The burden of cervical cancer varies considerably in the European Union (EU). In this article, we describe trends in incidence of and mortality from this cancer in the two most affected areas: the Baltic countries (Estonia, Latvia and Lithuania) and Southeast Europe (Bulgaria and Romania). Incidence data were obtained from the national cancer registries. Data on population and number of deaths from uterine cancers were extracted from the World Health Organization mortality database. Mortality rates were corrected for inaccuracies in the death certification of not otherwise specified uterine cancer. Joinpoint regression was used to study the annual variation of corrected and standardized incidence and mortality rates. Changes were assessed by calendar period and age group, whereas the evolution by birth cohort was synthesized by computing standardized cohort incidence/mortality ratios. Joinpoint regression revealed rising trends of incidence (in Lithuania, Bulgaria and Romania) and of mortality (in Latvia, Lithuania, Bulgaria and Romania). In Estonia, rates were rather stable. Women born between 1940 and 1960 were at continuously increasing risk of both incidence of and mortality from cervical cancer. Although some quality issues in the registration of cancer and causes of death cannot be ignored, the trends indicate increased exposure to human papillomavirus infection and absence of effective screening programs. Rising trends of cervical cancer in the most affected EU member states reveal a worrying pattern that warrants urgent preventive actions.
Recent estimates of the burden of cervical cancer in Europe revealed large variations in incidence and mortality rates between countries and regions.1 The world-age standardized incidence of invasive cervical cancer was estimated to be 10 for 2004 (expressed per 100,000 women-years) in the earlier 15 member states of the European Union (EU), situated in West and South Europe, but was 17 among the ten new member states predominantly situated in Central and Eastern Europe, which joined the EU in 2004.2 Moreover, in Bulgaria and Romania, the two newest member states that acceded to the EU in 2007, rates were still higher (age-standardized incidence in 2004): 20 and 22 per 100,000, respectively. The incidence of and mortality from cervical cancer in Romania was ∼5 and 12 times higher compared to Finland, the country in Europe with lowest cervical cancer burden at present. In Eastern Europe, cervical cancer is now the gynecological cancer associated with the highest incidence and mortality.3
We assessed trends of cervical cancer incidence and mortality in the two regions of the EU with the highest burden of cervical cancer: the Baltic countries (Estonia, Latvia and Lithuania) and two countries in Southeast Europe (Bulgaria and Romania). These five countries currently receive support from the European Commission through the EUROCHIP-3 Network to assess the situation and to increase adherence to organized cervical screening in accordance to European guidelines.4–6
Studying trends of cervical cancer mortality is complex because of inaccuracies in the certification of deaths causes,7, 8 because deaths from uterine cancer are often not specified whether the cancer origin is cervix or corpus uteri. We therefore used reallocation methods for uterine cancers with unspecified origin and discuss the corrected trends in terms of birth cohort effects, alongside the impact of screening and treatment of invasive cervical cancer.
Material and Methods
Source of data on mortality from cervical cancer
Data on the number of deaths from uterine cancers and the size of the female population, aggregated by calendar year, 5-year age group (except for the oldest women categorized as ≥85 years) and country, were extracted from the World Health Organization (WHO) mortality database (http://www.who.int/whosis/mort/download/en/) for the five member states of the EU, with the highest burden of cervical cancer: the three Baltic countries (Estonia, Latvia, Lithuania) and two countries in Southeast Europe (Bulgaria and Romania).1, 2
Among the cancers originating from the uterus (UT), the following anatomical subsites were distinguished: cervix uteri cancer (CVX), corpus uteri cancer (CRP), cancer from the uterus not otherwise specified (NOS) and some other cancers from the uterus (OTH) that are not CRP, CVX or NOS. Different codes for cause of death were used throughout successive ICD (International Codification of Diseases) editions (see Table 1). In the 8th ICD edition, corpus and uterus NOS cancers were combined (ICD-8 = 182). In addition, for several countries, other non-ICD codes were used in the WHO Mortality Database, where corpus and uterus NOS cancers were combined (CRPNOS) (last two columns of Table 1).
Table 1. ICD/WHO1 codes used to identify cancers at the different sites of the uterus
Source of data on cervical cancer incidence
The cancer registries of Estonia, Latvia, Lithuania, Bulgaria and Romania provided data files containing the number of newly diagnosed cases of cervix uteri cancer, corpus uteri cancer and uterus cancer not otherwise specified, by 5-year age group for available calendar years.
Reallocation rules to estimate the number of deaths from cervical cancer
We estimated the number of deaths from cervix uteri cancer (corCVX) from the number of deaths certified as originating from cancer of the uterine cervix (CVX) and other parts of the uterus (CRP, NOS, CRPNOS or CRPNOSOTH), using three different reallocation rules.9
According to Loos et al.,8 when the proportion of NOS of all uterus cancer was less than 25%, adjustments can be based using allocation Rule 1, assuming that the NOS death certification was allocated at random:
where the indices i and j correspond with age group and year at death, respectively. Only Lithuania (between 1993 and 2004) presented data allowing application of Rule 1.
Allocation Rule 2 was applied for areas where reallocation Rule 1 was applicable for certain periods (where pNOS < 25%) but not for the span of years where the proportion of NOS > 25% or because NOS was not available as a separate group. The corrected age-specific proportions among uterus cancers (pcorCVXij = corCVXij/UTij [UTij being the sum of the number of deaths from all parts of the uterus]) from periods corrected using Rule 1 were then imputed to estimate corCVXij of the other periods.9 Rule 2 could be applied to correct data for Lithuania before 1993.
In the reallocation Rule 3, reference areas were used as template to estimate the number of cervix cancers from total number of uterus cancers in nonreference areas. Corrected proportions of cervix cancer from the template countries (pcorCVXijt) were used for reallocation in the other countries: corCVXijc = UTijc * pcorCVXijt, where c refers to a country with low-quality data and t to its respective template (Rule 3). We accepted that Lithuania could be considered as template country to correct data from the other two Baltic countries. Corrected data from Hungary, also adjusted according to Rules 1 and 2 in an earlier study,9 were considered as sufficiently representative for Bulgaria and Romania.
Reallocation rules to estimate the number of new cases of cervical cancer
Because the proportion of uterus NOS cancer cases was always <25%, Rule 1 could be applied to estimate the annual number of incident cervical cancer cases for all the five countries.
Age standardization was performed using the world standard population.10 Joinpoint regression was used to analyze time trends of the standardized corrected incidence or mortality rates, as a linear function of year of cancer incidence or death.11 Joinpoint regression identifies periods with distinct linear slopes that can be separated by joinpoints, where the slope of the trends changes significantly.12, 13 For each linear segment, the average annual percentage of change (APC) and corresponding 95% confidence intervals were calculated. Rates that change at a constant percentage over time are presented as an exponential curve in a plot with a linear Y axis.
Age-specific trends were analyzed by 5-year calendar period and by 10-year birth cohort. Five-year periods were defined using years ending with zero or five as starting year. According to availability of data, first and last periods did not always span 5 years. Birth cohorts (k = calendar period − age) were identified by the median year within each category.14
Variations by birth cohort were summarized by the standardized cohort mortality (SCMR) or incidence ratio (SCIR). The SCMR represents the relative risk of a certain cohort of dying from cervical cancer compared to a reference cohort,15, 16i.e., the ratio of the number of observed deaths in a given cohort, k, over the number of deaths expected, when the age-specific mortality rates of the reference cohort are applied to cohortk. The 1940 cohort was chosen as reference because of its central position in the available cohorts for all the five countries.
We used software developed by the National Cancer Institute for Jointpoint regression.11 All other statistical analyses were performed with STATA (version 10.1, StataCorp, College Station, TX).
Age-standardized incidence and mortality
The observed age-standardized incidence and mortality rates as well as the fitted jointpoint trends are plotted in Figure 1, and the slopes and joinpoints are described in Table 2.
Table 2. Joinpoints, years where slopes of linear trends changed (including 95% CI around this year) and magnitude of the annual percentage of change (APC) in each linear segment and its 95% CI
In Estonia, the incidence was initially decreasing but became horizontal since the early 1980s (APC not statistically significantly different from zero). A steep drop in incidence in the earliest available years (1980–82) was noted in Latvia. We considered this dramatic decrease as spurious, and, therefore, we excluded the first 3 years from the joinpoint regression. Since 1983, the Latvian incidence was stable. In Lithuania, the incidence rate increased between 1992 and 2004 (APC = 3.4; 95% CI: 2.1–4.6). The incidence rates, expressed per 100,000 women-years, observed in the latest available years, were as follows: 15.4 in Estonia (2006), 12.3 in Latvia (2004) and 18.4 in Lithuania (2007).
The mortality rates were stable in Estonia but increased at a constant rate in Latvia (APC = 1.5, 95% CI: 0.9–2.1) and Lithuania (APC = 1.3, 95% CI: 0.9–1.7). The corrected mortality rates, observed in 2004, were as follows: 6.1 in Estonia, 7.4 in Latvia and 9.0 in Lithuania.
In Bulgaria and Romania, age-standardized incidence trends increased monotonically with APCs of 3.1 (95% CI: 2.8–3.4) and 1.7 (95% CI: 0.6–2.8), respectively. The latest rates were 21.4 in Bulgaria (2006) and 21.3 in Romania (2004).
For mortality, trends that were statistically significantly rising over a limited time period were observed in both countries: Bulgaria (APC = 3.0, 95% CI: 1.5–4.7, between 1981 and 1989) and in Romania (APC = 0.4, 95% CI: 0.2–0.6, since 1980). The most recent mortality rates in 2004 were as follows: 7.2 in Bulgaria and 11.1 in Romania.
Age-specific trends by period
Age-specific (by 10-year age groups) incidence and mortality trends are plotted against calendar period in Figure 2.
In the Baltic countries, before 1990, women aged 50–69 years displayed a decreasing or horizontal incidence trend, which tended to raise or become horizontal thereafter. No important incidence changes occurred among women of 70 or older. Women younger than 50 years showed a raising incidence trend, which even accelerated after 1990.
Mortality rates in women older than 60 were stable or slightly declining, at the exception of the oldest groups (>80 years) in Latvia and Lithuania where mortality increased. In Estonia and Latvia, mortality rates were increasing in the age groups 40–59, with also an increase in the age group 30–39 in Latvia after 1990. In Lithuania, mortality raised in all age groups in the range 30–59. After 1985, mortality rates raised in middle-aged women, in particular in the age groups 40–49 in Estonia, 30–49 in Latvia and 25–49 in Lithuania. Mortality was low in the youngest groups (<30 years) where no changes could be discerned.
In Bulgaria and Romania, incidence rates generally raised in all age groups older than 30 with a more steep increase in the latter country after 1940. For both countries, mortality data were available over longer periods (since 1959 for Romania and 1964 for Bulgaria). The mortality rates in these two countries decreased slightly for women aged 55 or older after 1985. For the age groups 20–59, mortality increased after 1985, but the trend was interrupted for the youngest age group (20–29). In Bulgaria, mortality trends were rather stable with a tendency to raise in age groups 30–59, in particular in the later periods. In Romania, mortality rates decreased in women of 40–79 years of age. However, mortality rates started increasing in age group 30–39 after 1975, in age group 40–49 after 1985 and in age group 50–59 after 1995.
Standardized cohort incidence and mortality ratios are plotted in Figure 3, and joinpoints are described in Table 3.
Table 3. Joinpoints, birth cohorts where slopes of linear trends of the standardized cohort incidence or mortality ratio changed (including 95% CI around this cohort) and magnitude of the annual percentage of change (APC) in each linear segment and its 95% CI
Statistically significant V-shaped cohort effects were observed (with, respectively, decreasing and increasing risks for women born before and after the 1940–1945 cohorts) for incidence and mortality for all the three Baltic countries.
Bulgaria and Romania displayed generally rising cohort incidence ratios with a mild joinpoint for Romania around 1940. For women born after 1965, the SCIR stabilized in Bulgaria but tended to increase further in Romania. Because of more available mortality data for Bulgaria and Romania, we can appreciate effects for cohorts born in the first decades of the 20th century with stable or increasing mortality risk in the earliest decades that decreased after 1925 until 1940–1945. As in the Baltic countries, the SCMR started to rise again among cohorts born after 1945.
The main etiologic factor for cervical cancer is persistent infection with sexually transmittable high-risk human papillomaviruses (HPVs).17 By well-organized screening and treatment of screen-detected high-grade cervical intraepithelial neoplasia, invasive cancer can be avoided.18 Therefore, trends in incidence of cervical cancer largely reflect coverage and quality of screening as well as changes in exposure to risk factors that are mainly related to sexual habits of successive cohorts.19, 20 We will subsequently discuss the elements that may have driven the trends in the five studied countries.
An important question is whether the applied correction for inaccuracies in the certification of death causes allows the study of the true rates of cervical cancer mortality. For Lithuania, the proportion of uterus NOS deaths was small (<25%), and therefore corrected rates can be considered as reliable. Even if the assumption of random allocation (applied in Rule 1) was incorrect, the error would be limited. The assumption that the Lithuanian proportions are applicable to those of Estonia and Latvia looks plausible given the common background risk and history of preventive health care.9 However, the application of proportions from Hungary to adjust data from Bulgaria and Romania could be considered as problematic. To find more reliable solutions to correct for NOS and CRPNOS cancer deaths, we propose further research, involving linkages between mortality and cancer registries.21–23 We tried alternative approaches to address inaccuracies in death cause certification, such as general application of Rule 1 (used by IARC to estimate the world-wide cancer burden),3 and restriction of total uterus mortality to women younger than 45 years (where nearly all uterus cancer originates from the cervix uteri).24 All these solutions resulted in trends of similar shape as those presented in this article (data not shown).
Incidence data received from the five national cancer registries suffered less from certification problems because the proportion of NOS cases among all uterine cancers was small. However, other biases may have intervened. In Latvia for instance, the abrupt drop in incidence between 1980 and 1983 (Fig. 1), followed by a stable incidence but increasing mortality trend indicating worsening survival, looks spurious. Exclusion of carcinoma in situ cases, abrupt in the first years and more gradual thereafter, could have hidden a rise in incidence of invasive cervical cancer. However, an alternative interpretation could be that Latvian incidence data suffered from underregistration in the years after 1983. The higher mortality/incidence ratio after 2000 (0.64) compared to the 1980s (0.48) indicating (improbable) worsening survival over time seems to corroborate the hypothesis of underreporting. Other possible data quality issues are the increased exhaustivity of cancer registration in Bulgaria and Romania and a certain degree of underestimation of cervical cancer mortality in Bulgaria. In Lithuania, since 2006, personalized information from death certificates is unavailable for the cancer registry, and death-certificated-only cases are not added to the incident cases anymore. This change in registration practice could be responsible for the observed incidence decrease in years 2006 and 2007.
In general, the risk of developing cervical cancer or dying from it decreased for women born between the two world wars, whereas cohorts born after 1940–45 expressed increasing risks. These cohort effects were also observed in many other industrialized countries.9 The decreasing risk before 1940 may be due to poorly understood etiological (co-) factors, linked to improved social conditions and access to health care.25 The greater SCIR and SCMR in the cohorts born between the 1940s and 1960s are most plausibly explained by changes in sexual behavior resulting in higher rates of HPV infection, which may be enhanced by increased frequency of smoking and oral contraception.26–29 It is also possible that some other factors such as early diagnosis of invasive cancer among younger women due to increased access to gynecological care may be responsible for cohort effects observed in the deaths rates. A particular finding was the rise in the SCIR in Bulgaria and Romania for women born before 1940. Whether this phenomenon should be ascribed to a continuously augmenting trend of HPV infection or to artifacts in early cancer registration is unknown. Biobank-based research using archived Pap smears or control biopsies could provide pertinent information to answer this question.30
In most West-European countries with either well-organized screening programs or widespread opportunistic screening, it was shown that the rising cohort effect (also observed for women born after 1940) was counterbalanced by a protective period effect. This period effect was strongly correlated with screening coverage.20, 31–35
The increasing trends of cervical cancer incidence and/or mortality observed in all the five countries are most plausibly explained by the absence of screening programs or by the poor quality and coverage of opportunistic screening practice since the last decades.
Recently, national cytology-based screening programs were initiated in the three Baltic countries, and a regional program was set up in the province of Cluj (Romania), whereas in Bulgaria, plans for organized screening are not started yet.18, 36, 37 All these programs suffer of understaffing, insufficient resources and management capacity and reach less than 20% of the target population. Obviously, more time, continued efforts and comprehensive EU support will be needed to bend cervical cancer trends downward.
Among the five countries studied, only Estonia and Lithuania are included in international survival comparisons.38–40 The average European 5-year age-standardized relative survival among cervical cancer patients diagnosed from 1990 to 1994 was 63%, whereas 53% for Estonia.40 The trend of the 5-year survival revealed a slow but steady increase of about 2% per year among cancer patients diagnosed in the period 1983–94 in Europe.38 No improvement was noted in the areas where survival was lowest (Central/Eastern Europe and UK). A more recent period-based analysis, over the years 2000–04, showed lowest survival rates for Lithuania (52%) and Poland (53%) without significant improvement.39 Low 5-year survival was also reported for patients with cervical cancer in Bulgaria for the period 1993–2002.41
Survival from cervical cancer is strongly determined by age and staging. Reduction of the case fatality can be expected by down staging through the expansion of screening and by improved treatment. Unfortunately, there is no systematic data currently available on the quality of cervical cancer treatment in Europe.
There is an elevated burden of cervical cancer in the three Baltic countries, Bulgaria and Romania. Moreover, incidence and mortality rates tend to increase or remain stable. Public health authorities should set-up well-organized cervical cancer prevention programs without delay as recommended by the European Council42, 43 according to the European Guidelines for Quality Assurance in Cervical Cancer Screening.6 It is particularly challenging for public health experts to define, in the future, how prophylactic HPV vaccination besides screening will contribute in tackling this preventable disease.