Comparing the sensitivities of two screening tests in nonblinded randomized paired screen-positive trials with differential screening uptake

Staged design recommendations for validating relative sensitivity of self-sample human papillomavirus tests for cervical screening

OBJECTIVES

To ensure that the emerging methods for human papillomavirus (HPV) testing on self-collected samples in cervical screening are evaluated robustly.

STUDY DESIGN AND SETTING

We assess paired study designs for relative sensitivity of self-collected vs. traditional clinician-collected samples in detection of high-grade cervical intraepithelial neoplasia.

RESULTS

Designs considered are (D1) both samples at screening, with clinical actions triggered by HPV positivity; (D2) offering a self-sample test to clinician-collected HPV-positive women; (D3) as D2 but using a repeat clinician-sample as comparator; (D4) offering a choice of self- vs. clinician-sampling, and the alternative test in HPV-positive women; (D5) paired samples at referral appointment. D1 is simple to analyze but requires the largest sample size and referral of self-sample positive, clinician-sample negative women. D2 requires a much smaller sample size, and no change to clinical practice, and could be used to rule-in a test because estimates are conservative (against self-sampling). D3 mitigates this bias but requires a second clinician sample. D4 is only manageable where self-sampling already occurs. The liberal D5 might be used to rule-out a self-sampling test.

CONCLUSION

A universal recommendation for an optimal study design is challenging. Staged validation might be useful with D5 as a gatekeeper for D1-D4.