Number Needed to Screen or Invite as Measures of Cancer Screening Efficiency
Chyke A. Doubeni,Norah L. Crossnohere,Carlo Senore
DOI: https://doi.org/10.1001/jamanetworkopen.2024.35615
2024-10-02
JAMA Network Open
Abstract:Colorectal cancer (CRC) is the second leading cause of cancer deaths worldwide, and improving delivery of screening across all populations is critical for further decreasing incidence and mortality. An invitation to screen is a commonly used population-based strategy to improve the reach of CRC screening. The number needed to invite (NNI) or screen (NNS) to prevent 1 CRC diagnosis or death, the reciprocal of the absolute risk difference (RD), is used to assess the efficiency of screening strategies or tests. This study by Brenner and colleagues 1 reanalyzed data from a randomized clinical trial (RCT) of invitation to screening colonoscopy to show how delayed reporting of cancer diagnosis and mortality by registries used for outcome ascertainment could lead to spuriously high NNI and NNS. 1 In the original RCT, Bretthauer et al 2 examined the invitation strategy in 84 585 presumed-eligible individuals aged 55 to 64 years in Poland, Norway, and Sweden. The RCT randomly assigned men and women to invitation (28 395 individuals) or usual practice (56 785 individuals) and ascertained CRC diagnosis and death over 10 to 15 years of follow-up using cancer registries. The analysis excluded people who were found to have CRC (221 individuals) or had died (373 individuals) prior to randomization and only confirmed after inclusion due to delays in registration. In the analysis, 10-year CRC cumulative incidences were 0.98% in the invitation group and 1.20% in the usual practice group (RD, 0.22%) in intention-to-treat (ITT) analysis. In per-protocol analysis, 10-year CRC cumulative incidences were 0.84% in the invitation group vs 1.22% in the usual practice group (RD, 0.38%). Based on the 221 people excluded due to registry reporting lags, Brenner et al 1 estimated a 2- to 3-year cancer registration delay. Accounting for reporting delay with an assumption of nondifferential ascertainment bias, they derived RDs of 0.44% for a 2-year delay and 0.88% for a 3-year delay in ITT analysis and 0.76% for a 2-year delay and 1.52% for a 3-year delay in per-protocol analysis. 1 The corresponding NNI and NNS for a potential 2-year reporting delay were 227 and 132, respectively, in contrast to 455 and 263 from the RCT results. 1 ,2 Accurate outcome measurement is critical to the validity of RCTs of cancer screening effectiveness. Therefore, many RCTs adjudicate outcomes to, for instance, confirm new diagnosis, pathologic characteristics, and cause of death. Those approaches may not be feasible at scale, and cancer registries are efficient data sources if completeness and quality are assured. The analyses by Brenner et al 1 raise questions about the time interval needed beyond the end of the study to assure completeness, which varies across registries. Lagged reporting of diagnosis and deaths to cancer and vital registries are well known and often determine the timing of dissemination. The RCT by Bretthauer et al 2 followed people for a median of 10 to 15 years, suggesting an adequate follow-up for complete ascertainment but neither affirmed nor tested those assumptions in their study. However, estimating potentially missed cases is not straightforward because of changes in risk over time. Improvements in CRC risk generally manifest after approximately 5 years of follow-up from initiating screening because detection of prevalent CRC leads to higher incidence in individuals screened during early follow-up periods. 3 The effect of screening colonoscopy extends beyond 10 years, and incidence and mortality curves across groups in the RCT may continue to diverge over time, and, as such, a precise cutoff date for a true effect is uncertain. 2 ,4 The 10-year 50% (95% CI, 23%-73%) mortality risk reduction from per protocol analysis was consistent with the 67% (95% CI, 48%-79%) reduction reported by an observational study that preceded the trial. 2 ,5 Thus, the overall findings from the RCT by Bretthauer et al 2 are not surprising. RCTs address specific, often narrow, questions. Viewed from the lens of the question on invitation to screening, the results of the study by Bretthauer et al 2 are encouraging, and the observed 42% participation rate is not atypical. Despite the modest participation rate, the strategy reduced the 10-year CRC risk by 18% (risk ratio [RR], 0.82; 95% CI, 0.70-0.93) although there was no significant reduction in mortality (CRC RR, 0.90; 95% CI, 0.64-1.16). The analysis by Brenner et al 1 analysis raises questions about the appropriateness of NNI and NNS outside individual RCTs. NNI and NNS are related to screening strategy or test effectiveness and the underlying disease risk or prevalence, as is positive predictive value. 6 The higher the cumulative risk, the lower the NNI and NNS. Thus, if the c -Abstract Truncated-
medicine, general & internal