• ENDOSCOPY
• SCREENING
• COLORECTAL CANCER

Morgan et al recently pointed out that, while colorectal cancer (CRC) remains a highly frequent cancer worldwide, trends in CRC incidence are decreasing markedly in some countries,1 which may largely be attributed to endoscopic screening.2 We; therefore, read with great interest the recently published follow-up results of the Nordic-European Initiative on Colorectal Cancer (NordICC) trial, suggesting a presumably smaller than expected protective effect of screening colonoscopy to prevent CRC. After 10-year follow-up, CRC incidence was reduced by 18% in the intention-to-screen analysis, comparing the intervention versus the control group, and by 31% in the per-protocol analyses, comparing participants screened versus controls.3

However, there is a strong reason to assume that the reported incidence reduction is significantly underestimated. In randomised screening endoscopy trials,3–8 the measured incidence shows an initial apparent increase in the screened group, which is attributed to the detection of prevalent, preclinical cancers. Such prevalent cancers cannot any longer be prevented, but only detected early (as they are already present), and thus bias estimates on the true CRC incidence reduction by screening endoscopy, with the impact of the bias decreasing with increasing length of follow-up. Thus, the commonly measured effects on CRC incidence in such trials do not quantify the true endoscopy impact on incidence, which is likely substantially underestimated.

For illustration, we quantified this expected underestimation by replicating NordICC with and without prevalent preclinical cancers at baseline in a comprehensive simulation approach. Details are provided in online supplemental material. Briefly, first, we recalibrated the COlorectal Cancer Multistate SImulation MOdel (COSIMO), a validated modelling tool which simulates the natural history of CRC based on the process of developing adenomas to preclinical and clinical cancer,9 to reproduce outcomes of NordICC by matching simulated subjects with reported numbers. Second, we derived the expected adjusted incidence in the screening and control group and the difference between the observed risk ratio (RR_OBS), that is, the RR including prevalent CRC cases at baseline, and the adjusted RR (RR_ADJ), that is, not including prevalent CRC cases, which is not directly observable in real-life studies. RR_ADJ was calculated by omitting the number of cases arising from prevalent preclinical cancers from the model calculation.

We found that the number of cases and the 10-year risks in both groups after 10 years as predicted by the recalibrated COSIMO were very similar to those observed for the actual NordICC trial, and all modelled RR_OBS point estimates were within the reported 95% CIs of the actual trial (table 1A). Patterns of (observed, biased) cumulative incidence in the simulated screening and control groups were in line with those reported in the real NordICC trial (figure 1A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Cumulative incidence (A) cumulative numbers of colorectal cancers cases (B) and risk ratios (C, D) in the actual and simulated NordICC trial. COSIMO, Colorectal Cancer Multistate Simulation Model; CRC, colorectal cancer; NordICC, Nordic-European Initiative on Colorectal Cancer.

Excluding prevalent preclinical cancers at baseline markedly changed the number of CRC cases. The relative share of prevalent cases among all detected cases was higher in the intervention group but exceeded 50% in the initial years of follow-up in both groups (figure 1B). While the difference between RR_OBS and RR_ADJ diminished with increasing length of follow-up, absolute incidence reduction was still underestimated by 13% and 31% in intention-to-screen and per-protocol analyses, respectively, after 10 years (table 1B, figure 1C,D). Finally, the adjusted cumulative incidence was consistently lower than the observed cumulative incidence, without overlapping curves (figure 1A).

These findings elucidate why screening colonoscopy had a presumably smaller than expected protective effect in NordICC and may contribute to improved risk communication with individual patients. However, it should be noted that on a population-level, CRC incidence could only be reduced to the magnitude estimated in our simulation if screening colonoscopy was initiated and perfectly adhered to prior to the development of prevalent CRC (eg, at age 45, as recommended by US guidelines,10 instead of ages 55–64 as in NordICC). Even though colonoscopy is highly efficacious at preventing CRC, the effectiveness of a screening programme depends on the specifics of its implementation, for example, starting age and uptake.

This study is limited by model simplifying assumptions and uncertainties related to input parameters, such as the proportion of colonoscopy users outside of the trial, the prevalences of CRC precursor lesions at baseline and their true transition rates, and the true adenoma miss rate within NordICC.

In conclusion, our study suggests that the preventive effect of screening colonoscopy is likely much stronger, and manifests much earlier than recently reported. Published findings of trials such as NordICC likely significantly underestimate the true preventive effective of screening colonoscopy due to inclusion of prevalent, preclinical cancers which cannot any longer be prevented. The apparently low effects of NordICC should not unduly discourage use of colonoscopic screening for CRC. Quantification of the impact of screening trials should take into account an adequate estimation of prevalent cancers.