Introduction

Type 2 diabetes mellitus (T2D) is one of the leading problematic non-communicable diseases worldwide. It is characterized by insulin resistance and chronic hyperglycemia, leading to multiple deleterious chronic microvascular and macrovascular complications. In addition to chronic vascular complications, T2D causes adverse effects on the skeleton. T2D drives a low bone turnover state1,2 deteriorates bone microarchitecture3,4 and increases the risk of fragility fractures5,6,7,8,9,10 leading to increased disability and mortality among patients with fractures. However, the fragility fractures related to T2D were underestimated by measures commonly used in clinical practice, including bone mineral density (BMD)7,8,10 and the fracture risk assessment tool named FRAX®11. T2D was documented for increasing risks of fracture despite preserved BMD7,8,10. Furthermore, FRAX® did not precisely estimate a 10-year probability of fractures in patients with T2D11 even with multiple proposed adjustment methods12. Therefore, the additional tools for predicting fractures in patients with T2D remain to be further elucidated.

Fibroblast growth factor 21 (FGF21) is a 22-kDa protein predominantly produced and released from the liver to function as a hormone by interacting with its specific receptor and coreceptor, named fibroblast growth factor receptor 1 (FGFR1) and β-Klotho (KLB), in multiple tissues. FGF21 enhances glucose utilization, insulin sensitivity, and energy expenditure in adipose tissue, as well as lessens sugar consumption by stimulating hypothalamic ventromedial neurons, influencing glucose homeostasis and body weight reduction. The circulating level FGF21 elevated in patients with T2D13,14,15 while the level of either FGFR115 or KLB14 expression suppressed in these patients, suggesting FGF21 resistance and compensatory secretion of FGF21 in patients with T2D. The circulating level of FGF21 was demonstrated as a biomarker of diabetes progression16as well as a biomarker for predicting diabetic kidney disease progression17,18 and cardiovascular diseases19,20,21 in patients with T2D.

FGF21 was shown to influence bone turnover in both primates22 and humans23. Even though molecular mechanisms of FGF21 in bone homeostasis have not been fully elucidated, multiple mechanisms have been demonstrated, including receptor activator of nuclear factor-κB ligand (RANKL) and insulin-like growth factor binding protein 1 (IGFBP1) induction that promoted osteoclast development, peroxisome proliferator-activated receptor gamma (PPARγ) activation that inhibited osteoblast development, and hepatocyte growth factor (HGF) upregulation that prevented osteoblast precursor cell apoptosis24,25. In humans, Talukdar and colleagues23 demonstrated that a pharmacological dose of long-acting FGF21 analog elevated bone resorption markers but suppressed bone formation markers in a dose-dependent manner, suggesting the role of FGF21 in controlling bone turnover. Consistent with data in humans, Andersen and colleagues22 demonstrated that recombinant FGF21 increased a bone resorption marker without changing BMD in rhesus macaque monkeys with diet-induced obesity. Even though the pharmacological dose of FGF21 was reported to enhance bone resorption markers and suppress bone formation markers in a dose-dependent manner in individuals with T2D and obesity23, the effects of the physiological enhancement of FGF21 on bone homeostasis and fracture outcomes in individuals with T2D have never been explored. Furthermore, the circulating level of FGF21 was documented as a biomarker for disease progression and vascular complications of diabetes; it is valuable to determine whether FGF21 is also a useful biomarker for predicting fractures in T2D. Therefore, this study aimed to determine whether FGF21 is associated with fragility fractures in individuals with T2D and to develop a potentially predictive model that included FGF21 as one of the clinical risk factors to predict fractures in T2D.

Results

Increased incidence of fragility fractures in individuals with type 2 diabetes mellitus

This prospective study enrolled 371 participants, of whom 203 (64.0%) participants had preexisting T2D (Table 1). Individuals with T2D were older (65.0 ± 8.6 years vs. 62.8 ± 11.1, P = 0.042) and had more prevalent obesity (65.0% vs. 50.0%, P = 0.003), central obesity (85.7% vs. 62.5%, P < 0.001), and metabolic syndrome (98.5% vs. 41.7%, P < 0.001) compared to those without diabetes (Table 1). In addition, individuals with T2D had higher systolic blood pressure (140.3 ± 19.6 mmHg vs. 131.2 ± 15.4 mmHg, P < 0.001) but lower HDL-C (48.3 ± 15.4 mg/dL vs. 52.6 ± 14.6 mg/dL, P = 0.008), and LDL-C (95.1 ± 36.8 mg/dL vs. 104.4 ± 38.9 mg/dL, P = 0.021) with higher rate of statin use (Table 1). Enrolled individuals with T2D had fair glycemic control with a mean FPG of 138.4 ± 56.8 mg/dL and a mean HbA1c of 7.6 ± 1.7%.

The ten-year probability of hip fracture (FRAX-H) and major osteoporotic fracture (FRAX-M) were estimated using the fracture assessment tool, FRAX®, calculated based on BMI. The probability of hip (BMI-based FRAX-H) and major osteoporotic fractures (BMI-based FRAX-M) derived from FRAX® were comparable in both groups of participants (Table 1) and classified as low risk of fractures. The prevalence of previous fragility fractures was also comparable between groups with T2D and non-diabetes. The timed up and go test (TUG) test was performed in 95 (46.8%) participants with T2D and 137 (81.5%) participants without diabetes. Individuals with T2D demonstrated longer time required to complete the test (13.2 ± 3.7 s vs. 11.5 ± 5.1 s, P = 0.003) and had more prevalent of slow TUG (60% vs. 20%, P < 0.001) compared to those without diabetes (Table 1), suggesting poorer balance or mobility in individuals with T2D.

Table 1 Baseline characteristics of all participants.
Full size table

The incidence of fragility fractures during the 10-year follow-up period was shown in Table 2, and the time to first fracture was illustrated as Aalen-Johansen plots (Fig. 1a and b). Individuals with T2D had significantly higher incidences of any fractures (10.3% vs. 3.6%, P = 0.012) and major fractures (9.9% vs. 2.4%, P = 0.003) compared to those without diabetes (Table 2). As shown in Fig. 1, individuals with T2D had higher risk of major fractures by 3.7 times [HR 3.65 (95% CI 1.24–10.71), P = 0.018] and tended to have higher risk of any fractures by 2.50 times [HR 2.50 (95% CI 0.99–6.33), P = 0.053].

Table 2 Ten years fracture outcomes in all participants.
Full size table
Fig. 1
figure 1

Aalen-Johansen estimated the cumulative incidence risk of the time-to-first fractures for 10 years followed-up. Panel a showed the time until the first occurrence of any fractures. Individuals with type 2 diabetes mellitus (T2D) tended to have higher probability for the first occurrence of any fractures [SHR 2.50 (95%CI 0.99–6.33, P = 0.053]. Panel b showed the time until the first occurrence of major fractures. Individuals with T2D had higher probability for the first occurrence of major fractures [SHR 3.65 (95%CI 1.24–10.71), P = 0.018]

Full size image

FGF21 as an essential predictor of fragility fractures in individuals with type 2 diabetes mellitus

Further analyses were conducted in diabetic subgroups to characterize associating factors specifically for fractures in the diabetic population. We categorized individuals with T2D into two groups by the occurrence of any fractures after enrollment. Most baseline characteristics of individuals in both groups were comparable, except for gender, preexisting cardiovascular disease, β-blocker use, and the levels of FGF21 and AST (Table 3). Factors previously documented as associating factors for fragility fractures in diabetes, including age26,27,28,29,30FRAX-H12,28FRAX-M12,28previous fracture9,28,29HbA1c31,32thiazolidinedione33,34,35sulfonylureas33and insulin28,30,32,33,34,35were similar between the two groups. TUG, a previously reported predictor for fractures in the elderly36,37,38was also comparable in both groups. Interestingly, individuals with incident fractures had significantly lower levels of FGF21 [213.6 (164.7-276.9) pg/mL vs. 276.4 (182.7-485.5) pg/mL, P = 0.028] compared to those without fractures (Table 3). Univariable and multivariable regression analyses demonstrated that FGF21 was an independent risk factor for incident fractures (Table 4). Every 10 pg/mL decline in FGF21 was associated with an increased risk of incident fracture by 6% (AOR 0.94, 95% CI 0.90–0.99). FGF21 was demonstrated as an essential predictive factor for incident fracture. As shown in Fig. 2, the model that included FGF21 had higher performance for predicting the occurrence of fractures than that without FGF21 [AuROC 0.84 (95% CI 0.77–0.91) vs. AuROC 0.76 (95% CI 0.65–0.86), P = 0.023).

Table 3 Baseline characteristics of individuals with type 2 diabetes mellitus classified by fracture outcomes.
Full size table
Table 4 Univariate analysis and multivariate logistic regression analysis in individuals with type 2 diabetes mellitus.
Full size table
Fig. 2
figure 2

Area under receiver operating characteristic (AuROC) curve for the prediction of fragility fractures in individuals with type 2 diabetes mellitus (T2D). The AuROC of two predictive models for fragility fracture in T2D were compared. The predictive model which was composed of age, gender, BMI-based FRAX-M and AST had AuROC of 0.75. (95% CI 0.64–0.86). With the combination of FGF21 into that predictive model, the AuROC significantly increased from 0.75 to 0.84 (95% CI 0.77–0.91) (AuROC 0.84 VS 0.75, P = 0.023).

Full size image

Predictive model and predictive risk score for fragility fractures in individuals with type 2 diabetes mellitus

Univariable and multivariable logistic regression analyses demonstrated four independent risk factors for incident fractures in T2D, including age, gender, BMI-based FRAX-M, and FGF21 (Table 4). Age, female gender, and FRAX-M calculated based on BMI were independent predictors with the AOR of 1.13 (95% CI 1.04–1.22, P = 0.005), 5.06 (95% CI 1.41–18.12, P = 0.013), and 0.77 (95% CI 0.60–0.99, P = 0.042), respectively. Even though age and gender are parameters used to estimate BMI-based FRAX-M, both showed no collinearity with BMI-based FRAX-M in our regression model. The predictive model constructed using these four independent predictors and AST demonstrated good performance for predicting incident fractures with an AuROC of 0.84, a sensitivity of 84.2%, a specificity of 74.5%, and an accuracy of 75.6%. AST was required in the predictive model to maintain its performance; specificity and accuracy were dropped down to 64.5% and 68.2% if AST was omitted.

For clinical application of this predictive model, a predictive risk score was created from the β-coefficient of each predictive factor identified in multivariable logistic regression analysis. The transformed score ranged from − 6 to 36 (Table 5). Total score higher than 103 demonstrated acceptable performance to classify patients as high risk for incident fracture with an AuROC of 0.79, a sensitivity of 84.2%, a specificity of 75.5%, and an accuracy of 75.6%.

Table 5 Predictive risk score for fragility fractures in type 2 diabetes mellitus.
Full size table

Discussion

This study demonstrated 10.3% of any fractures and 9.9% of major fractures in individuals with T2D over a 10-year follow-up period. We also illustrated a significantly increased risk of major fractures by 3.7 times and tended toward higher risk of any fractures by 2.5 times in individuals with T2D compared to those without diabetes. FGF21 was a novel independent risk factor for fractures identified in this study; every 10 pg/mL decline in FGF21 was associated with an increased risk of incident fragility fracture by 6%. In addition to FGF21, this study demonstrated three other independent risk factors for fractures in T2D, including age, gender, and BMI-based FRAX-M, which were previously described as predictors for fractures in T2D. A good performance predictive model and a predictive risk score were constructed by the combination of these four independent risk factors and AST. FGF21 significantly enhanced the performance of the predictive model that is composed of three previously described predictors.

An increased risk of fragility fractures was well established in individuals with T2D; multiple meta-analysis studies demonstrated a significantly increased risk of overall fragility fractures ranging from 5 to 70% in individuals with T2D30,39,40. In this study, we demonstrated a marked increase in any fractures and major fractures in individuals with T2D by 2.5 and 3.7 times compared to individuals without diabetes. Our study demonstrated a much higher incidence of fractures than previously described risk in those meta-analysis studies. However, the incidence of fractures in our study is consistent with the study performed in the Asian population41. Ha and colleagues41 conducted a retrospective longitudinal study using data from the Korean National Health Insurance Service dataset of preventive health checkups, which included more than 500,000 individuals with T2D. An incident fracture rate of 12.4% was demonstrated during a 7-year follow-up period. Participants included in their study seem to have similar baseline characteristics to our study, individuals with a mean age of 60.6 ± 9.9 years, which were approximately 50% female, 50% obese, and 70% centrally obese.

This study aimed to explore predictors for fragility fractures in T2D; therefore, further univariable and multivariable logistic regression analyses were conducted in patients with T2D by categorizing these individuals into two subgroups by incident fractures. Multiple factors previously documented as predictors for fragility fractures in T2D were comparable between the two groups, including previous fractures9,28,29 FRAX-H12,28HbA1c31,32thiazolidinedione33,34,35 sulfonylureas33and insulin28,30,32,33,34,35. Even though slow TUG was more prevalent in individuals with T2D compared to those without diabetes, slow TUG was comparable between T2D subgroups, either with or without incident fractures. Therefore, TUG was not an independent risk factor for fractures among individuals with T2D in this study. Nevertheless, the higher prevalence of slow TUG in individuals with T2D shown in our study may partly contribute to higher incident fractures in individuals with T2D compared to those without diabetes since the TUG was demonstrated as a predictor for fractures in the elderly36,37,38. This study demonstrated four independent risk factors for incident fractures among individuals with T2D, including age, gender, BMI-based FRAX-M, and FGF21. Consistent with the previous reports, older age26,27,28,29,30 and female gender26,28,29 were documented as independent risk factors of fractures in individuals with T2D. FRAX-M was previously demonstrated as a predictor for fractures in T2D; a higher FRAX-M led to a higher risk of fractures12,28. In contrast to the previous reports, the higher BMI-based FRAX-M was demonstrated as a protective factor for fracture in our study. FRAX-M was completely estimated based on BMI with the addition of rheumatoid arthritis as a proxy for the effects of T2D in our study, while FRAX-M was estimated based on BMD by Leslie and colleagues12 and was calculated based on either BMD or BMI by Kong and colleagues28. The difference in performing the FRAX-M calculation partly contributed to the differences between ours and those previous reports. A higher BMI would lower the number of FRAX-M, so obesity would affect the fracture risk calculated by BMI-based FRAX-M. Premaor and colleagues42 previously demonstrated that 10-year probability of hip and major osteoporotic fractures estimated by FRAX® based on BMI were significant lower in obese individuals than those with non-obese. Since more than half of our enrolled individuals with T2D were classified as obese, especially those with incident fractures, the BMI-based FRAX-M tended to be lower in T2D with incident fractures. This phenomenon may provide clinicians guidance to avoid using only BMI-based FRAX to predict fracture risk in T2D with obesity since it could mislead the fracture risk and the treatment strategies. However, BMI-based FRAX-M could be used to predict the risk of fracture in T2D with obesity when combined with other clinical risk factors and biochemical profiles, as demonstrated in this study.

In the present study, FGF21 was demonstrated for the first time as an independent risk factor for fractures in individuals with T2D. Every 10 pg/mL decline in FGF21 was associated with an increased risk of incident fragility fracture by 6%. FGF21 was elevated in individuals with T2D13,14,15while the expression of its specific receptor and coreceptor, FGFR115 and KLB14was suppressed, suggesting FGF21 resistance and compensatory secretion of FGF21 in individuals with T2D. The effects of this compensatory elevation of FGF21 may be critical to maintaining bone integrity in T2D; therefore, the failure to have this phenomenon entails the risk of fractures in T2D. To date, no studies directly explored the skeletal effects of compensatory elevation of FGF21 in individuals with T2D. Talukdar and colleagues23 determined the skeletal effects of FGF21 by using a pharmacological dose of FGF21. They demonstrated that long-acting FGF21 analog suppressed bone formation markers while elevated bone resorption markers in a dose-dependent manner. In addition, Andersen and colleagues22 demonstrated that recombinant FGF21 increased a bone resorption marker in rhesus macaque monkeys with diet-induced obesity. Since FGF21 was shown to influence bone turnover in both primates and humans, a failure in the compensatory secretion of FGF21 may partly link to the state of low bone turnover and increased fracture risk seen in T2D. This low bone turnover state was shown to relate to impaired bone quality, entailing increased fracture risk43 despite the preserved BMD in T2D. Our results were contrasted with the findings shown in non-primate T2D models. Li and colleagues44 demonstrated neutral effects of recombinant FGF21 on bone turnover markers, BMD, and bone microarchitecture in diet-induced obese mice, while Charoenphandhu and colleagues45 showed the deterioration of skeletal microarchitecture and decreased bone formation in diet-induced obese rats. It is possibly due to the differential functions of FGF21 between species46. Even though AST was not an independent risk factor demonstrated in our study, AST was essential to maintain the performance of our predictive model. FGF21 is mainly synthesized in the liver and is increased in response to metabolic stress13,14,15 or liver injury47. Several studies demonstrated a positive correlation between the levels of serum AST, a liver enzyme, and FGF2148,49. Cantero and colleagues48 demonstrated a positive correlation between serum AST and FGF21 levels in individuals with obesity. Consistent to individuals with obesity, Xiao and colleagues49 demonstrated a positive correlation between serum AST and FGF21 levels in individuals with hyperthyroidism. Since AST and FGF21 levels are correlated with each other, the requirement of AST in our predictive model may reflect an essential role of FGF21 as a predictive factor in our newly constructed model.

This study had multiple strength points. First, this study identified FGF21 as a novel factor for predicting fragility fractures in individuals with T2D. Compared to previous studies that showed controversial data of FGF21 on surrogate markers of bone strength, this study was straightforward in demonstrating the predictive ability of FGF21 on fractures, a hard outcome of bone strength. Second, this study was the first to construct a good performance predictive model for predicting fragility fractures in T2D by using FGF21, AST, and three other simple clinical risk factors, including age, gender, and BMI-based FRAX-M. Third, this study was the first to construct a predictive risk score and propose its cut-off point to classify T2D at risk for fragility fractures. However, there were some limitations to address. First, this study was a 10-year prospective cohort study that followed participants in clinics only for the first 5 years. After 5 years, participants were followed by telephone interviewing and medical record reviewing. Therefore, asymptomatic fractures and minor fractures that did not require medical care may not be completely retrieved during the last 5 years of the cohort, leading to an underestimation of incident fragility fractures. Nevertheless, our newly constructed predictive model and risk score would be useful to predict the occurrence of fractures that need medical care since these types of fractures were expected to be completely retrieved. Second, this study was a single-center study that focused on individuals with high cardiovascular risk. Therefore, our newly constructed predictive model and predictive risk score were derived from this specific subgroup of T2D and were not optimized for individuals with different backgrounds. External validation in individuals with different degrees of cardiovascular risk is mandatory to evaluate the generalizability of our predictive model and risk score in individuals with T2D. Third, some previously described contributing factors for fractures were not included in our study, such as falling, T2D duration, and vitamin D status.

In conclusion, this study demonstrated an increased risk of any fractures and major fractures in individuals with T2D. This study identified FGF21 as a novel independent risk factor of incident fragility fractures in T2D. Furthermore, a predictive model was constructed using AST and four independent risk factors identified in this study, including FGF21, age, gender, and BMI-based FRAX-M. For practical use in clinics, this study proposed a predictive risk score and its cut-off point to classify individuals with T2D at a higher risk for fractures. This newly developed predictive risk score may help clinicians decide on appropriate treatment for T2D with osteoporosis, especially in cases with equivocal results after using current tools, including BMD measurement and 10-year probability of fracture calculated by FRAX®.

Materials and methods

Study design and study population

This study is a sub-study of The Cohort Of patients at a high Risk for Cardiovascular Events (CORE) - Thailand registry, which is a prospective study involving Thai patients with high atherosclerotic risks. The CORE-Thailand registry protocol has been previously described in detail50. In brief, individuals with high cardiovascular risk were enrolled at the outpatient clinic at Maharaj Nakorn Chiang Mai Hospital in the period April 2011 to March 2014 and then reevaluated at 6, 12, 24, 36, 48, and 60 months. The criteria for high cardiovascular risk included individuals aged 45 years or older with established cerebrovascular disease, coronary artery disease, or peripheral arterial disease, or with at least 3 atherosclerosis risk factors, which are composed of T2D or impaired fasting glucose, hypertension, dyslipidemia, chronic kidney disease (proteinuria and/or an estimated glomerular filtration rate of less than 60 mL/min), current smokers, either males aged at least 55 years old or females aged at least 65 years old, and a family history of premature atherosclerosis. During the reevaluation visit in the regular protocol of the CORE-Thailand registry in the period of February 2015 to August 2017, 371 patients were randomly invited and then enrolled in this sub-study. After 60 months of reevaluation visit, the participants were followed yearly by telephone interviewing or medical record reviewing until the occurrence of the first fragility fracture or reaching the censoring date in June 2023.

The CORE-Thailand registry study protocol was approved by the Joint Research Ethics Committee and Ministry of Public Health, Thailand (Approval Number COA JREC 004/2011) and this sub-study protocol was approved by the institutional Ethics Committee of the Faculty of Medicine, Chiang Mai University, Chiang Mai, Thailand (Approval number MED-2557-02327, MED-2557-02520 and MED 2566 − 0373). All participants provided written informed consent before enrollment. This study was performed in accordance with the guidelines and regulations of our institute and adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for observational studies51.

Data collection and plasma FGF 21 measurements

Demographic data, physical examination including a TUG test, and blood samples were obtained at enrollment. The TUG test determines balance and mobility and predicts falling. The participants were recorded for times required to rise from sitting in an armless chair, 3-meter walk at a normal pace, turn around 180 degrees, walk back, and sit down again. Slow TUG was defined by the required times longer than 12 s to complete this procedure; this cut-off point was demonstrated for steeply increased risk of fracture in the elderly36. Fasting blood samples were collected to measure fasting plasma glucose (FPG), hemoglobin A1c (HbA1c), creatinine (Cr), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), triglyceride, aspartate transaminase (AST), alanine transaminase (ALT), alkaline phosphatase (ALP), and FGF21 levels. All biochemical parameters except FGF21 were assessed using standardized procedures at a central laboratory of Maharaj Nakorn Chiang Mai Hospital, Faculty of Medicine, Chiang Mai University. FGF21 levels were determined by a human FGF21 enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems Inc., Minneapolis, MN, USA). Glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) method. Fracture risk estimation was estimated from The Fracture Risk Assessment Tool (FRAX®) based on body mass index (BMI) using the Thailand database52 with further adjustment by adding rheumatoid arthritis as a clinical risk factor in all individuals with T2D12. Time to first fragility fracture and types of fracture were obtained by on-site interviewing, telephone interviewing or medical record reviewing. Fracture outcomes accounted only for fragility fractures, which were defined as non-traumatic fractures or low-trauma fractures, for example, falling from standing height. Major fractures were defined as clinical or morphological fragility fractures in any area except the skull, hand, foot, and ankle, while any fractures were defined as fragility fractures that occurred in any area.

Definition

Body mass index (BMI) and waist circumference (WC) were used to classify obesity and central obesity, respectively. BMI (≥ 25 kg/m2)53 and WC (≥ 80 cm in women or ≥ 90 cm in men)54 higher than cut-off points for the Asian population were defined as obesity and central obesity, respectively. Metabolic syndrome (MetS) was defined in participants who met any 3 out of 5 of the following criteria: elevated WC (≥ 90 cm in men and ≥ 80 cm in women); blood triglyceride level ≥ 150 mg/ dl or treated; HDL-C < 40 mg/dl in men or < 50 mg/dL in women; blood pressure ≥ 130/85 mmHg or treated; and FPG ≥ 100 mg/ dl or treated55.

Statistical analyses

All statistical analyses were performed using STATA (version 16.0, StataCrop LLC., College Station, TX, USA). Normally distributed continuous data was presented in mean and standard deviation (mean ± SD), while non-normally distributed data was presented in median and interquartile ranges (IQR). Categorical data was presented by counts and percentages. Continuous data was analyzed using an independent t-test and the Mann-Whitney U test as appropriate, while categorical data was analyzed by Chi-square or Fisher’s exact test. Statistically significant was defined as a P-value less than 0.05.

The Aalen-Johansen estimator was used to calculate the cumulative incidence rates for time to first fragility fractures, hereby adjusting for death as a competing risk. Differences between groups were determined with the Wald test. Competing-risk survival regression was used to calculate estimates of relative risk and 95% confidence intervals (95% CI) for the two study groups while accounting for death. Univariable and multivariable logistic regression analyses were conducted in individuals with T2D by categorizing these individuals into two subgroups by the occurrence of any fractures after enrollment. Logistic regression analysis was performed and reported as odds ratio (OR) or adjusted odds ratio (AOR), and 95% CI. The final predictive model was developed using backward elimination method by removing the factors with P-value > 0.3. To create an item score, the AOR was converted to β-coefficient. The β-coefficient of each factor was divided by the smallest β-coefficient in the final model and rounded to the nearest 0.5. The cut-off point for the risk levels was retrieved from the level using Youden’s index that yielded the highest sensitivity, specificity, and accuracy. The performance of the predictive model and predictive risk score were presented as the area under the receiver operating characteristic curve (AuROC).