Population-Targeted Risk Prediction to Guide Breast Cancer Screening in Cameroon - Project Summary/Abstract:
As cancer becomes an increasingly important cause of mortality in sub-Saharan Africa, nations must improve
their capacity to diagnose and treat the malignancies that most affect their populations. Breast cancer is the
most common cancer and cause of cancer-related mortality for women in Cameroon, where two thirds of
patients present with stage III or IV disease, and five-year survival rates are less than 30%. A longstanding
collaboration between the Cameroonian Ministry of Public Health, the University of Buea, and the University of
California, Los Angeles (the MBLA partnership) plans to conduct key stakeholder interviews in order to assess
the feasibility and acceptability of implementing a breast cancer screening program that is initially offered to
high-risk women. Risk-based screening aims to equitably target early screening efforts while ensuring that
diagnostic and treatment capacity are sufficient to manage lesions identified through screening. Such a
program requires a breast cancer risk prediction model that is applicable and acceptable to Cameroonian
women and feasible to evaluate through a community-based screening program. Studies in the United States,
Asia, and Nigeria demonstrate that breast cancer risk prediction models perform best when they are ethnic
group-specific, but no breast cancer risk prediction model has been developed for Cameroonian women. The
African Breast Cancer case-control Study (ABCS) contains breast cancer risk factor information for women
from Nigeria, Uganda, and a small subset of women from Cameroon. Traditional approaches to risk prediction
will likely suffer from small sample size in models trained on Cameroon data only or from bias in models
trained and validated on the full, ethnically diverse dataset. In Aim 1, a subgroup-specific cross-validation
method will be incorporated into the Super Learner ensemble prediction algorithm to develop a breast cancer
risk prediction model that incorporates all ABCS data but is optimized for Cameroonian women. Aim 2
addresses MBLA members’ concerns that certain risk factors from ABCS will be difficult to evaluate by
community survey. Targeted learning methods will be used to define metrics for comparing risk prediction
models including and excluding these variables so that Cameroonian stakeholders can evaluate whether to
include these risk factors in their breast cancer risk prediction model. In Aim 3, an R shiny app will be
developed, tested, and optimized in order to facilitate use of a Cameroon-specific breast cancer risk model in a
future screening program. The methods developed and tested in this project could help to optimize cancer risk
prediction models for other ethnic groups with limited data in sub-Saharan Africa and globally. This research
will be conducted under the mentorship of the MBLA collaboration, UC Berkeley’s leaders in the field of
targeted learning, and the UCSF Global Cancer Program. By providing protected time for training, research,
and career development, this grant will facilitate the applicant’s progress towards becoming a breast surgical
oncologist researching methods of improving access to cancer care in Africa.
