Maternity Clot Risk Model to Predict Women at Risk of VTE

External validation of Maternity Clot Risk model to predict women at risk of postpartum venous thromboembolism

This study revealed that the Maternity Clot Risk model is flexible in setting new treatment thresholds and that adopting it in clinical practice may help optimize the use of low-molecular-weight heparin (LMWH) postpartum to better target high-risk groups.

• In high-income countries, venous thromboembolism (VTE) is the leading cause of direct maternal mortality.
• In 2016, Sultan et al. developed a risk prediction score (“Maternity Clot Risk”) for estimating the risk of VTE during the first 6 weeks after childbirth.
  • However, external validation of risk prediction models is needed for effective utilization in everyday clinical practice.
• This study aimed to conduct an independent external validation of the predictive performance of the “Maternity Clot Risk” model and assess its performance across various groups of postpartum women from England.

• The study enrolled 535,583 women with 700,185 deliveries.
• First VTE events during 6-week post-delivery follow-up: 549 (absolute risk = 7.8 per 10,000 deliveries [95% confidence interval {CI}: 7.2–8.5]).
• Prediction of VTE risk using the Maternity Clot Risk model: Predicted VTE risk was 0 to 745 per 10,000 deliveries (maximum equivalent to 7.5% VTE risk; median predicted risk = 5.3 per 10,000, 10th percentile 3.0 per 10,000, 90th percentile 14.2 per 10,000)
  • 5.0%, 0.6%, and 0.1% women had a predicted VTE risk of ≥0.2%, ≥0.5%, and ≥1.0%, respectively.
• Overall model performance: C-statistic = 0.67 (95% CI: 0.65–0.70); calibration slope = 0.84 (0.74–0.94); calibration-in-the-large = 0.02 (−0.06 to 0.10)
• Performance by subgroup: The model had similar performance in women (1) from different socioeconomic groups, (2) from different geographic regions, (3) giving birth in different time periods, and (4) of different age groups, with minimal heterogeneity in C-statistic and calibration slope
• Comparison with the RCOG guideline: Decision curve analysis showed that although the net benefit was small, the model was better than a treat-all or treat-none strategy (risk threshold: 10–30 per 10,000 deliveries)
  • The model had higher net benefit than the RCOG guidelines (risk threshold: 5–30 per 10,000 deliveries).
  • For identifying the same proportion of women based on predicted risks, the Maternity Clot Risk model had a slightly higher sensitivity than the RCOG guidelines; but the difference was not statistically significant.
• Re-calibration results: After shrinking the original predictor coefficients by 0.79 (0.84 * 0.94) and re-estimating the intercept, the calibration slope was 1.00 (0.88–1.11)

• Electronic health data were used for the development and validation of risk prediction models; VTE events could not be individually validated.
• Misclassification of VTE events in data development and validation would attenuate the effect of predictor variables on VTE risk.
• Prophylactic doses of LMWH could not be separated from therapeutic doses.
• QResearch covered more recent data (2004 onward) compared to the data used for model development (1997 onward).
• Primary care prescriptions selected by the Maternity Clot Risk model were for women receiving 6-week prophylaxis due to a previous VTE.
• More than 17% women had missing values on pre-pregnancy body mass index and their baby's birth weight.
• VTE events around pregnancy may be coded differently in practices using SystmOne, limiting the generalizability of the model to these practices.
• Owing to some miscalibration in applying the model in the QResearch population, risk for women with high-predicted values may have been overestimated.
• The score cannot be applied to women with previous VTE or known high-risk hereditary thrombophilia.