The global rise in type 2 diabetes (T2DM) has profoundly impacted healthcare systems, especially in low- and middle-income countries where over 75% of diabetes cases occur [1].

Prediabetes — a condition marked by blood sugar levels higher than normal but not yet at diabetic levels — is a critical precursor to T2DM, often progressing to full-blown diabetes if left unaddressed [2][3]. Individuals with prediabetes face heightened risks for complications like nephropathy, neuropathy, and macrovascular diseases [3–5]. Studies show that up to 70% of those with prediabetes eventually develop T2DM, sometimes within five years [2][6]. The American Diabetes Association (ADA) defines prediabetes using HbA1c levels between 5.7% and 6.4% or fasting plasma glucose levels between 100 and 125 mg/dL [7–11]. However, there is no universal consensus on the HbA1c range that best identifies high-risk individuals, as different recommendations vary across organizations [12][13]. In the Asia-Pacific region, guidelines advocate for screening and intervention for those aged 35 and older or individuals at high risk, using laboratory tests such as FPG, HbA1c, and the 75-gram OGTT to support early detection and potential prevention of T2DM [14]. Despite these screening protocols, diabetes risk assessment tools differ significantly in sensitivity, specificity, and suitability across diverse populations [9][15–18].Previous research identifies several key risk factors for prediabetes including sex (female) [19–22], age (45 years or older) [19][21][23][24], Body Mass Index (overweight or obese) [21][25], waist circumference and family history of diabetes [26], high blood pressure [22], polycystic ovary syndrome in women [27][28], dyslipidemia [29], psychological factors such as stress or depression [6], lifestyle behaviors like smoking or tobacco use and alcohol consumption [22][30], poor dietary habits [28][31–33], and physical inactivity [34]. This scoping review aims to summarise existing research on diabetes risk screening tools systematically, identifying knowledge gaps to support potential shifts toward population-wide screening within community-based programs. Specifically, this review examined the characteristics, development methods, and effectiveness of diabetes risk assessment tools for identifying prediabetes and monitoring progression to diabetes in at-risk individuals.

This review systematically identified and synthesized studies on diabetes risk assessment tools (Fig. 1). The methodology followed PRISMA-ScR guidelines, accessible at http://www.prisma-statement.org/ and https://www.prisma-statement.org/scoping [35] (see PRISMA checklist in the supplementary file). The protocol was reviewed and revised with input from the advisory board of the Community Medicine Division at Thammasat University; a Principal Research fellow from Southampton Health Technology Assessments Centre (SHTAC), School of Healthcare Enterprise and Innovation, University of Southampton, UK; and members of the research team.

Eligibility Criteria:

1) Inclusion: Studies on diabetes risk tools for adults with prediabetes or T2DM risks (18+), examining factors affecting tool adoption, implementation, or validation.

2) Exclusion: Studies lacking relevant data, focusing on genetic factors, or not classified as original research. Also excluded were studies focusing on children, pregnant individuals, or unrelated conditions.

A systematic search across PubMed, ScienceDirect, and Google Scholar used Boolean operators, MeSH terms, and synonyms (e.g., "diabetes risk assessment", "prediabetes screening", "adults", "non-invasive tools") (Appendix A, B). The timeframe covered October 2012 to September 30, 2022, and October 1, 2022 to September 2024.

Studies were selected through:

Title & Abstract Screening: Based on predefined criteria.

Full-Text Review: To confirm eligibility.

Two reviewers independently extracted data using a standardized QUADAS-2 form. Discrepancies were resolved by a third reviewer. Of 843 identified studies, 14 met inclusion criteria (see Figure 1 for the PRISMA flow chart summarising the selection process). These studies included 26 risk models (e.g., THIARISK, CANRISK, FINDRISC, ADA-Risk). Data on AU-ROC, sensitivity, specificity, and validation efforts were summarized in tabular format for comparison.

As this is a scoping review, for which risk of bias assessment is regarded by JBI as optional rather than mandatory [36], we used a pragmatic approach to the risk of bias assessment. That is, we selected a subset of what we considered to be the most important risk of bias questions from a relevant risk of bias tool for screening studies. The QUADAS-2 tool [37][38], was used to assess bias across four domains: Patient Selection, Index Test, Reference Standard, and Flow and Timing. For comparative accuracy, QUADAS-C [39], was applied where relevant. Additional quality considerations included sampling and reporting bias.

Assessment Process: 1) Independent Review: Two reviewers evaluated study design rigor, using adapted QUADAS-2 criteria for diabetes risk assessment. 2) Study Categorization: Included 7 cross-sectional, 6 cohort, and 1 case-control study. 3) Bias Judgment: Each study was rated as low, high, or unclear risk in QUADAS-2 domains. 4) Comparison with Previous Reviews: Unlike earlier findings [9][18], with 87.8% high risk of bias, our review found all 14 studies had low risk of bias, highlighting their methodological robustness. All studies were deemed low risk of bias and included in the final analysis (Table 2 in Appendix C).

This section may be divided by subheadings. It should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn.

This review includes 14 studies conducted across 11 countries, spanning the years 2009 to 2018. These studies were drawn from five databases and nine community-based settings, providing a diverse geographic and methodological landscape. The study designs included one case-control study [40], seven cross-sectional studies [41–47], and six cohort studies [18][31][48–51]. Participants were adults aged 18 and older who were identified as being at high risk for diabetes or prediabetes.

The types of diabetes risk assessment tools varied, with the following tools and models being assessed (Table 1):

1) Risk Screening Tools: Thai-RISK, CANRISK (3 studies), FINDRISC (3 studies), ADA, IDRS, UDD, and Filipino risk scores.

2) Comparative Models: OGTT, CANRISK Original, FBG, ADA, CDA, FPG, and A1C were used to benchmark and compare the efficacy of these risk tools.

3) The "gold standard" criteria or diagnostic cut-off points applied in these studies included CANRISK Original, ADA, CDA, FPG, OGTT, and A1C, allowing for standardized evaluation of risk assessment performance.

Developing reliable diabetes risk assessment tools is essential for early detection and diagnosis in adults with high-risk impaired glucose, intermediate hyperglycemia, or prediabetes. Low-cost, simple tools such as paper-based risk questionnaires or anthropometric measurements allow for easier identification of at-risk individuals who may benefit from lifestyle interventions. Common tools for assessing prediabetes or diabetes risk include the CANRISK, FINDRISC, and ADA Risk Questionnaires. The development of these tools generally falls into three main categories:

Non-invasive risk score assessments: These tools estimate diabetes risk based on non-invasive parameters, such as BMI, age, and lifestyle factors [42–44][47].

Index tests for diabetes risk scores: These studies investigate the utility of diabetes risk scores as diagnostic tools [40][41][45][46][50].

Predictive models for incident diabetes: These models focus on predicting the future development of diabetes in individuals identified as at risk [18][51].

In summary, this review outlines three key findings regarding diabetes risk screening tools (Table 2):

These findings highlight the need for tailored, effective risk assessment tools that are both accessible and accurate, enabling early intervention in at-risk populations.

Across 14 studies, 8.69 million participants were analyzed, with 53% female and a mean age of 38.4 years (SD=7.7). Risk assessments included 5 to 15 factors (median: 11), categorized as follows (Table 3):

Among 6.49 million (74.7%) who completed risk assessments: 3.1% (198,968) were identified as high-risk for prediabetes, 2.8% (180,639) were diagnosed with diabetes during follow-up. Among diagnosed cases, the average diabetes duration was 5.5 years.

Key risk factors in diabetes screening tools include age, BMI, history of diabetes (HxDM), hypertension (HT), and waist circumference (WC). Some tools incorporate additional factors (Table 4):

Tool Performance:

Risk Stratification:

The tools that had more risk factors had lower % of people identified at high risk of pre-DM, suggesting that adding more risk factors doesn't improve prediction.

The 14 studies included a range of screening tools with different cutoffs and several different reference (i.e. gold) standards, with limited repetition of these combinations of tools, cutoffs and reference standards across the studies (Table 1 and Table 2). This makes it difficult to determine how generalisable the findings from individual studies would be. Overall, 26 screening tools were assessed across the 14 studies.

Across all studies and all combinations of the screening tools and their corresponding reference standards, sensitivity of the tools for detecting people with prediabetes or those at high risk of developing diabetes ranged from 54% to 98% whilst specificity ranged from 31.5% to 100% and the PPV ranged from 10% to 100% (Table 2). Thus, none of the tools studied optimised both sensitivity and specificity and, in all cases, the PPV (i.e. the probability that someone with a positive screening test result would actually have prediabetes or a high risk of developing diabetes) was relatively low. The screening tools with the highest sensitivity for identifying people with prediabetes or a high risk of developing type 2 diabetes were ADA Risk Tes with a ≥9  cutoff, FINDRISC ≥15 cutoff and IDRS with a ≥60 cutoff (94% and 92% sensitivity respectively) but in the corresponding specificity was relatively low (43%, 45% and 67% respectively) (Table 2).

There is a strong need for more sensitive and specific tools to identify prediabetes effectively. While CANRISK and FINDRISC show potential for early detection, greater internal validation is essential to ensure their reliability across populations. Strengthening validation efforts can enhance screening accuracy, guiding health providers and policymakers in developing targeted preventive strategies for high-risk groups.

This review analyzed one case-control study, seven cross-sectional studies, and six cohort studies focused on tool development and validation.

This scoping review provides a comprehensive synthesis of diabetes risk screening tools for identifying adults at risk of prediabetes or type 2 diabetes mellitus (T2DM), as follow each objective of the study;

Characteristics of Diabetes Risk Screening Tools

We identified 14 studies representing 26 risk models, including widely used tools such as CANRISK, FINDRISC, and ADA-Risk, with considerable variability in sensitivity, specificity, and applicability across populations [42–44][47]. The majority use indirect predictors such as BMI, family history of diabetes, age, waist circumference and physical activity [44][49]. Similarly, Rowan et al. [45]. Studies show that the standard CANRISK questionnaire, with a cut-off score of 33 points, achieves good accuracy, while a lower cut-off of 21 points significantly increases sensitivity [52]. However, adjusting BMI and waist circumference cut-off points for ethnicity did not enhance predictive accuracy. A risk score based on factors such as sex, age, waist circumference, hyperglycemia history, and family diabetes history, with scores ranging from 0 to 15, was deemed effective for assessing prediabetes or diabetes risk [40][42]. Women with impaired fasting glucose (IFG) are often underdiagnosed, and using OGTT may improve prediabetes detection, especially among women aged 45 and older [41]. Between simplicity and utility are some of the tools (eg, Thai-RISK and CANRISK) that were developed with restricted external validation so their extrapolation outside the study population should be with caution [40][41][45][46][50]. A model combining FBG and OGTT (C=0.70) proved effective for diabetes risk assessment [31]. However, models based on IFG (WHO criteria) had lower sensitivity (67.7%) and specificity (77.9%) than expected.

The effectiveness of various tools in predicting T2DM was assessed through their ability to classify individuals based on risk factors like impaired glucose tolerance (IGT) and impaired fasting glucose (IFG). For instance, the CANRISK tool has demonstrated success in identifying impaired glucose in Canada’s First Nations and Métis communities, as well as in Saudi Arabia [40][41][45][46][50]. Jieng’s study suggested that OGTT was superior to FPG for risk prediction [43], and Norway’s community pharmacies effectively utilized HbA1c testing to identify undiagnosed T2DM cases. The QDiabetes-2018 algorithm further showed strong predictive capabilities for 10-year risk of T2DM, confirming its feasibility for use in community screening [41][51]. Screening through FINDRISC in pharmacy settings proved feasible, highlighting the utility of accessible locations for reaching at-risk populations. Suggested model cut points include A-Q Diabetes (no FBG or A1c), B-FBG, and C-A1C [48]. For 10-year diabetes risk prediction, Model B_FBG provided the best performance, effectively identifying individuals needing intervention or more intensive follow-up [48]. Other studies found the risk score approach feasible and effective for assessing diabetes risk in community and pharmacy settings, allowing both pharmacists and participants to engage in the screening process [44][50]. Simple, cost-effective questionnaires are valuable for raising awareness and assessing type 2 diabetes risk [49]. Aside from the SADRISC tool and ADA Risk Test, which were each included in two studies, all other diabetes screening tools were only included in one study and so their sensitivity, specificity and PPV in a wider range of settings is unknown. Due to the different tools and reference standards used across the studies a quantitative meta-analysis of the studies is not feasible. Our risk of bias assessment using QUADAS2 suggested that the studies were broadly similar in their risk of bias, but the studies differed in a range of factors, including the country, study design, age range, and risk factors of the populations (Table 1). These factors likely contribute to the observed heterogeneity of the sensitivity, specificity and PPV results but are difficult to test for as explanatory variables in this evidence base due to the limited occurrence of each set of variables across the studies.

Cohort studies primarily examined the progression of individuals with prediabetes to diabetes, emphasizing IFG and Metabolic syndrome (MetS) as significant risk factors [18][31][51]. Kaneko et al. found that IFG held a higher population-attributable fraction (PAF) than MetS in predicting T2DM incidence among middle-aged Japanese participants [25], and the coexistence of IFG and MetS showed the highest risk [9]. This finding suggests that IFG could be a valuable marker for diabetes risk, especially when used in combination with MetS criteria. In Saudi Arabia and Algeria, basic assessment tools have effectively evaluated diabetes risk in the population. Across studies, socioeconomic factors such as age, education, and marital status, as well as biomarkers like blood pressure and lipid levels, were found to significantly impact diabetes progression [18][31][51].

In a five-year period, 15.9% of participants with impaired glucose tolerance (IGT) or coronary heart disease (CHD) progressed to diabetes, underscoring the importance of targeted interventions. A risk prediction model utilizing clinical variables readily available in routine practice can help estimate diabetes risk in specific populations, such as Chinese individuals with CHD or IGT. Despite some studies having small sample sizes, the general recommendation is for settings to raise diabetes risk awareness for individuals scoring 9 or higher on the ADA risk test [18][47]. If the intention is primarily to identify people with prediabetes or at high risk of developing type 2 diabetes then the relatively low specificity and PPV values may not be a concern, provided that there are no negative issues associated with false positive results (such as the costs of testing or of the management of patients who receive a negative diagnosis).

This review mapped the field of diabetes risk screening tools and emphasized the importance of more systematic validation, particularly in different populations. A range of different screening tools has been tested that could identify people with prediabetes or a high risk of developing type 2 diabetes. However, where sufficient evidence was available to compare tools across studies the performance of these tools was inconsistent. Several tools have only been investigated in single studies, with uncertainty around their wider generalisability. Clinicians or researchers wishing to screen people for prediabetes or a high risk of developing type 2 diabetes using any of these tools should be aware of their potential limitations.

Limitations: The limitations of this review are that it only covers English and Thai literature; possible publication bias due to the fact that grey literature is not included; and that no meta-analysis was conducted. Although this is to be expected in scoping reviews, the reporting of pooled performance ranges in our tables serves to narrow the knowledge gap and achieve higher interpretability.

Strengths: Diverse Geographic Coverage: Findings are generalizable across different populations. Comprehensive Tool Review: Analyzing 14 screening tools provides a broad comparative perspective. Quality Assessment: QUADAS-2 ensures structured evaluation, enhancing reliability. Clinical Relevance: Identifies practical, non-invasive risk factors (e.g., age, BMI) for easy application in healthcare settings.

Source of funding. This research received no external funding.

Conflict of interests. Authors declare no explicit and potential conflicts of interests associated with the publication of this article.

Authors involvement. P.W.: Conceptualization, methodology, investigation, resources, data curation; P.W., S.D.; validation, formal analysis, P.W.; S.D., and TTN.; writing-original; All authors read and final proof the manuscript.

Acknowledgements. We thank for our information specialists and all advisory board of the community medicine division, Thammasat University, for their help in developing the search strategy and for selecting databases. Special thanks to Dr. Geoff K. Frampton, the Principal Research fellow from the School of Medicine, University of Southampton, UK. and Associate Prof. Dr. Omid Dadras, Queensland Ambulance Services, Queensland Government Department of Health, Brisbane, Australia.