Colorectal Cancer Screening by Fecal Immunochemical Tests (FIT): Considerations on Sampling and Markers (Hb and Hb/Hp Complex) of Fecal Occult Blood (FOB)

Discussion

Given that CRC screening by FITs has a potential to achieve both the primary prevention (detection of precursors), and secondary prevention (early detection of cancer), the failure to reach global success in reducing CRC incidence and mortality rates (2, 3) must indicate that some steps in CRC screening are not adequately conducted at present. It is generally agreed that FIT testing is highly accurate in detecting invasive CRC lesions (4-6, 8, 9, 18-25). Detecting invasive cancers, however, is not an optimal goal of CRC screening. A major focus should be on detecting CRC precursor lesions for adequate treatment, thereby preventing the development of invasive CRC (2, 3). The major weakness in all CRC screening may be associated with inadequate detection of CRC precursor lesions [usual adenomas, sessile serrated polys, advanced adenomas (AAD)] (4-6, 8, 9, 19, 21, 23, 25, 29, 30).

The 10-year cumulative risk of developing CRC from AAD has been estimated to fall within the range from 25.4% to 42.9% among women, and from 25.2% to 39.7% among men (31). Surprisingly enough, the global prevalence of these high-risk lesions has been uncertain until a recent meta-analysis (including 70 population-based studies with 637,414 subjects) by Wong et al. in 2020 (30) provided global estimates on the prevalence rates of adenomas. The pooled prevalence of usual adenomas was 23.9% (95%CI=22.2%-25.8%), prevalence of AAD was 4.6% (95%CI=3.8%-5.5%), and that of CRC was 0.4% (95%CI=0.3%-0.5%) (30). These data including detailed subgroup analysis by sex, age, ethnicity, and geography, are very useful as quality indicators for existing CRC screening programs (30). These pooled global estimates on adenoma prevalence show a good match with our clinical FIT studies in two geographic regions (5, 6).

Typical to practically all screening settings, also the newly implemented CRC screening in Finland is subjected to verification (work-up) bias, because the gold standard reference test (colonoscopy and biopsies) is used for verification of only the subjects testing positive with the index test (FIT Hb) (14). When the index test-negative subjects are not verified by the reference test, one cannot calculate the DA indicators (SE, SP, PPV, NPV, AUC) directly from the screening data (26). This is called an incomplete study design, where exact numbers are available for TP and FP cases only (28). Ideally, on such occasions, one should invite a random sample of 5% of index test-negative subjects for verification by the reference test. When such data (FN and TN cases) from the random 5% of index test-negatives are at hand, statistical techniques correcting for the verification bias can be used to calculate the usual DA indicators (28).

In the present screening setting, however, with only 5% of the screened subjects testing FIT Hb positive (Table III), it is not feasible to request inviting a random 5% of FIT Hb-negative subjects for colonoscopy. Even a random 1% increase would mean >3,000 extra colonoscopies, which would cause a substantial increase in workload for the national colonoscopy capacity (26). Because the analytical performance of the FIT does not depend on the study setting (i.e., whether a sample is collected in a clinical or a screening setting), one can exploit the data from a study with complete design to calculate the DA indicators of the incomplete study design. In the present analysis, the FIT data from the verification cohort (Table I, Table II) were used to calculate the missing DA parameters (FN, TN) for the screening setting, using two different approaches. First, the PPV and NPV of the FIT test results of the validation cohort were translated to the screening setting (Table III, Table IV, Table V, Table VI, Table VII). Second, the data from the 2 × 2 contingency tables of the validation cohort (5) were exploited to perform the state-of-art correction for the verification bias, according to the method of Reichenheim et al. 2002 (28) (Table VIII). The corrected DA parameters obtained by these two different approaches were identical (Table III, Table V, Table VI, Table VIII), as will be discussed in detail later.

The present analysis examined the DA of FIT in detection of both adenoma and cancer, using the number of tested stool samples (one, two or three) and component of FIT (Hb and Hb/Hp complex) as independent variables. The estimates of TP, FN, FP, and TN shown in Table III indicate that the vast majority (>97,000 out of 111,000) of adenomas remain FN when only one sample is tested for stand-alone Hb. Interestingly, these estimates provide a 34.5% prevalence for adenomas, which falls within the range of global adenoma prevalence (30). When these parameters are used to calculate the DA indicators (27), SE is 12.5% and SP is 98.9%, with AUC=0.560 (Table III). Highly interestingly, these figures show an almost perfect match with the data reported by Gies et al. (2018), who compared nine FIT Hb brands in a head-to-head setting (29). When the Hb thresholds were adjusted to yield a SP of 99%, 97%, or 93%, the sensitivities of these nine FIT brands in detecting adenomas (14.4%-18.5%, 21.3%-23.6%, and 30.1%-35.2%, respectively) were practically identical. The same applies to the FIT Hb positivity rates for adenomas (2.8%-3.4%, 5.8%-6.1%, and 10.1%-10.9%, respectively) (29). Thus, with the positivity rate of 5% in the screening setting (16,300/323,400), the FIT brand (ColonView-FIT Hb) used in our validation cohort would give 98.9% SP and 12.5% SE for adenomas, similar to those nine FIT brands (29). This is another indication that no major differences in their DA exist between the FIT brands, as already suggested by two recent meta-analyses (8, 9).

Apart from adenomas, we also tested the applicability of this approach to the detection of CRC in the screening setting (Table IV). In the absence of exact numbers of CRC detected in the screening (26), we used the age-standardized incidence rates for the age groups covered by the screening (14). With the SE of 94.7% (in the validation cohort), the DA of FIT Hb for detection of CRC is outstanding (AUC=0.950) (Table VI). This is consistent with the reported clinical studies where the DA of Hb testing for CRC has been excellent in general (2, 3, 5, 6, 8, 9, 18, 22, 24). With regard to the raised points of concern, these data (Table III, Table IV) confirm that testing one stool sample for Hb alone is 1) accurate in detecting CRC, but 2) clearly not sensitive enough in detecting adenomas. Detecting invasive CRC but leaving most of the adenomas undetected is not an optimal practice to reach the goals of CRC screening (2, 3).

To improve the detection of colorectal adenomas by FIT, two apparent options are available: 1) increase the number of samples tested for Hb, or 2) to complement the FOB detection by using the Hb and Hb/Hp complex (8, 9, 18-25). In the present analysis, both were shown to be helpful to some extent, but the latter approach was clearly superior to the former (Table V and Table VI). When the number of samples tested for Hb is increased from one to three, the SE of the test increases from 12.5% to 19.4% (Table V). This increase in SE is not reflected in the overall DA because the AUC values increased from 0.56 to 0.59 only. By this approach, the number of adenomas to be missed (FN results) decreased from >97,000 (one sample) to 58,000 (three samples). As related to the total number of screened subjects (323,400), this number (18%) is still unacceptably high, and unlikely to improve the overall efficacy of the screening program.

The second option is to perform the FIT testing by using both the Hb and Hb/Hp complex when examining the samples (one, two, three). With this approach, the DA parameters of the screening setting improved remarkably (Table VI). Even one-sample testing with Hb and Hb/Hp complex gave a higher DA (AUC=0.60) than did three-sample testing of Hb alone (AUC=0.590). The most dramatic effect was seen when two samples were tested for Hb and Hb/Hp complex. Test SE increased from 20.6% to 47.5%, with a respective AUC=0.73 (Table VI) (p<0.0001). Importantly, adding the third sample did not increase these values any further. The number of FN results was reduced to 15,000 adenomas, which is substantially less (4.6%) than that achieved with the three-sample testing of Hb alone (18%). Regarding the detection of CRC, increasing the number of samples did not significantly improve the overall DA (Table VII). One-sample stand-alone testing for Hb still missed a few (n=4) invasive carcinomas, and this could be totally avoided by adding the second sample in the protocol.

Finally, we used a formal statistical correlation for the verification bias (28) to test the validity of the approach, where PPV and NPV values of the validation cohort (5) were used to estimate the DA indicators in the screening setting. The two extremes were compared: 1) one-sample testing with Hb, and 2) three-sample testing with Hb and Hb/Hp complex (Table VIII). Not unexpectedly, the results are dramatically different; AUC=0.56 and AUC=0.73 (p<0.0001). The estimates of DA indicators for one-sample Hb-testing obtained after correction for verification bias (28) in Table VIII are exactly the same as those shown in Table III and Table V, based on the “PPV-NPV approach”. The same is true for the three-sample testing with Hb and Hb/Hp complex shown in Table VIII and Table VI. These data implicate that both systems provide identical results when a complete design is used to calculate the missing DA parameters in an incomplete study design.

By definition, the statistical correlation for verification bias has an inevitable outcome of a decrease in SE and increase in SP, as discussed by Reichenheim et al. 2002 (28). In the present settings, the validation cohort values of SE (45.1%) and SP (93.9%) changed to 12.5% and 98.9% for Hb testing and those of Hb and Hb/Hp testing (SE=94.5%; SP=85.1%) changed to SE=47.5% and SP=99.1% (Table VIII). The respected changes in the overall DA values (AUC=0.69 to AUC=0.56) (p<0.0001) and (AUC=0.89 to AUC=0.73) (p<0.0001) also implicate the superiority of a complete design where all index-tested subjects are confirmed by the reference test, as compared with an incomplete design where only index test-positive cases are verified (28).

The intention of the present study was to explore the potential points of concern in the program that might impede reaching the program goals in this country. Because these two goals (reduction in incidence and mortality) have not been convincingly reached by any FIT-based CRC screening in any country as yet (2, 3), these points of concern gain a more global perspective while penetrating in the issues how CRC screening by FIT might be improved. The important managerial and programmatic issues related to all organized screening, such as national coverage, target age groups, attendance rates, screening intervals, training, laboratory performance, quality control and many others (32), were not addressed in this communication. One limitation of our approach is the fact that we were unable to provide the DA indicators adjusted to the Hb cut-off values (available in quantitative FITs) as a potential covariate impacting the SE of FIT, simply because our validation cohort was not based on a quantitative FIT (5, 6, 18-25).

Taken together, our calculations implicate that: 1) testing one stool sample for Hb alone has a low sensitivity in identifying colorectal adenomas (CRC precursors), but sufficient sensitivity in detecting invasive CRC; 2) the DA of stand-alone Hb testing for adenomas can be improved only slightly by increasing the number of samples from one to three; 3) two consecutive samples result in 100% sensitivity in detection of CRC by stand-alone Hb testing; 4) for adenomas, testing for Hb and Hb/Hp complex even in one sample is superior to three-sample testing of Hb alone; 5) the optimal DA for adenomas is achieved by testing two consecutive samples by Hb and Hb/Hp complex, which also detects 100% of CRCs.