Factors Related to Inadequate Cervical Cancer Screening in Two Brazilian State Capitals

Fatores associados ao rastreamento inadequado do câncer cervical em duas capitais brasileiras ABSTRACT OBJECTIVE: To analyze factors associated with cervical cancer screening failure. METHODS: Population-based cross-sectional study with self-weighted two-stage cluster sampling conducted in the cities of Fortaleza (Northeastern Brazil) and Rio de Janeiro (Southeastern Brazil) in 2002. Subjects were women aged 25-59 years in the last three years prior to the study. Data were analyzed through Poisson regression using a hierarchical model. RESULTS: The proportion of women who did not undergo the Pap smear Higher prevalence ratios of cervical cancer screening failure in both cities were seen among women with low education and low per capita income, old age, unmarried, who never underwent mammography, clinical breast examination, and blood glucose and cholesterol level testing. Smokers also had lower screening rates compared to non-smoker women and this difference was only statistically signifi cant in Rio de Janeiro. CONCLUSIONS: The study fi ndings point to the need of intervention focusing particularly women in worse socioeconomic conditions and access to healthcare, old-aged and unmarried. Education activities must prioritize screening of asymptomatic women and early diagnosis for symptomatic women and access to adequate diagnostic methods and treatment should be provided. Cervical cancer accounts for about 15% of all cancers in women, and it is today the second most common one in women worldwide. In developing countries, it ranks fi rst among all cancers in women while in developed countries cervical cancer is the sixth most common type of cancer. The highest incidence rates of cervical cancer are seen in Latin America and the Caribbean, parts of Africa, South and East Asia, while in North America, Australia and North and West Europe, the rates are lower. a In Brazil, between 1991 and 2001, age standardized incidence rates ranged from 14.3 per 100,000 women in Salvador to 50.7 per 100,000 women in the Federal District. b In 2005, cervical cancer accounted for 6.7% of all cancer deaths in Brazil, the fourth most common cause of death. c RESUMO OBJETIVO: Analisar fatores associados à não-realização do exame de Papanicolaou. participantes foram mulheres de 25 a 59 anos de idade nos três anos anteriores à pesquisa, nos municípios de Fortaleza (CE) e Rio de Janeiro (RJ). Os dados foram analisados por regressão de Poisson por meio de modelo hierárquico. no Rio de Janeiro. As maiores razões de prevalência para a não-realização do exame nas duas localidades foram …

Cervical cancer accounts for about 15% of all cancers in women, and it is today the second most common one in women worldwide.In developing countries, it ranks fi rst among all cancers in women while in developed countries cervical cancer is the sixth most common type of cancer.The highest incidence rates of cervical cancer are seen in Latin America and the Caribbean, parts of Africa, South and East Asia, while in North America, Australia and North and West Europe, the rates are lower.a In Brazil, between 1991 and 2001, age standardized incidence rates ranged from 14.3 per 100,000 women in Salvador to 50.7 per 100,000 women in the Federal District.b In 2005, cervical cancer accounted for 6.7% of all cancer deaths in Brazil, the fourth most common cause of death.c RESUMO OBJETIVO: Analisar fatores associados à não-realização do exame de Papanicolaou.

DESCRITORES: Esfregaço
For WHO, a coverage of 80.0% of screening test among women aged 35 to 59 years old would cause an impact on morbidity and mortality indicators, which can be seen in four years after the implementation of early detection.In countries where effective screening programs are in place, 65.0% to 70.0% reduction in mortality can then be expected.
Despite the growing knowledge in this fi eld, a more effective approach for cervical cancer management continues to be screening using the Papanicolaou (Pap) These Brazilian studies pointed out common variables seen among women who did not undergo the Pap test.In 1987 Nascimento et al 5 reported that being between 15 and 24 years of age, single, of low socioeconomic level, not having consulted a doctor in the year preceding the study, not having used oral contraceptives and not having performed breast self-examination in the previous year were factors associated with cervical cancer non-screening.Pinho et al 6 reported that the main reasons for not undergoing testing were: not having gynecological problems, being embarrassed or afraid of undergoing testing, and reported diffi culty in gaining access to a health care unit.In another study, Costa et al 3 concluded that the lower the social condition the less women would undergo cervical screening and that coverage was lower in older women.Quadros et al 8 showed that black and low income women were proportionally less likely to undergo screening.Finally, Cezar et al 2 reported that the highest prevalence ratios for non-screening were found among black or mixed skin color women, with less than 20 years of age, with low family income and low education, living alone and who had fi rst given birth after 25 years of age.
As more in-depth information on cervical cancer nonscreening is needed, the present study aimed to analyze factors associated with cervical cancer screening failure in two large Brazilian cities.

METHODS
The present study is a subset of a larger survey about risk behaviors and reported morbidity of noncommunicable diseases and damages carried out in 2002 and 2005 in 18 Brazilian cities.It is a population-based cross-sectional study and its details are described elsewhere.a Data on women aged 25-59 years, living in two Brazilian state capitals (Rio de Janeiro and Fortaleza) were included.The cities were chosen considering the deep socioeconomic differences between them: Rio de Janeiro is located in the most developed region (Southeast) of Brazil, has low illiteracy rate (4.4%), low proportion of poor people (15.4%), and good health insurance coverage (50.9%);Fortaleza is in the Northeastern region, has high illiteracy rate (11.2%), high proportion of poor people (42.4%), and inadequate health insurance coverage (25.7%).b,c In order to determine the sample size for each municipality, a 95% confi dence interval, a relative precision of 10.0% and the ability to adequately obtain a 27.0% prevalence -which is the estimate prevalence for smoking, one of the most important epidemiological risk factors for non-communicable diseases -were taken into consideration.Self-weighted two-stage cluster sampling was used in each municipality.Census tracts were systematically selected, proportionally to the number of private households.In each census tract, all households were recorded, classifi ed as occupied, closed, empty, occasional use and non-residential.Then they were selected in a systematic way.A total of 1,021 households in Fortaleza, and 1,716 in Rio de Janeiro were visited, and 760 and 987 women between 25-59 years old were included in the sample, respectively.All non-interviewed households and 30% sample of the interviewed ones were revisited by trained supervisors to ensure coverage and quality of information.It was not possible to obtain information in 4.6% and 14.0% of the households in Rio de Janeiro and Fortaleza, respectively, because people were not present, refused to participate in the study or other reasons.
A household questionnaire was used to collect sociodemographic information, and other information was obtained from a specifi c questionnaire directed to adult population.Questionnaires were tested before being applied.
The following predictive variables were analyzed: age (25 to 34 years; 35 to 49 years; and 50 to 59 years), schooling (zero to three years; four to seven years; eight to ten years; and 11 years and more), marital status (unmarried; and married or living with a partner), per capita income (≤3 monthly minimum wages [MMWs]; and >3 MMWs), cholesterol and blood glucose levels (if any), mammography and clinical breast examination (if any), and smoking habits.Marital status and schooling were defi ned according to the Instituto Brasileiro de Geografi a e Estatística (Brazilian Institute of Geography and Statistics, IBGE) criteria.a Smoking habits were defi ned according to CDC criteria, b i.e., cigarette smokers were defi ned as people who have smoked 100 or more cigarettes in their lifetime.
A software in Delphi, using the Oracle platform for data input, was developed.For data analysis Stata 8.0 was used as the "survey" module provides data analysis of epidemiological surveys with complex sampling strategies.Proportion and association estimates, and their respective 95% confi dence intervals (95%CI), were calculated using the cluster sampling plan.Statistical signifi cance was tested through Pearson's chi-square test, which, based on the sampling strategy, was converted to an F-test (Fisher-Snedocor).c To measure the rate of association between the predictive variables and the outcome, crude and adjusted prevalence ratios (controlled for possible interactions and confounders), and their respective confi dence intervals were calculated through the Poisson regression model, which uses the prevalence ratio as the effect measure.It allows more accurate interpretations since the odds ratio overestimates the magnitude in highly prevalent conditions. 1,11Some variables were recategorized or changed, as follows: schooling, classifi ed as incomplete elementary school (zero to seven years of schooling), and complete elementary school and more (eight years of schooling or more); smoking habits, defi ned as smoker or non-smoker (non-smoker or former smoker); and access to blood glucose and cholesterol testing, categorized as "yes" when both tests were performed, and "no" when none or only one was performed.The signifi cance of each variable in the model was assessed through the Wald test.c In the analysis, p-values lower than 0.05 were considered statistically signifi cant.In multivariate analysis, variables were included following a previously determined hierarchical framework based on the literature review. 4 this model some variables were assumed to mediate their effects through other variables as well as directly.Variables were selected based on biological plausibility and association (p-value lower than 0.05) with outcome according to the bivariate analysis.The fi rst level included the variable age; the second level included the variables schooling, marital status, and per capita income; and the third level included blood glucose and cholesterol levels, mammography, clinical breast examination, and smoking habits.
The study was carried out in a way to protect the individual's privacy, thus ensuring anonymous and voluntary participation.All participants signed a free informed consent.The study was approved by the Research Ethics Committee of the Brazilian Institute of Cancer, and conducted in accordance with all recommendations of the National Health Council.

RESULTS
The characteristics of the population studied are described in Table 1.The proportion of women not undergoing the Pap test within the previous three years was 19.1% (95% CI: 16.1;22.1)and 16.5% (95% CI: 14.1;18.9) in Fortaleza and Rio de Janeiro, respectively (p=0.18).The lower the education and per capita income, the higher the proportion of screening failure in both cities (Table 2).Moreover, smokers, unmarried women, those who had never undergone a mammography, clinical breast examination, cholesterol and blood glucose testing underwent less Pap tests in both cities.Age was the sole variable included in the model which was not statistically associated with the outcome.
The analysis of gross prevalence ratios (Table 3) showed that all variables, except age in Fortaleza, were associated with screening failure in both cities and thus were included in the multivariate model.Plausible interactions were tested.These interactions were not incorporated into the fi nal model since prevalence ratios lost precision, probably due to the low rate of subset outcomes.Changes in effect were seen only in the third level.
Women with incomplete elementary school, unmarried, and with lower per capita income showed higher prevalence ratios of screening failure in both cities.The same was seen among women who had never undergone a mammography, clinical breast examination, and blood glucose and cholesterol testing.In Rio de Janeiro, smokers showed a statistically signifi cant difference of screening failure compared to other women.

DISCUSSION
The present study showed that women with low education, low per capita income, older age, unmarried, smoker, who had never undergone a mammography, clinical breast examination, or blood glucose and cholesterol testing were less screened for cervical cancer in both cities.Smokers showed higher prevalence of screening failure, but this association was only statistically signifi cant in Rio de Janeiro.The results found for schooling, per capita income, and marital status are consistent with other Brazilian studies.The same is seen for age and clinical breast examination.Some studies have found lower rates of cervical cancer screening among younger women (15-24 years old, less than 20 years old, 15-29 years old, 25-29 years old), while others have found them among older age groups (50-69 years old, 60-69 years old, and 55-59 years old). 4Nascimento et al, 5 Costa et al, 3 and Quadros et al 8 found high rates of cervical cancer screening failure among women who had never undergone breast self-examination and did not attend medical visits in the year prior to the study, which was not evaluated in the present study.
One fi nding, to date not reported in other Brazilian studies, was that smokers underwent less Pap tests compared to other women.It may be explained by the fact that smokers have less healthy habits such as engaging in physical activities, healthy eating, and low alcohol consumption that can also be associated with failure of undergoing preventive examinations such as the Pap test.However, this behavioral indicator needs to be further investigated as a barrier to cervical cancer screening. 9other point must be highlighted: the coverage rates of Pap test in both cities were similar and did not show a statistically signifi cant difference.In regard to the predictive variables, both cities studied had distinct distributions: Rio de Janeiro had older women with higher income and education, higher coverage of mammography, blood glucose and cholesterol testing, which may be explained by high insurance coverage in the city.In 2005, health insurance coverage (proportion of benefi ciaries in the whole population) in Rio de Janeiro was 56.3% while it was 38.4% in Fortaleza.a In spite of these differences, factors associated with screening failure were similar in both cities, except for smoking, which showed a statistically signifi cant association only in Rio de Janeiro.This difference may be due to the sample size, biological differences among women and specifi c characteristics of cigarette consumption (type and number of cigarettes smoked, and age when they started smoking, which were not assessed in this study).Other Brazilian studies did not consider the potential interaction and confounders, which may have infl uenced the results.
As Cesar et al 2 and Pinho et al 6  mind that information based on women's self-reporting tend to overestimate the rate they undergo examinations, and to underestimate the time from the most recent examination.On the other hand, recall of examinations occurring in the previous three years has been reported to be more accurate than those occurring more than three years before the interview.Another aspect is that studies that deal with this subject assume that all women know what a Pap test is, which may not be true.For instance, women may confuse the Pap test with a gynecological consultation, and vice-versa.The present study sought to minimize this misunderstanding through the questionnaire's previous testing and interviewers '   training, besides including a note in the questionnaire to explain what a Pap test is.It is not possible to completely avoid such limitations through a cross-sectional design, although this fact does not invalidate the useful information provided for planning health actions.
An important methodological aspect that should be highlighted is regarding the sample size.The study used a 5.0% signifi cance level, an estimated prevalence of about 30.0%, and a relative precision of 10.0%.For this reason, lower prevalences, such as those included in this study, may have lost precision, generating wider confi dence intervals.

Table 2 .
Characteristics of women who did not undergo Pap test in the previous three years.Fortaleza and Rio de Janeiro,Northeast and Southeast Brazil, 2002-2005.
example, age at sexual initiation or number of partners), and utilization of health services (for example, hospitalizations or number of medical consultations in the last year).In addition, women who had undergone hysterectomy should not have been included in the survey since they do not need to undergo cervical cancer screening.But this data was not available for the entire formation about two Brazilian state capitals which was not studied before; the adjusted sampling model; and quality control of data collection and input.It should also be highlighted the modeling process, which used a hierarchical method and Poisson regression.This hierarchical method, based on a theoretical model, aimed to avoid that only statistical criteria would determine the

Table 3 .
Adjusted Poisson multiple regression analysis of predictive variables for cervical cancer screening failure in the previous three years.Fortaleza and Rio de Janeiro, Northeast and Southeast Brazil, 2002-2005.APR= adjusted prevalence rate; MMWs = monthly minimum wages.Adjusted models: Level 1: age; Level 2: age + schooling + marital status + per capita income; Level 3: age + schooling + marital status + per capita income + smoking habits + mammography + clinical breast examination + blood glucose and cholesterol levels