Clustering of adverse perinatal outcomes in women with multiple sensitising events: a data-driven approach using clinical and immunohematologic profiles - 12/11/25
, Garima Yadav b, Swati Asati c, Pratibha Singh bHighlights |
• | Cluster-guided surveillance: Cluster 1 requires iron optimisation, serial MCA Dopplers, and early fetal medicine referral. |
• | Hidden-risk Cluster 3: Cluster 3 seronegative women may benefit from closer intrapartum monitoring and placental testing. |
• | Decision-support integration: Rule engine is transparent, light, and integrable into EMR for automated risk flagging. |
• | This clustering framework, based on five routine antenatal parameters, identified two novel high-risk phenotypes. Cluster 1 included antibody-positive pregnancies with anaemia and sensitisation, while Cluster 3 comprised seronegative women with preserved haemoglobin but poor neonatal adaptation (Apgar <7). Both were independently associated with adverse perinatal outcomes, even after adjustment for comorbidities. The model was robust across analytic checks, offering a reliable alternative to titre-based screening. By enabling personalised risk stratification, it supports targeted care—iron optimisation and serial Dopplers for Cluster 1, and enhanced intrapartum monitoring with placental testing for Cluster 3. |
Abstract |
Background |
Red-cell alloimmunisation is a preventable driver of haemolytic disease of the foetus and newborn, yet most risk scores rely on single-parameter thresholds and overlook clinically important heterogeneity.
Objective |
To uncover latent phenotypes among sensitised pregnancies by clustering routinely collected clinical and immunohaematologic variables.
Methods |
We retrospectively analysed 2084 antenatal records (2020–2021). Five variables—maternal antibody status, haemoglobin (Hb) concentration, cumulative number of sensitising events, gestational age at first positive antibody screen, and parity—were multiply imputed and scaled. A rule-based approximation of Gaussian mixture modelling and HDBSCAN assigned women to five clusters. Internal validity was assessed with the silhouette coefficient (0.41) and Davies–Bouldin index (0.88). Multivariable logistic regression evaluated the association between cluster membership and a composite adverse perinatal outcome, adjusting for maternal age and comorbidities.
Results |
Five clinically coherent clusters emerged (Cluster 1 = 13, Cluster 2 = 848, Cluster 3 = 26, Cluster 4 = 36, Cluster 5 = 513; 648 records lacked sufficient data for assignment). Cluster 1 combined antibody positivity with marked anaemia (mean Hb 8.6 ± 1.3 g/dL) and showed the highest risk of adverse outcome (adjusted OR 4.3, 95 % CI 2.7–6.8, p < 0.001). Cluster 3—seronegative women with preserved Hb (≥12 g/dL) but neonatal depression (1-min Apgar < 7 in 100 % of cases)—represented an unexpected high-risk phenotype. Cluster 5 (antibody-negative, Hb ≥ 12 g/dL, 1-min Apgar ≥ 7) served as the low-risk reference.
Conclusion |
Unsupervised clustering of simple antenatal parameters reveals hidden risk profiles that outperform single-threshold screening. This data-driven phenotyping could refine surveillance intensity and transfusion strategies in sensitised pregnancies.
Le texte complet de cet article est disponible en PDF.Keywords : Alloimmunisation, Haemoglobin, Unsupervised clustering, Silhouette coefficient, Perinatal outcome, Risk stratification
Plan
Vol 32 - N° 4
P. 350-356 - novembre 2025 Retour au numéro
Article précédent
- Cryoprecipitate utilization review in light of recent guidelines: a single-center prospective analysis of haemato-oncology patients
- Joyisa Deb, Daljit Kaur, Gita Negi, Zikra Syed, Aswin K. Mohan, Priyanka Rathod, Ashish Jain, Uttam Kumar Nath
- Stratification of blood donors and understanding motivational and deterrent factors to develop effective recruitment strategies
- Azeez Mohammed, Shamee Shastry, Ganesh Mohan, Deepika Chenna, Deep Madkaiker
Bienvenue sur EM-consulte, la référence des professionnels de santé.
L’accès au texte intégral de cet article nécessite un abonnement.
Déjà abonné à cette revue ?