Date of Award

3-2025

Degree Type

Dissertation

Degree Name

Master of Education (MEd)

First Advisor

Dr. Felix Oindi

Second Advisor

Dr. Mukaindo Mwaniki

Third Advisor

Dr. Ingrid Gichere

Department

Obstetrics and Gynaecology (East Africa)

Abstract

Introduction 

Pre-eclampsia is a leading cause of significant maternal morbidity and mortality and complicates approximately 2%-4% of all pregnancies. Delivery is the only known definitive treatment. Pre-eclampsia can have unpredictable progression with sudden deterioration leading to severe adverse outcomes.

Currently, no adverse maternal risk prediction model is validated for use in sub-Saharan Africa. The new Pre-Eclampsia Integrated Estimate of Risk Machine Learning (PIERS-ML) model identifies women with pre-eclampsia who are at the lowest and greatest risk of severe adverse maternal outcomes within 48 hours of initial assessment and informs their care. Whereas, it has been validated for use in high and middle-income countries, there is need to validate this model in Kenya to confirm or refute its performance. Positive validation will support adoption for use in Kenya.

Objective 

To evaluate the performance of the new PIERS-ML model in predicting severe maternal outcomes in a Kenyan cohort of women with pre-eclampsia within 48 hours of their initial assessment.

Methodology 

This was a retrospective validation study involving pregnant and post-partum women admitted with pre-eclampsia at Aga-Khan University Hospital and Pumwani maternity hospital. Routinely collected data relevant to pre-eclampsia in the woman’s first admission were collected and adverse maternal outcomes within 48 hours up to a week since admission were recorded.

PIERS-ML model was assessed on its ability to correctly classify women into outcome and no outcome groups by the Area under Receiver Operator Characteristic (AUROC), and precision-recall curves using this dataset. Decision curve analysis was used to assess the clinical utility of the model. vi

Women were grouped into various risk strata based on pre-defined risk thresholds, very low risk defined as negative LR < 0.1, low risk (negative LR of 0.1 to 0.2), high risk (positive LR 5.0-10.0), and very high risk (positive LR >10).

Chi-square, Fischer’s exact, and Mann-Whitney U tests were used to compare categorical and continuous variables. The ANOVA (Analysis of Variance) test was applied to numerical variables. Statistical significance was set at P< 0.05.

Results 

A total of 2,010 women were eligible for the study. Majority were young with a median age of 29 years [range 15-50] and presented at term with a median gestation of 38 weeks [range 20-42]. Of these, 414 (20.6%) women suffered at least one of the composite adverse maternal outcomes within 48 hours of admission while 74 (3.68%) had outcomes after 48 hours and up to 7 days. There were 4 maternal deaths, and they all occurred within the first 48 hours.

PIERS-ML risk stratified well women in the study cohort into very high risk 9 (0.4%), high risk 820 (40.8%), moderate risk 1,181 (58.8%). None of the eligible women was classified as low risk. Adverse events were frequent in the very high- and high-risk categories with event rates of 77.8% and 31.7% respectively.

PIERS-ML performed well showing good discrimination with AUROC 0.71(95% CI 0.68-0.74), AUPRC 0.44. At treatment thresholds of >18.8 and >45.6% representing the high and very high-risk groups respectively, treatment was predicted to achieve a higher net benefit compared with treating all patients or treating none.

Conclusion 

The PIERS-ML model was able to identify women with pre-eclampsia at high and very high risk of adverse maternal outcomes within the first 48 hours of initial admission. It can, therefore, be used clinically for joint decision making among patients and their providers to support guidance on treatment to optimize maternal outcomes.

First Page

1

Last Page

51

Download

Share

COinS