Table 2.

Possible sources of bias in studies of rrBSO and breast cancer risk.

Type of biasDefinition and exampleImpactPossible mitigation strategies
Confounding by indicationConfounding by indication may be introduced if women who choose rrBSO have a different BC risk to those who do not have rrBSO. For example, within BRCA1 and BRCA2, there are areas of each gene, which, when mutated, increase TOC risk and decrease BC risk compared with mutations in other regions. Carriers with an inherently higher risk of TOC and lower risk of BC may be more likely to choose rrBSO because they have a stronger family history of TOC (46).The potential benefit of rrBSO on BC risk may be overestimated as women opting to undergo rrBSO may do so because of a strong FHx of TOC and may have been at comparatively lower risk of developing BC (46).The potential impact of confounding by indication may be mitigated by matched sibling cohorts, taking into account the age difference of siblings to prevent introduction of bias associated with start of follow-up (below; ref. 46); however, this strategy can substantially reduce sample size.
Adjusting for FHx in the analysis offers partial mitigation of this bias.
Survival bias from competing risk of TOCThis bias is closely related to confounding by indication and describes the observation that women who are at inherently higher risk of TOC than BC, who do not undergo rrBSO, may contribute fewer person-years at risk during follow-up if they die from TOC before censoring for another reason.Overestimation of the protective association between rrBSO and BC risk, further amplifying confounding by indication (46).As per confounding by indication.
If these women are overrepresented in the control group, the bias introduced by indication and survival is accentuated (46).
Informative censoringWhen a censoring event, for example, rrBM, depends on the study endpoint (BC risk) then the censoring becomes “informative.” Carriers with higher familial BC risk may be more likely to undergo early rrBM, before rrBSO, compared with carriers with lower familial BC risk. The censoring event, rrBM, is considered “informative” because the group of women who undergo rrBM were more likely to develop BC than women who proceed to either rrBSO or other risk-reducing options (4, 46).The potential benefit of rrBSO on BC risk may be overestimated due to an excess of lower risk women in the rrBSO group.This may be partially mitigated by adjusting for family history.
Cancer-induced testing biasCancer-induced testing bias explains the observation that diagnosis of BC often prompts genetic testing. Women who are then found to carry a BRCA1 or BRCA2 mutation may be recommended to undergo rrBSO for TOC risk reduction. Thus, an analysis of BC incidence before and after rrBSO may be enriched for BC events in the non-rrBSO period (4, 46).May lead to overestimation of the association between BC risk and rrBSO.Exclusion of women with a personal history of BC prior to genetic testing. Starting the observation period at the time of genetic testing (4, 46).
Immortal person-time biasImmortal person-time bias relates to the follow-up period that participants survived BC-free before rrBSO. This bias is introduced if the person-time before rrBSO is not allocated to the non-rrBSO group (4).Results in misallocation of observation time away from the non-rrBSO group and consequently, an increase in BC events per person-year in this group, biasing toward a protective association between rrBSO and BC (4).Consider rrBSO as a time-dependent variable and allocate the observation period between the date of genetic testing and rrBSO to the non-rrBSO group (4).
Confounding by other risk factorsConfounding by other risk factors for BC also needs to be taken into account when assessing the efficacy of rrBSO (46).May lead to over- or underestimation of the association between rrBSO and BC risk depending on risk factor.Adjustment for these confounders.
E.g., parity - parous women may be more likely to undergo rrBSO compared with nulliparous women. If parous carriers are also at lower risk of BC, the association between rrBSO and reduced BC risk may appear spuriously stronger.
Missing dataBecause of the nature of observational studies, it is not always possible to collect data points of interest on all patients (45).Depending on the volume of missing data and its relationship to the main study outcomes, missing data may affect the integrity of the results (45).Imputation methods (66, 67).
OtherAge at rrBSO – if the association between rrBSO and reduced BC risk only occurs for women who have early premenopausal rrBSO and not for those who have peri- or postmenopausal rrBSO (which is biologically plausible), then including women with peri- and postmenopausal rrBSO in the analysis will tend to weaken the association seen between rrBSO and reduced BC risk.Any association between rrBSO and reduced BC risk may be underestimated or missed.Analyses stratified by age at rrBSO.
cHRT – women who undergo premenopausal rrBSO may be more likely to receive subsequent cHRT than women who do not have rrBSO. If cHRT increases BC risk in carriers, any association between rrBSO and reduced BC risk may be spuriously weaker.Any association between rrBSO and reduced BC risk may be underestimated or missed.Adjustment for use of cHRT.
  • Abbreviations: cHRT, combined hormone replacement therapy; FHx, family history; TOC, tubo-ovarian cancer.