Abstract:
Background: Preeclampsia is one of the leading cause of maternal and perinatal morbidity and mortality in low-resource settings, including Ethiopia. Early identification, timely clinical decision, and care can prevent both maternal and neonatal morbidities. Therefore, a prognosis risk score for preeclampsia can be developed based on easily available predictors and used by health professionals as a simple clinical tool in their decision-making.
Objective: To develop and validate a prognosis risk score model for preeclampsia among pregnant women who had antenatal care visit at Debre Tabor comprehensive specialized hospital, Ethiopia. Methods: A retrospective cohort study was conducted on January 2nd - February 16, 2023 and a total of 1,100 pregnant women were included. We selected a client chart using a simple random sampling technique. Data were extracted using structured checklist prepared in the Kobo Toolbox and exported to STATA version 14 and R version 4.2.2 software for data management and further analysis. Stepwise backward multivariable analysis was done. A simplified risk prediction model was developed based on maternal characteristics using a binary logistic regression model and the model's performance was assessed by discrimination power and calibration. The internal validity of the model was evaluated by bootstrapping, which replicates the sample 2,000 times. Decision curve analysis was used to determine the clinical impact of the risk score.
Result: Incidence of preeclampsia was 13.27% (95% CI: 11.2, 15.2). Predictors such as age>35, primigravida, chronic hypertension, diabetic mellitus, multiple gestation, family history of preeclampsia and mean arterial pressure >90mmHg were kept in the reduced multivariable logistic regression. The simplified risk score model had a total score of 18. The discriminatory power of the model was 85.9% (95% CI: 0.823, 0.895), and the calibration test was at a p >0.05. The optimal cut-off for classifying risks as low or high was 3. At this cut point, the sensitivity is 72.6%, specificity is 83.6% and accuracy is 82.18%. It was internally validated using the bootstrapping method and has an optimism of 0.0005. The decision curve analysis revealed the model's net benefit is greater than the treating all or none approach.
Conclusion and recommendation: The developed risk score model had excellent discrimination performance. Therefore, it can be applied in clinical practice. We recommend that the researcher externally validate the model and for clinicians use a prognosis risk score.
