Table 5.

Potential biases, their influence and how this may be mitigated

Potential source of bias	Impact of this bias	Possible strategies to minimise this and other relevant considerations
Selection bias with regard to initial case ascertainment since cases were all service users and IDU were not selected at onset of injecting	Causes, consequences, natural history and duration of IDU amongst injectors who do not present to services may be different	Impossible to avoid however likely to be less of an issue than in studies where cases are recruited from specialist clinics as this involves additional level of selection. In addition, time from onset of injecting to recruitment in this study shorter than in most other cohorts.
Survival bias with regard to case follow-up	Patterns of association between the factors under study may have been different amongst living compared to dead cohort members	Information on most factors of interest was available through record linkage on both living and dead cohort members
Selection bias with regard to case follow-up. Cases successfully followed up were willing to be interviewed. Unwillingness to be interviewed may have reflected either more chaotic current circumstances or a reluctance to discuss long resolved drug problems.	Patterns of association between the factors under study may be different amongst those lost to follow-up. Outcomes of IDU may have either been over or underestimated	Impossible to avoid though loss to follow-up was relatively low and much was due to structural factors (e.g. GP unwillingness to recruit) unlikely to be related to participant characteristics
Selection bias with regard to control recruitment as controls were all attending a health facility	If controls were more likely to have health problems than the general population this may have diluted associations between some risk factors and outcomes in case-control comparisons	The majority of the population use primary care services relatively regularly often for reasons unrelated to a significant health problem and any "unhealthy participant" effect is therefore likely to be small
Selection bias with regard to control recruitment as controls may not have been a representative sample of service users	Potential controls declining recruitment may have been different from those agreeing with regard to the factors under study	Consecutive eligible service users were approached during control recruitment. Only 3% declined suggesting substantial bias is unlikely
Selection bias with regard to control recruitment as controls were not recruited at the same time as cases	To be recruited controls must be alive and resident in the practice area. Despite age and sex matching this may have introduced bias.	Impossible to avoid as control selection from reconstructed historical practice list was unfeasible (see text). Impact may not have been substantial since healthier controls would be both more likely to be living but may also have been more likely to leave practice area. These influences would tend to cancel each other out in terms of resulting bias.
Social desirability bias in relation to interview measures	Cases may have been more likely to disclose drug use and other socially sensitive behaviours and exposures leading to overestimation of the association between these factors and IDU	Assurances of confidentiality and good relationship with practice team should have mitigated this. Where possible objective corroboration with measures collected through linkage was sought
Recall bias in relation to interview measures	Case recall of some early life exposures may have been influenced by their own beliefs around causes of IDU leading to overestimation of the association between these factors and IDU. Substance use may also have impaired case recall of previous exposures leading to underestimation of the association between these factors and IDU.	Use of the life-grid approach should have mitigated this. Where possible objective corroboration with measures collected through linkage was sought
Strong association between disadvantage and IDU may lead to confounding of case-control comparisons	Some apparent effects of both IDU itself and possible risk factors may in reality be effects of other correlates of disadvantage	Recruitment of controls from the same community as cases should mitigate any bias of this type and measurement of individual social position allows further adjustment