Skip to main content

Estimating the optimal rate of adjuvant chemotherapy utilization for stage III colon cancer

Karim S, Booth CM, Brennan K, Peng Y, Siemens DR, Krzyzanowska MK, Mackillop WJ. Cancer Med. 2019; 8(12):5590-9. Epub 2019 Aug 12. DOI:

Background — Identifying optimal chemotherapy utilization rates can drive improvements in quality of care. We report a benchmarking approach to estimate the optimal rate of adjuvant chemotherapy (ACT) for stage III colon cancer.

Methods — The Ontario Cancer Registry and linked treated records were used to identify ACT utilization. Monte Carlo simulation was used to estimate the proportion of ACT rate variation that could be due to chance alone. The criterion-based benchmarking approach was used to explore whether socioeconomic or system-level factors were associated with ACT. We also used the "pared-mean" approach to identify a benchmark population of hospitals with the highest ACT rates.

Results — The study population included 2801 patients; ACT was delivered to 66% (1861/2801). Monte Carlo simulation suggested that the observed component of variation (15.6%) in ACT rates was within the 95% CI (11.5%-17.3%) of what could be expected due to chance alone; the nonrandom component of ACT rate variation across hospitals was only 1.5%. There was no difference in hospital ACT rate by teaching status (P = .107), cancer center status (P = .362), or having medical oncology on site (P = .840). Unadjusted ACT rates varied across hospitals (range 44%-91%, P = .017). The unadjusted benchmark ACT rate was 81% (95%CI 76%-86%); utilization rate in non-benchmark hospitals was 65% (95%CI 63%-66%). However, after adjusting for case mix, the difference in ACT utilization between benchmark and non-benchmark populations was significantly smaller.

Conclusions — We did not find any system-level factors associated with the utilization of ACT. Our results suggest that the observed variation in hospital ACT rate is not significantly different from variation due to chance alone. Using the "pared-mean" approach may significantly overestimate optimal treatment rates if case mix is not considered.

View full text