Predicting Incident Hip Fractures in Chinese People with Type 2 Diabetes using A to G’s: Development and Validation of a Prediction Tool

Abstract

Background and aims: FRAX is known to underestimate hip fracture risk in type 2 diabetes. Recently a hip fracture risk equation, comprising age, gender, BMI, eGFR and peripheral sensory neuropathy (PSN), was developed from a Caucasian-predominant cohort of the Fremantle Diabetes Study Phase I (FDS1) consisting of exclusively individuals with type 2 diabetes, validated in an independent Caucasian cohort. In view of significant regional differences in hip fracture risk profile and that PSN may not be readily assessed especially in the primary care setting, we aimed to develop and validate a Chinese Hip Fracture prediction equation (CHFracture) from a Chinese type 2 diabetes cohort, and compared its performance with the FDS1 hip fracture risk equation. Materials and methods: Between 2008 and 2012, we identified, from the electronic health records in Hong Kong, 84,359 Chinese individuals with type 2 diabetes aged 60-85 years who completed diabetic complication screening during this period and with ≥2 years of follow up. Individuals with prior hip fractures or on anti-osteoporosis medications at baseline were excluded. Incident hip fractures were recorded till the censor date on 31 May 2018. This cohort was randomly divided into the training set (80%, n=67,488) for the development of CHFracture and the testing set (20%, n=16,871) for internal validation. CHFracture was developed by using multivariable Cox regression of incident hip fractures with variables selected based on Bayesian information criterion. The discrimination and calibration of CHFracture was assessed by C-statistics, Greenwood-Nam-D’Agostino statistics, and calibration plot, in comparison with the FDS1 risk equation. Results: 84,359 individuals (53.7% women) were included, with mean age 70.7±6.97 years, BMI 25.3±3.74 kg/m2, HbA1c 7.28±1.23% and duration of diabetes of 9.84±7.92 years. 2,432 individuals had incident hip fractures during a median follow-up of 6.9 years (6.4-7.6), an incidence of 4.1 per 1000 person-years. The clinical characteristics were comparable between individuals in the training set and the testing set. Individuals who had incident hip fractures were more likely to be women, older, smoking, have longer duration of diabetes, lower BMI and eGFR, history of severe hypoglycaemia, fall, fractures, cardiovascular diseases and stroke (all p<0.001). Based on Bayesian information criterion, CHFracture comprised Age, BMI, Cigarette (smoking), Duration of diabetes, EGFR, history of Fall, Gender and history of Stroke (mnemonic: ABCDEFG-S), all readily available in standard diabetes management. CHFracture predicted mean 5-year incident hip fracture risk at 1.6% with good discrimination (C-statistics 0.744, 95% CI 0.733-0.754) and calibration (2=7.995, p=0.535) in the training set. CHFracture had a better C-statistics 0.754 (95% CI 0.733-0.776) than the FDS1 risk equation (C-statistics 0.729, 95% CI 0.706-0.752, p<0.001) in the testing set. Conclusion: CHFracture was developed based on 8 readily available parameters (‘A to G’s), which more accurately predicted incident hip fractures among Chinese individuals with type 2 diabetes, compared with the FDS1 risk prediction equation developed from Caucasian data. CHFracture had the potential clinical utility to identify high-risk individuals with diabetes, for a more aggressive approach to optimize their bone health.