It is evidenced that people with the family history of diabetes are 2–8 fold more likely to develop diabetes (23). This association was independent of other risk factors, such as obesity, insulin resistance, and lifestyle factors which means this risk factor solely is strong enough to predict diabetes (5). Therefore, in the present study we have focused on this at-risk subgroup of population and developed a reliable risk score for future development of diabetes and prediabetes.
In this study, higher WHR, Cholesterol, TG, Lipid index, and the OGTT-AUC at the baseline in those FDRs who developed diabetes after 16 years were related to the increased risk of developing diabetes. Unfavorable clinical factors to develop prediabetes after this period of time were age, waist circumferences, OGTT-AUC, HbA1c, cholesterol, TG, and the lipid index.
According to these factors, we developed two hazard models to predict the risks of developing diabetes and prediabetes, in which OGTT-AUC or FPG were used in each one (Figs. 3 and 4). Diabetes risk models included WHR, HbA1c, lipid profile and OGTT-AUC or FPG. The best diabetes predictive ability was obtained by a model, which included OGTT-AUC. The AUC of this model was 0.71 (0.66–0.77). The other model included FPG instead with the AUC of 0.69 (0.63–0.74). According to diabetes risk model 1, individuals with score values more than 5.57 were determined as high-risk for developing diabetes. Although the predictive efficiencies of both models were higher than other plasma glucose indices, the predictive ability of the OGTT-AUC alone was comparable to the best developed predictive model in our study. The OGTT-AUC cut point for diabetes prediction is blood glucose more than 7.8 and 7.2 mmol/L at 30 and 60 minutes, respectively.
Prediabetes risk model 1 included HbA1c, SBP, lipid, and OGTT-AUC with the AUC of 0.63 (0.59–0.67). Prediabetes risk model 2 was based on age, waist circumferences, HbA1c, lipid index, and FPG with the AUC of 0.60 (0.57–0.64). Based on prediabetes risk model 1, the score value more than 2.96 is a wakeup call for the onset of prediabetes. In comparison with other plasma glucose indices, the AUC of all prediabetes models were significantly higher than FPG and HbA1c. However, there was no considerable difference in predictability of the OGTT-AUC alone, and prediabetes model 1. Similarly, in previous studies (24, 25), no further improvement in model predictability of developing diabetes achieved by adding other clinical factors. For instance, in Framingham Offspring study (24), the initial model was based on age, gender, parental history of diabetes, BMI, waist circumference blood pressure, HDL, Triglyceride, FPG. They observed no further improvement in diabetes prediction ability by adding 2h-OGTT, fasting insulin level, log Gutt insulin sensitivity index, HOMA index, and C-reactive protein level to the models. Moreover, the ARIC’s (25) model for predicting diabetes, based on non-invasive parameters including waist, height, hypertension, blood pressure, family history of diabetes, ethnicity, and age, performed similar to fasting glucose alone (AUCs were 0.71 and 0.74, respectively). On the other hand, another model composed of the non-invasive parameter plus FPG (AUC 0.78) and the model including FPG, triglycerides and HDL (AUC 0.80) had better predictability. This diversity in influential risk factors in the final models might be resulted from the diversity in population. Diabetes risk scores demonstrated good predictability in the original populations in which they were derived. However, their predictive values were usually reduced in external populations (26). Therefore, it was suggested that to develop population-specific risk prediction tools (27).
The FINDRISC (12) is one of previous models, which followed their participants up to 10 years and also included blood-based metabolic characters in the model. FINDRISC was developed based on age, BMI, waist circumference, antihypertensive drug therapy, and history of high blood glucose levels. According to this model, the diabetes risk score value ranked from 0 to 20. The predictive value of the model was the AUC of 85% with 77% sensitivity, and 66% specificity at the score 9 (12). The data was collected through a yes/no questioner and self-reporting data. In the present study, biochemical tests have been conducted on all participants at the baseline and during the follow-up, which provide more accurate data.
In a previous study (26), we evaluated the validity of the concise FINDRISC to predict type 2 diabetes in our population (i.e. the first degree relatives of patients with type 2 diabetes who have normal glucose tolerance) (28). The predictive ability of the FINDRISC in our population was lower than Finish population. In this study we compared the ability of the FINDRISC and our diabetes models to predict the onset of diabetes and results confirm that the predictive performance of our diabetes models is more precise than the FINDRISC in our population.
The results of this study need to be interpreted in light of its strengths and weaknesses. The advantages of our study are: (1) the large sample size (n = 1765), (2) long-term follow-up, (3) valid diagnose of diabetes and prediabetes by FPG and OGTT criteria. The limitation of our study is that it was conducted in a single urban city in Iran. As risk factors, prevalence, and progression to diabetes may well differ in other cities and rural areas, so the results should be handled with caution before they can be generalized to the rest of the country.
In conclusion, developing simple assessment models for the target population is a first step to identify FDRs of patients with diabetes with an increased likelihood of developing diabetes or prediabetes. In the context of health checkups OGTT-AUC is strong enough for predicting the future risk of diabetes and prediabetes. Moreover, present study evidenced that the diabetes model 1 have the best performance in identification of future risk of developing diabetes compared to FINDRISC and diabetes model 2 in our population.