Focus on Epidemiology and Risk Models
Network - Summer 2011
Can experts predict who is more likely to develop cancer? We asked Carol Etzel, Ph.D., associate professor in the Department of Epidemiology, about the basics of designing a “risk model” for lung cancer.
Why and how do epidemiologists develop risk models?
We try to identify the factors that indicate risk, paying attention to what’s feasible. We choose, then evaluate factors based on:
- a literature review to identify what other researchers have reported, and
- discussions with experts to ensure that we consider all potentially important factors.
Then we develop a hypothesis to explore.
Can you describe the first risk models you developed for lung cancer?
In 2007, we published our first lung cancer risk prediction model. This was developed using the non-Hispanic white population of lung cancer cases and controls from a multiracial/ethnic lung study that’s been ongoing in the Department of Epidemiology for 20 years.
For this population, the risk factors were determined after evaluating participants’ ages, genders and lifestyle habits, with an emphasis on smoking, family history of cancer, medical conditions and workplace exposures. This model was arranged by smoking status: never, former or current.
What about the risk model you developed for African-Americans and lung cancer?
Our goal was to develop a comprehensive epidemiologic lung cancer risk prediction model just for African-Americans. We used data from the study mentioned above.
The model included:
- smoking-related variables, such as whether or not subjects smoked, and then for smokers or past smokers, how many packs a day they smoked, their age when they stopped smoking, and how many years since they stopped smoking (if they have);
- self-reported physician diagnoses of chronic obstructive pulmonary disorder (COPD) or hay fever; and
- exposures to asbestos or wood dusts.
We took our risk estimates from our risk model. Then, we factored in rates of lung cancer incidence and mortality from other causes. With this information, we produced 10-year absolute risk estimates for lung cancer for African-Americans.
Why did you design such a model?
In preliminary analyses, we had observed that though African-Americans share similar risk factors for lung cancer as non-Hispanic whites (such as smoking and COPD), their levels of risk were higher.
So, we felt that existing risk prediction models for lung cancer developed in white populations (including the one created by our group) might not be best for predicting risk among African-Americans.
How do the two models differ?
Our original model is arranged by smoking status: never, former or current. In the African-American model, all smokers are included in a single model. Exploring the interplay of smoking status, intensity of smoking and COPD status, we created a “composite” variable to evaluate risk for lung cancer.
We found that the highest risk is for African-American smokers (former or current) with a history of COPD and at least 26.4 pack-years of smoking exposure. So, African-Americans with COPD who’ve smoked — or used to smoke — for more than 20 years might discuss with their doctor their need to be screened for lung cancer regularly.