Statistics Seminar - 02/07/23

Abstract: 

Background: Female childhood cancer survivors are at much higher risk of premature menopause (a.k.a. primary ovarian insufficiency - POI), which causes infertility and the accompanied wide range of post-menopausal symptoms. To counsel survivors, accurate risk estimates at different ages are needed.
 
Methods: We used data from 7891 participants in the childhood cancer survivors study (CCSS) to develop the prediction algorithms and 1349 survivors in the St Jude Life study (SJLIFE) for external validation. Ovarian status was ascertained from longitudinal self-reported menstrual surveys (CCSS) or clinical assessment (SJLIFE). Demographic and treatment data were abstracted from medical charts. We used Cox PH, and age-specific logistic regression and age-specific XGBoost to develop separate POI risk algorithms as survivors aged from 21 to 40 years, using treatment data of three different precision levels. For a subset of 1985 CCSS participants with genotype data, we evaluated the contributions of polygenic risk scores (PRSs) from published general population genome-wide association studies of natural menopause age to predict POI risk by age 40.
 
Findings: The most influential predictors were minimum of the radiation doses to the right or left ovary (MORD), hematopoietic stem-cell transplant (HSCT), diagnosis age, and chemotherapy agents. As survivors aged from 21 to 40 years, the estimated POI prevalence increased from 7.9% to 18.6% (CCSS) and 8.8% to 16.8% (SJLIFE). Across this age range, the best performing algorithm AUROCs (area under ROC curve) were 0.76-0.80 in the CCSS cohort and 0.85-0.90 in the SJLIFE cohort, while the AUPRCs (area under precision-recall curve) increased from 0.45 to 0.60 in CCSS and 0.58 to 0.80 in SJLIFE. Adding PRSs improved calibration (Spiegelhalter-z decreased from 11.4 to 0.1), but not the AUROC/AUPRC in the subset of CCSS participants with genotype data. 
 
Interpretation: Given cancer diagnosis and treatment information, these POI risk prediction algorithms perform excellent and can inform decisions regarding fertility preservation interventions among female childhood cancer survivors. An APP has been developed to visualize POI risk as survivors age.

Bio:

Dr. Yuan is an associate professor at the University of Alberta School of Public Health, which she joined in 2011. She received a BSc. in Biochemistry (Nanjing University, P. R. China), a MSc. in Animal Behavior (Michigan State University, USA), followed by a Master in Biostatistics and a PhD in Statistics (University of Waterloo, Canada). Dr. Yuan's research program focuses on developing and applying biostatistical methods in cancer related population health and biomedical research using data from observational studies. Methodologically, her research interests are statistical prediction and classification, and developing appropriate metrics for quantifying the prediction performance. Dr. Yuan's ongoing applied health research focuses primarily on: 1) risk prediction of late effects in childhood, adolescent and young adult cancer survivors; 2) brain tumour surveillance and developing artificial intelligence tools for improving cancer surveillance.