Tel.: +358-9-13533265, Fax: +358-9-673108
Mortality audit of the Finnish cervical cancer screening program
Version of Record online: 12 OCT 2012
Copyright © 2012 UICC
International Journal of Cancer
Volume 132, Issue 9, pages 2134–2140, 1 May 2013
How to Cite
Lönnberg, S., Nieminen, P., Luostarinen, T. and Anttila, A. (2013), Mortality audit of the Finnish cervical cancer screening program. Int. J. Cancer, 132: 2134–2140. doi: 10.1002/ijc.27844
- Issue online: 18 FEB 2013
- Version of Record online: 12 OCT 2012
- Accepted manuscript online: 14 SEP 2012 04:55AM EST
- Manuscript Accepted: 31 AUG 2012
- Manuscript Received: 24 APR 2012
- Finnish Cancer Organizations
- European Union Seventh Framework Programme contract EUROCOURSE—Europe against Cancer: Optimization of the Use of Registries for Scientific Excellence in Research; Co-funding was provided by the European Union Public Health Programme (Project no. 2006322, European Cooperation on Development and Implementation of Cancer Screening and Prevention Guidelines [ECCG])
- mass screening;
- uterine cervical neoplasms;
- case-control studies;
Incidence-based evaluations of cervical cancer screening programs have suggested age-specific impacts and there is uncertainty regarding the effectiveness of screening outside the ages of 30–60 years. We audited the screening histories of cervical cancer deaths and conducted a case-control evaluation of the effectiveness of organized screening in different ages with mortality as outcome. We included all 506 cervical cancer deaths in Finland in 2000–2009 due to cancers diagnosed in 1990 or later, and 3,036 controls matched by age at diagnosis to the cases. Squamous cell carcinoma constituted 59% of the cases, adenocarcinomas 29%, and the remaining 12% were other specified and unspecified cervical malignancies. Most deaths (54%) were due to cancers diagnosed more than 5 years after last screening invitation, 24% were diagnosed among nonattenders and only 14% of deaths occurred among women who had attended invitational screening. The risk reduction associated with attending a single program screen at an age below 40 was nonsignificant (OR 0.70; 95% CI 0.33–1.48), while clear risk reductions were observed after screening at the age of 40–54 (OR 0.33; CI 0.20–0.56) and 55–69 (OR 0.29; CI 0.16–0.54). This study also provides some indication of a long-lasting additional effect of screening at the age of 65. Possible avenues for improving the effectiveness of the Finnish screening program include efforts to increase attendance and an extension of the target ages to include 65-to 69-year-old women. The potential benefit of increasing the sensitivity of the screening test or shortening the screening interval is smaller.
The effectiveness of cervical cancer screening by cytology has never been established in a randomized controlled trial, but numerous observational studies have confirmed the considerable potential for prevention of cancer incidence and mortality.1 However, recent research suggests that screening may not be equally effective in all age-groups currently targeted by screening programmes.2,3
In a case-control evaluation of the Finnish program, no effect on the risk of cancer in the following 5-year interval was found after participation at the age of 25–29.2 Risk reductions increased with increasing age up to 40, and remained fairly constant thereafter. Similarly, a case-control study from the UK found screening to be less effective in women aged 20–34 and that the efficacy decreases with decreasing age even within this age range.3 By contrast, a Swedish audit found screening to be equally effective in preventing cancer diagnosed in the ages 21–29 and 30–65.4 To our knowledge no mortality based effectiveness studies have yet addressed this issue.
The purpose of this study was to audit the screening histories of women who died of cervical cancer and evaluate the impact of participation in organized screening on the mortality from cervical cancer by morphology in different ages. We especially wanted to explore how screening at the age of 65 affects the mortality at advanced ages.
Material and Methods
The cervical cancer screening program in Finland was introduced in 1963 and achieved national coverage by the early 1970s. Women are invited for screening every 5 years within the recommended target ages of 30–60, some municipalities also invite women at the ages of 25 and 65. The screening test has been conventional cytology, with the exception of some municipalities in Southern Finland that participate in a randomized trial with primary HPV DNA testing.5,6 According to trend and cohort studies, the program has been very effective with decreases in both incidence and mortality of up to 80% by the early 90 s.7 Since then, some increase in incidence in the younger age groups has been observed.8
For this audit and impact evaluation study, all 545 deaths attributed to cervical cancer during the period 2000–2009 in Finland were collected from the cancer register and linked to the screening register. Even though screening has the greatest impact on squamous cell carcinomas (SCCs), these still constitute the majority (59%) of cervical cancers that cause death in Finland. As the purpose of this study was to evaluate the preventive effects of organized screening, screening history was constructed retrospectively from the date of incidence of the cancer through linkage by personal identifier to the screening register. The mass screening register database contains close to complete data on invitations, screening tests and diagnostic confirmation of findings from 1990 onwards. If not cured, cervical cancer usually causes death within 5 years of incidence.9 Even so, 39 cases were excluded from the study because their screening exposure mapped to the period before 1990 from which we did not have screening information. The mean time period between diagnosis and disease-specific death was 1,436 days among all 545 cases and the median was much shorter still at 518 days. For the remaining 506 cases, the clinical, pre-operative FIGO stage was derived from cancer notifications and grouped into IA (microinvasive), IB-IIA (localized) and IIB+ (advanced). The morphologies were classified into SCCs (59%), adenocarcinomas (29%) and other or unknown morphologies (12%). As expected, only a very small proportion of cancers (1%) that eventually caused death were staged microinvasive at presentation.
Each case was assigned a mode of detection with respect to invitational screening (Table 1). While opportunistic screening occurs in parallel to invitational screening, the national screening register does not contain data on tests outside the program. Mode of detection and screening exposure are based on invitational screening tests only. Those with a registered invitation in the 5 years (one screening interval) preceding the date of diagnosis were classified as invited. Those with a registered screening test were further classified as attenders. If diagnosis occurred within 12 months of a screening test that resulted in referral for colposcopy (positive screening test), the case was considered screen-detected. Those with negative or borderline last screens, or diagnosis more than 12 months after a positive screening test but before the next program invitation, were classified as interval cases. The uninvited were further classified into those diagnosed before first invitation, those diagnosed more than 5 years after last invitation and those lacking a registered invitation for other reasons. These other reasons included immigration or relocation within the country, nonadherence to screening recommendations by the home municipality and missing invitational information.
For the case-control analysis of screening effect, six controls matched for birth year and month were drawn from the population register for each of the 506 cases. Only women alive and not diagnosed with cervical cancer at the time of diagnosis of the case were eligible as controls.
A preventive screening exposure for both cases and their controls was derived from the screening status at the index screen. This index screening event was defined as the last age-group invitation and possible screening test within the 66 months that preceded the diagnosis. This time period corresponds to one screening interval and a 6 months allowance for test date variation within the invitational year. Screening events leading to the diagnosis of cancer (screening events of screen-detected cases) were disregarded and the previous age-group screening event within 12–78 months before diagnosis was used as index for that case and the corresponding controls. The majority (97%) of all age-group invitations were targeted at women in ages divisible by 5.
Conditional logistic regression was used to estimate odds ratios of the association between cervical cancer death and index screen participation in the program among invited women. The measure approximates the risk ratio of death from cervical cancer with diagnosis in the interval between screening invitations up to and including any screen-detected diagnoses in the following screening event. Attendance in the index program screening event was regarded as the exposure. We estimated a correction factor to account for self-selection bias by calculating odds ratios for those not responding to invitation compared to those not having been invited. This was possible because some municipalities have invited women at the age of 25 (overall 32% national coverage by invitation over the study period), some at the age of 65 (14% invitational coverage) and the invitational coverage of age cohorts 30 and 60 have also been incomplete during the study period (81 and 83%). For a more robust estimate of the self-selection bias factor, we used total incidence of fully invasive stage IB+ cervical cancers instead of only cancers leading to death. Odds ratio of stage IB+ cervical cancer in the following 5-year interval for those choosing not to attend screening after an invitation compared to those not invited was estimated at 1.45 (95% CI, 1.10–1.91). The corrected odds ratios for screening effect over all ages were estimated by applying formula (5) from Duffy et al.,10 using the participation rate of 71% of the source population i.e., those invited to screening in 1990–2009 in the cervical cancer screening program. Because there was considerable heterogeneity in the participation rates by age at invitation, the correction factor was adjusted accordingly for the individual 15-year invitational age groups. For the adjustments of the correction factor, we used the formula
where p is the overall participation rate 0.71, page is the age-specific participation rate, OR1 is the overall odds ratio of those not participating compared to those not invited (overall correction factor 1.45), OR2 is the odds ratio of those participating compared to those not invited (0.82, derived from the formula ) and OR1,age is the age-specific correction factor. We assumed that the higher risk observed in the nonparticipants is due to a small group of high risk women that do not participate at any age. The magnitude of the selection bias is thus dependent on the participation rate, that is, how many women at low risk are also not participating. When more women with low risk women fail to participate, risk difference is diluted and the correction factor approaches one. STATA/MP 11.0 software (StataCorp LP, College Station, TX) was used for all statistical calculations.
There were no deaths associated with cervical cancers diagnosed before the age of 25, only 9 cases diagnosed in the ages 25–29 and 13 in the ages 30–34 (Fig. 1). There were two peaks in the distribution of age at diagnosis. The first peak was associated mainly with an increase in adenocarcinomas after the age of 50 and the second peak with an increase in SCCs after the age of 70.
Accordingly, most of the deaths caused by SCCs (57%) but less than half of the adenocarcinomas (45%) were diagnosed more than 5 years after the last program invitation (Fig. 2). Only between 1 and 2 percentages of deaths caused by both of these types of cancer were diagnosed before first invitation. Adenocarcinomas were much more commonly diagnosed between screening rounds (22% vs. 8%). A larger proportion of the localized cancers (25%) than of the advanced cancers (13%) were diagnosed among screening attenders (Table 1).
There were clear differences in the screening status of cases and controls. In the whole case-control data, the proportion of women without invitation were similar among cases and controls, because lacking an invitation was largely age-dependent and so indirectly matched for, but invited nonattenders were more than twice as common among the cases than among the controls (Table 2). There were 64 cases of death (13%) with a negative screening test in the index round. Borderline cytology in the index round was equally common among cases and controls in total, but more common among cases as a proportion of those actually screened. There were no cases of death after colposcopy in the index round regardless of diagnostic outcome of the histological verification.
The odds ratios for the association of cervical cancer mortality and screening participation were found to differ across age groups. After correction for self-selection bias, the OR in the age group for invitation at 25–39 was estimated at a statistically non-significant 0.70 (CI 0.33–1.48) (Table 3). The ORs were estimated at 0.33 (CI 0.20–0.56) for the invitational ages of 40–54 and 0.29 (CI 0.16–0.54) for the invitational ages of 55–69. The overall effect of participation in one program screen in terms of reduction in the risk of cervical cancer death was estimated at 66% (OR 0.34; CI 0.14–0.49). The risk reduction in mortality due to SCCs only was 78% (OR 0.22; CI 0.13–0.36) whereas no significant risk reduction was observed for mortality due to cervical adenocarcinoma overall, nor stratified by age.
In an attempt to gain information about the duration of any protective effect of screening, we also explored the association of participation in the 5-yearly age-group program invitation preceding the index invitation among those with no screening test recorded at index (Table 3). In this analysis the screening exposure mapped to 5–10 years before diagnosis. The overall OR was 0.48 (CI 0.28–0.84) and after stratification by age the OR remained significant in the oldest invitational age group of 55–69 (OR 0.18; CI 0.05–0.62).
The effect of participation by age at last life-time invitation on the risk of cervical cancer death with diagnosis at any age above 65 was analyzed (Table 3). The effect of participation at 65 was a nonsignificant 0.28 (CI 0.03–2.47) after correction for self-selection. The effect of participation after a last invitation at 60 was similar (0.54) and nonsignificant. The number of cases with last invitation at 55 was insufficient for the estimation of associations.
By expanding the analysis to include also those not invited, statistically significant ORs for the association of being screened in individual age groups and mortality from cancer diagnosed at the ages of 66–80 and 71–80 were observed (Table 4). There was no correction for self-selection in this analysis as the comparison group did not exclusively consist of nonresponders. Hence, estimates should be cautiously interpreted as to their absolute value. Compared to no screening at the age of 55 or above, the point estimates for a last test given at 55, 60, or 65 were all below one, significantly so for last test at 60 and 65. The reduction in risk tended to be larger for last test at a later age, more clearly so when restricted to diagnosis in the ages of 71–80.
We estimated the effect of a single age-group invitational program screening test on the risk of death of cervical cancer with diagnosis in the next 5-year interval up to and including any screen-detected diagnoses in the following screening round. We observed a risk reduction of 66% for mortality from any cervical cancer across all ages but little or no effect for screening in the ages of 25–39. The screening histories showed that most of the remaining mortality (61%) arose in those without invitation to the screening program in the 5-year preceding diagnosis, mainly in the ages after last program invitation. The nonattenders formed the second largest group. A maximum additional reduction of 24% in mortality is thus theoretically achievable with perfect attendance. Mortality among attenders accounted for 14% of the cases, most of these were screening test sensitivity failures and only one was a work-up failure (in this case a failure to comply with referral to colposcopy).
Age-dependent effectiveness of screening has been documented in incidence-based case-control evaluations in the past few years.2,3,11 This study indicates that also when using mortality as outcome, screening in younger ages (below 40) has less impact than in older women. Because of a small number of deaths among invited women, it was not possible to estimate ORs for the associations of screening participation in the individual 5-year age categories. In particular, there were only two fatal cases in women invited for screening at the age of 25–29, both of which were nonattenders to the program. Historically, mortality rates from cervical cancer have decreased most in older women (age 40 or more at death), but there has also been a consistent decrease in the death rate in women below that age.12 Because of the small number of deaths, and the wide use of screening services outside the program also before the current study period, the study could not evaluate adequately the screening impacts in the youngest age groups.
The oldest age category in this study included invitations at age 65. Women of this age are invited by only a limited and declining number of municipalities as it is currently not recommended in the national guidelines. The effect of screening in the ages of 55–69 appeared equally effective as screening in the ages of 40–54. We attempted to estimate the value of screening specifically at the age of 65 by defining outcome as cervical cancer deaths with diagnosis at any point after 65 and estimating ORs for participation at last screen at 55, 60, or 65. The numbers were too small for statistically significant corrected ORs but the results were compatible with an increased risk reduction of last invitation at a later age. A similar additional effect of program participation at the age of 65 on the cancer risk was suggested in our recent evaluation of screening effectiveness using incidence as outcome.2
The duration of the protective effect was investigated by defining exposure as participation in the screening round (5 years) before the index round. There was a significant reduction in the mortality risk due to cervical cancer of 52% across all age groups. However, the age groups behaved differently with an absence of effect for those aged 25–39 at invitation, a small decrease in effect compared to index round participation for those aged 40–54 and a high effect similar to that in the index round for those 55 or more at invitation. These results are in line with a shorter duration of the protective effect of screening in younger ages, proposed previously in evaluations of other screening programmes.11,13 Also, the results indicate a long-lasting protective effect of screening in the older age group aged 55–69 at invitation. A decrease in cervical cancer mortality in Finland has been observed in the ages considerably above the last invitation for screening. Mortality has declined in the age group 70–79 from close to 20 per 100,000 women per year in the beginning of the 1980s to 5 per 100,000 in 2008. Similarly, women aged 80–89 have experienced a decline in mortality, beginning some years later than for the previous age group, from 24 per 100,000 in 1990 to 10 per 100,000 in 2008.12 Some of this decline may be due to improvements in clinical management and risk factors but it is likely that at least some of the observed reduction in mortality at these ages is due to a long-lasting protective effect of past participation in the screening program.
No significant reduction in mortality risk due to adenocarcinoma was observed. Small effects of screening on the incidence of adenocarcinoma have been demonstrated previously, and a similar or higher effect could be expected with mortality as outcome. Although non-significant, our estimate is compatible with the effect of around 30% found in recently published papers with incidence as outcome.2,14
It has been proposed that the duration of a preclinical phase of invasive carcinoma has to be excluded from the window of exposure when defining screening histories for subjects in a case-control evaluation.15 The duration of this phase is difficult to ascertain either collectively or individually and the exclusion of a period may bring other biases. However, in this study, there was a similar effect of screening when the exposure mapped to 5–10 years before diagnosis which excludes a preclinical phase of up to 5 years.
The strengths of the Finnish setting for screening evaluation include comprehensive nation-wide registers of program screening tests and cancers available for individual-level linkage by means of the unique personal identifier, 5-yearly personal invitations to all women in the target age-groups and a well-established screening program. The main limitation is the wide-spread opportunistic screening with no centralized register available. The magnitude of the opportunistic screening activity has been estimated to close to 60% of the total screening activity in Finland.16 Approximately half of those attending invitational screening have also been opportunistically screened in a 5-year interval. This proportion is fairly stable over the invitational ages. However, there is some indication that young women that do not attend program screening are more actively screened by opportunistic tests than women towards the upper age limit of the invitational program at 60 or 65. There is evidence that the effectiveness of opportunistic screening tests is lower than that of program screens, possibly because of weaker quality assurance and even stronger selection of women at low risk of cancer.17 Still, the impact of these tests is a subject that needs further evaluation but as there are only small differences in their distribution between ages 30 and 60, they cannot entirely explain the age-dependent impact of program screens in the population.
As a sensitivity analysis of the impacts of opportunistic screening activity on the level of screening effect observed, we randomly assigned attender status to 20% of nonattenders and recalculated self-selection corrected ORs for the association of attendance and cervical cancer mortality. Under these circumstances the association was somewhat diluted with an OR over all invitational ages of 0.50 (CI 0.35–0.70). For ages 25–39 the OR was 0.77 (CI 0.37–1.62), for 40–54 0.46 (CI 0.27–0.76) and for ages 55–69 0.49 (CI 0.27–0.86). In reality the use of smears outside of the program would have been more common, but the population-based effects of these opportunistic smears in preventing cervical cancer mortality smaller compared to programmatic smears. Possible differences in the risk in women that participated in the program only, or participated in both program and opportunistic screening, could not be contrasted with those not screened at all in the above calculation. Also, it is unlikely that opportunistic screening has been equally common among cases and controls. When attender status was assigned to 20% of nonattender controls only, the apparent effect of screening participation slightly increased. In this situation the overall corrected OR was 0.26 (CI 0.18–0.37). Stratified by age groups the OR was 0.54 (CI 0.24–1.16) for the invitational ages 25–39, 0.24 (CI 0.14–0.41) for ages 40–54 and 0.22 (CI 0.12–0.40) for ages 55–69. In essence, these two scenarios did not significantly change the estimates of effect.
To conclude, we have shown that the remaining cervical cancer mortality in Finland is mainly due to cancers developing in ages above the target ages for the invitational program and among those not responding to screening invitations. Cases diagnosed within 5 years of a program screen constituted a smaller proportion (14%) and the management of screen-positives appeared excellent in this evaluation. There are very few mortality cases with diagnosis before the first recommended program invitation at 30 and hence it proved impossible to evaluate the effectiveness specifically in this age-group. However, with mortality as outcome, screening effectiveness was found to be strongly correlated with age. The results strongly suggest, but do not conclusively prove, that the addition of a screening round at 65 would yield additional long-term benefits during the years that see the largest frequency of cervical cancer mortality.
- 1IARC. IARC handbooks of cancer prevention: cervix cancer screening, vol. 10. Lyon: IARC Press, 2005. 302.
- 12NORDCAN: cancer incidence, mortality, prevalence and survival in the nordic countries, Version 5.1 (March 2012): association of the nordic cancer registries. Danish Cancer Society., , , , , , , , , .
- 16Terveyden ja hyvinvoinnin laitos. Terveyden ja hyvinvoinnin laitoksen asettaman papilloomavirustautien torjuntatyöryhmän selvitys [Report of the study by the Expert Group on Papillomavirus Disease Prevention appointed by the National Institute for Health and Welfare]. Available at: http://www.thl.fi/thl-client/pdfs/94d6f45d-22e1-4b53-b615-2eea48d90e1c.