Introduction

The increasing number of patients with diabetes mellitus represents an escalating challenge worldwide [,]. Based on the recently published International Diabetes Federation (IDF) Diabetes Atlas, the recorded worldwide incidence of diabetes among the adult population (20‐79 years) in 2021 was 10.5%, while the prevalence in Southeast Asia was 8.7% []. It is estimated that by the year 2045, the global prevalence of individuals living with diabetes will increase to an estimated 783 million, corresponding to a prevalence rate of approximately 12.2% [], which indicates the urgent need for global diabetes prevention and control. Given the significant prevalence of patients with impaired fasting glucose or undiagnosed diabetes, it is imperative to prioritize the early screening and detection in these patients to mitigate the morbidity associated with diabetes, minimize healthcare expenditures, and prevent a decline in the quality of life [,].

Diabetes mellitus is a metabolic disorder, affecting both glucose and lipid metabolism [,]. Recent studies have investigated the correlation between glucose and lipid-related indexes and the prevalence of diabetes, including the triglyceride-glucose (TyG) index, metabolic score for insulin resistance (METS-IR), triglyceride to high-density lipoprotein cholesterol ratio (TG/HDL-C), and total cholesterol to high-density lipoprotein cholesterol ratio (TC/HDL-C) [-]. However, the accuracy of these indexes remains limited, and better prognostication indexes need to be found. Meanwhile, certain researchers have redirected their attention towards employing computational methods in order to predict diabetes at an early stage with greater precision [-]. The diabetes risk test created by the American Diabetes Association (ADA) serves as a screening tool designed to categorize individuals at a heightened risk within the community to enhance awareness regarding modifiable risk factors and promote the adoption of a healthy lifestyle []. Bang et al [] developed a new screening score using the National Health and Nutrition Examination Survey (NHANES) and a combined cohort of two community studies However, the utilization of diabetes prediction and risk assessment models has been limited due to their predominantly designed focus on White populations. Specifically, there is a scarcity of scoring systems tailored to Japanese populations.

Therefore, in this study, we aimed to develop and validate an easy, systematic index for predicting the diabetes risk in the Asian population using glucose and lipid metabolism parameters and anthropometric parameters, which will enable the clinician or lay persons to assess their own risk of undiagnosed diabetes.

MethodsStudy Population

The study participants were selected from the NAGALA (NAfld [nonalcoholic fatty liver disease] in the Gifu Area, Longitudinal Analysis) database—a human dock that serves as a repository for detecting chronic diseases and their linked risk factors, with the aim of advancing public health initiatives. The cohort has been comprehensively delineated in a prior investigation undertaken by Okamura et al []. The NAGALA cohort was initiated in 1994, encompassing individuals who underwent health examinations at Murakami Memorial Hospital. As a significant portion of the hospital’s screening participants, approximately 60%, return for annual or biennial check-ups, the NAGALA team implemented an extended follow-up study to monitor the emergence of diabetes and nonalcoholic fatty liver disease in this population over time. Furthermore, the available dataset relevant to the investigation cohort has been made accessible on the Dryad public database [] to facilitate the sharing of data.

The present study incorporated the data of 20,944 individuals who were enrolled in the NAGALA cohort between 1994 and 2016. After the further exclusion of participants who were diagnosed with diabetes, impaired fasting glucose, liver disease, excessive drinking, missing data on potential covariates, baseline medication, and unexplained withdrawal from the survey, a total of 15,464 participants were included as the analytical sample.

Ethical Considerations

The data used in this study were obtained from publicly available online databases. As these databases consist of deidentified and aggregated data, they do not involve human subjects or animals. All data were collected from publicly available sources; therefore, ethical approval was not required. Therefore, ethical review and approval were not required for this study in accordance with the ethical standards of the institutional and national research committees, as well as with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Measurement

Demographic data, examination and laboratory test results, medical history, and prescription medication information were used from the NAGALA data. The laboratory test levels included alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), total cholesterol (TC), triglyceride (TG), fasting plasma glucose (FPG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C). The examination and laboratory tests included the evaluation of the BMI, systolic blood pressure (SBP), diastolic blood pressure (DBP), and waist circumference (WC). Regarding exercise behaviors, the participants’ frequency of engaging in physical exercise per week was evaluated, with the criterion for the habit of exercise being defined as more than once per week. Alcohol consumption was categorized as heavy, moderate, light, and none, in line with the quantity and nature of participants’ weekly alcohol intake over the course of the preceding month []. Smoking status was categorized as current smoker, past smoker, and never smoker.

TyG index was calculated as follows: TyG index = Ln [fasting TGs (mg/dL) × fasting glucose (mg/dL) / 2] [].

TG/HDL-C was calculated as follows: TG/HDL-C ratio = TG/HDL-C [].

METS-IR was calculated as follows: METS-IR = Ln ((2 × FPG) + TG) × BMI) / (Ln (HDL-C)) [].

Gastroenterologists made a diagnosis of fatty liver by evaluating specific characteristics observed on abdominal Doppler ultrasonography; these characteristics included vascular blurring, deep attenuation, liver brightness, and hepatorenal echo contrast [].

The diagnosis of diabetes mellitus was established using the criteria set forth by the ADA. Participants were deemed to have diabetes if their HbA1c measurements were ≥6.5% or if their FPG levels were ≥7 mmol/L during the follow-up period. Furthermore, the analysis incorporated the participant’s self-reported diabetes diagnosis during the follow-up period.

Statistical Analysis

If the continuous variables followed a normal distribution, they were represented as the mean and the standard deviation. The variables exhibiting nonnormal distribution were presented as medians accompanied by their respective minimum and maximum values. To evaluate variations between the two groups, the statistical analysis employed the Pearson chi-square test for categorical variables, while the Mann-Whitney U test or Student t-test was utilized for continuous variables. One-way ANOVA was used to compare several groups of data that followed a normal distribution, while the Kruskal-Wallis test was used to analyze data that did not conform to a normal distribution.

A total of 15464 participants were randomly assigned to the training (n=9280, 60%) and internal validation (n=6184, 40%) sets. To establish the index for predicting the incidence of diabetes, univariate Cox regression was applied to detect the variables related to diabetes in the training set. Subsequently, the Cox regression model incorporating the least absolute shrinkage and selection operator (LASSO) was employed to ascertain the coefficients for constructing the model by utilizing an optimal log λ value. The index was established with the following formula: risk score = value of factor 1 × β1+ value of factor 2 × β2+…value of factor n × βn (β was the weighted coefficient of each factor).

Furthermore, an analysis of the receiver operating characteristic (ROC) curve was conducted, and areas under the curve (AUCs) were carried out to assess the predictive value of fasting stress hyperglycemia ratio, FPG, and HbA1c in relation to the risk of diabetes.

The patients were divided into three groups (tertile 1: <33rd percentile, tertile 2: ≥33rd to 67th percentile, and tertile 3: ≥67th percentile) according to the novel index tertiles. Cox proportional hazards regression models were employed to assess the link between the index and the incidence of diabetes. The models included several covariates, namely sex, alcohol consumption, smoking status, habit of exercise, BMI, fatty liver, ALT, AST, GGT, TC, SBP, DBP, TyG, TG/HDL-C ratio, and METS-IR.

The statistical analyses were conducted using SPSS version 23.0 (SPSS Inc.) and R version 3.6.0 (R Foundation for Statistical Computing). A two-sided P-value <.05 was deemed to be statistically significant.

ResultsStudy Population

The current investigation encompassed 15,464 diabetes-free participants, with a mean age of 42 (range 18-79) years, 8430 (54.5%) men and 7034 (45.5%) women, and a mean follow-up of 6.05 (SD 3.7) years. A 6-to-4 ratio was used to randomly allocate patients into the training (n=9280) and validation (n=6184) sets. Table S1 in displays the variations in baseline characteristics of the study population in both the training set and validation set.

Establishment of the Diabetes Incidence Index

A univariate analysis was conducted on the training dataset in order to ascertain the variables that were correlated with the occurrence of diabetes. Eight significant variables, namely age, BMI, WC, HDL-C, TC, TG, HbA1c, and FPG, were employed as predictive factors for progressing diabetes (P<.05; see Table S2 in ). The 8 clinical features were subsequently included in the LASSO regression analysis, which was performed for 1000 bootstrap iterations. From this analysis, 4 features were identified with nonzero coefficients and a minimum lambda value (Figure S1 in ). The findings of multivariate Cox regression analysis indicate that age, WC, HDL-C, TG, HbA1c, and FPG were identified as independent risk factors (Table S3 in ).

To develop a risk index, we constructed a clinical feature-based risk signature derived from the results of a multivariate Cox regression analysis. The risk score was calculated as follows:

[(FPG) × 1.525] – [(HDL-C) × 0.454] + [(TG) × 0.216] + [(HbA1c) × 3.298] + [(WC) × 0.044] + [(age) × 0.016].

We defined this new parameter combined with FPG, HDL-C, TG, HbA1c, WC, and age as the FHTHWA index based on the initial letter of each variable.

Baseline Characteristics

The baseline characteristics of participants were examined based on the FHTHWA index in both the training and validation sets.

For the training set, patients with increased FHTHWA index levels tended to be of advanced age and displayed characteristics indicative of fatty liver, with higher BMI, WC, ALT, AST, body weight, GGT, TC, TG, HbA1c, FPG, SBP, DBP, alcohol consumption, smoking, TG/HDLC ratio, TyG index, and METS-IR, as well as low HDL-C levels (; all P<.01).

Table 1. Baseline characteristics of the study population by tertiles of the FHTHWA index in the training set.CharacteristicsTertile 1 (n=3093)Tertile 2 (n=3093)Tertile 3 (n=3094)P valueSex, n (%)<.001 Male1106 (11.92)1808 (19.48)2174 (23.43)<.001 Female1987 (21.41)1285 (13.85)920 (9.91)<.001Age, years, mean (range)40 (18-69)43 (21-74)47 (19-79)<.001BMI, kg/m2, mean (SD)20.52 (2.36)21.88 (2.70)23.93 (3.28)<.001Waist circumference, cm, mean (SD)70.94 (7.21)76.36 (7.61)82.21 (8.59)<.001Alanine aminotransferase, IU/L, mean (range)15 (4-141)16 (4-129)21 (4-149)<.001Aspartate aminotransferase, IU/L, mean (range)16 (7-98)17 (3-150)19 (6-94)<.001Body weight, kg, mean (SD)54.70 (9.03)60.31 (10.29)66.77 (11.98)<.001Gamma-glutamyl transferase, IU/L, mean (range)12 (4-248)15 (5-399)19 (3-259)<.001High-density lipoprotein cholesterol, mmol/L, mean (range)1.58 (0.55-3.49)1.42 (0-3.21)1.24 (0-3.05)<.001Total cholesterol, mmol/L, mean (SD)4.86 (0.80)5.11 (0.83)5.38 (0.87)<.001Triglyceride, mmol/L, mean (range)0.55 (0.07-3.67)0.73 (0.11-6.75)1.03 (0.12-7.69)<.001Hemoglobin A1C, %, mean (SD)4.89 (0.23)5.15 (0.21)5.46 (0.24)<.001Fasting plasma glucose, mmol/L, mean (SD)4.83 (0.32)5.15 (0.31)5.49 (0.32)<.001Systolic blood pressure, mmHg, mean (SD)108.47 (13.36)114.51 (14.25)120.71 (14.80)<.001Diastolic blood pressure, mmHg, mean SD67.60 (9.42)71.41 (10.14)75.94 (10.28)<.001Fatty liver, n (%)<.001 Yes118 (1.27)394 (4.25)1134 (12.22) No2975 (32.06)2699 (29.08)1960 (21.12)Habit of exercise, n (%).570 Yes529 (5.70)547 (5.89)561 (6.05) No2564 (27.63)2546 (27.44)2533 (27.30)Alcohol consumption, n (%)<.001 Heavy69 (0.74)115 (1.24)126 (1.36) Moderate220 (2.37)301 (3.24)305 (3.29) Light318 (3.43)366 (3.94)367 (3.95) None2486 (26.79)2311 (24.90)2296 (24.7)Smoking status, n (%)<.001 Current536 (5.78)736 (7.93)816 (8.79) Past393 (4.23)627 (6.76)765 (8.24) Never2164 (23.32)1730 (18.64)1513 (16.30)TG/HDL-C ratio, n (range)0.79 (0.11-11.35)1.16 (0.13-15.48)1.89 (0.19-26.73)<.001Triglyceride-glucose index, mean (SD)7.68 (0.56)8.02 (0.57)8.42 (0.60)<.001Metabolic score for insulin resistance, mean (SD)27.45 (4.55)30.63 (5.47)35.52 (6.75)<.001FHTHWA index, mean (SD)0.20 (0.12)1.01 (0.43)12.60 (24.44)<.001Follow-up duration, years, mean (SD)6.79 (3.86)5.88 (3.79)5.47 (3.57)<.001Progressed to diabetes, n (%)<.001 Yes4 (0.04)19 (0.20)190 (2.05) No3089 (33.29)3074 (33.13)2904 (31.29)

aTG/HDL-C: triglyceride to high-density lipoprotein cholesterol.

For the validation set, patients with increased FHTHWA index levels tended to be of advanced age and displayed characteristics indicative of fatty liver, with higher BMI, WC, ALT, AST, body weight, GGT, TC, TG, HbA1c, FPG, SBP, DBP, alcohol consumption, smoking, TG/HDLC ratio, TyG index, and METS-IR, as well as low HDL-C levels (Table S4 in ; all P<.05).

Predictive Value of the FHTHWA Index

Time-dependent ROC curves were utilized in order to assess the predictive efficacy of the FHTHWA index, as well as other newly proposed predictive parameters such as TG/HDL-C, the TyG index, and METS-IR, in relation to the risk of developing diabetes in the future. shows the AUC of the FHTHWA index and other novel predictive parameters over time for the training set. Overall, the baseline FHTHWA index demonstrated strong predictive capabilities for future diabetes occurrence at various time intervals. The AUCs of the FHTHWA index were higher compared with the AUCs of the TG/HDL-C, TyG index, and METS-IR during the 12-year follow-up. shows the AUC of the FHTHWA index and other novel predictive parameters over time for the validation set. Overall, the baseline FHTHWA index demonstrated strong predictive capability for future diabetes occurrence at various time intervals. Furthermore, the FHTHWA index exhibited greater AUC values compared to the TG/HDL-C, TyG index, and METS-IR at all time points.

‎Figure 1. Time-dependent receiver operator characteristic curves of the FHTHWA index vary with time to predict the future risk of diabetes. (A) Area under the curve (AUC) comparison of the FHTHWA index, TG/HDL-C ratio, TyG index, and METS-IR in the training set. (B) AUC comparison of the FHTHWA index, TG/HDL-C ratio, TyG index, and METS-IR in the validation set. METS-IR: metabolic score for insulin resistance; TG/HDL-C: triglyceride to high-density lipoprotein cholesterol ratio; TyG index: triglyceride-glucose index. Relationship of the FHTHWA Index With Diabetes Risk

In the training set, 213 (1.38%) cases progressed to diabetes during the follow-up period. From quartile 1 to quartile 3 of the FHTHWA index, the diabetes rate increased gradually from 0.04% to 2.05% (). After multivariate adjustment, including sex, consumption of alcohol, smoking status, habit of exercise, BMI, fatty liver, ALT, AST, and GGT, an elevated FHTHWA index was markedly correlated with diabetes risk (). Multivariate adjusted hazard ratios and 95% CIs from the lowest to the highest FHTHWA index categories (tertile 1, tertile 2, and tertile 3) were 1.00 (reference), 4.509 (1.520‐13.373), and 32.141 (11.545‐89.476), respectively, for the incidence of diabetes (P<.001). The analysis of weighted Kaplan-Meier curves demonstrated significant dose-response patterns within the FHTHWA index strata in relation to the incidence of diabetes. These trends persisted even after a 12-year follow-up period. Furthermore, the risk of developing diabetes across the FHTHWA index strata displayed a noticeable divergence over time ().

Table 2. Hazard ratios and corresponding 95% CIs for diabetes incidence according to the FHTHWA index among the population without diabetes in the training set.FHTHWA indexP trendPer unit increment in the FHTHWA indexTertile 1Tertile 2Tertile 3Number of cases of diabetes mellitus/total4/309319/3093190/3094Crude1.000 (ref.)5.846 (1.988‐17.158)66.098 (24.542‐178.020)<.0011.019 (1.018‐1.021)Model 11.000 (ref.)5.720 (1.914‐16.856)64.129 (23.679‐173.677)<.0011.020 (1.018‐1.022)Model 21.000 (ref.)4.975 (1.684‐14.695)39.714 (14.459‐109.084)<.0011.017 (1.015‐1.019)

aModel 1: adjusted for sex, alcohol consumption, smoking status, and habit of exercise.

bModel 2: further adjusted (from Model 1) for BMI, fatty liver, ALT, AST, and GGT.

‎Figure 2. Weighted Kaplan-Meier curves for diabetes incidence by the baseline FHTHWA index strata. (A) Diabetes risk across the FHTHWA index strata in the training set exhibited divergence over time. (B) Diabetes risk across the FHTHWA index strata in the validation set exhibited divergence over time.

In the validation set, 160 (2.59%) cases progressed to diabetes during the follow-up period. From quartile 1 to quartile 3 of the FHTHWA index, the diabetes rate increased gradually from 0.16% to 2.25% (Table S4 in ). After multivariate adjustment, including sex, consumption of alcohol, smoking status, habit of exercise, BMI, fatty liver, ALT, AST, GGT, TC, TG, SBP, DBP, TyG, TG/HDL-C ratio, and METS-IR, an elevated FHTHWA index was markedly correlated with diabetes risk (Table S5 in ). Multivariate adjusted hazard ratios and 95% CIs from the lowest to the highest FHTHWA index categories (tertile 1, tertile 2, and tertile 3) were 1.00 (reference), 1.010 (0.419‐2.435), and 9.924 (4.830‐20.392), respectively, for the incidence of diabetes (P<.001). The utilization of weighted Kaplan-Meier curves demonstrated significant dose-response patterns across different strata of the FHTHWA index in relation to the incidence of diabetes, even after a follow-up period of 12 years. Moreover, the risk of developing diabetes across the FHTHWA index strata exhibited a noticeable divergence over time ().

Stratification Analyses

In the stratified analyses of the dose-response association between FHTHWA index levels and diabetes risk, participants were stratified by sex (male or female), habit of exercise (yes or no), BMI (<25 kg/m2 or ≥25 kg/m2), current smoker (yes or no), and current alcohol use (yes or no). In the training set, we discovered that a high FHTHWA index was linked with a higher diabetes risk, contrasted with the reference group in all subgroups (). In the validation set, the results were the same as our main results (Table S6 in ).

Table 3. Stratified analyses of the associations (hazard ratios, 95% CIs) between the FHTHWA index and diabetes incidence at the end of follow-up among people with diabetes in the training set.VariablesHazard ratio (95% CI)Event, n/NUnadjusted modelMultivariable-adjusted modelSexMale157/50881.019 (1.017‐1.021)1.020 (1.017‐1.023)Female56/41921.019 (1.017‐1.022)1.013 (1.009‐1.017)Habit of exerciseYes35/16371.036 (1.030‐1.042)1.028 (1.019‐1.037)No178/76431.019 (1.017‐1.020)1.015 (1.013‐1.018)BMI, kg/m2<25113/77521.020 (1.018‐1.023)1.020 (1.015‐1.025)≥25100/15281.016 (1.013‐1.018)1.015 (1.012‐1.017)Current smokerYes88/20881.032 (1.027‐1.036)1.030 (1.024‐1.036)No125/71921.019 (1.017‐1.020)1.015 (1.012‐1.017)Current alcohol useYes56/21871.017 (1.014‐1.020)1.024 (1.018‐1.030)No157/70931.020 (1.018‐1.022)1.014 (1.012‐1.017)

aHazard ratios (95% CIs) were assessed using weighted Cox proportional regression that was fully adjusted except for the stratification factor.

DiscussionStudy Findings and Comparison With Previous Works

Diabetes mellitus can lead to a range of complications that result in significant psychological and physical distress for individuals who carry the disease. Moreover, this places a substantial strain on the healthcare systems [,]. Nevertheless, the early detection of this condition is often hindered due to the absence of distinct symptoms and a limited focus on public healthcare. Early identification of the population with high risk for diabetes might help prevent the onset of diabetes or delay its progression []. In this study, we developed and validated a robust parameter, named the FHTHWA index, to forecast the development of the future diabetes likelihood depending on a comprehensive longitudinal cohort study of the Japanese physical examination population. Our results revealed that the higher the FHTHWA index, the higher the risk of developing diabetes. The correlation exhibited independence from conventional risk factors, such as sex, alcohol consumption, smoking status, habit of exercise, BMI, fatty liver, SBP, DBP, TyG, TG/HDL-C ratio, and laboratory parameters.

The early identification of individuals at high risk for diabetes is crucial for implementing preventive strategies and mitigating the burden of this disease. A growth of risk assessment tools for type 2 diabetes mellitus has been observed on a global scale, such as the ADA risk score [], the Canadian Diabetes Risk Score (CANRISK) [], the Finnish Diabetes Risk Score (FINDRISC) [], the Indian Diabetes Risk Score (IDRS) [], and Indonesian Undiagnosed Diabetes Mellitus (UDDM) scoring system []. Both the ADA and CANRISK tools have demonstrated efficacy in identifying individuals at risk within their respective communities; nevertheless, these products were specifically created to cater to the diverse populations residing in the United States and Canada, respectively. The validation of the IDRS took place in a low- to middle-income country. The IDRS is a valuable screening tool specifically for the mono-ethnic Indian population []. The scoring system originated in Indonesia. However, its accuracy has not been assessed through prospective evaluation []. Most of the predictive indices are genetic, demographic characteristics, and clinical parameters. With the intensive study of the pathogenesis of diabetes, many studies have begun to identify the genetic associations associated with the onset of diabetes. These studies not only reveal the genetic basis of diabetes but also provide new perspectives for understanding its pathogenesis. For example, recent genome-wide association studies identified multiple genetic variants associated with type 2 diabetes that may affect insulin secretion and islet β-cell function []. Furthermore, studies show that certain genetic variants are strongly associated with the age of onset of diabetes, and younger patients with the onset of diabetes often face a higher risk of cardiovascular disease and all-cause mortality []. The effects of genetic variation also vary in different populations. For example, studies in East Asian populations found that certain genetic variants associated with type 2 diabetes mellitus showed different effects in this population, underscoring the importance of genetic studies in different ethnic groups []. Moreover, studies targeting monogenic diabetes have also revealed genetic mechanisms associated with autoimmunity that may play an important role in polygenic diabetes []. It is very important to consider incorporating genes into predictors in future studies.

Moreover, most tools are based on the general population in defined regions and may not be generalizable to populations from different regions [,]. Currently, there is still a lack of widely used diabetes risk tools for the Japanese population []. The participants in this study were selected from a longitudinal cohort study that specifically targeted those undergoing health check-ups. This study commenced in 1994 and involved the continuous recruitment of individuals who underwent health examinations at Murakami Memorial Hospital. Thus, the study population typically included Japanese or Eastern Asian populations. The genetic constitution and lifestyle of Asian populations are different from those of European or American populations. Thus, the FHTHWA index might be suitable for the Japanese general population and other East Asian populations.

Insulin resistance (IR) is a medical condition characterized by reduced cellular responsiveness to insulin, independent of hyperinsulinemia []. IR serves as the primary pathophysiological mechanism underlying the developing type 2 diabetes mellitus, and it has the potential to manifest in individuals up to 2 decades prior to the diagnosis []. IR serves as a pivotal indicator of the imminent onset of diabetes. Evidence indicates that visceral adiposity indexes, such as BMI and WC, possess the capacity to effectively detect IR [,]. Moreover, evidence supports the effectiveness of classic lipid indices, including TC, TG, HDL-C, and LDL-C, in diagnosing IR [,]. However, the predictive capacities of these factors are limited []. Recently, the TyG index and TG/HDL-C have been recommended as diabetes risk indicators, both of which have better performance than traditional indicators (BMI, WC, TC, TG, HDL-C, and LDL-C) [,]. In this study, we performed a comparison of AUC values by time-dependent ROC analyses between the FHTHWA index, TyG index, TG/HDL-C, and METS-IR. It was found that the FHTHWA index had high accuracy, with the AUC value greater than 0.9 during the more than 12 years of follow-up. However, the AUC values of the TyG index, TG/HDL-C, and METS-IR mostly ranged from 0.7 to 0.8. Thus, the novel index has a higher prediction accuracy than the traditional parameters.

Study Strengths

This study’s primary strengths include the large sample size of 15464 individuals ranging from 18‐79 years of age, who underwent thorough examination and were subsequently monitored over an extended duration of up to 13 years. This sample was representative of the entire nation, enabling the findings to be extrapolated to the broader Japanese or East Asian populations.

Study Limitations

The present study has certain limitations. First, the primary focus of this study was the onset of diabetes as the outcome of interest. Nevertheless, it is important to acknowledge that the diagnostic criteria utilized in this investigation did not include the participants’ 2-hour postprandial blood glucose level. This has the potential to result in an underestimation of the incidence of diabetes. Second, due to the utilization of the NAGALA database, it was impossible to eliminate self-reported characteristics, resulting in a potential reporting bias. Consequently, the findings of this study can only be generalized to the Japanese or East Asian population. It is crucial to determine the degree to which these findings can be extrapolated to heterogeneous populations. Third, it is important to note that the primary outcome measure of this study is the occurrence of new-onset diabetes events. However, it is noteworthy to emphasize that the dataset utilized in this study does not include information on death events that may have occurred during the follow-up period. This limitation introduces a potential competing risk that could impact the findings of the current research.

Conclusion

The research has successfully developed the FHTHWA index, which demonstrates a high degree of accuracy in predicting the future risk of diabetes within the Japanese population without diabetes. This index is particularly valuable for clinical practitioners in Japan, as it is recommended for regular use in assessing the risk of diabetes and monitoring therapeutic progress among the Japanese population.

We would like to thank Prof. Okamura Takuro and his team at Kyoto Prefectural University of Medicine for sharing their data. This study was supported by grants from the Jiangxi Provincial Key R&D Plan (no. 20201BBG71006), the Natural Science Foundation of Jiangxi Province (no. 20224BAB216009), and Technology Plan of Jiangxi Health Planning Committee (no. 202210484).

All the data used in this study were obtained from the Dryad Digital Repository database (https://datadryad.org/stash/dataset/doi:10.5061/dryad.8q0p192).

Edited by Christian Lovis; submitted 01.08.24; peer-reviewed by Feier Song, Viswanathan Mohan; final revised version received 14.11.24; accepted 17.11.24; published 28.01.25.

© Jiao Wang, Jianrong Chen, Ying Liu, Jixiong Xu. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 28.1.2025.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on https://medinform.jmir.org/, as well as this copyright and license information must be included.