SILVEIRA, MARIÂNGELA F. MD, MS*; BÉRIA, JORGE U. MD, PhD†‡; HORTA, BERNARDO L. MD, PhD†; TOMASI, ELAINE MS†; VICTORA, CESAR G. MD, PhD†
WOMEN ACCOUNT FOR a growing proportion of HIV/AIDS cases: according to UNAIDS, at the end of 1999 they accounted for 46% of the global total of 34.3 million people living with HIV. 1 Women are more susceptible to sexually transmitted diseases (STDs) than men because of both biological and social characteristics. 2
In Brazil, women account for one-third of the patients with AIDS. 3 An analysis of reported AIDS cases suggests that the disease is more common among women aged less than 30 years and those with little schooling. 2,3 A thorough literature search for the period 1980 to 2000, however, did not yield any Brazilian studies specifically designed to investigate the factors associated with risk behaviors among women. Because findings from the international literature are not necessarily applicable to Brazilian women, local studies are needed to identify women with a higher risk of infection, in order to help target effective interventions.
The current study was aimed at identifying socioeconomic, demographic, and behavioral factors associated with risk behaviors for STD/AIDS in a population-based sample of urban women in the southern Brazil city of Pelotas.
The target population for the cross-sectional survey included all women aged 15 to 49 years living in the urban area of Pelotas (population, 323,000). The city is located in the extreme south of Brazil, near the Uruguay border, and the standard of living is relatively high in comparison with most of the country but typical of the southern region.
A systematic sample of 48 of the city's 281 census tracts was selected. In each tract, a random starting point was chosen and every third household was selected thereafter, until 44 households were included in each tract. An attempt was made to interview all women living in these 2112 households.
Demographic data indicated that each household would include an average of 0.7 women in the age range of interest; approximately 1500 women were thus expected to be included in the sample. This would allow detecting any association between a characteristic affecting 10% of the women and a risk outcome present in 30% of them, with α = 5%, statistical power of 95%, a risk ratio of 1.6, and a design effect of 1.4.
Thirty female interviewers, all with a secondary or higher education, were selected and trained. A pilot study was carried out, and data collection lasted from October 1999 to January 2000.
The questionnaire included three parts: a household module with socioeconomic data, an individual module including interviewer-administered questions, and a confidential, self-administered module. The latter was identified only by a serial number and was sealed by the respondent herself in an envelope. It was explained to her that this envelope would be opened only by the study coordinator, and that it would not be possible to identify individual women by name when this information was merged with the rest of the questionnaire. The University's ethics committee approved the project, and informed consent was obtained from all participants.
The following variables were collected via the open questionnaire: age (in full years); skin color (according to observation by the interviewer, classified as white or nonwhite); marital status (single, married, in consensual union, widowed, separated, or divorced); family income and personal income in the previous months (calculated in terms of minimum wages); per capita income (total family income, divided by the number of members); schooling (number of years completed); attendance at religious practices (yes/no); smoking (yes/no); and regular physical exercise in the previous 3 months (yes/no).
The confidential questionnaire included questions about the age at which sexual activity (vaginal, oral, or anal sex) began and the number of partners in the previous 3 months. For the most recent intercourse, information was collected on use of condom (yes/no); use of alcohol by the woman (yes/no) or her partner (yes/no); use of drugs by the woman (yes/no) or her partner (yes/no); and practice of anal sex (yes/no). A risk score was calculated by assigning one point to each of the above six risk behaviors. For age at onset of sexual activities, <18 years was considered as high risk, 4 as was having ≥2 sex partners; other variables were coded as 1 for yes and 0 for no. This variable became the outcome for subsequent analyses.
Quality control included the repetition of 5% of all interviews by trained supervisors. All questionnaire data were entered twice by different clerks, and range and consistency checks were carried out with Epi-Info software version 6.0 (Centers for Disease Control and Prevention, Atlanta, GA). Data analysis was carried out with SPSS 8.0 software (SPSS, Chicago, IL) and Stata 6 software (Stata, College Station, TX).
Among independent variables, family income accounted for the highest proportion of missing values: 2.1%. Missing values for all independent variables were recoded to the modal category. The risk score, however, accounted for 11% of missing values. The following approaches were used to reduce this proportion. A linear regression equation including age, schooling, and income showed a good fit for predicting age at first intercourse; this equation was used to eliminate the 5.2% missing values for this variable. The number of partners was not available for 4.9% of the women, but it was not possible to obtain an equation with an appropriate fit; therefore, the final score for these women was multiplied by 1.2, the average effect of adding this variable to the scores of the 95.1% of women with available data. After such adjustments, 3.4% of the women still remained whose risk score values were missing because of lack of data for other component variables; these were excluded from the analysis.
Ordinal regression was used in the data analysis. This procedure does not require the arbitrary choice of a single cutoff, as does logistic regression. The proportional odds model was used. In this model, the odds ratio estimates the risk of moving to the next higher risk behavior category, per unit increase in the independent variable. 5 The raw risk score was tested for proportionality of odds—an underlying assumption of the ordinal regression model. Because it did not satisfy this assumption, it was recoded into five cate-gories (0; 0.01–1.0; 1.01–2.0; 2.01–3.0; >3.0), after which the assumption was met. All analyses were adjusted for clustering. 6
In the multivariate analyses, independent variables were introduced according to predetermined causality levels, starting with distal determinants and adding, in a stepwise fashion, intermediate and proximate determinants. 7 Causality levels were based on a conceptual framework (available upon request).
The first level included the income variables (per capita and women's), schooling, age, and skin color. The second level included marital status and religious practice, and the third level, smoking and physical activity. A backward elimination regression model was used. Variables from the first level with P < 0.2 were kept in the model and those in the second level were added to it; second-level variables with P > 0.2 were excluded; and, finally, variables from the third level were added according to the same criterion. A high P value was used to ensure that potential confounders were kept in the model, even if not statistically significant at the 0.05 level. 8 In the final presentation of results, variables were considered as significant if the P value was <0.05 in the first equation when introduced in the model.
A total of 2112 households were visited and their inhabitants listed according to age and sex categories. For closed households, approximate age and sex information was obtained from neighbors so that it would be possible to characterize nonresponders. We identified 1851 women, aged 15 to 49 years, of whom 65 (3.5%) either refused or could not be interviewed after at least three separate attempts. All 1543 women who reported having ever had intercourse were included in the study.
Table 1 shows the socioeconomic and demographic characteristics of the sample. The lower frequency of adolescents is because many of those interviewed had not started their sex life. Four of every five women were white; 61% lived with a husband or partner; 1 in 5 had a per capita income of ≤0.5 the minimum wage; 41% did not have a personal income; more than half had <9 years of schooling; and 55% did not practice a religion.
In relation to STD/AIDS risk factors, 72% of the women reported that they had not used a condom during their most recent intercourse; approximately half of them had their first intercourse before 18 years of age; 14% of their partners and 7% of the women used alcohol or drugs before the most recent intercourse; 7% had had ≥2 partners in the previous 3 months; and 3% reported having anal sex during the most recent intercourse.
The distribution of the risk scores used in the ordinal regression is shown in Table 2. Almost 10% of the women reported no risk behaviors, 46% reported 1, and 3% reported >3.
Table 3 shows the crude and adjusted odds ratios for the ordinal regression. Of variables belonging to the first level of the postulated causal model, age and schooling remained significantly associated with the risk score, even after adjustment. For females aged 15 to 19 years in comparison with women aged 45 to 49 years, the odds ratio was 3.7. Schooling for <5 years was associated with a twofold increase in the odds ratio, relative to that for women who studied for ≥12 years. Skin color and income variables were no longer significant after adjustment for the remaining variables.
Of variables included in the second equation (Table 4), marital status became significant after adjustment for age and schooling, and religious practice had no effect. In the third level of determination (Table 4), regular physical exercise was no longer significant after adjustment, but smoking was associated with a higher risk in both the crude and adjusted analyses.
The population-based nature of the survey provides a representative estimate of the frequency of reported risk behaviors and of their main determinants in a middle-sized city in southern Brazil. The representativeness of the study is enhanced by the low frequency of nonrespondents (only 3.5%).
Possible limitations of the study include the difficulty in reporting sensitive behaviors, even through a confidential, self-applied questionnaire, and the fact that the interviews took place in the respondent's home, which may have affected adolescents living with their parents. Another limitation is that most behaviors were assessed for the most recent intercourse. The alternative, however, was to ask about longer periods of time with the possibility of inaccurate recall or of obtaining reports of ideal rather than actual behaviors. Last, the low educational level of some women may have affected the accuracy of the self-applied instrument. These possible biases could lead to an underestimation of the risk score. It is reassuring, however, that both adolescents and less-educated women reported higher frequencies of risk behaviors than other females.
Another possible limitation was the fact that all six behaviors were given equal weights in the risk score. This was because there were no objective criteria for weighting these behaviors. An attempt was made to use factor analysis for this purpose, but the results were not consistent. Further research is needed to arrive at a better understanding of risk behaviors and how these may be combined into a score.
Almost half of the women (47%) had their first intercourse before 18 years of age. Several other studies in Latin America and Brazil have yielded similar findings. 9,10 In the United States, early sexual initiation was associated with use of illegal drugs, a larger number of partners, and less condom use. 4 The high incidence of teenage pregnancies in Brazil—about a third of all women have their first child before the age of 20 years 9 —confirms that unprotected sex is common among adolescents.
Anal sex during their most recent intercourse was reported by 3% of the women. Studies in the United States and England showed frequencies of 9% and 7%, respectively, for the year prior to the survey, whereas a previous school-based Brazilian study showed a rate of 7% among girls aged 12 to 19 years for the most recent intercourse. 11,12 It is difficult to assess the reliability of this information because of its sensitive nature.
Condom use during the most recent intercourse was reported by 28% of the women. In a previous school-based study in Pelotas, 42% of girls aged 12 to 19 years reported use of a condom during the most recent intercourse, 12 which is in agreement with our finding of 47% for the age range of 15 to 19 years in the current study. A large-scale, population-based Brazilian study showed that 35% of women aged 16 to 25 years reported condom use at any time in the previous year. 13 These findings show that most women do not use condoms, but a welcome sign is that adolescents are reporting higher use rates than older females.
Having more than one sex partner in the previous 3 months was reported by 7% of the women. Studies in the United States have shown rates ranging from 6.6% to 8.4% for unmarried females aged 15 to 44 years. 11,14 We found no other reports of Brazilian studies on this topic.
The literature on factors associated with STD/AIDS will now be reviewed in light of the current findings on possible determinants of risk behaviors. Being younger than 20 years of age, in the current study, was associated with an odds ratio of 3.7 for the risk score, in comparison with women aged 45 to 49 years. In the United States, the risk of acquiring STD or AIDS goes down with age. 11 For most Brazilian women with AIDS, the diagnosis was made before the age of 30 years, suggesting that contamination took place in adolescence. 3
The incidence of STD and AIDS is associated, in the United States, with low income, little schooling, and being black. 4,15,16 In Brazil, the AIDS incidence was 53% higher among women with up to 8 years of schooling than among those with ≥12 years of schooling. 2 In the current study, education level had a strong effect, whereas the effects of income and skin color disappeared after adjustment.
Risk was lower for women whose marital status was single and highest for those who were separated or divorced. The effect of this variable increased after adjustment for age, showing that negative confounding was in place. The apparently paradoxical finding for single women may be explained by their greater use of condoms. No other reports of Brazilian studies were located.
Regular physical exercise was associated with lower risk scores in the crude analysis, but this effect disappeared after adjustment. The crude results were confounded by age because older women were more likely to have regular exercise.
The odds ratio was increased by 50% for smokers, even after adjustment for other variables. This has been reported earlier. 17 It is postulated that smoking is a marker for more risky behavior, indicating less concern about health. Smokers tend to be more active sexually, have earlier sexual initiation, and have more partners. 17
In summary, the results of the current study show that most women reported one or two risk behaviors. The most common were early sexual initiation, nonuse of condoms, and use of alcohol or drugs by the partner before intercourse. Risk behaviors were more frequent among young women, those with little schooling, those who were separated or divorced, and smokers. This information is helpful for designing and targeting public health interventions to prevent STD and AIDS.