These analyses provide quantitative estimates of the benefit of different screening intervals at different ages. We have considered cancers regardless of histological type because it is not possible to set one screening interval for the prevention of squamous cell carcinoma and another for adenocarcinoma. Although cytological screening may be less effective against adeno-carcinoma, it appears to have a substantial impact (Sasieni and Adams, 2001
It is inappropriate to compare the time since the most recent negative smear in screen-detected cases with that in randomly selected controls (IARC, 1986
; Sasieni, 2001
), because for a screen-detected case the time since her last negative smear will be her normal screening interval plus a short period to allow for diagnosis. It is not possible to identify which cases are screen-detected using routine data, because approximately 50% of women screened in England in the mid-1990s did not attend in response to an invitation (Department of Health, 1996
). Even if GP notes were examined, it is doubtful whether one could distinguish between symptoms reported during a routine screening examination and symptoms that lead to the consultation. By considering only frankly invasive (stage 1B+) cancers, we hoped to eliminate screen-detected cancers, but in a series of 327 such cancers from southern England, 78 (24%) were classified as screen-detected (Herbert et al, 2001
Screening intervals in this study were primarily either just over 3 or just over 5 years. Hence, if screen-detected cancers are included, the RR for 0–3 years will be artificially low, but it will increase substantially for 3–3.5 years. This is precisely the pattern observed (). It is beyond the scope of this study to quantify the extent to which women with screen-detected stage 1 cancers benefited from screening.
Although we have used these RRs to estimate the absolute reduction in stage 1B+ cervical cancer that might result from 1-, 3- and 5-yearly screening, these should not be used in formal calculations of the cost per cancer prevented since negative smears do not prevent cancer. This method will overestimate the proportion of cancers prevented since not all women with abnormal smears will be prevented from developing cancer (Sasieni et al, 1996
). Nevertheless, it does provide valuable information on the natural history of cervical precancer and of the relative benefit of different screening intervals.
Relating incidence to the time since the most recent screen estimates the extent to which a woman who has been screened recently is prevented from developing cancer. However, it is difficult to determine whether a smear was taken for screening purposes: a high proportion of cases had at least one smear in the months preceding diagnosis and many of these women had no other smears on record. For that reason, we excluded all smears taken during the 6 months immediately preceding (pseudo-) diagnosis. A period of 6 months was chosen by examining the proportions whose most recent smear was within various intervals from diagnosis (
). Half of all cases had a smear within 3 months of diagnosis compared to just 5% 6–12 months prior to diagnosis. In fact, the 6-month exclusion (used in ) is not long enough to take account of all ‘diagnostic smears’ – the relative proportions last screened 6–12 months, compared to 1–6 years, prediagnosis is greater in cases than controls (P=0.002).
Proportion of women aged 20–69 years whose most recent smear was within different intervals
Another difficulty with identifying the last screening test is that once a woman has been treated for a cervical lesion, she could be put on indefinite annual follow-up. Our solution of censoring screening histories at the first abnormal smear is not ideal and it potentially introduces a small bias in favour of screening. Further, the fact that the RRs (in ) are greater in the first time period than the second suggests that the 6-month exclusion is not quite sufficient and that these RRs should not be overinterpreted. Despite these caveats, we believe that this approach does provide reasonable estimates of the efficacy of 3- and 5-yearly (but not annual) screening ().
In younger women, the risk of disease in those whose last smear was more than 5.5 years ago was greater than in those who had no smears (). This suggests that those who opt out of screening altogether are at a lower underlying risk of cervical cancer than those who are screened occasionally. Opportunistic screening of women seeking contraceptive advice and those attending STD clinics could account for such a trend. What then is the appropriate baseline for estimating the RRs? We have used those with no smears, but use of those with no recent smear would have made the estimated effect of screening considerably greater in young women. The RR in women aged 20–39 years, whose most recent operationally negative smear was 3.5–4.5 years ago, was 1.06 relative to those with no such smear (). However, relative to those whose most recent negative smear was more than 6.5 years ago, it is 0.45 (=1.06/2.37). In our opinion, such adjusted RRs are inappropriate: (ignoring the effect of screening) it is more likely that those who were last screened many years ago form a high-risk subgroup than that those who are never screened are at low risk. But this needs to be tested by a larger, more detailed study in young women in which risk factors for the acquisition and persistence of HPV infection are collected along with screening histories.