Population Attributable Factor (PAF) Project

Maarit Laaksonen, Claire Vajdic

For the PAF project, we received funding from WCRF / Cancer Australia to investigate the following cancers: prostate (aggressive), breast, colorectal, UADT, kidney, and pancreas. We plan to submit funding applications for the remaining cancers.

Cancer is a leading cause of disease burden and death globally. One of the principal strategies for reducing this burden is to target the key preventable causal factors that have a strong association or where the exposure is common.

The disease burden measure population attributable fraction (PAF) can be used to estimate the proportion of cancers that could be prevented if exposure to its risk factors were removed or reduced. PAFs are increasingly used to evaluate the national, regional and global burden of cancer and to advocate for changes in public health policy and activity settings to reduce the prevalence of causal risk factors.

However, most prior cancer-PAF studies have relied on published exposure-cancer associations, and risk factor interaction and population subgroup analyses are rarely available. Most studies that have estimated PAFs for combined effects of risk factors have assumed independence between carcinogenic exposures. Yet, in reality, modifiable lifestyle-related risk factors can interact to cause cancer and their effect may be higher for certain subgroups.

PAFs are best estimated from cohort studies, which can allow for ascertainment of multiple outcomes related to an exposure and thus permit analyses to account for potential competing risks, such as death. In addition, most previous cancer-PAF cohort studies have estimated exposure prevalence from the cohort population even when it has not been sampled to be representative of the target population of interest. This hinders both generalisation and comparison of the findings. Confidence intervals (CIs) for PAF estimates are often not provided, precluding an evaluation of their precision and also differences between population subgroups.

Our overarching goal is to estimate PAFs and their CIs for cancer incidence, allowing analysis of the simultaneous effects of multiple factors and accounting for competing risk of death, to the Pooling Project of Prospective Studies of Diet and Cancer (DCPP) in the National Cancer Institute Cohort Consortium and representative external exposure prevalence data.

To calculate fractions of cancers causally related to smoking, alcohol consumption, BMI, physical activity, fruit consumption, vegetable consumption, red and processed meat consumption, multivitamin use, oral contraceptive use, menopausal hormone therapy use, and breastfeeding. The cancers we seek to investigate include prostate (aggressive), breast, colorectal, lung, UADT, kidney, bladder, pancreas, stomach, liver, ovary, thyroid, multiple myeloma, NHL, and gallbladder.

To achieve this, we will leverage the existing resources of the Pooling Project of Prospective Studies of Diet and Cancer (DCPP) in the National Cancer Institute Cohort Consortium. We will be able to estimate PAFs for each of these cancers in the DCPP because this consortium has an existing data repository to support evaluation of different cancer sites and has examined previously associations with several of the proposed cancers. Most of the data needed for this project has already been harmonised. We will estimate the relative risks between lifestyle-related risk factors (e.g., smoking, alcohol consumption, diet, overweight, exercise, and reproductive and hormonal factors) and the cancers of interest. Repeated measures of exposure such as body weight and alcohol consumption will be utilized, if available. We will estimate the population-level prevalence of the risk factors from the latest representative national health surveys (such as NHANES in the U.S., NHS data in Australia, etc). Using the RRs and the prevalence estimates for each exposure we will calculate PAFs for different lengths of follow-up. These PAFs will represent the fractions of cancer that are preventable during such follow-up time if the exposure to the risk factors in the population were removed or reduced. We will analyse potential between-cohort heterogeneity, and investigate reasons for heterogeneity. We will also evaluate and compare the overall importance of the risk factors across cancers.

Note: We are aware there are other approved analyses that overlap with this proposal, including a PAF-pancreatic cancer project, a PAF-physical activity project, and other cancer-exposure analyses. We will consult with the lead investigators of those projects regarding the overlap and will not publish results on these specific results before those groups publish their already planned projects.

By leveraging on the DCPP, this will allow comparisons of the RR and PAF estimates between individual cohorts and countries, to evaluate the joint effects and to explore PAFs in defined subgroups. The use of DCPP data will increase the statistical power to detect exposure-cancer associations, especially for less common cancers and for risk factor interactions.

The minimum number of cases for inclusion are based on criteria used in the DCPP for other projects and range from 25 to 200.
aggressive prostate cancer: 50
breast cancer: 200
colorectal cancer: 100
lung cancer: 50
UADT cancer: 25
kidney cancer: 50
bladder cancer: 50
pancreatic cancer: 50
stomach cancer: 100
liver cancer: 25
thyroid cancer: 25
multiple myeloma: 25
NHL: 50
gallbladder: 25

We request to use all available incidence data of the cancers and mortality data (including cause of death where available) in the DCPP. These data are being or have already been collected by the DCPP.

We request all available questionnaire data for the modifiable exposures:
smoking, alcohol consumption, BMI, physical activity, fruit consumption, vegetable consumption, red and processed meat consumption, multivitamin use, oral contraceptive use, menopausal hormone therapy use, and breastfeeding. Nearly all of these variables are being or have already been collected by the DCPP.

We request all available questionnaire data to examine potential covariates (e.g. age, gender, height, country of birth, marital status, education, socioeconomic status, reproductive history and personal family history of cancer. Nearly all of these variables are being or have already been collected by the DCPP.

No