Data Sharing for the Public Good

One-half million people joined the UK Biobank, described in the consent form as a “public good resource” with an overarching goal to support research to “promote health throughout society” (1). The study by Arthur et al. (2) in this issue of the Journal of the National Cancer Institute, which used this shared resource, is a positive step towards fulfilling the contract the UK Biobank made with participants to serve the public good. The investigators used available genetic and questionnaire data to examine whether behavioral factors may modify inherited genetic risk for breast cancer and found that a healthier lifestyle was associated with a lower risk of breast cancer, especially among those with the highest polygenic risk score. The analysis highlights the societal benefit of data sharing and lends support to further defining synergistic models to inform population guidance.

Arthur et al. leveraged the available data to calculate polygenic risk scores associated with breast cancer risk and a healthy lifestyle index for 146 326 women. By pairing combinations of single nucleotide polymorphisms (SNPs) (via a polygenic risk score) and combinations of modifiable risk factors (via a heathy lifestyle index), the results of the study lend support for guidance on risk factors that can be modified or controlled, even among those at high risk based on genetic factors. This may help to alleviate some of the potential concerns associated with using a genetic-based risk score if there is the perception that nothing can change the risk. The results of the combined approach in this study can inform how risk may be communicated with a more integrated understanding of both nonmodifiable and modifiable risk factors.

Caution should be exercised in interpreting the results of this study, particularly with respect to use of risk scores. Applying risk prediction models to the individual and extrapolating models to other populations, for example non-European populations, may not be appropriate. A comprehensive approach to breast cancer risk prediction is required to reflect the complexity of cancer etiology that includes environmental, lifestyle, constitutional, and genetic factors. Risk prediction models for the general population continue to evolve. The initial model to have widespread use for the general population of women is the Breast Cancer Risk Assessment Tool (3), also known as the “Gail Model” (4). The model has been widely validated (5) and is based on six known risk factors, including age, age at first menstrual period, age at first live birth, family history of breast cancer, history of breast biopsy, and race/ethnicity. Over the years, other models have been developed, including additional risk factors, expanding family history, and adding SNPs, with only modest improvements in accuracy of prediction and discrimination (6). The polygenic risk score used in this study focused on genetic factors and was developed among European populations and thus may not be generalizable to non-European populations or individuals.

Although this study demonstrates the value of access to large research resources, when using existing data, limitations exist. For example, in this case, data were not available to classify breast cancer subtypes. Additionally, modifications were required to calculate the polygenic risk scores (7) and the healthy lifestyle index (8) due to missing data elements. Data were available for 304 of the 313 SNPS included in the study by Mavaddat et al. (7) that developed and validated a risk score based on SNPs from genome-wide array data for individuals of European ancestry from 69 studies, including the UK Biobank. To examine key modifiable risk factors, several types of indices have been designed (8, 9), and, more recently, the World Cancer Research Fund (WCRF) Expert Third Report (10) led to the development of a standardized scoring system, the WCRF Score, that could be applied in large cohort studies (11). The WCRF Score includes eight components (weight; physical activity; whole grains, vegetables, fruits, and legumes; fast foods; and processed foods; red and processed meat; sugar-sweetened drinks; alcohol; and breastfeeding [optional]). For this analysis, the UK Biobank further modified the WCRF Score to create a healthy lifestyle index with five of the eight lifestyle factors: diet (a summary score only), alcohol, physical activity, weight (for postmenopausal women only), and the additional construct of smoking. These modifications may influence results when the scoring approaches are less comprehensive than originally designed (ie, specific aspects of diet to limit such as fast foods and processed foods and sugar-sweetened drinks), when different constructs are included (ie, smoking), and when the scoring strategy is altered (ie, subjective population-based tertiles). Additionally, changes in scoring may affect the ability to compare across studies and populations. Thus, it is important to clearly note the methods used when modifying scores.

These notes of caution do not diminish the positive message that following a healthy lifestyle is an important public health measure for all regardless of their genetic risk. Although some studies have taken a more reductionist approach that focused on only multiple SNPs or only lifestyle behaviors, both the availability of data and development of more comprehensive tools, indices, and methods will allow for enhanced models to better consider the complexity of cancer etiology and prevention. The use of scores and indices here provides one example of the importance of taking an integrative approach to cancer control and prevention, and future research can further extend these efforts.

The study by Arthur et al. demonstrates that the information provided by the participants, and broadly shared, can inform public health measures and serve the “public good.” The willingness of the individuals who took part in the UK Biobank to provide and widely share their data, with no clear benefit to them individually, is lauded. Other opportunities are currently underway, including All of Us, which aims to gather data from 1 million Americans (12). Large-scale data sharing and population studies will benefit the communities they represent, as scientific evidence is translated into actionable next steps that allow for improved strategies to reduce and manage risk at the individual and population levels. Analyses such as this one provide important context to add to the “evidence-to-action” cycle and support large-scale data sharing, research to continue to refine scores, and further optimize research applications of scores and translation of results into population guidance.

Notes

The views expressed are those of the author and do not necessarily reflect the views of the National Cancer Institute.

The authors have no conflicts of interest to disclose.