Table 3 Non-randomised controlled study options and key difficulties

Option 1: Intervention group: engage^a with PAS

Comparison group: active exposure but no engagement with PAS^b

Systematic difference between intervention and comparison group. IB recipients who engage with PAS are likely to be healthier and closer to a return to work than those who are aware of PAS but are not referred (by GP or self).

Selection bias leading to overestimate of health effects

Option 2:

Intervention group: engage with PAS^a

Comparison group: not exposed to PAS^d

Need to identify suitable comparison group with respect to eligibility for referral to PAS (only 20% of those in the comparison GP practices would be 'comparable' to the intervention group)

How to determine suitable controls: option to ask GP in comparison practices to tell us who, in principle, they would refer to PAS. This requires a high level of involvement by GPs not offering PAS

Possible selection bias depending on ability to match controls

Cluster level influences need to be accounted for

Option 3:

Intervention group: engage with PAS^a

Comparison groups:

(1) Not exposed to PAS^d

(2) passive exposure to PAS^c

Need to identify suitable comparison group with respect to eligibility for referral to PAS (only 20% of those in the comparison GP practices would be 'comparable' to the intervention group)

How to determine suitable controls (this will be aided by information from the group who are exposed but do not engage)

Possible selection bias depending on ability to match controls

Cluster level influences need to be accounted for