Model for accurate prediction of postpartum metabolic syndrome developed

Model for accurate prediction of postpartum metabolic syndrome developed | Image Credit: © mi_viri - © mi_viri - stock.adobe.com.

A model has been developed for accurate prediction of metabolic syndrome within 2 to 7 years following delivery, as published in the American Journal of Obstetrics & Gynecology.¹

Cardiovascular disease (CVD) accounts for 22% of deaths in female individuals, making it the leading cause of death in the United States. Metabolic syndrome, a series of CVD risk factors, has increased in prevalence over time in the United States from 16.2% to 21.3% in those aged 20 to 39 years between the 2011 to 2012 and 2015 to 2016 periods.²

CVD risk may be lessened through the treatment of metabolic syndrome.¹ This highlights the need for a machine model including a wide variety of factors to predict future metabolic syndrome development.

To develop machine learning models for metabolic syndrome prediction, investigators conducted a secondary analysis of data from the Nulliparous Pregnancy Outcomes Study. Participants included patients with singleton pregnancy recruited between 6- and 13-weeks’ gestation.

Patients who completed the study were invited to participate in the secondary analysis if they were aged at least 18 years, agreed for further contact, and had available obstetric delivery data. These patients were visiting without pregnancy at 2 to 7 years since the index pregnancy and at least 6 months since the previous pregnancy.

Metabolic syndrome was reported as the primary outcome of the analysis. This outcome was defined by the presence of at least 3 criteria including waist circumference of at least 88 cm, fasting glucose of at least 100 mg/dL, high-density lipoprotein (HDL) cholesterol under 50 mg/dL, triglycerides of at least 150 mg/dL, and blood pressure over 130/85 mm Hg.

Risk factors during pregnancy were assessed and categorized into demographic, intrapartum, social determinants of health, and serum analytes. Further variables were selected through an exploratory factor analysis. Variables were then compared between patients with metabolic syndrome and those without metabolic syndrome.

Data regarding metabolic syndrome was identified in 4225 individuals aged a mean 27 years, with 17.8% developing metabolic syndrome while 82.2% did not. Seventy of the selected variables had sufficient variance among these patients for further analysis.

The most important variables in the prediction model included HDL level, insulin level, high-sensitivity C-reactive protein, hip circumference, neck circumference, third trimester systolic and diastolic blood pressure, years lived in the United States, second trimester diastolic blood pressure, and systolic blood pressure.

These variables were included in the forest model. Similar variables were used in the lasso model, alongside first trimester Perceived Stress Scale score, maternal age, and family income. Neck circumference, years living in the United States, and first trimester blood pressure were not included in the top 10 variables in this model.

The forest model had an area under the receiver operating characteristic curve (AUROC) of 0,878 when including all 70 variables. In comparison, the AUROC of the lasso model was 0.850, indicating a higher AUROC from the forest model. This led to the creation of forest models using the top 10, 5, 4, 3, and 2 variables.

Inferiority was reported for the forest model using the top 2 variables compared to the original model, but not for the models using the top 3, 4, 5, and 10 variables. The final model selected used the top 3 variables and had an AUROC of 0.867.

The optimal cutoff point of 18% had a sensitivity of 0.78, specificity of 0.76, positive likelihood ratio of 3.23, and negative likelihood ratio of 0.29. Random forest models and lasso models both displayed superiority vs treating all individuals when the threshold probability was from 0% to 80%.

These results indicated accuracy from the validated models toward the predication of metabolic syndrome from 2 to 7 years following delivery. Investigators concluded these models may be used for identifying the risk of metabolic syndrome in postpartum individuals.

References

1. Kawakita T, Greenland P, Pemberton VL, et al. Prediction of metabolic syndrome following a first pregnancy. Am J Obstet Gynecol. 2024;231:649.e1-19. doi:10.1016/j.ajog.2024.03.031
2. Hirode G, Wong RJ. Trends in the prevalence of metabolic syndrome in the United States, 2011-2016. JAMA. 2020;323(24):2526-2528. doi:10.1001/jama.2020.4501