Diagnostic accuracy of various anthropometric indexes for assessing insulin resistance in the Peruvian population: An analytical cross-sectional study

Study design

This analytical, cross-sectional study used data from the PERU MIGRANT cohort study collected from November 2007 to April 2008.

Population and sample

The PERU MIGRANT study enrolled participants of both sexes aged 30 years and above, excluding individuals with a history of mental illness or those who were currently pregnant, and residing in Peru. Participants were recruited from both rural and urban settings using a single-stage random sampling method based on the 2006 and 2007 censuses. Rural residents were selected from San Jose de Secce, Huanta, Ayacucho, while urban residents and rural-to-urban migrants were chosen from Las Pampas de San Juan de Miraflores, Lima. Migrants were defined as individuals born in Ayacucho who had resettled permanently in Las Pampas de San Juan de Miraflores at the time of data collection, while urban participants were those born and continuously residing in this urban area. Comprehensive details on eligibility criteria, variables assessed, sampling procedures, and response rates have been previously described in the original study protocol. ¹⁷

For the present study, participants without glucose-impaired metabolism were included (n = 930). A fasting glucose ≥ 126 mg/dL or glycosylated hemoglobin ≥6.5% or Type-2 diabetes mellitus diagnosed by a physician was the criterion for defining glucose-impaired metabolism. Those with missing data on any of the variables of interest were excluded (n = 22). Finally, data from 908 participants were analyzed.

Variables and measurements

Index tests

Various anthropometric indicators were assessed as index tests. These tests were based on waist circumference, and skinfold thickness measurements as well as DXA regression equations. Different cut-off values by sex and for each anthropometric index test were analyzed. Then using the specific cut-off values obtained, each index test was dichotomized as high or low separately according to sex. All anthropometric indexes tested are shown in Table 1.

OPEN IN VIEWER

Table 1.

Various anthropometric indexes tested for insulin resistance diagnosis.

Anthropometric indexes based on	Brief description
Waist circumference
Waist circumference (cm)	Circumference at the midpoint between the lower rib and the iliac crest
Waist to hip ratio (cm)	Waist (cm)/hip (cm)
Waist-to-height ratio	Waist (cm)/height (cm)
Skinfolds thickness
Biceps (mm)	Skinfold thickness in the bicipital site
Triceps (mm)	Skinfold thickness in the tricipital site
Supra spinal (mm)	Skinfold thickness in the supra-spinal site
Subscapular (mm)	Skinfold thickness in the subscapular site
Upper limb (mm)	The sum of biceps + triceps skinfolds (mm)
Trunk (mm)	The sum of supra spinal + subscapular skinfolds (mm)
Trunk to upper limb ratio	(supra spinal + subscapular)/(biceps + triceps)
Sum of skinfold (mm)	The sum of upper limb and trunk skinfolds
DXA regression equations
Fat mass index (kg/m2)	Fat mass (kg)/height (m) 2
Fat mass percentage (%)	[Fat mass (kg)/weight (kg)] * 100
Fat-to-lean mass ratio	Fat mass (kg)/lean mass (kg)

Statistical analysis

STATA v17.0 was used for all analyses. Participants with missing data for any variables of interest were excluded from the analysis. Descriptive analyses were presented as absolute frequencies and percentages for categorical variables and median and interquartile ranges for numerical, not normally distributed variables. Differences in frequency distribution between the participants’ characteristics and sex were conducted using a Chi-squared test for categorical variables and the Mann–Whitney test for numerical variables, respectively. The diagnostic accuracy of each anthropometric index was assessed using receiver operating characteristic (ROC) curves calculating the area under the curve (AUC) and its corresponding 95% confidence intervals (CI). The ROC curves were stratified according to sex. Sensitivity, specificity, positive and negative predictive values, as well as positive and negative likelihood ratios, were calculated. The sex-specific optimal cut-off point for each anthropometric index was determined using the Youden Index.

For examining the association between the best anthropometric indexes and insulin resistance, separate forward stepwise logistic regressions were conducted stratified by sex considering p < 0.05 and p ≥ 0.10 as the threshold for including and excluding the variables from the model, respectively. All anthropometric indexes (categorized as high or low based on previously determined cut-offs), along with covariates (age, migration status, smoking, alcohol intake, physical activity level, educational level, and wealth index), were included as candidate variables in the forward stepwise logistic regression. Model performance was assessed through the Hosmer–Lemeshow test and Nagelkerke R². Unadjusted and adjusted odds ratios (OR) with their respective 95% confidence intervals (CI) were calculated. Only variables meeting the p-value criteria were retained in the final models. The selection was fully automated, and no covariates were forced into the models. The final set of covariates varied by sex, with only educational level retained in men, and no sociodemographic variables retained in women.

Main findings

The present study analyzed the diagnostic accuracy of various anthropometric measurements for IR, providing insights that extend beyond traditional indicators like BMI, which fails to differentiate between fat and lean mass. Our findings highlight the FMI as the most accurate measure, with AUC values above 0.80 for both sexes, offering robust sensitivity and specificity. Moreover, high WC, subscapular skinfold, upper limb skinfold, and FMI were significantly associated with IR with odds ratios ranging from 2.50 to 4.25. Among men, subscapular skinfold and FMI showed even stronger associations, with odds ratios of 3.75 and 11.70, respectively. These findings underscore the importance of incorporating diverse and precise anthropometric indicators tailored, allowing for a more comprehensive and tailored assessment of IR risk.

Comparison with other studies

The prevalence of IR reports contrasts among countries. The overall IR prevalence in this study was lower than reported in the USA and Mexico. Using a HOMA-IR cut-off of ≥2.5, these studies found a prevalence of 39.60% and 22.3% respectively.^9,26 Also, in urban regions of Venezuela, the prevalence of IR reached 46.70%. ²⁷ These variations may be explained by the fact that populations in different countries are undergoing distinct stages of epidemiological and nutritional transition. ²⁸

Our study revealed some differences between men and women. Findings showed higher IR prevalence among women, this result is consistent with diabetes mellitus prevalence over the years in Peru, where a higher proportion of women have been diagnosed with this condition. ²⁹ In addition, it is noteworthy that the highest prevalence of IR was observed in urban residents, aligned with evidence that urbanization promotes sedentary lifestyles and dietary patterns rich in energy-dense foods, which diminish insulin response. ³⁰ These findings are similar to Brazilian results that showed that people with poor diet quality and limited physical activity had a higher risk of IR. ³¹ These disparities underscore the interplay between environmental and behavioral factors in the manifestation of IR.

Our data found statistically significant differences in nearly all anthropometric indicators between participants with and without IR. Similar findings have been reported globally, as most of these indicators assess adiposity. ¹⁰ Excess adiposity, particularly visceral fat, contributes to systemic inflammation and metabolic disruption by releasing adipokines and free fatty acids. These factors impair insulin receptor signaling, thereby reducing glucose uptake in peripheral tissues. ⁴

Some studies have previously assessed the association and accuracy of anthropometric indicators for IR,^9–11 revealing variations across ethnic groups. ¹⁰ Most of these investigations have used more traditional measures, such as BMI and WC. A study conducted using data from the United States National Health and Nutrition Examination Survey (NHANES) and the Korea National Health and Nutrition Examination Survey (KNHANES) revealed that the ability of BMI and WC to diagnose IR and diabetes was highest in Whites (BMI AUC = 0.8324, 95% CI: 0.8322–0.8325; WC AUC = 0.8468, 95% CI: 0.8467–0.8469), in comparison to other ethnicities such as Hispanics (BMI AUC = 0.7998, 95% CI: 0.7995–0.8001; WC AUC = 0.8012, 95% CI: 0.8009–0.8015). ¹⁰ In our study, WC demonstrated a similar AUC, which was among the highest AUCs. However, the FMI was the anthropometric measurement that exhibited the highest AUC among the evaluated indicators with AUCs of 0.80 in women and 0.81 in men. In contrast to other anthropometric measurements, FMI incorporates in its calculation both fat mass and squared height, offering a more precise assessment of adiposity relative to body structure. Interestingly, FMP and FLMR also performed well, with slightly lower AUC values. This suggests that indicators incorporating both fat and lean mass may be valuable for IR detection, as they reflect the metabolic impact of excess adiposity and the protective role of lean mass for cardiometabolic diseases. ³²

Sex-specific FMI cut-offs of 11.70 kg/m² for women and 7.52 kg/m² for men demonstrated high sensitivity (>80%) and specificity (>70%). These thresholds align closely with previous findings in a Colombian population for detecting FMI and metabolic syndrome, which reported cut-offs of 11.8 kg/m² for women and 6.9 kg/m² for men. ³³ Notably, although the Colombian study sample consisted of younger individuals (18–35 years), their reported cut-offs are comparable to those observed in our study. This similarity might be explained by the fact that the Colombian thresholds were developed for metabolic syndrome—a condition associated with more advanced metabolic complications than insulin resistance alone. It is possible that higher fat mass accumulation acts as a precursor, exacerbating metabolic disturbances and triggering these complications, which could explain the close alignment of cut-offs across studies. This idea is further supported by population-based analyses, such as those using NHANES data, which identified higher FMI thresholds for obesity classification: 13 kg/m² for women and 9 kg/m² for men. ³⁴ However, ethnic differences should also be considered when interpreting these results. For instance, a cross-sectional study in a Chinese population reported significantly lower FMI cut-offs for metabolic syndrome diagnosis in women, at 7.9 kg/m². ³⁵ These discrepancies may reflect underlying ethnic differences in fat distribution. For instance, East Asian populations tend to exhibit a higher proportion of visceral fat relative to subcutaneous fat, even at lower levels of total fat mass.^36,37 Visceral fat is more strongly associated with insulin resistance and metabolic dysfunction compared to subcutaneous fat. ³⁸ This pattern might explain the need for lower FMI thresholds for diagnosing metabolic dysfunctions in these populations. Altogether these findings underscore the importance of tailoring FMI thresholds to the specific condition and population under study. Furthermore, it is noteworthy that although the equations for FM used in this study were originally developed in a different population, ²¹ our results demonstrate their strong diagnostic performance in the Peruvian context. This suggests that these equations could be appropriate and useful in Latin American settings where direct access to DXA measurements is limited.

The results from the tests used to evaluate the logistic regression model indicate a good model for both sexes, with women showing a slightly higher ability to explain the variability in insulin resistance. Furthermore, the adjusted logistic regression demonstrated significant associations between several anthropometric indicators and IR, varying by sex. Among women, WC, subscapular skinfold thickness, upper limb skinfold thickness, and FMI remained significantly associated with IR after adjustment. Notably, the upper limb skinfold exhibited the strongest association, suggesting the relevance of peripheral adiposity as a key factor in this population. Central and total adiposity, as indicated by WC and FMI, also demonstrated significant associations, though their magnitudes were comparatively smaller. Subscapular skinfold, reflecting trunk fat distribution, showed a moderate association, further emphasizing the role of adiposity distribution in women. In men, FMI and subscapular skinfold thickness remained significantly associated with IR after adjustment. FMI exhibited the strongest association, highlighting the importance of total adiposity in this population. The subscapular skinfold also remained significant, emphasizing the role of central fat distribution in men. The higher odds ratio observed for FMI in men compared to women suggests potential sex differences in how overall adiposity influences IR risk. However, the wider confidence intervals observed in men, particularly for FMI, reflect greater variability in the data, which should be considered when interpreting these associations. Notably, while educational level was retained in the final model for men, it was not retained for women. This sex-specific pattern aligns with findings from a Peruvian subsample of a large urban health study, which reported no significant association between education and diabetes among women, but a positive association in men. ³⁹

Beyond the associations, the diagnostic performance of these indicators provides further insights into their utility. While FMI demonstrated the highest diagnostic accuracy for IR in both sexes, highlighting its potential as a reliable diagnostic tool for identifying IR at a population level, the relatively wide confidence intervals observed in the logistic regression models, particularly in men, suggest some variability in the association estimates. However, this variability does not negate the strength of the diagnostic performance but rather emphasizes the need for cautious interpretation when using FMI in individual-level analyses. Taken together, these findings suggest that FMI and other anthropometric indicators are valuable tools for identifying IR in both sexes. However, their application should balance the observed strengths with the limitations in variability and sample-specific differences. Sex-specific strategies and larger studies with more diverse populations are needed to strengthen the evidence base for these indicators as practical tools in clinical and public health contexts.

Implications of the findings

Our findings hold significant public health and clinical implications, particularly for middle-income countries like Peru. IR is widely recognized as a precursor to a range of cardiometabolic conditions, including type 2 diabetes, cardiovascular disease, and non-alcoholic fatty liver disease. ⁴⁰ In Peru, as in other developing countries, the economic and healthcare systems face a high-cost burden of diabetes and cardiovascular diseases linked to IR. Therefore, early detection and monitoring of IR are critical for preventing these associated conditions, especially in resource-limited settings where laboratory-based methods could be costly or difficult to access for the entire population. Our study demonstrates that anthropometric indicators, such as waist circumference, fat mass index, and upper limb skinfolds, are strongly associated with IR. Compared to laboratory-based measures like HOMA-IR, anthropometric parameters are non-invasive, cost-effective, and readily available, making them valuable tools for screening and monitoring IR at the population level, as demonstrated in studies assessing other health outcomes. ²⁶

The relevance of these findings is amplified by their focus on the Peruvian population, contributing evidence from a developing country, where the prevention of IR is vital. While other indicators, such as biochemical markers, ⁴¹ are currently being proposed as components of models predicting insulin resistance, tailored interventions that use simple, accessible measures like FMI and WC could help mitigate the progression of IR and its related diseases. Such strategies are essential for reducing healthcare costs and improving population health outcomes in countries like Peru, where the double challenge of rising non-communicable diseases and limited resources demands innovative, preventative approaches. By integrating these measures into routine healthcare processes, clinicians and policymakers can effectively identify at-risk individuals early and implement targeted interventions to alleviate the burden of IR and its complications.

Limitations

Naturally, our study has some limitations. First, as both the exposure and the outcome variables were assessed at the same time, this study adds little evidence of a casual association between the variables. This is a common limitation of cross-sectional studies. In addition, there was no information available on the family history of IR and certain socio-economic factors that may introduce residual confounding. Also, information on physical activity, alcohol consumption and smoking were self-reported that may lead to information bias. Regarding alcohol consumption, the categorization was based on the frequency of heavy episodic drinking, which may not fully capture other relevant consumption patterns potentially associated with insulin resistance. The threshold values for HOMA-IR reported in the literature vary significantly, which, combined with the lack of standardization in insulin measurement, complicates the accurate assessment of the relationship between anthropometric parameters and IR. ⁴² Following the WHO’s recommendations, IR is typically defined as HOMA-IR values above the 75th percentile. ⁴³ Based on recent studies, a HOMA-IR value of 2.8 has been adopted as the cut-off for diagnosing IR in the Peruvian population. It is worth noting, however, that this cut-off was established specifically to predict type 2 diabetes risk. Different thresholds—both lower and higher—have been reported in other ethnic groups and clinical contexts, particularly when assessing risks for instance cardiovascular disease, ⁴⁴ cancer and mortality. ⁴⁵ Therefore, while 2.8 is appropriate for Peruvian population, it may be limited to type 2 diabetes mellitus. Additionally, the use of data collected between 2007 and 2008 may not fully reflect the current epidemiological landscape or recent demographic shifts. However, in the absence of substantial changes in the determinants of insulin resistance, the magnitude of its association with body composition indicators, and the diagnostic accuracy of anthropometric measures, is expected to remain consistent over time. Furthermore, the PERU MIGRANT study remains one of the few population-based sources in Peru that includes both anthropometric and biochemical markers, which underscores the value and relevance of this analysis despite the time gap. As this was a secondary analysis of the PERU MIGRANT study, no formal sample size calculation was performed for the present research question. Nonetheless, the available sample was sufficient to detect statistically significant associations. However, it should be acknowledged that the diagnostic performance and optimal cut-off points of anthropometric indicators may vary across age groups. Given the sample size and the distribution of participants, the study lacked sufficient statistical power to conduct stratified analyses by age, which precludes drawing age-specific inferences. Finally, the study population consisted of participants from Lima and Ayacucho provinces, thus, the results may not be generalizable for the overall Peruvian population.