Twenty-nine studies were included in the review (3,373 patients, sample size range 12 to 501): 18 randomised trials (2,716 patients) and 11 observational studies (657 patients). Follow-up ranged from 28 to 365 days. Two studies reported blinding of patients to the use of electronic monitoring, 20 reported that they did not blind patients and seven did not report any blinding-related information.
Crude pooled adherence estimates were highest when the lenient taking definition was assessed (dosing frequency range 80.1% to 93.1%) and lowest when the strictest timing definition was assessed (dosing frequency range 50.4% to 76.3%). Significant statistical heterogeneity (Ι²>85%) was reported for once-daily and twice-daily regimens for all three adherence definitions and for the timing analysis of three times-daily regimens. Publication bias could not be ruled out for once-daily regimens for all three adherence definitions and for the taking analysis of twice-daily regimens.
Compared with once-daily regimens, the adjusted weighted mean percentage adherence for twice daily and three times daily regimens were significantly lower regardless of the adherence definition used (p<0.01 for all). Results were reported fully in the paper.