Thoughtful data use should go into lung cancer screening eligibility

Thoughtful use of race and ethnicity data in prediction models may help reduce disparities in lung cancer screening eligibility, according to research published September 18 in JAMA Network Open.

A team led by Rebecca Landy, PhD, from the National Cancer Institute in Bethesda, MD found that removing race and ethnicity from the life-years gained from screening CT (LYFS-CT) model reduced such eligibility for African Americans. However, under counterfactual eligibility, no one became ineligible and African American eligibility increased.

“These findings suggest that when race and ethnicity is truly an independent predictor of risk and its inclusion in prediction models increases predictive accuracy, excluding race and ethnicity can exacerbate racial and ethnic disparities,” Landy and colleagues wrote.

While current guidelines in lung cancer screening aim to mitigate racial disparities in eligibility, the researchers pointed out that these eligibility points do not account for lung cancer risk among different racial and ethnic groups. Previous studies suggest that African Americans have higher lung cancer risk despite smoking less than white individuals and develop cancer at younger ages.

Prediction models could help improve this cause, but the researchers also noted that careful consideration must be taken on using flawed data or biased presumptions. The LYFS-CT model explicitly incorporates life expectancy when measuring lung cancer mortality in individualized risk assessment.

Landy et al wanted to investigate the potential effects of different approaches to incorporating race and ethnicity in the LYFS-CT model on screening eligibility. One approach excluded race and ethnicity. The other is a counterfactual eligibility approach that recalculates life expectancy for racial and ethnic minority individuals using the same covariates. However, it also substitutes white race and uses the higher predicted life expectancy, in order to not penalize historically underserved groups.

Using these approaches, the team examined screening eligibility in participants of the 2015-2018 National Health Interview Survey (NHIS). The participants were between the ages of 50 and 80 and had a history of smoking.

For the study, the team included 25,601 survey participants. Of these, 2,769 were African American, 649 were Asian American, 1,855 were Hispanic American, and 20,328 were white.

The researchers found that the approach that removed race and ethnicity from risk modeling underestimated lung cancer risk and all-cause mortality in African American participants. This included a ratio of expected to observed number of outcomes [E/O] of 0.72 and 0.9, respectively.

This approach also overestimated mortality in Hispanic American (E/O, 1.08) and Asian American individuals (E/O, 1.14).

However, the “no race” approach also increased Hispanic American and Asian American eligibility by 108% and 73%, respectively. It also reduced African American eligibility by 39%.

The counterfactual approach however had better maintained calibration across the participant groups and increased African American eligibility by 13%. It did so without reducing eligibility for Hispanic American and Asian American individuals, the researchers reported.

The researchers also noted that counterfactual eligibility for both lung cancer death and all-cause mortality together could not identify any additional African American individuals for eligibility.

“This happened because substituting white race into the model reduced lung cancer death risk, and hence the benefit from screening for African American individuals, which in turn reduced the estimate of life-years gained,” the study authors wrote. “This highlights the importance of careful consideration of the quantity to consider as counterfactual.”

The team also called for more efforts to incorporate social determinants of health into prediction models when such information can be more routinely collected in electronic health records.

“Race and ethnicity can represent risk pathways, such as environmental factors or other social determinants of health, that are difficult to ascertain in the clinic,” Landy and colleagues added. “Thus, appropriate use of prediction models that include race and ethnicity can provide a powerful tool for reducing disparities.”

The full study can be found < a href="" target="_blank">here. 

Page 1 of 648
Next Page