This study confirms our previous findings that cervical screening in women aged 20-34 is less effective than in older women. By studying the effect of screening in smaller age groups, we have shown that the efficacy of screening decreases with decreasing age, even within the age range 20-34.
On average, participation in the UK cervical screening programme by a woman aged 35-64 reduces her risk of cervical cancer over the next five years by 60-80% and her risk of advanced cervical cancer by about 90%. The benefit of screening for women aged 25-34 is more modest. Screening in women aged 20-24 has little or no impact on the incidence of cervical cancer under the age of 30. This applied whether we looked at all cancers or restricted analysis to frankly invasive (that is, stage IB or worse) squamous carcinoma, or even to stage II or worse (fig 3). Because stage IB+ cancer is rare in young women, however, the confidence intervals are wide and our data do not rule out the possibility of screening in women aged 20-24 being effective in reducing stage IB+ cancer in women aged 25-27.
It has also been argued that women should have their first two smear tests close in time to minimise the impact of an initial false negative result.13
Our results provide no evidence that women screened aged 20-22 and then again at 23-25 are better protected than those screened only at age 23-25.
A careful review of the screening histories of women aged 20-24 with a diagnosis of cervical cancer suggests that few (if any) of the cancers occurred through a lack of screening. Indeed only five of these 73 women had not been screened previously.
Strengths and weaknesses
We used prospectively recorded screening data and selected controls at random, thus eliminating both recall bias and selection bias (data were obtained for all selected controls). We believe this design to be the most appropriate given that a randomised controlled trial was not possible. Other papers have analysed trends in incidence of cancer (or mortality, or both) before and after the introduction of screening to estimate the impact of screening in the population.14 15 16 17
Such analyses rely on comparison of observed rates with estimates of what would have happened in the absence of screening; they are subject to trends in other factors such as the quality of the cancer registry data. For women aged 20-29, who have been offered screening from the age of 20, it is not possible to reliably estimate what their rates would have been in the absence of screening.
It is always possible to criticise observational studies as women who attend screening might differ from those who do not so that any observed effect might not be causal. For the observed difference in the benefit of screening at different ages to be caused by confounding, however, there would have to be differences in the way confounders affect the results at different ages. We know of no evidence to support such an interaction and suggest that differential benefits with age are not caused by confounding but reflect the true effects of screening. We think there are few biases in this analysis and are comfortable in viewing the associations as causal and using the term “protection offered by screening” to describe the odds ratios.
This large study of the impact of cervical screening on invasive cervical cancer contains more cases than all other studies with detailed screening information combined. Some of the data presented have been published previously, but over 45% are new. The results of analyses limited to the new data are qualitatively similar to those using all the data, suggesting that there have been no changes in the impact of screening in young women on rates of cervical cancer despite improvements in the quality of screening in the UK.
The new analysis considers the association between screening in one age group (for example, 25-29) and cervical cancer in the subsequent five years (at ages 30-34 for this example). With this approach, the exposure is close to the usual definition of screening coverage used by many screening programmes: the proportion of eligible women screened in the past five (or three) years. Furthermore, this approach more closely reflects what one could estimate by prospectively following a cohort of women. With a coverage interval of three years and a follow-up of five years, however, it could be as much as eight years between the last screen and a diagnosis of cancer. Thus the benefit of regular three (or five) yearly screening could be considerably greater than that implied by this model. Nevertheless, the approach does attempt to measure the protection achieved with screening (from the treatment initiated by a positive smear result) rather than the low risk periods associated with a negative smear result.
Comparison with other studies
Much of the earlier literature aiming to study the protection afforded by cervical screening did not consider the possibility of different levels of protection depending on age. An important recent paper reported on the results of an audit of cervical cancer in Sweden.11
The odds ratios in that paper are similar for all age groups with an odds ratio of 0.42 (1/2.37) for the effect of three yearly screening on incidence of cancer at ages 21-29. It is important to consider the methodological differences between the analyses when interpreting these results. They consider a woman (aged 20-52) to have been screened if she had a smear test between 3.5 years and 6 months before (the date of the case’s) diagnosis. They include stage IA cancers, most of which will have been screen detected, as well as screen detected stage IB cancer. Consider a screen detected cancer in a woman who has two smear tests 3.5 years apart. If the smear test that led to diagnosis is within six months of diagnosis she will be classed as unscreened. A control woman who is screened every 3.5 years will have a chance of 86% (3.0/3.5) of having had a smear test in the particular interval width of three years. Thus the inclusion of screen detected cases introduces a considerable bias in favour of screening. As the proportion of cancers that are stage IA or screen detected stage IB is greater in young women, the bias is particularly strong in young women. To illustrate the point, the same analysis applied to women aged 20-29 in our study yields an odds ratio of 0.46 (0.38 to 0.56).
A case-control study in New South Wales, Australia, found that screening every two years seemed to be more protective in women over the age of 30 than in those aged 20-29.10
The more favourable results for screening in women aged 20-29 in that study could be because their controls were selected from women who had been for screening (albeit possibly only after the date of diagnosis of the case).
Interpretation of the results
As we designed the study to eliminate most of the biases that affect case-control studies, our observed associations are almost certainly either causal or the result of confounding. The heterogeneity of association in different age bands within the same study argues strongly that these effects are real. There might be biological reasons for cervical screening working better in older women. Undoubtedly the specificity of screening is less in younger women because human papillomavirus (HPV) infections are so much more common. This does not, however, explain why the sensitivity of screening should be less. We favour the explanation that, by necessity, a cancer in a woman aged 25 (infected at, say, 15 years) will have progressed from HPV infection through cervical intraepithelial neoplasia grade III to cancer faster than in a woman aged 55 (infected at perhaps 25). This means that the opportunities for detecting the small proportion of cervical intraepithelial neoplasia grade III in women in their early 20s that will progress to cancer within at most a few years are small. It is an extreme example of length bias: most cases of cervical intraepithelial neoplasia grade III detected will be slow growing and could safely be left for several years; but the rare cases that are progressing rapidly will probably be missed.
It has been argued that screening from age 20 could prevent more cancers in women aged 25-34 than screening from age 25. Our study has little power to detect such an effect. Policy decisions should be based on balancing the benefits and harms of screening and the need to take into account the underlying risk of cervical cancer at different ages. Such an analysis is beyond the scope of this paper. We have provided more accurate estimates of the benefits of cervical screening in different age groups, which should aid policy makers in making their decisions. As screening undoubtedly leads to the detection of many cases of stage IA cancer in women aged 20-24, one might think it must be doing good by preventing more advanced cancers. If this were the case, we should have found that screening at ages 20-24 leads to a reduction in stage IB+ cancers. That we found no such reduction suggests that most of the stage IA cancers detected by screening women aged under age 25 would still be stage IA at ages 25-26 and could be picked up by screening at age 25 without adverse consequences.
This study is based on cancers diagnosed in the UK between 1990 and 2008 and smears taken within the UK from 1988 to 2008. Cervical cancer rates in women in their 20s have been relatively high compared with those in other countries and abnormal cytology results have been more common in those aged 20-34 than in older women throughout this period.18 19
We believe that the standard of smear taking, cytology reading, and fail safe procedures for cervical screening in this study have mostly been high. Since the early-mid 1990s the UK screening programmes have put great emphasis on quality assurance, and there is evidence that by the late 1990s, UK cytology was as good as (or better than) anywhere in the world.20
Thus our finding that cervical screening in women aged 20-24 has at best a modest effect on the incidence of cervical cancer at ages 25-29 is almost certainly generalisable to other countries. We have no reason to believe that it would be substantially more effective elsewhere.
Any decision on when to start screening women will have to weigh up benefits and harms and might depend on the local status quo. In a setting where screening is offered to women aged 20-24 policy makers might decide to continue this policy as we have not shown that the harms exceed the benefits. By contrast, where screening is not offered to women aged 20-24, the lack of evidence of any benefit from screening in this age group dictates that the policy should not change.
Unanswered questions and future research
Our study does not consider the harms of screening at different ages nor do we take into account the absolute rate of cervical cancer in screened and unscreened women of different ages. Such a synthesis of research is necessary for rational policy making. Undoubtedly, however, decision making will be complicated because of the uncertainty in many of the estimates of harm and benefit. As we have seen, the confidence intervals for the impact of screening at ages 20-24 on stage II+ cervical cancer are extremely wide. We have not even attempted to estimate the added impact of starting screening five years earlier on cancer at ages 30-44.
The most common harms of screening are the anxiety caused by abnormal test results and the trauma of treating cervical intraepithelial neoplasia that would never have progressed to cancer. These can be easily estimated. Treatment might be associated with premature delivery during subsequent pregnancies. If the association observed in several studies21 22
is causal then screening might do serious harm, but the association might simply be because of confounding. These issues require careful study.
The question of screening women aged 20-24 will decrease in importance as the cohort of women vaccinated against HPV types 16 and 18 reach their 20s. If it is questionable whether screening is worth while in unvaccinated women aged 20-24, there can be no doubt that the risk of cancer in women aged under 25 who are vaccinated before exposure to HPV will be low enough to make screening at such an age unjustifiable.
What is already known on this topic
- Cervical screening has had a substantial impact on the incidence and mortality of cervical cancer in many developed countries
- Most of the benefit from screening comes from the prevention of cervical cancer, but it can also lead to downstaging
- The relative protection against cervical cancer might be less at ages 20-34 than in older women
What this study adds
- Cervical screening in women aged 20-24 is substantially less effective in preventing cancer (and in preventing advanced stage tumours) than is screening in older women
- The new methods for the analysis of case-control studies of screening avoid some of the biases associated with previously used statistical methods