Has Cytology Become Obsolete as a Primary Test in Screening for Cervical Cancer?

The report by Sankaranarayanan et al., ¹ published in N Engl J Med in 2009, is central to the ongoing debate as to whether human papillomavirus (HPV) testing should replace cytology as the primary screening test in cervical screening. The large significant decrease in mortality from cervical cancer observed in this cluster-randomized screening trial after a single round of HPV screening, but not after a single round of cytology screening, is considered to have ‘immediate and global’ implications for widespread implementation of HPV testing, and for refraining from establishing cytology-based screening in developing countries. ² These data may thus have profound consequences for the health of women globally, so they deserve careful scrutiny.

The validity of conclusions from a randomized trial rests on balanced randomization of individuals into the separate trial arms. The authors found no difference in terms of household type, religion, education, occupation, marital status and the number of pregnancies, ¹ suggesting that the trial arms were adequately balanced. Some readers, however, identified variation between the trial arms in the underlying risk for cervical cancer, ³ while others suggested that there may have been some variation in follow-up care. ⁴ Such sources of variation could theoretically exaggerate the measured difference between the screening tests – but did they have any practical implications in this trial? Based on the reported data, we tried roughly to quantify these potential biases. We calculated 95% CI for relative risks (RR) for the HPV versus the cytology arm by assuming that the logarithms of RRs were approximately normally distributed.

Firstly, despite randomization women in the cytology arm (n = 32,058) tended to have a higher background incidence for cervical cancer than women in the HPV arm (n = 34,126). This can be illustrated with women who did not undergo screening, as the same percentage, 20.3%, among those invited in either of the two arms failed to attend (6509 in the cytology arm and 6934 in the HPV arm). Among these women, the risk of cervical cancer was 6.5 per 1000 (=42/6509) in the cytology arm, and 4.6 per 1000 (= 32/6934) in the HPV arm, ¹ resulting in a non-significantly increased risk of 40% in the cytology arm, RR 1.40 (95% CI 0.88–2.21). This tendency to an elevated background risk in the cytology arm may have also played a role among the screened women; however, this cannot be independently quantified because the effectiveness of the respective screening tests would then have to be taken into account.

Secondly, in the cytology arm 54 deaths from cervical cancer were observed among the 152 diagnosed patients, while 34 deaths from cervical cancer were observed among the 127 diagnosed patients in the HPV arm. The number of deaths depends on the length of follow-up of the patients, and on the stage distribution of the patients at the time of diagnosis. Women in the two arms of the trial were followed for the same eight-year period. A slightly higher proportion of patients in the HPV arm, 69% (=87/127), than in the cytology arm, 58% (=88/152), were diagnosed following a positive screening test, χ² = 3.33, df = 1, P = 0.07. Patients in the HPV arm therefore tended to spend somewhat more time with the cancer diagnosis than patients in the cytology arm, so variation in the length of follow-up of patients cannot explain why more deaths were observed in the cytology arm before the reported eight-year follow-up ended. Further, patients in the two arms had the same stage distribution at time of diagnosis, χ² = 4.04, df = 3, P = 0.26. One would therefore expect that roughly equal proportions of cervical cancer patients would die from their disease during the follow-up period. Applying the death proportion of 34/127 from the HPV arm ¹ on the 152 patients from the cytology arm would result in an expected number of 40.7 cervical cancer deaths in the cytology arm. Thus, had the estimated 40.7 instead of 54 deaths been observed, no difference would have been found between the risk of dying from cervical cancer with 34 cervical cancer deaths during the observed 268,674 person-years in the HPV arm, and the expected risk of 40.7 cervical cancer deaths during 251,144 person-years in the cytology arm, RR 0.78 (95% CI 0.50–1.23).

To sum up, the significantly decreased mortality from cervical cancer in the HPV arm compared with the cytology arm, RR 0.59 (95% CI 0.38–0.90), observed in the randomized trial in India may not necessarily have been a consequence of differences in the effectiveness between the two screening tests. A possibly higher background incidence rate in the cytology than in the HPV arm, as well as relatively poor survival among the patients in the cytology arm may also have contributed to the study outcome. This trial, therefore, ‘does not negate the benefit found from primary HPV screening’; ³ however, it does on the other hand not provide sufficient grounds for concluding that a single round of cytology is substantially less effective in preventing cervical cancer mortality than a single round of HPV testing. Until we manage to resolve the problem of a considerably higher rate of false-positive tests at HPV than at cytology screening, ⁵ cytology will still be an attractive primary test for cervical screening.

Conflict of interest statement: The authors declare no conflict of interest.

Sources of funding: Olga and Esper Boels Fund, Augustinus Fonden, Fabrikant Mads Clausens Fond, Familien Hede Nielsens Fond and Købmand Sven Hansen og hustru Ina Hansens Fond. None of these had any role in the study design, the collection of data, the analysis, the interpretation of the data, the writing of the paper and in the decision to submit the paper for publication.