A first trimester prediction model and nomogram for gestational diabetes mellitus based on maternal clinical risk factors in a resource-poor setting

Gestational Diabetes Mellitus (GDM) poses a significant health concern during pregnancy, impacting both maternal and foetal outcomes [1]. While its prevalence is increasing globally and in Africa [2, 3], resource-poor countries face unique challenges in managing and preventing this condition. In Nigeria, a highly populated country in sub-Saharan Africa with a higher proportion of its people in the sub-urban and rural areas, the burden of GDM is high [4], necessitating targeted and practical approaches for early identification and intervention.

Gestational Diabetes Mellitus (GDM), one of the most common complications of pregnancy, arises due to alterations in maternal glucose metabolism and insulin sensitivity which occurs as a result of physiological changes during pregnancy for which the beta-cells of the pancreas are unable to compensate [5, 6]. In the early stages of pregnancy, this hyperglycemia stimulates increased expression of parathyroid hormone-related protein (PTH-rP) and its receptor (PTH-R1), along with vascular endothelial growth factor (VEGF) and CD31 leading to substantial disruptions in placental function and angiogenesis, which are crucial for maintaining a healthy feto-maternal environment during pregnancy, potentially leading to adverse pregnancy outcomes [7, 8]. Considering the impact of GDM and its complications on both maternal and fetal well-being, the need for universal screening via Oral Glucose Tolerance Test (OGTT) becomes evident. Due to its proven benefit over risk factor-based screening, universal screening is highly recommended, even in early pregnancy, given availability of financial, material, space, and human resources [9, 10]. However, implementing oral glucose tolerance testing during pregnancy for all women, and more than once in some, is both operationally challenging and costly, hence the continued emphasis on the need to improve risk factor-based screening approach in most developing countries especially in sub-Saharan Africa.

The first trimester of pregnancy has been explored as a crucial window for predicting the risk of GDM development [11, 12]. This is necessary because hyperglycaemia in early pregnancy may induce pathologic changes in the foetus prior to the diagnosis of GDM later in pregnancy [13, 14]. However, existing first trimester prediction models, which often involve elaborate biochemical testing [15‐17], stem predominantly from high-income settings and may not be directly applicable or feasible in resource-constrained environments. Also, certain parameters needed for effective application of the current suggested models may be unavailable, come at prohibitive costs or result in a delay in decision making. Therefore, there is a pressing need for a setting-specific prediction model that relies on easily accessible maternal clinical risk factors, allowing for easy implementation even in primary care settings with limited healthcare resources.

This study aimed at developing a GDM prediction model based on maternal clinical risk factors easily assessable in the first trimester of pregnancy in a cohort of Nigerian women with singleton pregnancies. It involved the development of a maternal clinical risk factors-based prediction model and a user-friendly nomogram that is easily interpretable and implementable, to facilitate timely interventions and improved outcomes for both mothers and their infants. The study also aims at contributing to mitigating the impact of GDM in sub-Saharan Africa and similar settings worldwide where there is a paucity of studies on early prediction of the disease.

A total of 345 pregnant women were recruited to this study but only 253 underwent a 75-g OGTT for GDM evaluation and were included in the analysis. The results from the procedure led to the diagnosis of GDM in 52 (20.6%) participants, while 201 (79.4%) were normoglycemic pregnancies. Table 1 shows the maternal demographic and clinical characteristics of the cohort with selected risk factors by GDM and non-GDM participants. Compared with the non-GDM women, those who developed GDM were significantly older (31.4 ± 4.6 years versus 28.8 ± 4.7 years; p-value < 0.001), had higher BMI (31.8 ± 5.5 kg/m2 versus 27.7 ± 4.7 kg/m2; p-value < 0.001) and were significantly more likely to have first-degree relatives with diabetes mellitus (46% versus 10%; p-value < 0.001) and a history of foetal macrosomia in previous pregnancies (67.3% versus 6.5%; p-value < 0.001).

A multivariable logistic regression analysis was used to establish a predictive model for GDM risk in the study population, as outlined in Table 2. The model equation was: LogitP = 6.358 − 0.066 × Age − 0.075 × First trimester BMI − 1.879 × First-degree relative with DM − 0.522 × History of foetal macrosomia. The initial model (AUC = 0.827; 95% CI: 0.712–0.942; p-value < 0.001) was employed for training and adjustments (internal validation). Subsequently, the model derived from the trained set (AUC = 0.814; 95% CI: 0.751–0.877; p-value < 0.001) demonstrated consistent performance when applied to the learning set (AUC = 0.816; 95% CI: 0.743–0.890, p-value < 0.001) supporting the potential utility of the validated model in predicting gestational diabetes mellitus (GDM) risk. A user-friendly and implementable scoring system for identifying high-risk women in clinical practice was established through the derivation of a nomogram from the predictive model (Fig. 2).

The diagnostic performance of the prediction model in the cohort was assessed by ROC analysis (Fig. 3). It had a strong diagnostic accuracy with an area under the ROC curve (AUC) of 0.814 (95% CI: 0.751–0.877; p-value < 0.001). At the predicted probability threshold of 0.745 associated with the maximized Youden’s index of 0.537 and a selected cut-off of 0.5, the model had a sensitivity of 79.2% and specificity of 74.5%, demonstrating a well-accepted predictive and discriminative performance. Hosmer-Lemeshow goodness-of-fit test of 0.595 indicates that the model’s predicted probabilities correspond with the actual outcomes.

In this study, increasing maternal age, higher BMI, a family history of diabetes mellitus in first-degree relative and previous history of foetal macrosomia were found to be the major predictors of GDM in the study population. These factors coupled with the increased risk due to ethnicity (black race) were mostly responsible for the high incidence of GDM in this homogenous cohort. Although there were no significant differences in GDM risk amongst the various tribes in the study population, their uniform racial and ethnic features predisposed them to high risk of GDM.(21).

Previous studies have established the pathological basis for the existence of a strong link between increasing maternal age and higher BMI. This may be due to an age-related mitochondrial dysfunction leading to increased production of reactive oxygen species (ROS) which impair the insulin receptor complex in skeletal muscles amongst many other mechanisms [22], as well as the obesity-induced adipose tissue dysfunction and subclinical inflammation [23], which lead to insulin resistance. Also, having a family history of diabetes mellitus and a previous history of foetal macrosomia have also been found to be significant independent predictors of GDM in previous studies carried out in similar populations within the region [3, 24]. The family history of diabetes mellitus may suggest genetic predisposition to GDM while the history of foetal macrosomia may indicate a possible presence of GDM in the previous pregnancy.

The model in this study compares with previously suggested models in a recent systematic review with acceptable discriminatory ability (AUCs > 0.7) highlighting the reliability and efficacy of our model in predicting GDM risk during the first trimester [11]. Similar to our findings, maternal age, BMI, family history of DM, and history of GDM were the common major predictors in most studies. However, our study excluded the history of GDM while including the history of foetal macrosomia, which often associated with GDM [25], may function as a surrogate marker. Unlike most previous studies employing formula-given models, this study employed a user-friendly visual representation (nomogram) which enhances its practical utility in clinical settings, simplifying risk assessment for healthcare providers. Also, a recent Tanzanian study demonstrated better performance (AUC = 0.970) than our model [26]. This exceptional performance may be due to the use of mid-upper arm circumference (MUAC) and body fat percentage which are better indicators of fat mass than BMI as used in our model. They also ensured enhanced clinical utility of the model by developing a risk score. However, despite its promising nature, the unavailability of a bioelectrical impedance analyzer in most ANC settings in the region limits its application.

Previous studies assessed various early pregnancy prediction models including use of foetal heart rate, maternal clinical parameters and biochemical markers of which some had higher AUC compared to the index study [27‐29], but these were mostly in non-African populations and may not be applicable in resource-constrained environments. In the sub-Saharan African region, there is a scarcity of studies on first-trimester prediction of GDM. A prior Nigerian study, employing the biochemical predictor sex hormone-binding globulin (SHBG), demonstrated superior predictive ability (AUC = 0.874) compared to our model (AUC = 0.814) [12]. However, the practical applicability of the SHBG model is compromised by limited accessibility of SHBG assay in many prenatal centers especially in rural and suburban areas, lack of standardization of the assay, its high cost and prolonged turn-around times. Despite being slightly less predictive, our model prioritizes practical feasibility, considering our study objectives.

Our study has some limitations that warrant consideration for future research and application. Firstly, variations in demographic characteristics and risk factor prevalence among study populations may influence model performance. As our study is specific to a Nigerian population, some risk factors might have varying degrees of influence in different populations emphasizing the need for cautious interpretation in diverse settings. Additionally, the inclusion of clinical variables in models varies, and our focus on easily accessible maternal clinical risk factors might differ from variables included in other models, affecting overall comparability. Also, differences in procedures for GDM diagnosis, including variations in glucose tolerance tests, can contribute to performance variations, emphasizing the importance of standardization, and expansion of the study to a wider, multi-ethnic scale could enhance the model’s generalizability. A shared limitation with many other existing models is the need for extensive external validation, emphasizing the necessity of validating the model in diverse populations to ensure its broader applicability, particularly in resource-poor settings.

In this study, we developed a first trimester prediction model based on key maternal clinical risk factors that are easily accessible in a population of Nigerian pregnant women. The model showed a satisfactory diagnostic performance, and the accompanying nomogram enhances its practical utility, promising improved clinical outcomes by identifying high-risk women, particularly in resource-poor settings were facilities for implementation of universal screening are limited.