Monitoring and evaluating the performance of the UK NHS Cervical Screening Programme: monitoring performance by using cytology outcomes adjusted for population characteristics.

Abstract

Current quality assurance measures used in the NHS cervical screening programme (NHSCSP) include a review of laboratories with percentages of moderate/severe and borderline/mild smear results outside the 10th-90th percentiles. The method is limited by the fact that many of these outlier smear percentages may reflect laboratories covering populations with low or high risk and/or short or long average screening intervals. This paper outlines a new approach to aid the detection of outlier laboratories, by using data collected at the primary care trust (PCT) or health authority (HA) level and making allowances for population characteristics and screening interval. The setting is the NHSCSP in England using annual data provided by HAs. Data from the screening year 2000-01 is used to illustrate the methodology, although the methods can also be applied to data at the PCT level (now being collected for 2002-03 onwards). Percentages of smear results have been analysed against a series of explanatory variables using logistic regression models. These explanatory variables include Townsend deprivation index, uptake-corrected ethnic minority composition, a measure of screening interval, area type and region. An expected percentage of borderline/mild and moderate/severe smears is estimated from the models and an observed : predicted ratio (OPRmod/sev and OPRbord/mild) calculated. Low values are suggestive of relative undercalling and high values overcalling, after allowance for population characteristics. Analysis of data for 2000-01 showed that the OPRmod/sev for the 99 HAs varied from 0.68 to 1.44. Laboratories with low percentages of moderate/severe smears, but associated with PCTs or HAs with OPRmod/sev values closer to unity may not need to be investigated as their observed rates are consistent with predicted rates based on population characteristics. The method could also be directly applied to laboratories if further information on the population covered by each laboratory were routinely collected.