)
Second quintile	861 (26.97%)	118 (29.57%)
Third quintile	652 (20.43%)	85 (21.30%)
Fourth quintile	470 (14.72%)	51 (12.78%)
Highest quintile	346 (10.84%)	36 (9.02%)

High altitude was defined as a district with an altitude higher than 1500 m

SMD Standard mean difference

BMI Body mass index

We examined the balancing of matched data. After matching, the standardized mean differences of all covariates were within a 0.1 threshold (Supplementary Fig. 1, Table 1). The distance distribution in the two groups was similar and had substantial overlap (Supplementary Fig. 2). These results demonstrate that the matched dataset was balanced and met the requirements for further analysis.

Effect of high-altitude exposure on fall and hip fractures

Over a follow-up period of 7 years, 91 (4.6%) and 23 (5.8%) participants in the low-altitude and high-altitude groups experienced hip fractures, (Table 2). Kaplan–Meier curves showed an increased incidence hip fractures (p = 0.025) in the high-altitude group (Fig. 1). The Cox regression results showed that exposure to high altitudes significantly increased the risk of hip fractures. The hazard ratio (HR) for hip fracture was 1.65 (95% confidence interval [CI]: 1.06–2.57, p = 0.03), (Table 3).

Table 2.

Frequency of outcomes after follow-up

Outcomes	Low altitude N = 3192	High altitude1 N = 399	p-value
Hip fracture	129 (4.0%)	23 (5.8%)	0.11

High altitude was defined as a district with an altitude higher than 1500 m

Fig. 1.

Survival curves of high-altitude exposure and fractures

Table 3.

Relationship of high-altitude exposure and outcomes (N = 5,478)

Characteristic	HR1	95% CI2	p-value
High-altitude exposure	1.65	1.06–2.57	0.027

HR Hazard ratio

CI Confidence interval

Subgroup analysis

Subgroup analysis showed that exposure to high altitude affects women (HR: 1.997, 95% CI: 1.142–3.491, p = 0.025), non-smokers (HR: 2.142, 95% CI: 1.280–3.584, p = 0.007), non-alcohol users (HR: 1.759, 95% CI: 1.061–2.916, p = 0.040), and individuals with overweight (HR: 2.703, 95% CI: 1.186–6.165, p = 0.034) had an elevated risk of hip fractures when living in the highlands (Fig. 2).

Fig. 2.

Subgroup analysis results

Sensitivity analysis

Sensitivity analysis using different sampled datasets, matching methods, and analysis methods showed similar results to those previously described (Supplementary Fig. 3–6, Table S2–S5), suggesting the results are robust and do not rely on certain data or analysis methods.

Discussion

In this study, we found that exposure to high altitudes increased the risk of hip fractures and falls. The subgroup analysis revealed that women, patients with overweight, non-smokers, and non-alcohol users were vulnerable to this effect.

Temporarily or permanently residing in highland areas has been shown to negatively impact BMD. Basu et al. conducted two studies on participants who stayed in highland areas for 4–12 months [20, 21]. They found a 2% decrease in the ultrasonic speed of sound, an index for approximating skeletal Z-score, after a temporary residency in a highland area. Zuo et al. analyzed data from the CMEC study, which included 73,974 participants. Their results showed that after adjusting for smoking, drinking status, the intakes of vitamin D, calcium, and dairy, residing in an area with an altitude higher than 1500 m resulted in a 0.373 times SD decrease in the quantitative ultrasound index. However, owing to limited access to devices, none of these studies used dual-energy X-ray absorptiometry (DXA), the gold standard and most widely used method for evaluating BMD and diagnosing osteoporosis according to guidelines [22, 23]. Instead, ultrasonic devices were used because of their relatively low cost and portability. Considering the weak correlation between ultrasonically assessed BMD and DXA measurements, DXA BMD or fracture risk cannot be predicted [24, 25]. Evidence is still insufficient for assessing the consequences of bone loss due to high-altitude exposure. Nevertheless, many studies exist in animal research suggesting hypoxia decreases BMD and affects biomechanical performance [9, 26, 27]. Our findings address this knowledge gap in population studies, confirming that exposure to high-altitude areas may lead to bone loss, as revealed in previous studies, and increase the risk of hip fractures, the most serious consequence of osteoporosis [28].

Our subgroup analysis indicated that women and participants with overweight were vulnerable to the impact of high-altitude exposure on the risk of hip fractures. Individuals with overweight have been shown to experience more severe symptoms in high-altitude areas with lower oxygen saturation. Moreover, excess weight leads to various pathophysiologic disorders [29], and overweight is also a risk factor for osteoporosis [29, 30]. The interaction between more severe high-altitude sickness and the increased risk of hip fractures due to overweight may lead to a more pronounced effect of high-altitude hypoxia on hip fracture risk in these individuals. The more pronounced effect of high-altitude exposure in women is likely due to similar reasons [31]. Additionally, our results showed that people using tobacco or alcohol were not vulnerable to the effects of high-altitude exposure. We advise a cautious interpretation of these results and propose two potential explanations. First, the limited sample size of the subgroup might have resulted in inadequate statistical power, leading to false-negative results. Second, the substantial impact of smoking and alcohol consumption on the risk of hip fractures might have obscured the effect of hypoxia on fracture risk within the subgroup. Further research is needed to explore these possibilities and provide a more comprehensive explanation.

A previous investigation showed that approximately 40% of patients who experienced hip fractures had one or more undiagnosed vertebral fractures [32]. Considering that patients experience a rapid decline in BMD and an increased risk of refracture [33–35], it is reasonable to infer that undiagnosed vertebral fractures increase the risk of hip fractures and play an important role in their incidence. We hypothesized that the elevated fracture risk in highland residents may also be caused by undiagnosed vertebral fractures because these individuals have more severe bone loss. Implementing opportunistic screening of osteoporotic fractures based on X-ray images or computed tomography may help identify patients at high risk of refracture. Furthermore, initiating interventions earlier may be important and helpful for people living at high altitudes to prevent hip fractures.

To the best of our knowledge, this study is the first to report an elevated risk of hip fractures in people living in highland areas based on a large population and longitudinal database. However, this study has several limitations. Owing to the design of the CHARLS research questionnaire, we lacked data regarding the medication usage of the participants. Consequently, we could not check medical records, such as the use of corticosteroids, vitamin D, and calcium supplements, which could have impacted bone mass in our analysis. Because of privacy concerns, we could only analyze participants’ residencies at the district level, which limited the accuracy of our high-altitude exposure information. In addition, the CHARLS workgroup did not perform any BMD measurements; thus, the analyses were based on self-reported results. In the future, it will be essential to establish specific cohorts for measuring bone density and conduct a longitudinal follow-up of hip fracture incidence to further substantiate the results of this study.

Conclusions

Using the CHARLS dataset, we found that exposure to high altitudes elevated the risk of fractures. This indicates that preventive interventions in high-altitude districts should receive greater attention.

Authors’ contributions

SY Wang: conceptualization, data curation, formal analysis, methodology, visualization, writing original draft, review and editing; SY Zhu: data curation; FY Zhang, YH Guo, Y Zhong, DD Hao: supervising, review and editing; CH Zhang: funding acquisition, supervision, review and editing; YH Wu: funding acquisition, project administration, resources, supervision, review and editing.

Funding

This study was supported by the Tibet Autonomous Region Science and Technology Program, the central government guides the local science and technology development project under Grant Nos. XZ202202YD0011C, XZ202102YD0026C, and Tibet Autonomous Region Science and Technology Plan, Joint Funding Project, XZ202303ZY0011G, and XZ202301ZY0049G.

Data availability

The datasets analyzed during the current study are available in CHARLS program repository, (https://charls.charlsdata.com/)

Declarations

Ethics approval and consent to participate

The CHARLS survey project was approved by the Biomedical Ethics Committee of Peking University, and all participants provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.