Cancer
Target trial emulation of dynamic surveillance strategies for cancer survivors: an application to non-muscle invasive bladder cancer Emma E. McGee* Emma E. McGee Xabier García de Albéniz Barbra A. Dickerman Kendrick Yim A. Heather Eliassen Mark A. Preston Miguel A. Hernán
More than 60,000 U.S. adults are diagnosed with non-muscle invasive bladder cancer each year. These patients are recommended to undergo surveillance via repeated cystoscopies. However, the ideal frequency of surveillance remains unknown.
We used observational data from Surveillance, Epidemiology, and End Results (SEER)-Medicare to emulate 3 target trials comparing 10-year risks of bladder cancer mortality under cystoscopy every 3, 6, and 12 months among low, intermediate, and high risk patients. Risks were estimated using inverse probability weighted dynamic marginal structural models.
There were 16,104 low, 18,965 intermediate, and 22,839 high risk eligible individuals. Over 10 years, 5,288 bladder cancer deaths occurred. Among high risk patients, 10-year bladder cancer mortality risks were 21.3% under the 3 month strategy, 22.3% under the 6 month strategy, and 23.5% under the 12 month strategy. Risk differences (95% CI) were 1.0 (-0.9, 2.8) for 6 vs. 3 and 2.1 (0.1, 4.0) for 12 vs. 3 months. For intermediate risk patients, mortality risks were 12.4% under the 3 month strategy, 12.8% under the 6 month strategy, and 13.7% under the 12 month strategy. Risk differences were 0.4 (-1.1, 2.3) for 6 vs. 3 and 1.4 (-0.4, 3.3) for 12 vs. 3 months. For low risk patients, mortality risks were 7.6% under the 3 month strategy, 7.0% under the 6 month strategy, and 8.6% under the 12 month strategy. Risk differences were -0.6 (-2.3, 1.1) for 6 vs. 3 and 1.0 (-1.1, 3.1) for 12 vs. 3 months. When we mirrored the approach of prior observational studies which deviated from a realistic target trial, the direction of effect estimates switched.
We estimated that more frequent cystoscopy may result in reductions in bladder cancer mortality among high risk patients. Among low and intermediate risk patients, surveillance every 6 months may not meaningfully increase mortality as compared with every 3 months. Analyses which failed to emulate a target trial produced starkly different results.
