Sesión de Biometría

La Sociedade Portuguesa de Estatística (SPE) junto con la SGAPEIO organizan la siguiente sesión paralela especial de Biometría el viernes 27 de octubre, 12:30 - 13:30 en el Salón de Grados 2.

  • Impacto da recidiva na supervivencia en cáncer colorrectal: enfoque multiestado

    • Vanesa Balboa Barreiro1,2, Sonia Pértega Díaz1,2, Teresa García Rodríguez2, Cristina González Martín1,2, Teresa Seoane Pillado1,2

      • 1 Universidade da Coruña, Rheumatology and Health Research Group, Department of Health Sciences, Faculty of Nursing and Podiatry, Ferrol, Spain

      • 2  Instituto de Investigación Biomédica de A Coruña (INIBIC), Nursing and Health Care Research Group, A Coruña, Spain.

  • Comparative Evaluation of Survival Analysis Models: Cox Proportional Hazards versusRandom Survival Forest Using Real-World Data

    • Ana Isabel Coelho Borges1, Mariana Carvalho2

      • 1 CIICESI, ESTG, Politécnico do Porto

      • 2  CIICESI, ESTG, Politécnico do Porto

  • Disentangling Hospitalisation Trajectories

    • Rita Gaio1,2, Daniel Cordeiro1, Bárbara Peleteiro3,4,5, Lucybel Moreira3, Elsa Guimarães3, Raquel Cadilhe3, Ana Azevedo3,4,5

      • 1 Department of Mathematics, Faculty of Sciences, University of Porto

      • 2  Centre of Mathematics of the University of Porto

      • 3 Centre of Hospital Epidemiology, Centro Hospitalar Universitário de São João

      • 4 Department of Forensic and Public Health Sciences, and Medical Education; Faculty of Medicine, University of Porto

      • 5  Public Health Institute, University of Porto

 

Impacto da recidiva na supervivencia en cáncer colorrectal: enfoque multiestado

RESUMEN:

O cancro é unha das enfermidades máis frecuentes a nivel mundial, encontrándose entre as principais causas de morbilidade e mortalidade, sendo o cancro colorrectal (CCR) a segunda causa de morte por tumor. Estudáronse 994 casos incidentes de CCR sometidos a resección con intención curativa co obxectivo de analizar o efecto de diferentes factores prognósticos e o impacto da recorrencia no prognóstico da enfermidade, mediante o desenrolo de modelos multiestado (MSM). Na análise multiestado, incluíuse a recorrencia como un estado intermedio na progresión da enfermidade. Ao final do seguimento, segundo as probabilidades de ocupación, entorno ao 50% dos pacientes estaban vivos e sen recorrencia, o 4,2% vivos con recorrencia e o 8,8% mortos por CCR. A recorrencia ten un impacto negativo no prognóstico do CCR. O estadio avanzado de cancro no diagnóstico de CCR identificouse como o factor prognóstico máis alto para as mortes sen recorrencia.

 

Comparative Evaluation of Survival Analysis Models: Cox Proportional Hazards versus Random Survival Forest Using Real-World Data

RESUMEN:

Survival analysis is a crucial statistical tool used in various fields, including medicine, finance, andengineering, to analyse time-to-event data. In this study, we aimed to compare the performance of threedistinct survival analysis models: the Cox Proportional Hazards (PH) model, Random Forest SurvivalAnalysis (RFSA) with axes-based splitting and Accelerated Oblique Survival Random Forest (AOSRF). Our objective is to assess the predictive accuracy of these models using real-world data, with a focus onmetrics such as the Area Under the Curve (AUC), Brier score, and Integrated Prediction Accuracy (IPA). Our preliminary results indicate that the choice of survival analysis model significantly impacts predictiveperformance. While the Cox PH model demonstrated interpretability, it struggled with capturing complexnon-linear relationships. RFSA with axes-based splitting and AOSRF exhibited promising results,outperforming the Cox PH model in terms of AUC and Brier score. In conclusion, our study contributes to the understanding of survival analysis modelling bycomparing the Cox PH model with machine learning-based approaches like RFSA and AOSRF. We shedlight on the trade-offs between interpretability and predictive accuracy by utilising real-world data andevaluating multiple metrics.

 

Disentangling Hospitalisation Trajectories

RESUMEN:

The efficient management of patient flows in healthcare settings is a complex and critical issue. This work presents an inventive approach to analyse pediatric hospitalisation trajectories at Centro Hospitalar Universitário de São João, a portuguese public teaching tertiary care hospital in the city of Porto.
The analysis comprises data from 3133 hospitalisation episodes of newborns, infants, children and adolescents, aged between 0 months and 18 years old, from December 2021 to November 2022, excluding newborns who were discharged within 48 hours of birth. Each hospitalisation trajectory is represented by a sequence of a maximum of nine paediatric services and the associated lengths of stay in each of them.