IARC HANDBOOKS COLORECTAL CANCER SCREENING VOLUME 17 This publication represents the views and expert opinions of an IARC Working Group on the Evaluation of Cancer-Preventive Strategies, which met in Lyon, 14–21 November 2017 LYON, FRANCE - 2019 IARC HANDBOOKS OF CANCER PREVENTION 3. STUDIES OF COLORECTAL CANCER SCREENING 3.1 Methodological considerations end-point of the RCT can be the incidence of the cancer of interest. Methods for colorectal cancer (CRC) screen- The observed effect of screening in RCTs is ing include endoscopic methods and stool-based dependent on, among other things, the partic- tests for blood. The two primary end-points for ipation in the intervention group and the limi- endoscopic CRC screening are (i) finding cancer tation of contamination of the control group. at an early stage (secondary prevention) and Low participation biases the estimate of effect (ii) finding and removing precancerous lesions towards its no-effect value, and therefore it must (adenomatous polyps), to reduce the incidence be evaluated and reported. Screening of controls of CRC (primary prevention). The primary by services outside of the RCT also dilutes the end-point for stool-based tests is finding cancer effect of screening on CRC incidence and/or at an early stage. Stool-based tests also have some mortality. If the screening modality being eval- ability to detect adenomatous polyps; therefore, a uated is widely used in clinical practice in secondary end-point of these tests is reducing the the region or regions where the RCT is being incidence of CRC. conducted, then contamination may be consid- erable, although it may be difficult and/or costly 3.1.1 Randomized controlled trials of to estimate its extent. The standard evaluation colorectal cancer screening of the primary end-point is an intention-to-treat analysis. Methods to adjust for contamination A randomized controlled trial (RCT) that and low participation, called per-protocol anal- compares a screening arm with a non-screening yses, have been proposed (Cuzick et al., 1997; arm (or with a screening arm with a different Baker et al., 2002). [Note that these methods screening modality) is considered the reference differ from those that compare participants who standard in evaluating the cancer-preventive were actually screened with those who were not, effects of a screening test. For screening modal- which is an invalid method with a high potential ities that only detect cancerous lesions (such for selection bias.] as those for breast cancer or lung cancer), the primary end-point of a screening RCT is gener- ally mortality from the cancer of interest. For screening modalities that have the potential to detect and remove cancer precursors (such as those for CRC and cervical cancer), a co-primary 79 IARC HANDBOOKS OF CANCER PREVENTION – 17 3.1.2 Observational studies on the preventive incidence or mortality rates in a screened cohort effects of colorectal cancer screening with concurrent (or past) population rates is also prone to the selection biases described above and Data from observational studies, in which may also be biased by variation in temporal and study participants are not randomly allocated regional trends. to be screened or not screened, should be used Ecological studies that compare incidence with caution, because of their greater propen- or mortality rates in a region or country where sity to bias. Comparison of the survival, or stage screening has been implemented with rates in distribution, of screen-detected cases with that the same region or country during a previous of cases diagnosed outside of screening is noto- time period, or with contemporaneous rates in riously flawed, because of lead-time bias, length- a neighbouring region or country, may be useful time bias, and overdiagnosis bias. Lead time is but are subject to the standard caveats of ecolog- the period between when a cancer is found by ical studies. screening and when it would have been detected There are several early observational studies from clinical signs and symptoms (not directly that assessed the effects of endoscopy and stool- observable) in the absence of screening. Survival based tests for blood on CRC incidence and time, by definition, is measured from the date of mortality; most of these studies have major diagnosis to the date of death. Lead-time bias is methodological issues. Lead-time bias, length- the overestimation of survival time due to earlier time bias, selection bias, and confounding factors detection by screening than clinical presentation; in these studies could lead to overestimation or it is very difficult to distinguish from earlier diag- underestimation of the effect when not adjusted nosis leading to a real extension of life. Length- for. Misclassification of the outcome, CRC inci- time bias reflects disproportionate detection by dence or death (ascertainment bias), can also screening of indolent tumours, which may reside bias the effect estimates. In the Effectiveness of in a preclinical state for longer time periods Screening for Colorectal Cancer in Average-Risk than aggressive tumours do. Overdiagnosis is Adults (SCOLAR) nested case–control study, an extreme form of length-time bias in which a Goodman et al. (2015) stressed the challenges in tumour that would never have been diagnosed observational studies of colonoscopy screening, without screening is detected by screening. including distinguishing between indications for Using cohort or case–control designs to having had a colonoscopy (screening or investi- compare the mortality and/or incidence rates of gation of symptoms). Some observational studies a group receiving (or invited to receive) screening use, for example, a 6-month window to exclude with those of a group not receiving (or not invited index examinations that were done to investigate to receive) screening can avoid the above-men- symptoms of CRC. tioned biases, but it will usually involve some In light of the published RCTs of sigmoi- type of selection bias, because the decision to doscopy screening, the Working Group estab- be screened is not random and may be related lished two criteria for observational studies to to factors that predispose to, or against, devel- be included in the evaluation of effectiveness: oping the cancer. Careful adjustment for known the study (i) must be performed in a screening CRC risk factors may partially alleviate this setting and (ii) must not exclude cancer detected bias. Cohort studies are generally more reliable at the baseline endoscopy. Moreover, the Working than case–control studies for assessing the effi- Group established six considerations to weigh cacy of screening, although well-designed case– the impact of individual studies on the overall control studies may be useful. Comparison of the estimate: (i) there must be a concurrent control 80 Colorectal cancer screening group, (ii) there must be an adequate length of comparing individuals invited to screening with follow-up, (iii) the sample size must be large those not invited to screening. enough to detect relevant effects, (iv) the study It is possible that reassurance from a nega- must be conducted in a setting with contempo- tive screening result (whether true-negative or rary methods, (v) the outcome (and exposure) false-negative) could lead to delay in presentation ascertainment must be reliable, and (vi) potential for investigation of symptoms that develop in the confounder data must be available and adjusted interval between scheduled screening tests, with for in the analysis. consequent late diagnosis of an interval cancer and possibly death from it. Such an impact 3.1.3 Evaluation of the adverse effects of on mortality would form part of the overall colorectal cancer screening mortality measured in the screening arm of an RCT of CRC screening and would not be meas- An adverse effect of screening is defined as urable separately from it. However, its presence any negative effect on individuals or populations within the results of the RCT would remove any that results from being involved in the screening need to account for it separately when estimating process compared with not being involved in a benefit–harm ratio from the RCT. screening. It is important to quantify not only No physical harms are incurred with stool- the frequency of the harm but also its magnitude. based tests for blood. Potential harms of endo- It is also important to recognize that the harms scopic tests include pain, physical damage to the in an RCT may be different from the harms in a bowel due to endoscopy, possible hospitalization, screening programme. and the need for surgical repair. The frequency (a) Definitions of harms and severity of such harms can be estimated from a representative screened cohort. Harris et al. (2014) proposed a taxonomy Potential harms associated with the manage- of the harms of screening. They proposed four ment of a positive screening result include phys- domains of harms: physical effects, psychological ical harm from the workup and treatment and effects, financial strain, and opportunity costs. harm from the psychological response to knowl- Financial strain and opportunity costs [the indi- edge of the result and any aspect of its manage- rect effect of screening on health-related activ- ment. Apart from the possible harms of treating ities] are not considered in this review. Here, a screen-detected cancer, the harms experienced harms are classified on the basis of where in the are largely independent of whether the screening screening cascade they occur: harms associated result is true-positive or false-positive. Some with (i) the screening process, (ii) the screening people restrict the definition of a true-positive to test itself, and (iii) the management of a positive invasive cancer, whereas others include advanced screening result. adenoma in the definition, and still others may Potential harms of the screening process include detection and removal of precancerous include anxiety, caused by the invitation to lesions (polypectomy).