INTRODUCTION

The use of anthropometric indicators to relate it to diabetes has gained relevance in the academic field. Many investigations link anthropometric measurements and biochemical markers to predict the onset of diabetes. A PURE study design (Population Health Research Institute) was carried out in some communities in Colombia to evaluate the association between waist circumference, grip strength, body weight, and age¹. Waist circumference was the leading risk factor for a high MetS score.

A longitudinal study in children with type 1 diabetes mellitus, from 5 to 18 years of age in a Hospital in the northwest of Brazil; to anthropometric measurements, triglyceride, and glycated hemoglobin values were added; the triglyceridemic waist phenotype (CHTG), have a higher body mass index and glycosylated hemoglobin with a significance of p <0.05².

The waist circumference, measured in a horizontal plane, between the lower margin of the rib and the upper edge of the iliac crest, according to the guidelines of the World Health Organization³, continues to show that it is a better predictor for diabetes than body mass index⁴.

The objective of the study was to evaluate the discriminative ability to predict diabetes with anthropometric and biochemical indicators and antecedents. Anthropometric, biochemical, and background indicators turn out to be good discriminators for predicting diabetes mellitus, as well as having a good level of sensitivity and specificity⁵.

METHODS

RESULTS

There are many scales to predict diabetes, but the most used is the Findrisc scale, which allows assessing the risk of developing the disease in 10 years, using a score greater than 14, which represents a slightly high risk. With the ROC and AUC curves (reference test: the FINDRISC score, which divides the sample into two groups, low risk (<14) and high risk (> 14)) of the following anthropometric indicators: Age, systolic pressure and diastolic, family and personal pathological history, physical activity, glycemia, cholesterol, triglycerides, weight, height, BMI, abdominal circumference, waist-hip index, waist-height index and percentage of body fat.

Better results were obtained with glycemia, abdominal circumference and waist-height index (Figure 1). The abdominal circumference obtained greater discriminative power to predict diabetes (AUC = 0.747; p <0.001; CI: 0.624-0.870), compared to glycemia (AUC=0.749; p <0.001; CI: 0.645-0.852) and the index waist-height (AUC=0.737; p=0.001; CI: 0.638-0.836). Table 1 shows the Youden index, the sensitivity and specificity of the tests.

Figure 1: COR curve of anthropometric parameters 

The cut points allowed dichotomizing the variables (those who had a high risk of having diabetes in the future and those that do not) to contrast them with a Gold Standard (the values of the STOP-NIDDM test, with at least 10 points to determine high risk) through a confusion matrix. The best results were obtained in the variable glycemia and abdominal circumference. For glycemia, True Positive= 6, False Positive= 16, False Negative= 6 and True Negative= 76 were obtained. In the same way, the following values were obtained for abdominal girth, TP = 2, FP = 6, FN = 10 and TN = 86. The Diagnostic test calculator with these values was used to obtain the diagnostic test characteristics for both variables (Table 2).

Fagan nomograms for blood glucose (left) and abdominal girth (right) are shown in Figure 2.

Figure 2 Fagan's nomogram 

In a second stage, it was evaluated which variables could enter the multivariate logistic regression analysis. To the qualitative variables, the test was applied X2 (with the phi value) to measure the association of the dependent variable with each independent variable. The variables with the best results were: personal pathological history (APP, hereinafter: p = 0.001; phi = 0.312) and family pathological history (APF: p = 0.042; phi = 0.2), while those ruled out were gender (p = 0.319) and physical activity (p = 0.267).

In the case of quantitative variables, statistical tests were applied (Student's t for variables with normal distribution and Mann-Whitney U for variables that did not have a normal distribution) to determine the difference of means and medians between the groups formed by the dependent variable. The variables chosen were systolic blood pressure (SBP, p = 0.018), diastolic blood pressure (DBP, p = 0.026), glycemia (p = 0.031), cholesterol (p = 0.032), triglycerides (p = 0.006). The rejected variables were age (p = 0.256), BMI (p = 0.1), abdominal girth (p = 0.018 but Levene = 0.519), hip measurement (p = 0.3), waist / hip ratio (p = 0.022 but Levene = 0.678), waist / height ratio (p = 0.079) and% fat (p = 0.759).

In a second stage, a univariate logistic regression was performed and they entered the final model: APP (p = 0.04), APF (p = 0.07), SBP (p = 0.117), DBP (p = 0.035), glycemia (p = 0.039), cholesterol (p = 0.077) and triglycerides (p = 0.034).

Multivariate logistic regression was applied to six variables APP, APF, DBP, glycemia, cholesterol, and triglycerides, using the Wald forward method. The model performed 5 iterations until the best was found, with a -2LL0 = 74.386 and afinal -2LL = 66.437. The omnibus test of the model’s coefficients yields X2= 7.950 (p = 0.005), which allows evaluating the fit of the model. Cox and Snell's R-squared is 0.074, while Nagelkerke's R-squared = 0.144. At the same time, the RL2= 0.10 and the Pseudo-R2= 0.07.

Finally, the variable that enters the equation was APP (β = -1.897; p = 0.004; Exp (β) = 0.150; 95% CI = 0.042-0.542). Therefore, the logistic regression equation is finally defined as follows:

DISCUSSION

The World Health Organization (WHO) mentions that the number of people who have diabetes in Latin America could reach 32.9 million by 2030¹⁴. In this sense, it is important to determine the risk factors for Diabetes in health personnel through effective, sensitive, and specific indicators. The present study shows similar results with a longitudinal study in health workers that was carried out to relate the degree of occupation and risk factors for type 2 diabetes¹⁵.

With the analysis of the ROC curves, the abdominal circumference measurement had greater discriminative power to predict diabetes (AUC = 0.747; p <0.001; CI: 0.624-0.870). This agrees with various studies by Roos¹⁶, Yoon¹⁷, Darsini¹⁸, in which this indicator was effective in predicting risk. In a cohort study with different cut-off points, the risk of acquiring diabetes was evaluated¹⁹, and it was determined that the abdominal circumference was the most important.

The results of the ROC curve analysis and those of the diagnostic test calculator show that the blood glucose test (0.522 and 0.495, respectively) has a medium level of sensitivity. This result is similar to the measurement of the alteration of fasting glucose (> 100 mg/dl) in a health insurance personnel in Korea to determine the risk of diseases²⁰.

When performing the bivariate correlation, those with the highest correlation were personal pathological history (APP, hereinafter: p = 0.001; phi = 0.312) and family pathological history (APF: p = 0.042; phi = 0.2). In turn, the multivariate logistic regression analysis showed that APPs have predictive power (Nagelkerke's R squared = 0.144)²¹. Although it is true, this value is relatively low, it is important to note that this indicator allows predicting the risk mathematically. In the same prediction model, it was found that APPs have a prediction probability of 40%.²².

LIMITATIONS OF STUDIES

One of the limitations is not having carried out the frequency of food consumption to investigate another prediction risk. Another limitation was not determining baseline insulin if it is directly related to abdominal girth and being another predictor for diabetes.

CONCLUSION

The analysis of the ROC curves shows that blood glucose and abdominal circumference are the best predictors of diabetes. It is known that glucose is used to diagnose the presence of diabetes in patients. For its part, the abdominal circumference is an effective indicator to predict diabetes, as shown in this study and others carried out in Latin America and the Caribbean latitudes.

Multivariate logistic regression analysis places family medical history as an important variable when predicting the probability of having diabetes in the future. The difference between this equation with the FINDRISC tests and similar ones is that the equation gives a probability as a single variable. In contrast, the other tests require lengthy questionnaires and measurements of many variables. Therefore, a logistic equation is more versatile when it comes to forecasting probabilities. The best predictors of diabetes risk are glycemic index, abdominal circumference, and personal medical history, each with its potential, depending on the approach with which the test is used.