Associations of body mass index and diabetes with hip fracture risk: a nationwide cohort study

We retrospectively reviewed quadrennial data from the National Health Interview Survey in Taiwan (NHIST) for the years 2001, 2005, and 2009. The survey participants’ height and weight data were used to calculate the baseline BMI. We excluded pregnant women and participants younger than 40 years. We linked the NHIST participants’ individual data with the National Health Insurance Research Database (NHIRD), which includes nationwide data on all citizens in Taiwan. We extended the washout period to January 1st, 2000 for our participants in this study. All participants who had been diagnosed with hip fracture before NHIST survey were excluded from this study to make sure the temporal relationship between BMI/diabetes and hip fracture. We included a total of 22,048 participants and monitored hip fracture incidence for the period of 2000 to 2013.

The NHIST was conducted nationally and quadrennially by the Taiwan Health Promotion Administration. The information available from surveys include personal information, personal health status, knowledge about disease prevention, utilization of medical services, personal health behaviour, self-rated health status, and work and economic status, among others. The survey results provide a reference for developing and implementing healthcare policies in Taiwan [18].

This study was reviewed and approved by the research ethics committees of China Medical University (IRB No.: CMUH 103-REC3–109). We deleted all personal identification from the data analysed in this study to protect the patients’ personal identities. Taiwan’s National Health Insurance program was launched in March 1995, and as of 2013, the nationwide coverage rate was 99.68% [19]. This compulsory public health insurance program provides comprehensive information such as demographic data and data on all medical services, including prescription drugs, surgical treatments for outpatients, emergency care, and inpatient care. The NHIRD includes medical information on all citizens covered by insurance, including treatments for diabetes, hip fractures, and other conditions [20, 21]. The comprehensiveness and accuracy of the NHIRD have been confirmed by the Ministry of Health and Welfare, and the database has been used in numerous studies [22, 23].

The variables examined were BMI, personal basic characteristics (sex, age), environmental factors (urban or rural residential areas), socio-economic status (monthly salary), health status (Charlson Comorbidity Index [CCI] and diabetes complication severity index [DCSI]), health behaviour (weekly energy expenditure through exercise), and diabetes status. The WHO has developed a classification of BMI for international use, but the index for overweight Asian adults is lower than the world average. Thus, many Asian countries have developed their own criteria for overweight and obesity. We used the BMI classification criteria of the Taiwan Health Promotion Administration and divided the participants into five categories: low BMI (BMI < 18.5), normal BMI (18.5 ≦ BMI < 24), overweight (24 ≦ BMI < 27), mild obesity (27 ≦ BMI < 30), and moderate obesity (BMI ≧ 30) [24].

In healthcare, diagnosis codes are used as a tool to group and identify diseases, disorders, symptoms, poisonings, adverse effects of drugs and chemicals, injuries, and other reasons for patient encounters. In the NHIRD, diagnosis codes are collected using the ICD-9-CM code (The International Classification of Diseases, Ninth Revision, Clinical Modification). We defined diabetes patients as those who received a diagnosis of diabetes (ICD-9-CM: 250) and at least three outpatient treatments or one hospitalization during the year of the interview survey or within 365 days before or after the survey [25]. We excluded patients with type 1 diabetes, gestational diabetes, neonatal diabetes, or impaired glucose tolerance (ICD-9-CM: 6488, 7751, 7902, 6480). We defined hip fracture as a diagnosis of femoral neck fracture, intertrochanteric fracture, or subtrochanteric fracture (ICD-9-CM: 820.XX) and having received one of the following surgical treatments: partial hip replacement (ICD-9-CM: 81.52), open reduction of fracture with internal fixation of the femur (ICD-9-CM: 79.35), or closed reduction of fracture with internal fixation of the femur (ICD-9-CM: 79.15).

We divided residential areas into seven levels from most urban and to least urban [26]. We calculated the severity of comorbidities based on the CCI revised by Deyo et al. [27] and divided them into groups with scores of 0, 1–3, and  ≧ 4. We calculated the DCSI based on seven types of diabetes complications (retinopathy, nephropathy, neuropathy, cerebrovascular, cardiovascular, peripheral vascular disease, and metabolic) as classified by Young et al [28] and used different weight scores (0 or  ≧ 1) to represent different severities.

In terms of health behaviour, we calculated the weekly energy expenditure through exercise according to the method proposed by Wen et al. [29] using the NHIST. Each type of exercise corresponds to a different Metabolic Equivalent of Task (MET, a unit of exercise intensity) according to the breathing status during exercise. One MET is defined as the oxygen uptake in ml/kg/min when sitting quietly (3.5 ml/kg/min). The weekly energy expended in exercise is calculated as follows: MET * frequency of exercises over the past 2 weeks (times) * each exercise duration (hours) * body weight (kg) * 7/14. We used the MET to collect and validate the weekly energy expenditure in kilocalories (kcal) for specific exercises, based on which participants were divided into three groups according to the expenditure per week: no exercise, < 500 kcal/week, and  ≧ 500 kcal/week.

For the inferential statistical analysis, a Cox proportional hazard model was used to investigate the effects of BMI and diabetes on hip fracture risk after controlling for variables such as personal basic characteristics, environmental factors, social and economic status, health status, and health behaviour. Further, in order to examine whether diabetes has a same effect on risk of hip fracture in those with different BMI, we also examined the interaction relationship between diabetes status and level of BMI on the risk of hip fracture. A stratified analysis was further performed to investigate the effects of diabetes on hip fracture risk in those with different BMI if there was an interaction relationship between diabetes status and level of BMI. Finally, the Cox proportional hazard model was used to investigate the effects of BMI on hip fracture incidence in participants with diabetes.

A total of 22,048 subjects were eligible for inclusion in this study (Table 1), of which 3508 had diabetes and 315 had hip fractures. Among the different level of BMI groups, we found that the higher of the BMI, the higher prevalence of diabetes. When the participants had moderate obesity (BMI ≧ 30), diabetes risk was as high as 33.33%. There was significant difference between diabetes and non-diabetes patients in risk of hip fracture (p < 0.05). Additionally, participants with low BMI (BMI < 18.5) showed a higher hip fracture rate (3.56%) than other BMI subgroups. There were significant differences between participants with diabetes and those without diabetes in BMI, hip fracture, sex, age, urbanization of residence area, monthly salary, CCI, DCSI and weekly energy expenditure through exercise (P < 0 .05). In Table 2, there were significant differences in hip fracture incidence between the participants in terms of variables, including BMI, diabetes status, age, monthly salary, CCI, and DCSI (p < 0.05).

We also used the Cox proportional hazard model to analyze the effects of BMI and diabetes on hip fracture risk. The results of four models of are shown in Table 3. The first model is the univariate analysis of diabetes and hip fracture with unadjusted results, the second is for diabetes without BMI, the third is for BMI without the diabetes variable, and the final model is for both variables together. In the final model, we found that hip fracture risk in diabetes patients was 1.64 times the risk in non-diabetes patients (95% CI: 1.30–2.15, p < 0.05). Patients with low BMI showed a higher hip fracture risk (Adjusted hazards ratio [Adj. HR]: 1.75, 95% CI: 1.17–2.61) than those with normal BMI (reference). Additionally, patients who were overweight, mildly obese, or moderately obese had lower hip fracture risk than patients with normal BMI, but the differences were not statistically significant (p > 0.05). Hip fracture risk in female patients was 1.29 times that in male patients (95% CI: 1.03–1.62, p = 0.027). Compared with a reference group (aged 40–49 years), older patients showed a higher hip fracture risk (p < 0.05): among subjects  ≧ 80 years old, hip fracture risk was as high as 52.16 times the baseline risk. Separately, participants with higher CCI scores showed a higher hip fracture risk than the reference group (CCI = 0), and hip fracture risk in those who expended  ≧ 500 kcal/week in exercise was 0.67 times lower than in those who did not exercise (95% CI: 0.50–0.89).

We used stratified analysis to examine the relative risk of hip fracture between diabetes and non-diabetes patients at different level of BMI (Table 4). After relevant variables were controlled in Cox proportional hazard model, hip fracture risk in diabetes patients was greater than that in non-diabetes patients regardless of BMI. Among the five BMI groups, compared with non-diabetes patients, only diabetes patients with a normal BMI (18.5 ≦ BMI < 24) showed a statistically significant difference in hip fracture risk (Adj. HR: 2.13, 95% CI: 1.48–3.06, P < 0.05). It means that diabetes increases risk of hip fracture but the magnitude of risk varies with the BMI level.

A Cox proportional hazard model was used to analyze diabetes patients (Table 5 & Fig. 1). Compared with a reference group (normal BMI, 18.5 ≦ BMI < 24), those with overweight (24 ≦ BMI < 27) or obesity (BMI ≧ 27) showed a lower hip fracture risk (Adj. HR: 0.49 vs. 0.42, p < 0.05). Compared with the reference group (aged 40–49 years), older patients showed a higher hip fracture risk, but statistically significant differences were only observed in patients  ≧ 60 years old (p < 0.05). Among diabetes patients, those with higher CCI or DCSI scores were associated with a higher hip fracture risk. As the weekly energy expended in exercise increased in diabetes patients, hip fracture risk decreased compared with diabetes patients without exercise. In particular, when the weekly energy expenditure was  ≧ 500 kcal/week, hip fracture risk in diabetes patients was significantly decreased to 0.54 times (95% CI: 0.31–0.94, p < 0.05).

There were several limitations to our analyses. Data from the NHIRD were used for the analysis, so not all health behaviours and other factors were included in the analysis, such as eating habits, body composition/muscle mass, muscle function/sarcopenia, and history of falls history. Moreover, the duration of diabetes in all subjects and their blood glucose control status were not known.