1. Introduction
Diabetes mellitus is one of the most prevalent chronic diseases worldwide, and its global burden continues to rise. According to the International Diabetes Federation, an estimated 589 million adults aged 20 to 79 years were living with diabetes in 2024, a figure projected to reach 853 million by 2050.[1] Beyond hyperglycemia, diabetes is associated with major complications, including cardiovascular disease, chronic kidney disease, retinopathy, and neuropathy.[2]
Gestational diabetes mellitus (GDM), defined as diabetes first recognized during pregnancy, affects approximately 10% to 20% of pregnant women depending on the population and diagnostic criteria.[2,3] GDM is associated with increased maternal risks, including preeclampsia, cesarean delivery, and later type 2 diabetes.[4] For the offspring, it is linked to macrosomia, neonatal hypoglycemia, respiratory distress syndrome, and an increased long-term risk of obesity and type 2 diabetes.[5,6] GDM therefore has implications not only for pregnancy outcomes but also for the long-term metabolic health of both mother and child.
Several risk prediction models have been developed to identify women at risk for GDM. Most incorporate maternal age, body mass index (BMI), family history of diabetes, obstetric history, and laboratory parameters.[7,8] However, these tools are often population-specific, perform inconsistently across ethnic groups, and are considered too complex for routine use, underscoring the need for simple, reliable, and generalizable screening tools.[9,10]
The Finnish Diabetes Risk Score (FINDRISC) is a widely used, validated, and inexpensive questionnaire originally designed to estimate the 10-year risk of type 2 diabetes.[11] It is based on readily available information including age, BMI, waist circumference, physical activity, diet, antihypertensive medication use, history of high blood glucose, and family history of diabetes and has been applied successfully in community and clinical settings.[12] Its potential use for GDM has been suggested, but the evidence remains limited to small studies.[13,14]
The aim of this study was to evaluate the ability of FINDRISC, administered at the time of routine GDM screening at 24 to 28 weeks of gestation, to discriminate between women with and without GDM in a large multicenter cohort in Türkiye, and to explore whether it could serve as a simple triage tool within a 2-step screening strategy.
2. Materials and methods
2.1. Study design and setting
This multicenter cross-sectional study was conducted between June 1 and September 30, 2024, in 3 tertiary training and research hospitals located on the Anatolian side of Istanbul, Türkiye. The study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology statement for cross-sectional studies and, for the diagnostic accuracy analyses, the Standards for Reporting of Diagnostic Accuracy Studies 2015 guideline.[15,16] The study was approved by the Ethics Committee of Umraniye Training and Research Hospital (approval number: B.10.1TKH.4.34.H.GP.0.01/60; date: February 27, 2024). Written informed consent was obtained from all participants, and the study adhered to the principles of the Declaration of Helsinki.
2.2. Participants and sample size
All pregnant women who underwent routine GDM screening in the obstetrics outpatient clinics of the participating hospitals during the study period were assessed for eligibility. Inclusion criteria were age ≥ 18 years, gestational age 24 to 28 weeks, ability to communicate in Turkish, and scheduled routine GDM screening. Exclusion criteria were preexisting diabetes mellitus, a history of GDM in a previous pregnancy, unknown prepregnancy weight, and psychiatric disorders that could interfere with data collection.
All consecutive women who underwent GDM screening during the study period, met the eligibility criteria, and agreed to participate were included (consecutive sampling), yielding a final sample of 797 women (Fig. 1). There were no missing data for FINDRISC items or GDM status among the 797 included women; therefore, all analyses were performed on complete cases.
No a priori sample size calculation was performed, because all eligible women during the study period were included. Post hoc, with 104 women with GDM and 693 without, the study had approximately 93% power (2-sided α = 0.05) to detect an AUC of 0.60 against the null value of 0.50.
2.3. Data collection
Data were collected through face-to-face interviews conducted by trained researchers using a standardized questionnaire comprising a personal data form and the Turkish version of FINDRISC. The personal data form, developed by the researchers on the basis of the literature, recorded maternal age, gestational week, prepregnancy weight, and family history of diabetes (including the affected relatives). GDM screening test type and results were obtained from medical records after the interview.
2.4. FINDRISC assessment
FINDRISC consists of 8 items: age, BMI, waist circumference, daily physical activity of at least 30 minutes, daily consumption of vegetables or fruits, antihypertensive medication use, history of high blood glucose, and family history of diabetes. Total scores range from 0 to 26.[11] The Turkish version recommended by the Society of Endocrinology and Metabolism of Türkiye was used.[17] In this version, the antihypertensive item asks whether the respondent has ever been prescribed antihypertensive medication or been told that she has high blood pressure, and the blood glucose item asks whether she has ever been told that her blood glucose was high or borderline, including at a checkup, during an illness, or during pregnancy.
Because the study was conducted during pregnancy, BMI was calculated from self-reported prepregnancy weight and height measured at the study visit, and waist circumference was self-reported as the prepregnancy value measured at the level of the umbilicus.
Items were scored according to the original FINDRISC scoring.[11] Accordingly, all women younger than 45 years received 0 points for the age item (the Turkish form offers separate response options for <35 and 35–44 years, both scored 0). For family history, women reporting diabetes in a parent, sibling, or child were classified as having a first-degree relative (5 points), and those reporting only grandparents, aunts, uncles, or cousins as having a second-degree relative (3 points); women reporting both were classified as first-degree. Before analysis, all total scores were recalculated from the individual item responses, and participants were classified into the standard FINDRISC risk categories: low (<7), slightly elevated (7–11), moderate (12–14), high (15–20), and very high (>20).[11]
2.5. Diagnosis of gestational diabetes mellitus
GDM was diagnosed according to the 2019 guidelines of the Turkish Society of Endocrinology and Metabolism using either the one-step approach (75-g oral glucose tolerance test [OGTT]; ≥1 value ≥ 92, ≥180, or ≥153 mg/dL at fasting, 1 hour, or 2 hours) or the 2-step approach (50-g screening followed, when required, by a 100-g OGTT; ≥2 values ≥ 95, ≥180, ≥155, or ≥140 mg/dL at fasting, 1, 2, or 3 hours).[17]
Overall, 370 women (46.4%) underwent the one-step approach and 427 (53.6%) the 2-step approach; of the latter, 89 underwent a 100-g OGTT. GDM was diagnosed in 59 women (15.9%) with the one-step approach and in 45 women (10.5%) with the 2-step approach.
2.6. Statistical analysis
Analyses were performed using IBM SPSS Statistics for Windows, version 25.0 (IBM Corp., Armonk), and Python 3 (SciPy, statsmodels, and scikit-learn) for confidence intervals and the DeLong test. The normality of continuous variables was assessed with the Kolmogorov–Smirnov and Shapiro–Wilk tests. Because neither age nor FINDRISC score was normally distributed, both mean ± standard deviation and median (interquartile range [IQR]) are reported, and groups were compared with the Mann–Whitney U test.
Categorical variables are presented as n (%) and were compared with the Pearson chi-square test or Fisher exact test, as appropriate.
The association between the total FINDRISC score and GDM was evaluated by univariable binary logistic regression. To identify which components contributed independently, a multivariable logistic regression model including all 8 FINDRISC items was fitted (enter method). Because only 2 women were aged 45 to 54 years, age was dichotomized as <35 vs ≥35 years.
Results are reported as odds ratios (ORs) with 95% confidence intervals (CIs). Model calibration was assessed with the Hosmer–Lemeshow test, and multicollinearity was checked with variance inflation factors.
Discrimination was evaluated using receiver operating characteristic analysis. The area under the curve (AUC) and its 95% CI were calculated using the DeLong method, and the AUC of the multivariable model was compared with that of the total score using the paired DeLong test. The optimal cutoff was defined by the maximum Youden index (sensitivity + specificity − 1).
At each cutoff, a 2 × 2 table was constructed, and sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and positive and negative likelihood ratios (LR+ and LR−) were calculated. Wilson score 95% CIs were calculated for proportions and log-method 95% CIs for likelihood ratios. To evaluate the potential role of FINDRISC as a triage tool, we additionally report cutoffs with high sensitivity, together with the number of OGTTs that could be avoided and the number of GDM cases that would be missed.
Sensitivity analyses were performed separately for the one-step and 2-step diagnostic approaches and after removing the high-blood-glucose item, which may partly overlap with glucose results from the current pregnancy. A 2-sided P-value < .05 was considered statistically significant.
3. Results
3.1. Participant characteristics
A total of 797 pregnant women were included (Fig. 1). The mean age was 29.4 ± 5.5 years (median 29 [IQR 25–34]), and the mean FINDRISC score was 7.6 ± 4.5 (median 7 [IQR 4–11]).
The distribution of responses to the FINDRISC items is presented in Table 1.
Of the participants, 79.2% were younger than 35 years, and 74.5% had a prepregnancy BMI ≥ 25 kg/m2. Only 23.6% reported at least 30 minutes of daily physical activity. A prepregnancy waist circumference ≥80 cm was reported by 27.5%.
Although 73.0% consumed vegetables or fruits every day, 27.0% did not. A history of high blood glucose was reported by 15.3%, and 62.7% reported a family history of diabetes (first-degree, 36.0%; second-degree only, 26.7%).
According to the FINDRISC classification, 44.9% of women were at low risk, 34.5% at slightly elevated risk, 12.5% at moderate risk, 7.9% at high risk, and 0.1% at very high risk (Table 2).
3.2. Comparison of women with and without GDM
GDM was diagnosed in 104 women (13.0%). Comparisons between women with and without GDM are presented in Table 3. Women with GDM were older (median 31 [IQR 28–36] vs 28 [25–33] years; P < .001) and were more often aged ≥35 years (37.5% vs 18.3%; P < .001).
A prepregnancy BMI > 30 kg/m2 was more common in the GDM group (46.2% vs 30.2%; P = .001). Non-daily vegetable/fruit consumption was more frequent in women with GDM (37.5% vs 25.4%; P = .010), as was a family history of diabetes in a first-degree relative (49.0% vs 34.1%; overall P = .007). There were no significant differences in waist circumference, physical activity, antihypertensive medication use or hypertension, or history of high blood glucose (all P > .05).
The FINDRISC score was significantly higher in women with GDM than in those without (8.94 ± 4.49 vs 7.40 ± 4.45; median 8 [IQR 6–12] vs 7 [4–10]; P = .001).
3.3. Logistic regression analyses
In univariable logistic regression, the total FINDRISC score was significantly associated with GDM: each 1-point increase was associated with 8% higher odds of GDM (OR 1.08; 95% CI, 1.03–1.13; P = .001; Table 4A).
In the multivariable model including all FINDRISC items, age ≥ 35 years (adjusted OR 2.40; 95% CI, 1.51–3.80; P < .001), BMI > 30 kg/m2 (adjusted OR 2.88; 95% CI, 1.40–5.95; P = .004), and non-daily vegetable/fruit consumption (adjusted OR 1.70; 95% CI, 1.08–2.68; P = .022) were independently associated with GDM (Table 4B). A first-degree family history of diabetes was associated with GDM in crude analysis (OR 1.62; 95% CI, 1.02–2.57) but not after adjustment (adjusted OR 1.48; 95% CI, 0.90–2.43; P = .125). Waist circumference, physical activity, antihypertensive medication use or hypertension, and history of high blood glucose were not independently associated with GDM.
The model showed acceptable calibration (Hosmer–Lemeshow P = .350), and no multicollinearity was detected (all variance inflation factors < 2.3).
3.4. Discriminative and diagnostic performance
The AUC of the FINDRISC score for GDM was 0.599 (95% CI, 0.542–0.655; P < .001), indicating poor discrimination (Table 5; Fig. 2). The Youden index was maximal at a cutoff of ≥8 points (i.e., >7).
At this cutoff, 389 women (48.8%) scored positive, of whom 65 had GDM; 408 women scored negative, of whom 39 had GDM (Table 5). Sensitivity was 62.5% (95% CI, 52.9–71.2), specificity 53.2% (95% CI, 49.5–56.9), PPV 16.7% (95% CI, 13.3–20.7), NPV 90.4% (95% CI, 87.2–92.9), LR+ 1.34 (95% CI, 1.13–1.58), and LR− 0.70 (95% CI, 0.54–0.91). Thus, 39 of the 104 women with GDM (37.5%) would have been classified as low risk at this cutoff.
Performance across alternative cutoffs is shown in Table S1, Supplemental Digital Content 1. A cutoff of ≥4 points achieved a sensitivity of 88.5% but would have spared OGTT in only 175 women (22.0%) while missing 12 GDM cases (11.5%); a cutoff of ≥3 points achieved a sensitivity of 95.2% but would have spared OGTT in only 105 women (13.2%) while still missing 5 cases. Likelihood ratios were close to 1 at all cutoffs (LR− 0.33–0.70).
The apparent AUC of the multivariable model with re-weighted FINDRISC items was 0.694 (95% CI, 0.645–0.742), which was higher than that of the total score (P = .002, DeLong test); this estimate was not internally or externally validated and is likely optimistic.
3.5. Sensitivity analyses
Results were consistent across sensitivity analyses (Table S2, Supplemental Digital Content 2). The AUC was 0.611 (95% CI, 0.534–0.688) among women screened with the one-step approach and 0.589 (95% CI, 0.505–0.673) among those screened with the 2-step approach. Removing the high-blood-glucose item did not change discrimination (AUC 0.599; 95% CI, 0.544–0.655), and excluding women who reported a history of high blood glucose yielded an AUC of 0.594 (95% CI, 0.532–0.656).
4. Discussion
In this multicenter cross-sectional study of 797 pregnant women in Türkiye, the FINDRISC score, completed at the time of routine GDM screening at 24 to 28 weeks, was higher in women with GDM and was associated with GDM in logistic regression (OR 1.08 per point). However, its discrimination was poor (AUC 0.599). At the Youden-optimal cutoff of ≥8 points, sensitivity was 62.5% and specificity 53.2%; more than one-third of women with GDM would have been classified as low risk, and likelihood ratios close to 1 indicate that a FINDRISC result changes the probability of GDM only marginally.
Overall, these findings do not support the use of FINDRISC as a stand-alone tool or as a means of replacing universal OGTT screening in this population.
The NPV of approximately 90% should be interpreted with caution. With a GDM prevalence of 13.0%, the pretest probability of not having GDM is already 87.0%; the NPV at≥8 points (90.4%) therefore represents only a small gain over the prior probability, consistent with an LR− of 0.70. Similarly, when a high-sensitivity cutoff (≥3 or ≥4 points) was considered to minimize missed cases, only 13% to 22% of women would have been spared OGTT, while 5 to 12 GDM cases would still have been missed.
Accordingly, the potential of FINDRISC to safely exclude GDM within a 2-step strategy appears limited, and any such use would require prospective evaluation of its clinical and economic consequences.
Early and effective screening is critical for improving obstetric outcomes and reducing long-term metabolic risks for mothers and offspring. Clinical risk scores for GDM aim to reduce reliance on time-consuming and costly tests such as the OGTT; however, the discriminative ability of models based solely on clinical variables is often moderate. Teede et al reported an AUC of approximately 0.70 with a sensitivity of approximately 61% for a model developed in early pregnancy,[8] and van Leeuwen et al found an AUC of approximately 0.64 in external validation, with a high NPV (96–97%) but low PPV (8–10%), indicating that selective screening could reduce the burden of OGTT at the cost of missed cases.[18] Biomarker-based approaches may improve performance: Nanda et al showed that adding first-trimester adiponectin and sex hormone–binding globulin to maternal factors increased the AUC from approximately 0.79 to 0.84.[19]
FINDRISC, developed by Lindström and Tuomilehto to predict the 10-year risk of type 2 diabetes, has shown good accuracy in numerous studies,[11] although its performance varies across populations. In the Greek DE-PLAN study (n = 869), FINDRISC had an AUC of 0.724 for undiagnosed type 2 diabetes,[20] and in the Spanish SPREDIA-2 study (n = 1426), an AUC of 0.72.[21] Evidence for GDM is sparse. In a Turkish thesis study by Paydaş (n = 290), the optimal cutoff for GDM was ≥ 11 (AUC 0.713),[14] and in a small Indonesian study (n = 50), a score ≥9 was considered high risk.[13] The lower AUC in our larger multicenter sample may reflect differences in population characteristics, diagnostic criteria, and the exclusion of women with previous GDM, as well as the greater precision of estimates from a larger sample.
Several features of FINDRISC limit its performance in pregnancy. First, the age item assigns 0 points to all women younger than 45 years, so this item contributed almost no variability in our cohort, even though age ≥35 years was one of the strongest independent correlates of GDM (adjusted OR 2.40). Second, waist circumference and physical activity were not associated with GDM; self-reported prepregnancy waist circumference may be inaccurate, and waist circumference measured during pregnancy would mainly reflect the gravid uterus.
Third, the history-of-high-blood-glucose item, which carries the largest weight (5 points), was not independently associated with GDM, possibly because women with previous GDM were excluded; its removal did not change discrimination. Fourth, first-degree family history lost significance after adjustment. Notably, re-weighting the FINDRISC items in a multivariable model increased the apparent AUC to 0.69, suggesting that the item weights derived for type 2 diabetes are not optimal for GDM.
However, this estimate was derived and evaluated in the same data, has not been validated, and should be regarded as hypothesis-generating.
Our exclusion of women with previous GDM deserves comment. A history of GDM is one of the strongest risk factors for GDM in subsequent pregnancies, with recurrence reported in a substantial proportion of women. These women were excluded because they are already considered at high risk and are referred for testing irrespective of other factors.
Consequently, our results apply to women without previous GDM; including them would likely have increased the apparent discrimination of the history-of-high-blood-glucose item and of the total score but would not have changed clinical management for this subgroup.
Pregnancy-adapted risk scores that incorporate previous GDM, parity, early-pregnancy BMI, HbA1c or fasting glucose, and potentially adipokines may be more appropriate than adaptations of type 2 diabetes tools. Such models should be developed and validated prospectively in early pregnancy when risk stratification could influence lifestyle interventions or the timing of testing.
4.1. Strengths and limitations
This study has several strengths. It included a relatively large sample of pregnant women recruited from 3 tertiary centers. All participants underwent standardized GDM screening at 24 to 28 weeks, and FINDRISC was administered by trained interviewers using the validated Turkish version.
Total scores were verified from item-level responses, and performance was evaluated comprehensively using univariable and multivariable regression, receiver operating characteristic analysis with DeLong CIs, 2 × 2 tables, likelihood ratios, triage scenarios, and sensitivity analyses.
Several limitations should be considered. First, FINDRISC was administered at the same visit as GDM screening at 24 to 28 weeks; the study therefore evaluates concurrent discrimination, and the performance of FINDRISC when administered in early pregnancy, when it would be clinically most useful, remains unknown. Second, prepregnancy weight and waist circumference were self-reported and may be subject to recall bias.
Third, the high-blood-glucose item may include glucose results from the current pregnancy (e.g., an earlier screening test), which could introduce partial incorporation bias; however, removing this item did not change the results. Fourth, 2 diagnostic approaches with different criteria were used according to clinic practice, which may have influenced GDM prevalence; nevertheless, discrimination was similar with both approaches. Fifth, the study was conducted in tertiary centers in Istanbul, and the findings may not be generalizable to primary care settings, other regions of Türkiye, or other ethnic populations.
Sixth, women with previous GDM were excluded. Seventh, only women who agreed to participate were enrolled, and selection bias cannot be completely excluded. Finally, the multivariable model was not validated, and the cross-sectional design precludes causal inference.
5. Conclusion
When administered at 24 to 28 weeks of gestation, FINDRISC showed poor ability to discriminate between women with and without GDM. Its NPV largely reflects the low prevalence of GDM; more than one-third of GDM cases scored below the optimal cutoff, and no cutoff combined high sensitivity with a meaningful reduction in OGTTs. FINDRISC should therefore not be used as a stand-alone screening tool or to replace universal OGTT screening.
Pregnancy-specific risk models that incorporate factors such as previous GDM, parity, early-pregnancy glycemic markers, and biomarkers, evaluated prospectively in early pregnancy and validated in multicenter cohorts, are needed.
Acknowledgments
The authors thank the staff of all participating hospitals for their contributions to data collection and the participants for their time and cooperation.
Author contributions
Conceptualization: Selma Dagci, Ekmel Burak Ozsenel.
Data curation: Selma Dagci, Volkan Kizilay, Melike Punduk Yilmaz.
Formal analysis: Yasar Sertbas, Volkan Kizilay, Melike Punduk Yilmaz.
Writing – original draft: Selma Dagci, Yasar Sertbas, Melike Punduk Yilmaz, Meltem Sertbas.
Writing – review & editing: Selma Dagci, Yasar Sertbas, Volkan Kizilay, Melike Punduk Yilmaz, Meltem Sertbas, Ekmel Burak Ozsenel.