Development and validation of an osteoporosis risk-assessment model for Chinese older adults: a large-scale retrospective study

Abstract

Objective

To identify the determinants of osteoporosis (OP) among older adults in China and develop a nomogram-based risk-assessment tool for clinical screening.

Methods

We retrospectively enrolled 8,243 consecutive older adults who underwent routine physical examinations at the Health Management Center of the First Affiliated Hospital of the University of Science and Technology of China between January 2023 and June 2025. After classifying participants into OP (n = 4,302) and non-OP (n = 3,941) groups according to bone mineral density measurements, we collected baseline demographic information, clinical history, and routine biochemical indices. Variables selection were selected via LASSO regression; subsequently, multivariable logistic regression was applied to identify independent risk factors for OP. A screening nomogram was constructed from these variables and internally validated through 1,000-replicate bootstrapping; discrimination (ROC curve), calibration (Hosmer–Lemeshow test and calibration plot), and clinical utility (decision curve analysis [DCA]) were assessed.

Results

Multivariable analysis revealed that female sex, increasing age, diabetes mellitus, and prior fracture history were independently associated with higher odds of OP (OR > 1 and all P < 0.05). Conversely, higher body mass index (BMI), serum uric acid, serum calcium and vitamin D levels were linked to reduced odds (OR < 1 and all P < 0.05). The nomogram achieved an AUC of 0.799 (95% CI: 0.789–0.808); calibration plots aligned closely with the ideal line (Hosmer–Lemeshow P = 0.796), while DCA indicated net clinical benefit between risk thresholds of 13% and 89%. After bootstrapping, the AUC remained stable at 0.788 (95%CI: 0.779–0.797).

Conclusion

Female sex, advancing age, diabetes mellitus, prior fracture history, BMI, serum uric acid, serum calcium and vitamin D are independent determinants of OP in older adults.The nomogram built on these factors offers satisfactory screening accuracy and may assist clinicians in identifying individuals at an elevated risk for osteoporosis to guide personalized preventive strategies.

Introduction

Osteoporosis (OP) is a systemic skeletal disease characterized by reduced bone mass and deteriorated bone microstructure, leading to increased bone fragility and an elevated risk of fractures [1]. Osteoporotic fractures of the hip or vertebral body are associated with long-term disability, chronic pain, and a reduced quality of life; they also increase patient mortality and impose significant medical and economic burdens on families and society [2]. As populations age, the incidence of OP-related fractures continue to increase, especially among women over 55 years and men over 65 years [3]. Postmenopausal women experience decreased estrogen levels, which accelerate bone loss and confer a higher risk of OP than that observed in men [4]. Therefore, the early identification of OP risk in older adults—particularly women—supports preventive strategies, improves health management, and may help mitigate medical costs.

Currently, OP screening relies mainly on bone mineral density (BMD) testing; however, its high cost and limited availability limit its feasibility for large-scale community screening [5]. Moreover, multiple factors—including age, nutritional status (such as calcium or vitamin D deficiency), and regional characteristics—contribute to OP [6–8]. Existing OP risk-assessment tools in China are largely derived from Western cohorts, leaving an unmet need for a model tailored to older Chinese adults [9]. Consequently, developing an accessible and cost-effective risk-screening tool aligned with Chinese demographic features is essential for timely intervention.

In this study, we analyzed physical-examination data from 8243 healthy older adults attending the Health Management Center at the First Affiliated Hospital of the University of Science and Technology of China. BMD measurements were used to diagnose OP, and demographic characteristics, chronic-disease histories, and biochemical indicators were systematically collected. LASSO regression identified candidate variables, which were then evaluated using multivariable logistic regression to determine independent determinants. A nomogram was constructed to provide clinicians with an intuitive, individualized OP-risk estimate. The tool aims to facilitate early identification and targeted management of high-risk individuals, with the goal of reducing both OP-related morbidity and the associated fracture burden.

Objects and methods

Subjects

A total of 8,243 older adults who had undergone routine physical examinations at the Health Management Center of the First Affiliated Hospital of the University of Science and Technology of China between January 2023 and June 2025 were retrospectively enrolled. Missing data were handled using a complete-case analytical approach; individuals with missing values in any of the 31 candidate variables were excluded from the final analytical cohort to ensure the integrity and robustness of the multivariable model.

The inclusion criteria were as follows: (1) age ≥ 60 years. The selection of individuals aged ≥ 60 years as the cutoff aligns with the Law of the People's Republic of China on the Protection of the Rights and Interests of older adults, which defines individuals aged 60 and above as older adults. Furthermore, this threshold is consistent with China’s National Basic Public Health Service Project, which provides standardized health management and free annual physical examinations for residents starting at age 60. Using this cutoff ensures that our screening nomogram is highly applicable to the target population of China’s national primary care screening programs and reflects local demographic and policy norms; for women, the occurrence of natural menopause lasting ≥ 12 months or bilateral ovariectomy [10]; (2) clear consciousness allowing effective communication; (3) complete physical examination data.

Exclusion criteria included: (1) a diagnosis of psychiatric disorders or cognitive impairment; (2) the use within six months of drugs influencing bone metabolism (e.g., glucocorticoids or calcitonin); (3) an inability to cooperate with BMD measurement; (4) the presence of malignant tumors or other severe systemic diseases.

This study protocol was approved by the Medical Ethics Committee of the First Affiliated Hospital of the University of Science and Technology of China (No. 2025-RE-292). Because of its retrospective design, the requirement for informed consent was waived by the same committee.

Data collection methods

The Health Management System (HMS) and Hospital Information System (HIS) of the First Affiliated Hospital of the University of Science and Technology of China were used to retrospectively extract complete records for all individuals meeting the inclusion criteria from January 2023 to June 2025. Extracted variables comprised demographic characteristics, clinical history, laboratory results, and bone mineral density (BMD) reports.

BMD measurement and diagnostic classification followed established protocols: (1) Measurement: Lumbar spine BMD (L1–L4) was assessed by dual-energy X-ray absorptiometry (DXA; GE Lunar Prodigy, USA), performed by a single trained technician using standardized procedures. (2) Classification: Results were interpreted according to WHO criteria based on T-scores: normal (T-score ≥ –1.0), osteopenia (–2.5 < T-score < –1.0), and osteoporosis (T-score ≤ –2.5) [11].

Data comprised three domains: (1) Demographics: age; sex; height; weight (used to calculate BMI = weight/height²); waist circumference; systolic and diastolic blood pressure; pulse. (2) Medical history and lifestyle: chronic diseases—hypertension, diabetes mellitus, coronary heart disease, ischemic stroke, hyperthyroidism (identified via ICD-10 codes)—and any previous fracture (verified from medical records). (3) Laboratory parameters: fasting plasma glucose; two-hour post-load glucose; glycated hemoglobin (HbA1c); total cholesterol; triglycerides; LDL-C; HDL-C; serum creatinine; blood urea nitrogen; uric acid; albumin; calcium; phosphate; vitamin D [25(OH)D]; complete blood count (hemoglobin, white-blood-cell count, red-blood-cell count, platelet count).

Statistical methods

SPSS 25.0 and Stata 17.0 were used for all analyses. The distribution of continuous variables was assessed for normality using the Kolmogorov–Smirnov test. Continuous variables exhibiting a normal distribution are reported as mean ± SD, and differences between the two groups were assessed with an independent-sample Student’s t-test. For variables with a non-normal distribution, data are presented as median (interquartile range [Q1, Q3]) and were compared using the Mann–Whitney U test. Categorical variables are presented as n (%), and group differences were examined with the chi-square test.

To identify candidate variables of osteoporosis among older adults, LASSO regression was applied first; the final set of independent risk factors was determined by multivariable logistic regression incorporating only those variables retained after LASSO screening. To address the baseline variations between the osteoporosis and non-osteoporosis groups, multivariable logistic regression was specifically employed to adjust for these confounding variables and identify independent risk factors. Multicollinearity among the independent variables was assessed using Variance Inflation Factors (VIFs), with VIF < 10 indicating no significant collinearity.

A clinical nomogram was then built with Stata’s “nomolog” package; its discriminative accuracy was quantified by the area under the ROC curve, and the incremental screening value of the nomogram over basic clinical models (e.g., age, sex, and BMI) was evaluated using the DeLong test. To provide more clinically meaningful effect sizes, continuous variables such as blood calcium and vitamin D were rescaled (per 0.1 mmol/L and 10 ng/mL increment, respectively) before being entered into the final model. Its agreement between expected and observed probabilities was evaluated with the Hosmer–Lemeshow test (P > 0.05 indicates an acceptable fit), and net clinical benefit was assessed through decision-curve analysis (DCA). To mitigate over-fitting, internal validation was performed with 1,000 bootstrap resamples. A two-sided P-value < 0.05 was regarded as statistically significant.

Results

Participant selection and data completeness

A total of 8,720 older adults were initially assessed for eligibility. As illustrated in the recruitment flowchart (Fig. 1), 477 individuals (5.47%) were excluded due to missing values in one or more of the 31 candidate variables. Consequently, a final analysis cohort of 8,243 participants with complete data was included in the study, comprising 4,302 individuals in the osteoporosis group and 3,941 in the non-osteoporosis group.

Fig. 1.

Flowchart of participant selection and exclusion criteria

Comparison of relevant data between the osteoporosis group and the non-osteoporosis group in the older adults

Among the 8,243 older adults enrolled (mean age 68.60 ± 6.95 years), 3,173 (38.5%) were male and 5,070 (61.5%) were female; further baseline characteristics are listed in Table 1. The analysis sample comprised 3,941 participants without osteoporosis and 4,302 participants diagnosed with osteoporosis. Normality testing indicated that all continuous variables, with the exception of triglycerides, followed a normal distribution. Compared with the non-osteoporosis group, the osteoporosis group differed significantly regarding age; BMI; diastolic blood pressure; fasting blood glucose; glycated hemoglobin; total cholesterol; triglycerides; low- and high-density lipoprotein; serum creatinine; blood urea nitrogen; uric acid; hemoglobin; white-blood-cell count; red-blood-cell count; platelet count; serum calcium; vitamin D levels; sex distribution; and the prevalence of hypertension, coronary heart disease, diabetes mellitus, ischemic stroke, hyperthyroidism or prior fracture (all P < 0.05). Differences in waist circumference; systolic blood pressure; pulse rate; two-hour post-prandial glucose; serum albumin; and serum phosphorus did not reach statistical significance (P > 0.05).

Table 1.

Comparison of relevant data between the osteoporosis group and the non-osteoporosis group in the older adults

Variables	Total (n = 8243)	Non-osteoporosis group (n = 3941)	Osteoporosis group (n = 4302)	Statistic	P
Age (years), Mean ± SD	68.60 ± 6.95	67.43 ± 6.66	69.67 ± 7.04	t = −14.81	< 0.001
Body mass index (kg/m2), Mean ± SD	24.73 ± 2.76	25.02 ± 2.81	24.46 ± 2.69	t = 9.16	< 0.001
Waist circumference (cm), Mean ± SD	90.01 ± 7.00	89.97 ± 6.91	90.05 ± 7.09	t = −0.47	0.636
Systolic blood pressure (mmHg), Mean ± SD	138.60 ± 12.59	138.71 ± 13.66	138.51 ± 11.53	t = 0.70	0.484
Diastolic blood pressure (mmHg), Mean ± SD	79.59 ± 7.07	79.94 ± 7.85	79.27 ± 6.24	t = 4.31	< 0.001
Pulse (beats/min), Mean ± SD	78.65 ± 8.08	78.62 ± 8.12	78.68 ± 8.04	t = −0.33	0.742
Fasting blood glucose (mmol/L), Mean ± SD	5.68 ± 0.91	5.73 ± 1.07	5.64 ± 0.74	t = 4.26	< 0.001
Two-hour postprandial blood glucose (mmol/L), Mean ± SD	9.66 ± 1.10	9.66 ± 1.13	9.66 ± 1.07	t = 0.12	0.904
Glycated hemoglobin (%), Mean ± SD	6.08 ± 0.77	6.10 ± 0.80	6.05 ± 0.73	t = 2.96	0.003
Total cholesterol (mmol/L), Mean ± SD	5.15 ± 0.85	5.10 ± 0.93	5.20 ± 0.77	t = −5.40	< 0.001
Triglyceride (mmol/L), M (Q1, Q3)	1.58 (1.40, 1.68)	1.56 (1.21, 1.71)	1.59 (1.47, 1.67)	Z = −4.392	< 0.001
Low density lipoprotein (mmol/L), Mean ± SD	3.14 ± 0.54	3.12 ± 0.64	3.16 ± 0.44	t = −3.67	< 0.001
High density lipoprotein (mmol/L), Mean ± SD	1.37 ± 0.20	1.34 ± 0.23	1.40 ± 0.17	t = −13.43	< 0.001
Blood albumin (g/L), Mean ± SD	43.91 ± 2.71	43.91 ± 2.61	43.90 ± 2.81	t = 0.05	0.959
Blood creatinine (μmol/L), Mean ± SD	72.76 ± 13.58	75.22 ± 15.22	70.50 ± 11.43	t = 15.80	< 0.001
Blood urea nitrogen (mmol/L), Mean ± SD	5.91 ± 1.12	5.95 ± 1.31	5.88 ± 0.90	t = 2.83	0.005
Serum uric acid (μmol/L), Mean ± SD	343.14 ± 55.19	353.78 ± 64.45	333.40 ± 42.81	t = 16.76	< 0.001
Hemoglobin (g/L), Mean ± SD	140.37 ± 11.62	142.38 ± 12.03	138.53 ± 10.91	t = 15.17	< 0.001
Total WBC (10⁹/L), Mean ± SD	5.89 ± 1.23	6.00 ± 1.33	5.79 ± 1.12	t = 7.66	< 0.001
Erythrocyte count (1012/L), Mean ± SD	4.65 ± 0.60	4.69 ± 0.55	4.60 ± 0.65	t = 6.76	< 0.001
Platelet count (10⁹/L), Mean ± SD	203.75 ± 36.26	202.61 ± 41.40	204.81 ± 30.79	t = −2.72	0.007
Blood calcium (mmol/L), Mean ± SD	2.35 ± 0.04	2.35 ± 0.04	2.35 ± 0.03	t = 6.32	< 0.001
Blood phosphorus (mmol/L), Mean ± SD	1.16 ± 0.08	1.16 ± 0.08	1.16 ± 0.08	t = −0.86	0.391
Blood vitamin D (ng/mL), Mean ± SD	40.99 ± 9.25	41.29 ± 9.76	40.71 ± 8.75	t = 2.83	0.005
Gender, n(%)				χ2 = 2053.45	< 0.001
Male	3173 (38.49)	2517 (63.87)	656 (15.25)
Female	5070 (61.51)	1424 (36.13)	3646 (84.75)
Hypertension, n(%)				χ2 = 90.48	< 0.001
No	6836 (82.93)	3106 (78.81)	3730 (86.70)
Yes	1407 (17.07)	835 (21.19)	572 (13.30)
Coronary heart disease, n(%)				χ2 = 22.31	< 0.001
No	8111 (98.40)	3851 (97.72)	4260 (99.02)
Yes	132 (1.60)	90 (2.28)	42 (0.98)
Diabetes, n(%)				χ2 = 7.26	0.007
No	7779 (94.37)	3691 (93.66)	4088 (95.03)
Yes	464 (5.63)	250 (6.34)	214 (4.97)
Ischemic stroke, n(%)				χ2 = 11.01	< 0.001
No	8199 (99.47)	3909 (99.19)	4290 (99.72)
Yes	44 (0.53)	32 (0.81)	12 (0.28)
Hyperthyroidism, n(%)				χ2 = 9.31	0.002
No	8216 (99.67)	3936 (99.87)	4280 (99.49)
Yes	27 (0.33)	5 (0.13)	22 (0.51)
Previous fracture history, n(%)				χ2 = 6.06	0.014
No	8181 (99.25)	3921 (99.49)	4260 (99.02)
Yes	62 (0.75)	20 (0.51)	42 (0.98)

Reference ranges: Fasting blood glucose, 3.9–6.1 mmol/L; HbA1c, 4.0%–6.0%; Total cholesterol, < 5.2 mmol/L; Triglycerides, < 1.7 mmol/L; LDL-C, < 3.4 mmol/L; HDL-C, > 1.0 mmol/L; Albumin, 35–55 g/L; Creatinine, 44–106 μmol/L; BUN, 2.1–7.1 mmol/L; Uric acid, 150–416 μmol/L (male), 89–357 μmol/L (female); Hemoglobin, 115–175 g/L; WBC, 4.0–10.0 × 10⁹/L; RBC, 3.8–5.8 × 10¹²/L; Platelets, 100–300 × 10⁹/L; Calcium, 2.1–2.6 mmol/L; Phosphorus, 0.8–1.5 mmol/L; Vitamin D (sufficient), > 30 ng/mL

Bold typeface indicates statistical significance at the P < 0.05 level

t: t-t t-test, χ² Chi-square test, Z Mann Whitney U test, SD Standard deviation

Screening of osteoporosis characteristic variables in the older adults

We applied LASSO regression with tenfold cross-validation to reduce the dimensionality among the 31 candidate variables listed in Table 1, selecting λ = 0.003, which corresponded to the minimum cross-validation error (Fig. 2). At this threshold, 21 variables retained non-zero coefficients: sex; age; hypertension; coronary heart disease; diabetes mellitus; ischemic stroke; hyperthyroidism; prior fracture history; BMI; systolic blood pressure; diastolic blood pressure; fasting plasma glucose; glycated hemoglobin; total cholesterol; high-density lipoprotein cholesterol; serum creatinine; serum uric acid; total white-blood-cell count; red-blood-cell count; serum calcium; and vitamin D.

Fig. 2.

Results of LASSO regression analysis for screening of osteoporosis characteristic variables in the older adults. A Coefficient paths; B Cross-validation plot with 1 SE bounds

Results of multivariate logistic regression analysis on the occurrence of osteoporosis in the older adults

Multivariable logistic regression incorporating all 21 LASSO-selected variables identified four independent risk factors for osteoporosis—female sex, advancing age, presence of diabetes mellitus, and prior fracture history—each associated with an odds ratio(OR) greater than 1 (all P < 0.05). To ensure the robustness of our estimates, we assessed multicollinearity using Variance Inflation Factors (VIFs); all VIF values were below 10 (Table 2), indicating no significant collinearity among the independent variables. Furthermore, to address concerns regarding extreme effect sizes, continuous variables were rescaled to clinically meaningful increments. For instance, the OR for blood calcium was 0.70 per 0.1 mmol/L increase, and for serum vitamin D, it was 0.94 per 10 ng/mL increase (Table 3). Four variables—higher BMI, elevated serum uric acid, higher serum calcium, and higher vitamin D—showed ORs below 1 (all P < 0.05) and can therefore be considered protective factors (Table 3).

Table 2.

Variance Inflation Factor (VIF) analysis for candidate variables

Variables	VIF	Variables	VIF
Gender	1.230	Fasting blood-glucose	1.420
Age	1.050	Glycosylated hemoglobin	1.340
Hypertension	1.120	Total cholesterol	1.240
Coronary heart disease	1.040	High-density lipoprotein	1.410
Diabetes mellitus	1.200	Serum creatinine	1.320
schemic stroke	1.020	Serum uric acid	1.350
Hyperthyroidism	1.010	Total white blood cells	1.080
Prior fracture history	1.000	Red blood cell count	1.050
Body mass index	1.110	Blood calcium	1.010
Systolic pressure	1.150	Blood vitamin D	1.000
Diastolic pressure	1.140

VIF < 10 for all independent variables suggests no significant multicollinearity

Table 3.

Results of multivariate logistic regression analysis of osteoporosis in the older adults

Variables	β	S.E	Z	P	OR (95%CI)
Gender
Male					1.00 (Reference)
Female	2.27	0.06	37.36	< 0.001	9.66 (8.58 ~ 10.88)
Age	0.06	0.00	15.67	< 0.001	1.06 (1.06 ~ 1.07)
Hypertension
No					1.00 (Reference)
Yes	−0.07	0.08	−0.95	0.340	0.93 (0.80 ~ 1.08)
Coronary heart disease
No					1.00 (Reference)
Yes	−0.4	0.24	−1.68	0.093	0.67 (0.42 ~ 1.07)
Diabetes
No					1.00 (Reference)
Yes	0.31	0.13	2.42	0.015	1.36 (1.06 ~ 1.75)
Ischemic stroke
No					1.00 (Reference)
Yes	−0.42	0.4	−1.06	0.29	0.66 (0.30 ~ 1.43)
Hyperthyroidism
No					1.00 (Reference)
Yes	0.77	0.54	1.44	0.151	2.16 (0.75 ~ 6.19)
Previous fracture history
No					1.00 (Reference)
Yes	1.00	0.33	3.07	0.002	2.72 (1.44 ~ 5.14)
Body mass index	−0.06	0.01	−6.05	< 0.001	0.94 (0.92 ~ 0.96)
Systolic pressure	0.00	0.00	−1.32	0.185	1.00 (0.99 ~ 1.00)
Diastolic pressure	0.00	0.00	1.14	0.256	1.00 (1.00 ~ 1.01)
Fasting blood-glucose	−0.04	0.04	−1.12	0.265	0.96 (0.90 ~ 1.03)
Glycosylated hemoglobin	−0.07	0.04	−1.73	0.084	0.93 (0.87 ~ 1.01)
Total cholesterol	−0.05	0.03	−1.51	0.131	0.95 (0.89 ~ 1.02)
High-density lipoprotein	0.23	0.15	1.49	0.135	1.26 (0.93 ~ 1.71)
Serum creatinine	0.00	0.00	−0.76	0.447	1.00 (0.99 ~ 1.00)
Uric acid	−0.01	0.00	−4.03	< 0.001	0.99 (0.99 ~ 0.99)
Total white blood cells	−0.04	0.02	−1.62	0.106	0.96 (0.92 ~ 1.01)
Red blood cell count	0.05	0.04	1.09	0.277	1.05 (0.96 ~ 1.14)
Blood calcium	−0.35	0.08	−4.50	< 0.001	0.70 (0.60 ~ 0.82)
Blood vitamin D	−0.06	0.03	−2.08	0.037	0.94 (0.89 ~ 0.99)

To provide more clinically meaningful effect sizes, blood calcium was rescaled per 0.1 mmol/L increment, and blood vitamin D was rescaled per 10 ng/mL increment

Bold typeface indicates statistical significance at the P < 0.05 level

OR Odds Ratio, CI Confidence Interval

Construction of a nomogram model for the risk of osteoporosis in the older adults

A risk-assessment nomogram for older adults was constructed from the eight independent variables identified by multivariable logistic regression, as illustrated in Fig. 3.

Fig. 3.

Nomogram model of osteoporosis risk in the older adults

Analysis of clinical applicability of the nomogram model

Decision-curve analysis indicated that when the threshold probability ranged from 0.13 to 0.89, the use of this nomogram conferred a higher net clinical benefit than either the “treat-all” or “treat-none” strategies (Fig. 4).

Fig. 4.

DCA curve of the nomogram model

Evaluation of calibration and discrimination of the nomogram model

Calibration curves showed close agreement between expected probabilities and observed frequencies (Fig. 5A), corroborated by Hosmer–Lemeshow testing (χ² = 4.630, P = 0.796). Discrimination was assessed via ROC analysis: AUC = 0.799 (95% CI: 0.789–0.808) (Fig. 5B). Internal validation with 1,000 bootstrap resamples yielded an AUC = 0.788 (95% CI:0.779–0.797) (Fig. 5C), indicating stable screening performance.

Fig. 5.

Calibration curve and ROC curve of the nomogram model. Note: A Calibration curve of the nomogram model; B ROC curve before internal verification of the nomogram model; C ROC curve after internal verification of the nomogram model

Comparison of screening performance between the nomogram and basic clinical models

To evaluate the incremental value of the multivariable nomogram, we compared its screening performance with two baseline models: Basic Model 1 (age and sex) and Basic Model 2 (age, sex, and BMI). As shown in Table 4 and Fig. 6, the comprehensive nomogram (AUC = 0.799) exhibited significantly higher discriminative power than Basic Model 1 (AUC = 0.597) and Basic Model 2 (AUC = 0.612). According to the DeLong test, the AUC of the nomogram was significantly higher than the AUCs of both Basic Model 1 and Basic Model 2 (both P < 0.001). Furthermore, the nomogram demonstrated superior sensitivity (0.830) and accuracy (0.751) compared to the basic models. These results indicate that incorporating additional clinical and biochemical variables—such as diabetes mellitus, prior fracture history, and serum metabolic markers—substantially enhances the tool’s ability to identify osteoporosis in the Chinese older population, thereby justifying the complexity of the proposed nomogram.

Table 4.

Comparison of screening performance between the nomogram and basic clinical models

Models	AUC	Sensitivity	Specificity	Youden Index	Optimal Threshold	Accuracy
Nomogram	0.799	0.830	0.664	0.494	0.556	0.751
Age + sex	0.597	0.673	0.476	0.149	0.496	0.579
Age + sex + BMI	0.612	0.649	0.519	0.168	0.498	0.587

Fig. 6.

Comparison of ROC curves between the nomogram and basic models. Note: Red line: Nomogram; Blue line: Age + Sex model; Light red line: Age + Sex + BMI model

Based on the maximum Youden index, we identified an optimal clinical decision threshold (cutoff) of 0.556 for the nomogram. At this threshold, the model exhibited a sensitivity of 0.830, a specificity of 0.664, and an overall accuracy of 0.751. To further evaluate its real-world reliability, we calculated the predictive values: the positive predictive value (PPV) was 73.0%, and the negative predictive value (NPV) was 78.2%. These metrics provide a clear operating characteristic for clinical implementation, ensuring that the majority of high-risk individuals are identified while maintaining a manageable referral rate.

Discussion

Osteoporosis is a chronic skeletal condition characterized by a prevalence that rises sharply with age [12]; its pathogenesis reflects the interplay among multiple biological determinants and environmental exposures rather than any single mechanism alone [12]. By using complete electronic health record data from 8,243 community-dwelling older adults attending routine health checks at the Health Management Center of the First Affiliated Hospital of the University of Science and Technology of China, we applied systematic analyses that identified female sex, advancing age, prevalent diabetes mellitus, prior fracture history, lower BMI, lower serum uric acid, lower serum calcium, and lower vitamin D concentration as independent influencing factors associated with osteoporotic status.

This study confirms female sex (OR = 9.66) and advancing age (OR = 1.06) as major determinants of osteoporotic risk [13–16]. The heightened vulnerability among women—particularly after menopause—is well documented [13]; the worldwide lifetime fracture risk after age fifty is about one-third for women versus one-fifth for men [15]. Chinese data mirror this gradient: osteopenia affects 43.94% versus 35.58%, and osteoporosis affects 11.72% versus 2.51%, respectively [16], while post-menopausal women exhibit a prevalence (14.05%) roughly five-fold higher than pre-menopausal women (2.62%) [16]. Declining estrogen is central to this disparity [14], yet sex-chromosome variation [13] and hormone-mediated modulation of oxidative stress [17] also contribute. Advancing age likewise correlates directly with osteoporotic incidence [18]. Bone loss accelerates after fifty [14]; international guidelines therefore recommend routine bone-mineral-density screening from sixty-five years onward [14], while fracture rates rise sharply after seventy [18] and periprosthetic fractures occur in up to 3.5% of octogenarians undergoing arthroplasty [19]. Age-related mechanisms include Men1-mediated osteoblast dysfunction [20] alongside micro-architectural deterioration driven by cellular senescence [21, 22].

Diabetes (OR = 1.36) and a previous fracture (OR = 2.72) were identified as independent risk factors for osteoporosis. Chronic hyperglycemia undermines osteoblast activity, suppressing bone formation and repair, which is a primary pathway through which diabetes elevates fracture risk [23, 24]. Additionally, although bone mineral density may appear normal or even elevated in type 2 diabetes, deterioration of the micro-architecture—evidenced by fewer trabeculae and greater cortical porosity—compromises mechanical strength [25, 26]. Moreover, a diabetes duration exceeding ten years or the presence of microvascular complications is associated with a further, dose-dependent rise in fracture incidence [27, 28]. Both insulin deficiency (type 1 diabetes) and insulin resistance (type 2 diabetes) disturb the balance of bone turnover [26, 29]. A prior fracture signifies that bone strength has already been compromised, and a low-trauma fracture is a clinical indicator of reduced bone mass [30]. Post-fracture immobilization and the inflammatory response accelerate bone resorption, while the remodeling process during healing may not fully restore structural integrity. These alterations create mechanical weakness that predisposes individuals to subsequent fractures [31]. Thus, an initial fracture serves as a marker of diminished bone strength and predicts later osteoporotic fractures with moderate reliability [32].

In this study, higher BMI (OR = 0.94), serum uric acid (OR = 0.99), blood calcium (OR = 0.03) and 25-hydroxyvitamin D (OR = 0.99) were associated with reduced odds of osteoporosis. Greater body mass increases mechanical loading on the skeleton, thereby stimulating bone formation and slowing age-related bone loss [33]. Uric acid acts as an endogenous antioxidant that can attenuate oxidative stress in the bone micro-environment; nevertheless, marked hyperuricemia may exert adverse renal effects that warrant attention [34, 35]. When circulating calcium falls, parathyroid hormone secretion rises, enhancing osteoclastic resorption to restore calcium homeostasis and, if uncorrected, leading to progressive bone loss [36]. Vitamin D supports bone by facilitating intestinal calcium absorption, suppressing parathyroid hormone release, and exerting direct regulatory effects on osteoblast and osteoclast activity [37].

The nomogram developed in this study achieved satisfactory discriminative ability (AUC = 0.799; bootstrap-corrected AUC = 0.788) and good calibration (Hosmer–Lemeshow P = 0.796). Unlike models derived from Western cohorts such as FRAX [30], the present tool was trained on urban Chinese older adults, thereby reflecting local epidemiological patterns. Furthermore, compared with traditional screening tools like the Osteoporosis Self-Assessment Tool for Asians (OSTA), which relies solely on age and weight, our nomogram incorporates a broader range of clinical determinants, including diabetes mellitus, prior fracture history, and key biochemical markers (e.g., serum calcium and vitamin D). While OSTA is highly accessible for self-screening, it may overlook the systemic complexity of bone metabolism in older adults. Our model achieves a more comprehensive risk profile, offering higher granularity for clinicians to tailor interventions beyond simple anthropometric measurements. Decision-curve analysis indicated that, across threshold probabilities of 13%–89%, the application of the model conferred a net clinical benefit over the “treat-all” or “treat-none” strategies, making it suitable for community-level screening or individual risk estimation.

While international clinical practice has increasingly emphasized absolute fracture risk (e.g., FRAX) as a primary management endpoint, identifying individuals with DXA-defined osteoporosis remains fundamentally important in the Chinese healthcare setting. According to the Chinese Journal of Osteoporosis and Bone Mineral Salt Disease guidelines, a BMD T-score ≤ −2.5 is a primary diagnostic criterion for initiating pharmacological intervention [1]. Furthermore, our model serves as a complementary precursor to fracture-risk tools. Since many fracture-risk algorithms exhibit improved accuracy when BMD data are included, our nomogram facilitates the early identification of "silent" osteoporosis before any clinical fracture occurs. By prioritizing high-risk individuals for DXA, this tool bridges the gap between general screening and definitive diagnosis, ensuring that preventive strategies are implemented early in the disease course. By converting a small number of readily collected variables into a visual risk score, the nomogram offers clinicians—especially in resource-limited settings—a straightforward means of identifying high-risk individuals. In the current Chinese healthcare context, while DXA remains the gold standard for a definitive diagnosis and a prerequisite for initiating certain potent anti-osteoporotic agents (such as bisphosphonates), the distribution of DXA technology is highly uneven between tertiary hospitals and primary care centers. Our nomogram is strategically designed as a “gatekeeper” or triaging tool rather than a direct diagnostic replacement. Specifically, by using this model, primary care physicians can effectively identify high-risk older adults who should be prioritized for referral to larger centers for DXA confirmation. Furthermore, for individuals identified as high-risk in settings where DXA is strictly unavailable, the model provides a clinical basis for early interventions that do not always require a DXA diagnosis, such as intensive lifestyle education, falls prevention, and standardized calcium and vitamin D supplementation.

To facilitate real-world implementation, we propose a standardized clinical pathway based on the identified cutoff of 0.556. For older adults attending health check-ups, the nomogram probability should first be calculated. Individuals with a probability ≥ 0.556 (high-risk) should be prioritized for DXA referral to confirm the diagnosis and receive immediate primary prevention interventions. Given the high NPV (78.2%), individuals with a score < 0.556 can be considered at lower immediate risk and managed with routine follow-up and lifestyle guidance. This evidence-based threshold allows for more efficient allocation of bone densitometry resources and ensures that high-risk patients are managed early in the disease course. This targeted approach ensures that limited medical resources are directed toward those with the greatest clinical need. While the cross-sectional design limits our ability to predict the future incidence of OP, the model serves as a robust screening tool for identifying current high-risk patients, especially in primary care settings where bone densitometry is not readily available.

However, the study has several limitations. First, its retrospective design and reliance on data from a single medical center may introduce selection bias. Furthermore, the study population was drawn from a cohort of older adults undergoing routine health examinations, who are typically more health-conscious and potentially healthier than the general population. This “healthy volunteer effect” may limit the representativeness of our findings and reduce the generalizability of the nomogram to the broader population, particularly those with higher comorbidity burdens or restricted access to healthcare. Consequently, the performance of the model in more frail or less health-seeking populations requires further evaluation. Second, lifestyle variables—such as physical activity, smoking, and alcohol intake—were not captured, limiting the ability to assess their influence on osteoporosis risk. Future work could incorporate bone turnover markers (e.g., β-CTX, P1NP [38]) and genetic factors (e.g., VDR polymorphisms [39]) to refine screening accuracy, and model-guided intervention trials could be carried out to estimate their effect on fracture prevention.

Third, the diagnosis of osteoporosis in this study relied solely on lumbar spine (L₁–L₄) BMD measurements. In older adults, lumbar BMD can be artifactually elevated by degenerative changes, such as osteophytes and aortic calcification, which may lead to an underestimation of the prevalence of osteoporosis compared to hip measurements. We focused on the lumbar spine because trabecular bone is metabolically more active and often shows osteoporotic changes earlier than cortical bone in the hip; furthermore, hip BMD data were not systematically recorded for the entire cohort in our health management system's retrospective database. Consequently, our model may be more sensitive at identifying individuals at risk for vertebral bone loss, and future studies incorporating both spine and hip BMD are needed to improve diagnostic comprehensiveness.

Fourth, although we utilized 1,000-replicate bootstrapping to mitigate over-fitting and ensure internal stability, the absence of external validation remains a significant limitation. While we employed bootstrapping because it is statistically recognized as an efficient method for estimating model optimism and providing stable, bias-corrected estimates in large datasets, we acknowledge that alternative strategies such as a predefined training/validation split or k-fold cross-validation could also have been utilized to characterize internal performance. However, without true external validation in an independent cohort, the generalizability of our nomogram cannot be fully established. The screening performance of the current nomogram may vary when applied to cohorts from different geographical regions or clinical settings. Future multicenter prospective studies are warranted to externally validate the screening value and generalizability of this model across diverse Chinese populations.

Fifth, the present study utilized a single pooled nomogram for both sexes rather than developing sex-stratified models. Given the substantial effect size observed for female sex (OR = 9.66), it is plausible that the underlying risk structure and the relative weight of certain clinical determinants may differ between men and women. We chose a pooled approach to provide a unified, simplified screening tool that is easily applicable for rapid triaging in busy primary care and health management settings. However, we acknowledge that a single model may partially obscure sex-specific risk patterns. Future research involving larger, multi-regional cohorts should explore whether sex-stratified models offer significantly improved accuracy and clinical meaningfulness for personalized risk assessment.

Conclusion

This study identified eight factors associated with osteoporosis in Chinese older adults: sex, age, diabetes, prior fracture, BMI, serum uric acid, serum calcium, and 25-hydroxyvitamin D. The nomogram built from these variables showed satisfactory discrimination and could assist community-level screening and individual risk management, offering a reference for early osteoporosis prevention.

Abbreviations

OP

Osteoporosis

BMI

Body mass index

ROC

Receiver operating characteristic

LASSO

Least absolute shrinkage and selection operator

DCA

Decision curve analysis

Authors’ contributions

M.X. conceived the study, X.N. and L.Z. designed and supervised the study. L.T. contributed materials and analysis tools. L.Z. analyzed the data. M.X. drafted the manuscript. X.C. supervised the study and revised the manuscript. All authors reviewed and approved the final version of the manuscript.

Funding

None.

Data availability

The data utilized and examined in this study are available upon reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

All procedures performed in this study involving human participants were in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. This study protocol was approved by the Medical Ethics Committee of the First Affiliated Hospital of the University of Science and Technology of China (No. 2025-RE-292). Due to the retrospective nature of the study, the requirement for informed consent was waived by the same committee.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.