Clinical screening for GCK-MODY in 2,989 patients from the Brazilian Monogenic Diabetes Study Group (BRASMOD) and the Brazilian Type 1 Diabetes Study Group (BrazDiab1SG)

ABSTRACT Objectives To evaluate the accuracy of routinely available parameters in screening for GCK maturity-onset diabetes of the young (MODY), leveraging data from two large cohorts – one of patients with GCK-MODY and the other of patients with type 1 diabetes (T1D). Materials and methods The study included 2,687 patients with T1D, 202 patients with clinical features of MODY but without associated genetic variants (NoVar), and 100 patients with GCK-MODY (GCK). Area under the receiver-operating characteristic curve (ROC-AUC) analyses were used to assess the performance of each parameter – both alone and incorporated into regression models – in discriminating between groups. Results The best parameter discriminating between GCK-MODY and T1D was a multivariable model comprising glycated hemoglobin (HbA1c), fasting plasma glucose, age at diagnosis, hypertension, microvascular complications, previous diabetic ketoacidosis, and family history of diabetes. This model had a ROC-AUC value of 0.980 (95% confidence interval [CI] 0.974-0.985) and positive (PPV) and negative (NPV) predictive values of 43.74% and 100%, respectively. The best model discriminating between GCK and NoVar included HbA1c, age at diagnosis, hypertension, and triglycerides and had a ROC-AUC value of 0.850 (95% CI 0.783-0.916), PPV of 88.36%, and NPV of 97.7%; however, this model was not significantly different from the others. A novel GCK variant was also described in one individual with MODY (7-44192948-T-C, p.Ser54Gly), which showed evidence of pathogenicity on in silico prediction tools. Conclusions This study identified a highly accurate (98%) composite model for differentiating GCK-MODY and T1D. This model may help clinicians select patients for genetic evaluation of monogenic diabetes, enabling them to implement correct treatment without overusing limited resources.


SUPPLEMENTARY METHODS
The logistic regression models were devised using either GCK-MODY versus T1D or GCK-MODY versus NoVar as the binary outcome.Significant variables from univariate analyses were entered as predictors in logistic models in three ways: (A) all significant continuous predictors, (B) all significant predictors (both continuous and categorical), and (C) backward elimination (progressive exclusion of predictors with the highest p value until only significant predictors remained).The models were named Models 1, 2, and 3 (GCK versus T1D as the binary outcome) and 4, 5, and 6 (GCK versus NoVar as the binary outcome).Fitted values of regression models were computed for each patient.The Akaike Information Criterion (AIC) was calculated as described below, and R 2 was computed for each model.Universitário João de Barros Barreto, Pará: Joao Felício Soares*, Flavia Marques Santos; Hospital Universitário Getúlio Vargas, Hospital Adriano Jorge: Deborah Laredo Jezini*.

SUPPLEMENTARY METHODS
The logistic regression models were devised using either GCK-MODY versus T1D or GCK-MODY versus NoVar as the binary outcome.Significant variables from univariate analyses were entered as predictors in logistic models in three ways: (A) all significant continuous predictors, (B) all significant predictors (both continuous and categorical), and (C) backward elimination (progressive exclusion of predictors with the highest p value until only significant predictors remained).The models were named Models 1, 2, and 3 (GCK versus T1D as the binary outcome) and 4, 5, and 6 (GCK versus NoVar as the binary outcome).Fitted values of regression models were computed for each patient.The Akaike Information Criterion (AIC) was calculated as described below, and R 2 was computed for each model.Equation 1. Probability calculation employing the logit obtained from logistic regression:

SUPPLEMENTARY METHODS
The logistic regression models were devised using either GCK-MODY versus T1D or GCK-MODY versus NoVar as the binary outcome.Significant variables from univariate analyses were entered as predictors in logistic models in three ways: (A) all significant continuous predictors, (B) all significant predictors (both continuous and categorical), and (C) backward elimination (progressive exclusion of predictors with the highest p value until only significant predictors remained).The models were named Models 1, 2, and 3 (GCK versus T1D as the binary outcome) and 4, 5, and 6 (GCK versus NoVar as the binary outcome).Fitted values of regression models were computed for each patient.The Akaike Information Criterion (AIC) was calculated as described below, and R 2 was computed for each model.

Equations for MODY probability calculation derived from multivariable models.
Continuous predictors should be entered as numeric values and categorical predictors should be entered as either 0 (absent) or 1 (present) in the equations.Results obtained from the equations below are logits or log-odds.Fitted probabilities should be calculated with the formula provided above, in Supplementary Methods.Abbreviations: DKA, diabetes ketoacidosis; HbA1c, glycated hemoglobin.
This is an open-access article distributed under the terms of the Creative Commons Attribution License