Assessment of the performance of algorithms for cervical cancer screening: evidence from the Ludwig-McGill cohort study.

OBJECTIVE

There are currently multiple tests available for cervical cancer screening and the existing screening policies vary from country to country. No single approach will satisfy the specific needs and variations in risk aversion of all populations, and screening algorithms should be tailored to specific groups. We performed long term risk stratification based on screening test results and compared the accuracy of different tests and their combinations.

METHODS

A longitudinal cohort study of the natural history of HPV infection and cervical neoplasia enrolled 2462 women from a low-income population in Brazil. The interviews and cervical screening with cytology and HPV DNA testing were repeated according to a pre-established protocol and the subjects were referred for colposcopy and biopsy whenever high grade lesions were suspected. We compared the specificity, sensitivity and predictive values of each screening modality. Long term risk stratification was performed through time-to-event analyses using Kaplan-Meier analysis and Cox regression.

RESULTS

The best optimization of sensitivity and specificity was achieved when using dual testing with cytology and HPV DNA testing, whereby the screening test is considered positive if either component yields an abnormal result. However, when allowing 12months for the detection of lesions, cytology alone performed nearly as well. Risk stratification revealed that HPV DNA testing was not beneficial for HSIL cases, whereas it was for ASCUS and, in some combinations, for negative and LSIL cytology.

CONCLUSION

Our results suggest that some high risk populations may benefit equally from cytology or HPV DNA testing, and may require shorter intervals between repeat testing.