Evaluación de la probabilidad clínica de embolia pulmonar en obstetricia a través de las puntuaciones de GENEVA y YEARS adaptadas al embarazo: Un estudio multicéntrico de cohorte

Abstract

Introducción: El tromboembolismo pulmonar (TEP) es una causa importante de mortalidad en pacientes obstétricas. Debido a su heterogeneidad clínica constituye un reto diagnóstico; los modelos tradicionales de predicción no han realizado una adecuada estratificación de probabilidad diagnostica en la población obstétrica. Sin embargo, los modelos adaptados a el embarazo GENEVA y YEARS mostraron un mejor rendimiento diagnóstico, facilitando la toma de decisiones y descartando casos de forma segura. Estos fueron validados en países de alto ingreso cuyas características poblacionales difieren de los países de menores ingresos. En Colombia el TEP ha ocupado el cuarto puesto de causas de mortalidad materna, por lo que es importante contar con este tipo de herramientas que permitan mejoras en la atención. Objetivos: Evaluar y validar los modelos GENEVA y YEARS adaptados al embarazo para la predicción diagnostica de TEP en la población obstétrica en diferentes regiones de Colombia. Métodos: Se realizó un estudio de cohorte multicéntrico, ambispectivo, donde se incluyeron pacientes obstétricas con sospecha de TEP, se evaluaron los modelos GENEVA y YEARS para cada sujeto y los resultados se compararon con los hallazgos del estudio imagenológico. Se evaluó la capacidad discriminativa mediante realización de curvas ROC. Para la calibración se utilizaron 3 métodos: grafica de calibración, prueba de Hosmer-Lemeshow y puntaje de Brier. Finalmente se estableció su utilidad clínica a través de análisis de beneficio neto. Se realizaron análisis por separado con la base de datos original y la que incluía datos imputados. Resultados: Se recolectaron 95 pacientes en 6 instituciones del país, dentro de las cuales 26 tuvieron TEP evidenciado por imagen. Los resultados de los análisis de la base original y la de datos imputados mostraron ser similares. En general ambos modelos mostraron una alta sensibilidad (>84%) pero baja especificidad (<30%). El modelo YEARS tuvo una mayor AUC (0.734; IC = 0.553–0.914) comparado con

Introduction: Pulmonary thromboembolism (PTE) is a significant cause of mortality in obstetric patients. Due to its clinical heterogeneity, it constitutes a diagnostic challenge; traditional prediction models have not adequately stratified diagnostic probability in the obstetric population. However, the pregnancy-adapted GENEVA and YEARS models have demonstrated better diagnostic performance, facilitating decision-making and safely ruling out cases. These models were validated in high-income countries, whose population characteristics differ from those of lower-income countries. In Colombia, PTE has ranked as the fourth leading cause of maternal mortality, making it important to have tools that can improve patient care. Objectives: To evaluate and validate the pregnancy-adapted GENEVA and YEARS models for the diagnostic prediction of PTE in the obstetric population in Colombia. Methods: A multicenter, ambispectivo cohort study was conducted, including obstetric patients with suspected PTE. The GENEVA and YEARS models were evaluated for each subject, and the results were compared with imaging study findings. Discriminative capacity was assessed using ROC curves. Calibration was evaluated using three methods: calibration plots, the Hosmer-Lemeshow test, and the Brier score. Finally, clinical utility was determined through net benefit analysis. Separate analyses were performed using both the original database and the database with imputed data. Results: A total of 95 patients were collected from six institutions in the country, of which 26 had imaging-confirmed PTE. The analyses of both the original and imputed data sets yielded similar results. Overall, both models demonstrated high sensitivity (>84%) but low specificity (<30%). The YEARS model had a higher AUC (0.734; 95% CI = 0.553–0.914) compared to GENEVA (0.606; 95% CI = 0.438–0.773) and showed a significant association with PTE diagnosis (OR = 2.21, 95% CI: 1.00–4.86, p = 0.048), unlike GENEVA. The calibration plot showed high variability and discrepancies for both models, particularly for GENEVA. However, the Hosmer-Lemeshow test yielded p-values >0.05 for both GENEVA and YEARS (chi² = 1.78; p = 0.619 and chi² = 0.18; p = 0.594), indicating good model fit to the observed data, with a better Brier score for YEARS (0.18 vs. 0.19). From a clinical utility perspective, both models demonstrated an adequate net benefit at low probability thresholds; however, YEARS maintained positive net benefit values up to thresholds of 30%, while GENEVA reached negative values at thresholds as low as 25%. Conclusions: The pregnancy-adapted GENEVA and YEARS prediction models show good sensitivity for diagnosis and adequate calibration. However, GENEVA appears to have limitations in its discriminative capacity, in contrast to YEARS, which demonstrated better discriminative ability and greater net benefit, especially in cases of lower probability suspicion.