Automation-assisted versus manual reading of cervical cytology (MAVARIC): a randomised controlled trial

Background The standard for reading cervical cytology is for a cytoscreener to manually search across an entire slide for abnormal cells using a conventional microscope. Automated technology can select fields of view to assess abnormal cells, which allows targeted reading by cytoscreeners. In the Manual Assessment Versus Automated Reading In Cytology (MAVARIC) trial, we compared the accuracy of these techniques for the detection of underlying disease. Methods For this randomised controlled trial, women aged 25-64 years undergoing primary cervical screening in Manchester, UK, were randomly assigned (1:2) to receive either manual reading only or paired reading (automation. assisted reading and manual reading), between March 1, 2006, and Feb 28, 2009. In the paired arm, two automated systems were used the ThinPrep Imaging System and the FocalPoint GS Imaging System. General practices and community clinics were randomised to either ThinPrep or to SurePath (for the FocalPoint system) liquid-based cytology with block randomisation stratified by deprivation index. Samples were then individually randomised to manual reading only or paired reading only. Laboratory staff were unaware of the allocation of each slide and concealment was maintained until the end of the reporting process. The primary outcome was sensitivity of automation-assisted reading relative to manual reading for the detection of underlying cervical intraepithelial neoplasia grade 2 or worse (CIN2+) in the paired arm. This trial is registered, number ISRCTN66377374. Findings 73 266 liquid-based cytology samples were obtained from women undergoing primary cervical screening; 24 688 allocated to the manual-only arm and 48 578 to the paired-reading arm. Automation-assisted reading was 8% less sensitive than manual reading (relative sensitivity 0.92, 95% CI 0.89-0.95), which was equivalent to an absolute reduction in sensitivity of 6.3%, assuming the sensitivity of manual reading to be 79%. Specificity of automation-assisted reading relative to manual reading increased by 0.6% (1.006, 95% CI 1.005-1.007). Interpretation The inferior sensitivity of automation-assisted reading for the detection of CIN2+, combined with an inconsequential increase in specificity, suggests that automation-assisted reading cannot be recommended for primary cervical screening