Rebecca Landy, PhD

Credit: dceg.cancer.gov

A counterfactual approach to the ‘life-years gained from screening–computed tomography’ (LYFS-CT) lung cancer screening model can maintain the model’s accuracy and reduce disparities for African-Americans, according to recent findings, without reducing eligibility for both Hispanic and Asian Americans.¹

The new findings came from a recent study conducted to determine the most effective strategies that are able to incorporate race and ethnicity into the LYFS-CT lung cancer prediction model, a model recommended by the American College of Chest Physicians for personalizing patients’ shared decision-making in screening for lung cancer.²

Screenings for lung cancer involve asymptomatic patients, and the harms of screening for these individuals must not exceed the potential benefits. Merely lowering the eligibility risk- or benefit-thresholds and thus allowing for more minority patients could, consequently, be a bad strategy if those individuals would likely not benefit.

Current guidelines usually recommend 5 - 10 years minimum life expectancy in order to prevent those more likely to die from comorbid conditions to be negatively affected.³ That said, some racial groups often lack access to health care and preventing them may unjustly penalize those projected to have lower life expectancy only due to their race or ethnicity.

The new research into the LYFS-CT model was conducted to address this, and it was led by Rebecca Landy, PhD, from the Division of Cancer Epidemiology and Genetics at the National Cancer Institute in Bethesda, Maryland.

“LYFS-CT is particularly interesting to examine from the perspective of fairness because, unlike risk models, it explicitly incorporates life expectancy,” Landy and colleagues wrote. “Here, we examined the effect of different approaches to incorporating race and ethnicity in the LYFS-CT model on screening eligibility for a contemporary representative US population.”

Background and Findings

The investigators set out to assess lung cancer screening eligibility within a contemporary representative US population, and they used 2 different approaches. In the first approach, the team refitted the LYFS-CT screening model by excluding patients’ ethnicity and race.

The second approach used by the investigators involved a counterfactual eligibility method which would recalculate life expectancy for racial and ethnic minority patients. It would substitute the White racial group and utilize the higher predicted expectancies of this group to prevent historically disadvantaged racial and ethnic groups from being underserved.

The research team used a comprehensive analysis with 2 submodels within LYFS-CT, and the 2 submodels were recalibrated and then validated without considering patient ethnicity or race. The team’s first submodel looked at risk of cancer death and looked at those drawn from a large clinical trial in the period between 1993 - 2001, with a smoking history (n = 39,180).

The second submodel focused on all-cause mortality among patients, and it was developed through the use of data taken from the National Health Interview Survey (NHIS) from 1997 - 2001. It covered individuals in the age range of 40 - 80 years who may have ever smoked (n = 74,842), and used follow-up data from until the last day of December 2006.

The investigators assessed individuals’ screening eligibility among NHIS participants from 2015 - 2018, in the age range of 50 - 80 years with a history of smoking. The data analysis from the team’s entire study was done between June 2021 and September 2022.

The research team’s primary outcomes were determined to be related to calibration, with comparisons of the LYFS-CT ‘NoRace’ model and the counterfactual approach, with a focus placed on the ratio of expected to observed (E/O) patient outcomes. The team additionally looked at the effects of such approaches on eligibility for screening and predicted the potential rise in life expectancy resulting from screening as estimated by the LYFS-CT model.

Overall, with the team’s NHIS 2015-2018 dataset covering former smokers in the age range of 50 - 80, the elimination of race or ethnicity from the submodels resulted in underestimated risks for lung cancer death and all-cause mortality in African Americans and overestimated mortality risks in those identified as Asian and Hispanic Americans.

Consequently, the investigators found that the LYFS-CT NoRace screening model ended up improving eligibility for Hispanic and Asian Americans by 108% and 73% but reduced it by 39% for African Americans.

They also found that the LYFS-CT using the counterfactual all-cause mortality model was able to better maintain calibration, with the team noting an increase in African American patient eligibility by a rate of 13% without reduction of eligibility for Asian or Hispanic American individuals.

“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,” they wrote. “Thoughtful use of race and ethnicity data may account for important historical factors that have led to disparities. Thus, appropriate use of prediction models that include race and ethnicity can provide a powerful tool for reducing disparities.”

References

1. ^{Landy R, Gomez I, Caverly TJ, et al. Methods for Using Race and Ethnicity in Prediction Models for Lung Cancer Screening Eligibility. JAMA Netw Open. 2023;6(9):e2331155. doi:10.1001/jamanetworkopen.2023.31155.}
2. ^{Cheung LC, Berg CD, Castle PE, Katki HA, Chaturvedi AK. Life-gained-based versus risk-based selection of smokers for lung cancer screening. Ann Intern Med. 2019;171(9):623-632. doi:10.7326/M19-1263.}
3. ^{Wilt TJ, Harris RP, Qaseem A; High Value Care Task Force of the American College of Physicians. Screening for cancer: advice for high-value care from the American College of Physicians. Ann Intern Med. 2015;162(10):718-725. doi:10.7326/M14-2326}