Table 4

Risk of bias in the included observational studies

Study nameDeveloping and applying appropriate eligibility criteriaMeasurement of exposureMeasurement of outcomeControlling for confoundingCompleteness of data
Boltri et al,10 2002Low risk
Physicians and residents in the control and exposed groups were from the same pool
Low risk
Policy applied across the clinic
Low risk
Data collection was based on medical records, and carried out by a research assistant blinded to study design and hypothesis
Low risk
‘Logistic regression was then performed to adjust the odds ratio for the relation of physician type, prescribing patterns, and time’
Low risk
No missing data reported
Spurling and Mansfield,12 2007Low risk
Diaries chosen at random for a 1-month period. A random week was chosen for auditing doctors’ prescribing
Low risk
Policy applied across the clinic
Unclear risk
Not clear whether the survey instrument was validated
High risk
According to the authors, the possibility of confounding cannot be ruled out
Low risk.
All except one returned the completed questionnaire
Hartung et al,11 2010Unclear riskLow risk
Policy applied across the clinic
Unclear risk
Use of claim data; however, validity of the data not described
Low risk
They include ‘a contemporaneous control group of patients or clinicians also experiencing this potential confounder’ (confounding resulting from secular changes in prescribing)
Low risk
‘Although it is possible that some prescriptions would not have been captured by using data from only one pharmacy, it seems unlikely that this subset would have introduced any systematic bias or loss of generalisability’