Performance of the Pre-Eclampsia Integrated Estimate of Risk-Machine Learning (PIERS-ML) model in a Kenyan cohort of women with pre-eclampsia- a retrospective test derivation validation study.

OBJECTIVE

Pre-eclampsia is a cause of significant maternal morbidity and mortality, with delivery initiating resolution. The Pre-eclampsia Integrated Estimate of RiSk-machine learning (PIERS-ML) tool provides individualized risk estimates to guide joint decision-making for women with pre-eclampsia. While it has been externally validated in the UK, our objective was to test PIERS-ML performance in Kenya.

DESIGN

Retrospective cohort validation study.

SETTING

Two tertiary hospitals in Nairobi, Kenya.

POPULATION: Women admitted with pre-eclampsia who had not experienced any element of the main outcome measure.

METHODS

Test performance was assessed by stratification capacity, area under the receiver-operator curve (AUROC), area under the precision-recall curve (AUPRC), and decision curve analysis.

MAIN OUTCOME MEASURES: Any component of the PIERS primary outcome of maternal death or major maternal organ dysfunction within 48 h of admission.

RESULTS

Among 2,002 women with pre-eclampsia, 408 (20.4%) experienced an adverse maternal outcome within 48 h of admission (including 4 deaths) and a further 74 (3.7%) between 3-7 days. Missingness was substantial for most laboratory variables, particularly at the public hospital. Despite this, individual level imputation enabled model assessment. PIERS-ML demonstrated good discrimination (AUROC 0.68; AUPRC 0.40) and clinically meaningful stratification: high-risk women had doubled outcome rates and the single very high-risk woman experienced an event. Decision curve analysis showed greater net benefit than treating all or none. Patterns of missingness and more severe outcomes suggested a higher risk Kenyan case mix.

CONCLUSION

In a high-morbidity Kenyan cohort, the PIERS-ML tool accurately identified personalised risk in women admitted with pre-eclampsia.