Four RCTs (n=1,084) and one controlled before-and-after study (n=1,979) were included.
All 4 RCTs scored more than three on the Jadad scale and used adequate methods of allocation concealment. Two RCTs described an adequate method of randomisation. All 4 RCTs reported using intention-to-treat analysis, but three did not describe how they handled missing data. The rates of follow-up ranged from 78 to 95% across all studies.
Reduction in prescriptions used or collected: all studies showed that delayed prescription statistically significantly reduced antibiotic use or collection compared with control; the RR ranged from 0.25 (95% CI: 0.19, 0.34) to 0.77 (95% CI: 0.73, 0.81). In the 3 studies in which patients returned to the physician's office to collect prescriptions, the RR ranged from 0.25 (95% CI: 0.19, 0.34) to 0.45 (95% CI: 0.36, 0.56). In the 2 studies in which patients were given the prescription at the time of consultation, the RRs were 0.54 (95% CI: 0.41, 0.7) and 0.77 (95% CI: 0.73, 0.81).
The RRs in 2 studies of children with otitis media differed: the RR was 0.25 for the RCT and 0.77 for the before-and-after study.
Symptoms and signs: one of the 4 studies reported a decrease in symptoms, while the other 3 RCTs found an increase in symptoms in the delayed prescription group. The results were also reported for other specific signs and symptoms.
Satisfactions with the consultation: the results were inconsistent. Two of the 4 RCTs found significantly reduced patient satisfaction with the consultation where delayed prescriptions were given. Other results were also reported.
Beliefs about antibiotics: two of the 3 studies reported a significant reduction in the belief that antibiotics were effective in the delayed group compared with the immediate prescription group.
Cointerventions: the results differed. One study found no significant difference in paracetamol use between treatment groups, while another study reported higher use in the delayed prescription group.
The funnel plot showed no evidence of publication bias, but it was difficult to assess given the small number of studies.