Correlates of unrealistic risk beliefs in a nationally representative sample

The Gail model is based on population-level data (Gail et al. 1989) and may not adequately predict breast cancer risk at the individual level (Rockhill 2001; Schonfeld et al. 2010), particularly for African American women (Adams-Campbell et al. 2009). However, the model was used to determine individual participant eligibility for the randomized clinical trials of tamoxifen and raloxifene (Fisher et al. 1998; Vogel et al. 2006), is mentioned in the Federal Drug Administration approval of tamoxifen for primary chemoprevention of breast cancer (Vogel et al. 2006), and is posted on heavily visited cancer risk assessment websites (for a review see Waters et al. 2009). Thus, it is not unreasonable to use the Gail model to obtain objective risk estimates.

We also examined unrealistic optimism on the group level. In accordance with established practice (Weinstein 1987), the comparative perceived risk variable was recoded as: −1 (less likely), 0 (about as likely), and +1 (more likely). The mean comparative risk for the entire sample was −0.24 (SE = 0.01), which was significantly different from zero (P < .0001). This demonstrates that, as a group, women perceived their breast cancer risk as lower than the average woman’s risk (i.e., unrealistically optimistic). Readers who are interested in the correlations between the raw comparative risk perception measure and the outcome variables should contact the corresponding author.

Sensitivity analyses indicate that these findings are reasonably robust whether the RR cutoffs for determining whether participants’ objective risk is average, above average, or below average are based on tertiles, quartiles, or quintiles, even after the sequential Bonferroni correction for multiple comparisons is conducted (i.e., P < .003 or P < .004 for unrealistic optimism and unrealistic pessimism, respectively). Analyses based on tertiles differed slightly from the quartile analyses. In particular, age² had an inverted U-shaped relationship for unrealistic pessimism (OR = 0.99, 95% CI [0.98, 1.00], P = .0003), such that younger and older women were less pessimistic than women in their middle years. Analyses based on quintiles also differed slightly. In contrast to the reported (quartile) analysis, people without a high school degree were not less likely to be unrealistically optimistic than people with a degree (OR = 0.88, 95% CI [0.76, 1.02], P = .10), having more than a 4-year degree was not associated with reduced unrealistic pessimism (OR = 0.70, 95% CI [0.50, 0.98], P = .04), and being non-Hispanic Black was not associated with increased unrealistic pessimism (OR = 1.25, 95% CI [1.04, 1.49], P = .015). It should be noted that all of the OR reported in this footnote either fall within or nearly within the confidence intervals reported in Table 3. For instance, the nonsignificant OR of 0.70 for the tertile-based relationship between having more than a 4-year degree and unrealistic pessimism is within 0.01 points of the CI for the same relationship using quartiles (i.e., 0.35–0.69; see Table 3). This suggests that, although the significance levels changed very slightly based upon the objective relative risk cutoffs, the sizes of the relationships—and hence the overall message—are consistent.

Although the paper is focused primarily on unrealistic optimism and unrealistic pessimism across the lifespan, exploratory analyses examined their relationships with appropriate mammogram use. The dataset was restricted to women aged 42–74 without any cancer history, and women were classified according to whether or not they had obtained a mammogram within the previous 2 years (National Cancer Institute 2010). The omnibus F-test indicated that mammography use in the past 2 years was not related to risk perception category overall, P = .81. Readers who wish to obtain additional details about this sub-analysis should contact the corresponding author.