Estimates of familial risks from family data are biased when ascertainment of families is not independent of family history
- Centre for Molecular, Environmental, Genetic and Analytic Epidemiology, The University of Melbourne, Parkville, Victoria, Australia
- Correspondence to Dr Mark A Jenkins, Centre for Molecular, Environmental, Genetic and Analytic Epidemiology, School of Population Health, Level 1, 723 Swanston Street, The University of Melbourne, VIC 3010 Australia;
In estimating familial risks from family data one must be careful that ascertainment of families does not depend on family history, otherwise the estimates might not be applicable even to the setting from which the families were ascertained. In this regard, Boparai et al1 reported the largest study to date examining the risk of colorectal cancer (CRC) and hyperplastic polyposis syndrome (HPS) for first-degree relatives (FDRs) of HPS patients. They found that the FDRs of the studied HPS patients had five times the incidence of CRC (relative risk, RR=5.4; 95% CI 3.7 to 7.8), and 39 times the incidence of HPS (RR=39; 95% CI 13 to 121) compared to people from the general population. They interpreted the RR estimates as applying to relatives of people found to be HPS patients in the clinical setting. These RRs are substantial, given that FDRs of patients with CRC are, on average, only twice as likely to be diagnosed with CRC than those with no family history.2 3 We suggest that these estimates are potentially biased and suggest how valid estimates could be derived.
The authors recruited HPS patients from four medical centres in The Netherlands but did not state the reason for the patients' attendance at the clinics nor how they were selected for the study. If, at either of these two stages, patients were preferentially ascertained because of family history (to any degree of genetic relatedness) of CRC, the RR estimates of CRC risk for FDRs of HPS patients will accordingly be inflated. The authors were aware that such ascertainment bias might have existed by stating that the HPS patients in their study ‘may have…higher risk of (familial) CRC than other unidentified asymptomatic HPS cases’. Nevertheless, they claimed that as ‘the aim of our study was to analyse the risk of CRC and polyps in FDRs of all identified patients with HPS, we believe our findings are relevant for the management of patients with HPS and their relatives in a clinical setting’. We disagree with their logic. If ascertainment of HPS patients was not independent of family history, their RR estimates based on comparison with population incidences would be upwardly biased. They would not be valid for any defined set of people, clinic attendees or not. That is, based on the reported estimates from this study alone, it is not possible to state with confidence that an unaffected FDR of a symptomatic HPS case is at increased risk of CRC, let alone by how much.
Comparison with the general population incidence is inappropriate when one suspects that the active investigation could be related to the outcome of interest. One way to obtain an unbiased estimate of CRC risk for relatives of ‘all identified patients with HPS’, as the authors claim was their aim, would be to study all patients investigated for HPS and compare the family histories of the clinic attendees identified with HPS to that of the clinic attendees found not to have HPS; that is, an informative comparison should be within the same ‘study base’.4 We wonder if the authors could undertake that comparison, making sure that the investigation of family history is as thorough for clinic attendees investigated and found not to have HPS as it is for whose found to have HPS.
The same concerns apply to the estimated RR of HPS for FDRs of HPS patients. If ascertainment was not independent of family history of HPS, their estimate will be biased upwards.
In conclusion, analysis of disease risk for relatives of cases must take into account the ascertainment of cases with respect to family history to reduce the potential selection bias, especially based on comparison with population incidences.
Competing interests None.
Provenance and peer review Not commissioned; not externally peer reviewed.