Mayer SE, Edwards JK, Lund JL, Hallas J, Sturmer T. Modeling drug treatment effects on breast cancer in the setting of differential screening. Poster presented at the Virtual ICPE 2021 Conference; August 2021. [abstract] Pharmacoepidemiol Drug Saf. 2021 Aug 23; 30(S1). doi: 0.1002/pds.5305

BACKGROUND: Medications may have direct effects on cancer risk (initiation) or promote growth of existing tumors. Measuring these effects on screen-detectable cancers is complex in settings in which receipt of cancer screening varies between treatment groups.

OBJECTIVES: We sought to model cancer initiating and promoting treatment effects and their impacts on observed breast cancer incidence under differential and nondifferential screening scenarios.

METHODS: We used discrete event simulation to model lifetime risk ofbreast cancer initiation, tumor growth, detection via symptoms or screen-ing, and competing mortality in 1,000,000 women aged 20-84. Treatment receipt was modeled as a binary variable with initiation (or matched indexdate) occurring at age 50 to remove the impact of screening prior to treatment. Cancer initiating and promoting effects increased age-specific cancer rate and tumor size, respectively, by 1.2 times in all subsequent years. Two-year screening coverage among women aged 50-74 was var-ied between 0-100%. Cancer detection was modeled probabilistically as a function of tumor size and screening; in settings with differential screen-ing, treatment and screening were indirectly associated through an unobserved "access" parameter. We used Aalen-Johansen risk estimation to calculate cumulative incidence of breast cancer initiation and detection by treatment under each screening regimen, and computed risk differ-ences and ratios (RD, RR) at all time points.

RESULTS: In the nondifferential initiation setting, all screening-based esti-mates underestimated the true treatment effect, with the 100% screening estimates being least biased; lower screening coverage further attenuated the RD and RR. The largest promotion effects in the nondifferential screening setting were observed with 0% screening. In settings where screening varied by treatment group, 25% screening led to the greatest overestimation of the RD and RR relative to nondifferential screening. The25-year RDs for 25 vs. 0 and 100% screening were 0.018 vs. 0.012 and0.006, respectively, for initiation, and 0.010 vs. 0.004 and 0.001 for pro-motion. RRs were highest early in follow-up: 1-year RRs for 25 vs. 0 and100% screening in the initiation setting were 1.31 vs. 0.97 and 0.97,respectively; for promotion they were 1.61 vs. 1.37 and 1.07. 

CONCLUSIONS: This study illustrates the potential impact of a treat-ment's etiological mechanism on observed risk of screen-detectable cancers, and highlights the sensitivity of results to the level of screen-ing and its distribution in the population. The RR in particular is highly sensitive to the level of screening that occurs early in follow-up, when incidence is low.

Share on: