Strengthening of reproductive health and family planning services in developing countries is repeatedly highlighted as a priority for reducing maternal mortality and improving maternal and child health. Effective implementation of appropriate services requires an understanding of the factors affecting reproductive outcomes among women at risk and their patterns of behaviour. In Brazil, despite relatively high levels of contraceptive use, there appears to remain a large unmet need for family planning, particularly in the poorer areas of the country. Evidence from the Demographic and Health Survey (DHS) suggests that if all Brazilian women who wanted to limit their fertility were protected by effective contraception, the total fertility rate (TFR) in 1996 would have stood at 1.8 children per woman – one-third lower than the observed rate 1. This proportion was about the same as that observed ten years earlier, despite the occurrence of rapid fertility decline (Figure 1). Moreover, important differences persist across sub-regions in reproductive health-related behaviours and outcomes.

Differences in availability, accessibility, and acceptability of the range of contraceptive technologies may mean that not all methods are favoured at the same time. Evidence from a number of developing countries reveals that the method mix tends to continuously evolve. It has been estimated that at least half of the women using contraceptives switch methods within a five-year period 2. A study conducted in six developing countries showed that about one third of women using reversible contraception had stopped using their method within 12 months 3. Such findings underlie the increasing importance of monitoring trends and determinants of method choice, as family planning and reproductive health programs must adapt to meet users' changing needs and preferences.

In Brazil, reversible methods account for less than half of the method mix 1. Oral contraceptives are the most commonly used reversible method, the choice of 16% of women of reproductive age. Use of other reversible methods is about 10%. Analyses of DHS results have revealed a number of socio-cultural variables, including women's exposure to the mass media and religion, to have significant influence on contraceptive method choice 4. Research using multilevel modeling has also pointed to community-level influences on method choice (in particular, significant random effects at the municipal level on adoption of sterilization), likely reflecting influences of the service environment such as presence of hospitals 5. Brazil does not have an official national family planning program, although in recent years some family planning-related services have been incorporated into the country's maternal and child health program. Evidence has pointed to important constraints in the availability of and access to family planning and reproductive health services, as well as severe deficiencies in quality of care 6.

In order to fully understand women's family planning choices, it is imperative to investigate the components of contraceptive dynamics such as contraceptive discontinuation, failure and switching. Contraceptive failure, resulting in unintended pregnancy, may be due to method failure, user error or provider failure 7. The impact of contraceptive discontinuation and switching on reproductive outcomes depends to a great extent on both the woman's decision to use another method and the effectiveness of that method. Of particular interest is switching to a less effective method or to no method, which is likely to increase the overall level of fertility. Switching between methods of similar effectiveness may hold less important demographic impact, although any switching may potentially increase the risk of an unintended pregnancy, as women are more likely to experience a method failure in the first months of use when they are not fully familiar with the new method. A study of contraceptive dynamics suggested that at least one-third of the TFR in 15 less developed countries (including Brazil) was associated with either a contraceptive failure or a contraceptive discontinuation for reasons other than a desire to get pregnant 8. It has been argued that contraceptive continuation rates could be raised substantially by eliminating discontinuation due to non-method- and method-related reasons 9. The contraceptive continuation rate has been suggested as a useful summary measure of the overall effectiveness of program services in enabling clients to sustain contraceptive use even though they may switch from one method to another 10.

At the same time, Brazilian society has been marked by sharp regional inequalities that have characterised the country since the colonial period 11. Much attention has been paid to assess the disparities between the poverty-stricken Northeast and more affluent Southeast, the two most populous regions, together comprising about 70% of the total population. For example, in terms of income levels, the proportion of workers earning less than one legal minimum wage is 2.4 times higher in the Northeast (58%) than in the Southeast (24%). A similar tendency can be observed in terms of rates of adult illiteracy (40% versus 9%) 1. Such differences are seen to hold important implications for demographic and health outcomes. This can be noted through regional variations in the TFR, from a low of 2.1 in the Southeastern state of Rio de Janeiro to a high of 3.1 in the Northeast 1. Many studies have previously examined the implications of contraceptive continuation on fertility outcomes and on family planning program performance measures at the national level or across countries; however, little research has been conducted at the sub-national level.

The objective of this paper is to analyze the factors associated with contraceptive discontinuation, failure and switching across regions of Brazil, drawing on data from the 1996 DHS. We examine the demographic and socio-cultural influences of contraceptive use dynamics across reversible methods, focusing special attention on the reported reasons for method discontinuation. Identifying the predictors of method failure and discontinuation in the context of rapid and profound changes in reproductive behaviours could draw the attention of policymakers on regional disparities and the difference in quality of reproductive health care provision and eventually assist in improving services.

The DHS is one of the largest programs collecting quantitative data on reproductive health knowledge, attitudes and practices in the developing world. Surveys are carried out using standardized instruments, methods of training, data collection and data processing 12. The most recent DHS in Brazil, the 1996 Pesquisa Nacional sobre Demografia e Saúde, collected information through personal interviews with 12,612 women aged 15–49, selected through a two-stage random sampling process designed to represent 95% of the country's population at the national and sub-regional levels (some rural areas in the North and Centre-West regions were excluded) 1.

In addition to core questions for measuring basic indicators for population and health program monitoring and evaluation, some DHS surveys include additional modules designed to obtain specialized information on specific topics. The present analysis takes advantage of the "calendar" module of reproductive events. The calendar records detailed information (i.e. month-by-month) about the timing of a number of events–including marital unions, residential mobility, births, and contraceptive use (including method or reason for discontinuation)–occurring in the five calendar years preceding the survey. This retrospective method of measurement makes heavy demands on the memory of respondents, but recall is aided by timing events in relation to one another. Overall, the quality of information obtained through this approach has been evaluated as superior to alternative retrospective data collection techniques for longitudinal information 1314. A relatively less-exploited module among the DHS surveys, the calendar has become increasingly important in monitoring contraceptive dynamics and has greatly facilitated researchers' capabilities to conduct analyses of discontinuation and switching in particular 15.

For this study, a discrete-time competing risks hazard model is used to estimate the probability of a woman making a specific transition at a given duration of use. A discrete-time competing risks model is a multinomial logistic model in which the observations are repeated according to the duration of use until the event occurs or is censored. This approach allows incorporation of time-varying covariates as compiled in the calendar (such as woman's age, marital status, and parity at the time of use). Our main objective is to describe the patterns and explain the independent determinants of contraceptive discontinuation, failure and switching among women at risk across the main regions of the country.

Included in the model are all women who were ever sexually active and who initiated use of a reversible method of contraception over the period covered by the calendar. The units of analysis are the episodes of contraceptive use (i.e. continuous use, month-to-month). Observations in the three-month period immediately before the survey are excluded, to reduce bias in the estimation of use-failure rates, given that some women may have been unaware of their very early pregnancy 8. Likewise excluded are episodes of use that began before the calendar period, as the date of initiation would not have been recorded.

We consider episodes of use of oral contraceptives, injections, condoms, and traditional or natural family planning methods (periodic abstinence and withdrawal). Uses of other modern reversible methods (such as IUD, diaphragm, and spermicides) are excluded, due to the small number of episodes observed in the survey and since it was not considered appropriate to aggregate these methods into a single category as discontinuation and failure rates can vary substantially across methods. Sterilization is also excluded given that discontinuation of this method occurs infrequently.

In examining the patterns of contraceptive use dynamics, four categories were created for the response variable: (i) failure; (ii) abandonment of the method while still in need of family planning; (iii) switching to another reversible method; and (iv) continuing use of the method. Contraceptive failures include any (presumably unintentional) occurrence of a pregnancy while using the method. Episodes where the woman reported having discontinued use for non-method-related reasons–such as a desire to get pregnant, marital separation, or infrequent sexual intercourse–were included under the fourth category, as they were not considered to have ended while in need of family planning. Note that these categories should be interpreted as approximate, given that self-reported reasons for contraceptive discontinuation may be somewhat unreliable 16.

A number of episode-specific and woman-specific variables were included in the model as potential confounding factors, including contraceptive intention and duration of use as well as woman's age, marital status, number of living children, education, ethnicity, place of residence (according to the residential history in the calendar), and mass media exposure (as assessed through television viewing habits). These covariates have been previously considered relevant to assess mechanisms that influence contraceptive use, method choice and/or discontinuation in Brazil and other developing countries 234151617. In particular, television programming, notably the highly popular soap operas (telenovelas), has been credited with playing a substantial role in promoting ideological change in Brazil with respect to reproductive behaviours by portraying lifestyles that favour smaller families 18. Moreover, in the Brazilian context, particular attention is paid to differentials across sub-regions.

Of further substantive and methodological interest, our study uses a multilevel approach. Standard regression models assume that observations are independent. However, given the hierarchically nested structure of the data being used here, multilevel modeling becomes necessary to allow for controlling for any unobserved correlation between observations within hierarchical levels. At the first level, in modeling women's episodes of contraceptive use, an individual may contribute more than one segment of use to the sample. At the second level, the DHS sampling scheme entails selection of households and individuals within enumeration clusters 19. Individuals from the same sampling cluster are considered likely to exhibit similar demographic and behavioural characteristics (because of a variety of unmeasured and unmeasurable factors) compared to those selected from different clusters. The multilevel model is thus used to compensate for assumed intra-woman and intra-cluster dependence of observations. Moreover, a cluster can be considered a proxy for neighbourhood or community, and reflects local service environment as well as local "culture". It has been argued that women in the same community often talk to each other and, therefore, are more likely to exhibit similar behaviours regarding contraceptive use 20.

The multilevel discrete-time competing risks model is used to assess regional disparities in contraceptive failure, abandonment and switching, conditioned for the set of fixed- and random-effects. The formulation of the model is as follows:

ln ⁡ ( λ r t i j k λ 4 t i j k ) = α r t + β ′ r x t i j k + u r j k + v r k , r = 1 , 2 , 3 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaafaqabeqacaaabaGagiiBaWMaeiOBa4MaeiikaGYaaSaaaeaaiiGacqWF7oaBdaWgaaWcbaGaemOCaiNaemiDaqNaemyAaKMaemOAaOMaem4AaSgabeaaaOqaaiab=T7aSnaaBaaaleaacqaI0aancqWG0baDcqWGPbqAcqWGQbGAcqWGRbWAaeqaaaaakiabcMcaPiabg2da9iab=f7aHnaaBaaaleaacqWGYbGCcqWG0baDaeqaaOGaey4kaSIaf8NSdiMbauaadaWgaaWcbaGaemOCaihabeaakiabdIha4naaBaaaleaacqWG0baDcqWGPbqAcqWGQbGAcqWGRbWAaeqaaOGaey4kaSIaemyDau3aaSbaaSqaaiabdkhaYjabdQgaQjabdUgaRbqabaGccqGHRaWkcqWG2bGDdaWgaaWcbaGaemOCaiNaem4AaSgabeaakiabcYcaSaqaaiabdkhaYjabg2da9iabigdaXiabcYcaSiabikdaYiabcYcaSiabiodaZaaaaaa@6798@

where λ_rtijk is referred to as the hazard of a transition of type r at time t for the use interval i of woman j from cluster k. The baseline hazard is represented by α_rt, a function of time. β_r is the vector of parameters for transition r, with x_tijk the associated set of covariates (the same for each of the three types of contrasts against continuation of method use). The estimators u_rjk and v_rk measure the random variations at the woman and cluster levels respectively. They are assumed to be mutually independent and normally distributed with mean zero and variances σrjk2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOCaiNaemOAaOMaem4AaSgabaGaeGOmaidaaaaa@33BE@ and σrk2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOCaiNaem4AaSgabaGaeGOmaidaaaaa@3261@ respectively.

The final sample for the study consisted of 6,027 episodes of contraceptive use. The analysis was carried out using the MLwiN statistical software program 21. In order to facilitate interpretation, the results from the multilevel competing risk hazard model were applied to estimate twelve-month cumulative probabilities of contraceptive discontinuation, using the multiple classification analysis (MCA) table 22. First, conditional probabilities of method discontinuation were calculated for each month. In order to calculate the effect of a specific covariate on the cumulative probability, the others were held at their mean.