Using ultrasonography, laboratory blood tests and maternal characteristics to predict Pre-eclampsia and adverse pregnancy outcomes at St. Mary’s Hospital Lacor, Northern Uganda

Abstract

Background: Preeclampsia causes 12% to 19% of maternal deaths in Uganda. Complications include preterm birth, stillbirth and low birth weight. Early diagnosis and timely delivery improve pregnancy outcomes. Nevertheless, due to poor infrastructure in northern Uganda, early prediction and diagnosis with eventual treatment may save lives. Therefore, we set out to predict pre-eclampsia and adverse pregnancy outcomes using maternal history, laboratory characteristics and uterine artery Doppler indices in northern Uganda. Methods: This prospective cohort study recruited 1,285 pregnant mothers at 16-24 weeks. Participants' history, physical findings, blood tests (full haemogram, renal and liver function) and uterine artery Doppler indices were recorded. One thousand four (1,004) enrolled pregnant mothers had complete delivery records. Preeclampsia, preterm birth, stillbirth and low birth weight were the desired outcomes. We built models in RStudio for predicting preeclampsia, preterm birth, stillbirth and low birth weight. Statistical analysis: t-tests, Mann-Whitney tests and Pearson’s chi-square were used to compare means, medians, and proportions, respectively. We calculated incidences of low birth weight at term, pre-eclampsia, preterm birth and stillbirth. We identified from maternal history, physical examination, uterine artery Doppler indices and blood tests, maternal risk factors for preeclampsia, preterm birth, stillbirth and low birthweight at term using the logistic regression models in RStudio. We re-processed the data using the ROSE package to produce synthetic data (test data) to evaluate the (original) model performance and validated the models using K-fold cross-validation. We weighed each variable contribution in the prediction model. Results: The incidence of pre-eclampsia, preterm birth, stillbirth and low birth weight at term were 4.3%, 11.6%, 2.5% and 5.7%, respectively. The predictors of these adverse pregnancy outcomes were Maternal age ≥ 35 years, nulliparity, personal history of preeclampsia, tertiary level of education, BMI ≥ 26.5Kg/m2, diastolic hypertension, bilateral end-diastolic notch, lateral placental location, serum GGT ≥30 IU, serum ALT 12 – 49 IU, white blood cell count ≥ 11,000 cells/µl, lymphocyte count of 800-4000 cells/µl, haemoglobin level ≥ 12.1g/dL and serum ALP <98 IU. The models had a good fit if McFadden's pseudo-R2 was between 0.2–0.4. Maternal history, laboratory tests and uterine artery Doppler sonography predicted pre-eclampsia with 84.9% AUC and McFadden’s pseudo-R2 of 0.30. The variables with weights up to ≥6.0 predicted adverse pregnancy outcomes by ≥60% AUC and ≥ 50% accuracy. Conclusion: The prediction models for preeclampsia had AUC of 71.4% to 84.9%. Since the patients present to prenatal clinics with different predictors, the variable weights adding up to ≥6.0 predicted adverse outcomes by ≥60% AUC. These may help to develop prenatal screening tools for preeclampsia in Uganda. We recommend incorporating the prediction of preeclampsia into prenatal care and strengthening the referral pathways for those found to be at risk.