Sir

SIR — Your Editorial 'Early warnings' (Nature 458, 679; 2009) points out some pitfalls in the effectiveness of cancer screening. I would like to draw attention to problems associated with gynaecological screening in resource-poor countries.

Disease prevalence can cause a large variation in the positive predictive value (PPV) of tests whose sensitivity and specificity are comparable. If 10% of the population is likely to have a disease and 1,000 people are screened for it, then by using a test with 100% sensitivity and 99% specificity, we should pick up 100 true positives and 10 false positives; the PPV will be 100 true positives to 110 total positives, or 91%. But for a disease with an incidence of 0.1%, a test of the same sensitivity and specificity will generate one true and 10 false positives per 1,000 screened, with a resultant PPV of one true positive/11 total positives, or 9%.

Take, for example, screening tests for ovarian and cervical cancers, to identify precancerous conditions and early-stage disease. Although these cancers are widespread killers, they have a low incidence compared with, say, tuberculosis. Ovarian cancer has a worldwide incidence of 0.01% (International Agency for Research on Cancer, http://www-dep.iarc.fr), meaning that one screen, even with 99.9% specificity, will give a PPV of 9%. The resulting surgery will be unnecessary for 10 out of 11 women who undergo it.

Such high specificity is very difficult to achieve, but anything lower is useless. Even multivariate systems combining many markers for ovarian cancer have failed to increase specificity above 99.9%. Despite this wastefulness, sponsors and governments are under pressure to organize large-scale trials, as your Editorial mentions.

The value of screening is also affected by the feasibility of further investigations and treatment. Take cervical cancer, which overall is more than twice as common as ovarian cancer (although its incidence varies widely around the world): the position of the cervix means that even a low PPV will not undermine well-established screening programmes, as results can easily be verified by biopsy rather than a major operation.

However, in many resource-poor countries, where only about 1% of women may be screened for these cancers, the confounding false negatives and negative predictive values caused by low sensitivity present a big problem. Many cases will not be discovered until the disease is too far advanced for successful treatment. In such countries, huge populations and restricted resources mean that screening efforts may need to be abandoned in favour of inexpensive universal treatment or prophylaxis. Newer, targeted approaches for ovarian cancer (T. A. Yap et al. Nature Rev. Cancer 9, 167–181; 2009) and a cheaper vaccine for cervical cancer may prove helpful.