BACKGROUND: Low-dose computed tomography (LDCT) screening has been shown to reduce mortality from lung cancer but at a substantial cost in diagnostic activity. The objective of this study was to investigate the characteristics of screening programmes associated with recall rates, detection rates and positive predictive values (PPVs). DESIGN: We conducted a systematic review of randomised trials and observational studies on LDCT screening for lung cancer. A meta-regression using random-effect logistic regressions was carried out to assess factors influencing recall rates for further investigation, cancer detection rates and PPVs of recall. RESULTS: We used data from 63 372 prevalent screens from 16 studies of LDCT screening for lung cancer and 79 302 incident screens from nine studies. In univariable analysis, the use of a cut-off size to define nodules warranting further investigation at prevalent screens reduced recall rates [odds ratio (OR) = 0.44, 95% confidence interval (CI) 0.24-0.82 and OR = 0.42, 95% CI 0.21-0.84 for cut-off sizes of 3-4 and 5-8 mm, respectively], without significant changes in detection rates and PPVs. The number of readers (1 or ≥2) was not associated with changes in recall rates, detection rates and PPVs at prevalent and incident screens. Using the volumetry software at incident screens significantly increased the PPV (OR = 5.02, 95% CI 1.65-15.28) as a result of a decrease in recall rates (OR = 0.25, 95% CI 0.12-0.51), without significant changes in detection rates. CONCLUSION: These results highlight the value of using a cut-off size for nodules warranting further investigation with lower recall rates at prevalent screens, whereas the volumetric assessment software at incident screens results in lower recall rates and higher PPVs. The presence of positron emission tomography in the work-up protocol might be associated with lower rates of surgical procedures for benign findings, although this hypothesis deserves further investigation.