Oportunidades began in 1997 as a national program designed to relieve extreme poverty in Mexico. The Oportunidades program combines a cash-transfer program with financial incentives for positive behavior in health, education, and nutrition. During its first 3 years, Oportunidades extended benefits to many eligible families in rural areas. Starting in 2001, the program expanded to urban areas. By 2004, Oportunidades covered approximately 5 million families in all 31 states of Mexico; it has been used as a model for initiatives in Argentina, Brazil, Colombia, Ecuador, Honduras, Jamaica, Nicaragua, Peru, Turkey, and the United States.

At its inception, Oportunidades determined household eligibility in 2 stages, first by identifying low-income communities and then by choosing low-income households within those communities (23). Low-income communities were selected on the basis of the proportion of households in those communities living in poverty, according to data from the 1995 national census. Households were selected according to an index of objective characteristics, such as housing materials, water and sanitation facilities, education, and family structure, which were shown to be good proxies for annual income. On average, 78% of the households in selected communities were eligible for program benefits, and 97% of these households enrolled in the program (24). Once enrolled, households received benefits for a minimum of 3 years, conditional on meeting the program requirements; new households were not able to enroll until the next certification period.

Program benefits were distributed only if family members complied with a series of behavioral changes. The requirements included prenatal care; well-baby care and immunization; nutrition monitoring and supplementation; preventive checkups; and participation in educational programs on health, hygiene, and nutrition. Adult family members were required to attend a biannual health checkup and were encouraged to participate in regular educational sessions at which health, hygiene, and nutrition issues and best practices were discussed. Oportunidades verified that households completed the required health care visits by having medical providers at participating public health clinics provide certification of participation.

Oportunidades families received 2 types of cash transfers every month. The first was a universal cash amount for all families, typically worth approximately 20%–30% of household income, equivalent to an average of approximately $25 per month. The second cash gift was given to households with school-aged children if the children were enrolled in and attended school. This amount varied depending on the number of children attending school and was greater for girls than for boys. Approximately 1% of households were denied the cash transfer because of noncompliance.

At the inception of Oportunidades, and for the purposes of conducting a rigorous evaluation of the program, eligible communities in rural areas of 7 states were randomly assigned to intervention and control groups. A baseline survey was conducted in all households in these groups in 1997 and 1998, which included information about socioeconomic status and household demographic composition but no information about anthropometry or health status. However, because the original control group was incorporated into the program 18 months after the program began, both the control and intervention groups were consolidated into one intervention group.

In 2003, a new control group was added as part of the 5-year follow-up survey of the original communities. This new group consisted of 151 control communities selected from the original 7 evaluation states. Data from the 2000 census were used to select new control communities that matched the old ones as closely as possible. Specifically, communities were selected that had not yet been incorporated into Oportunidades and that most closely matched the originals by propensity score matching methods applied to sociodemographic and infrastructure characteristics (P. Todd, unpublished data, 2004). Matching by using the propensity score is equivalent to matching intervention and control observations on the basis of a large number of characteristics (25). After completing the propensity score matching, we then chose the nearest community to each original community, in terms of propensity score, to be included in the comparison group. A proportion of the households in the new comparison communities was eventually invited to participate in Oportunidades in 2005 when the program was expanded to all rural areas. Thus, to reduce bias, we restricted the sample for analysis to just those families who were invited to participate in Oportunidades and subsequently enrolled in the program.

The survey reported here was conducted in 2003 in low-income households (income below the 20th percentile, mean daily per capita expenditure of less than $2) from 323 rural communities (defined as towns of <2500 inhabitants) in 7 Mexican states. The households included in the sample were a mix of those recruited in 1997 as part of the original sample and those from new control communities first surveyed in 2003. The final sample used for analysis consisted of an intervention group of 5280 adults from households living in the original communities recruited in 1997 and 1998 and a control group of 1063 adults living in the new comparison areas at the time of the 2003 survey (these 1063 adults were subsequently enrolled in Oportunidades in 2005).

Data were collected during house-to-house interviews with all participants. The interviews occurred on all days of the week except Sundays between 8 am and 6 pm. Interview teams visited the homes without appointments and returned to each home at least 3 times to try to locate household members. After identifying the head of household — or spouse of the head of household if the head of household was not available — the interviewers obtained written consent to conduct the medical assessment and the interview. The interviewers then measured and weighed each available adult and measured blood pressure. During this same visit, a questionnaire was administered to obtain information about demographic, socioeconomic, and other factors.

BMI and obesity

Height and weight were measured during the interview by trained personnel in duplicate by using standard techniques (26). If the two measurements differed by more than 5% for any outcome, survey personnel obtained a third measure and used the two closest measurements. Weight was measured in light clothing without shoes to the nearest 100 g on a digital scale (Tanita Mother-Baby scale, model 1582, Tanita Corp, Arlington Heights, Illinois). Height was measured in standard position with a portable stadiometer (Road Rod, model 214, seca corp, Hanover, Maryland) and recorded to the nearest millimeter. Obesity was defined as BMI ≥30.0 kg/m², and overweight was defined as BMI ≥25.0 kg/m².

Blood pressure and hypertension

Blood pressure was measured by trained nurses with mercury sphygmomanometers. Uncontrolled hypertension was defined as SBP ≥140 mm Hg or diastolic blood pressure (DBP) ≥90 mm Hg (27). Participants were also asked about symptoms of hypertension (e.g., headaches, dizziness, buzzing in the ears, seeing lights without apparent reason, nosebleeds without apparent reason).

Self-reported health

Questionnaires were administered by survey personnel to obtain information about self-reported health status, self-reported fitness, ability to participate in ADL, and socioeconomic status. Specifically, participants were asked if they could participate in medium-effort ADL, such as working on a farm or in a garden or sweeping. They were also asked if they were able to participate in heavy-effort ADL, such as running or lifting a heavy object. Participants were asked how many sick days they had had in the previous 4 weeks and were also asked the number of days of "inability" (not being able to perform daily activities) they had experienced in the past 4 weeks.

Demographic and other household-level control variables

The following individual- and household-level variables were obtained through the household questionnaire: age, sex, educational attainment and occupational status of all household members, whether head of household was married, whether head of household spoke an indigenous language, whether any household member was self-described as disabled, land use (whether a household owned and used any piece of land), ownership of farm animals, ownership of an animal other than a farm animal, amount of land owned by the household, presence of dirt floor, presence of bathroom, presence of electricity, number of large assets (including television, washing machine, gas heater, and refrigerator), number of small assets (including blender, electric kettle, radio, stereo, video cassette recorder, and fan), and ownership of car or other vehicles.

An additional questionnaire was applied to the new comparison households, which asked families retrospectively about household demographic structure and ownership of assets in 1997 and 1998. The goal of the retrospective survey was to collect information that could be easily and accurately recalled and could be incorporated into the analysis. This section of the questionnaire was extensively pilot-tested to ensure that adults in households from comparison communities could report on their household-level socioeconomic and demographic structure from 5 years before. Only questions for which survey respondents were confident in their accuracy and recall were included in the final questionnaire.

Community-level covariates

A community survey was also administered to obtain detailed information about community characteristics. These questions included the following variables: proportion of the population that was indigenous, proportion of the village supplied with electricity, proportion of the village with a drainage system, availability of a public or private telephone, availability of a preschool, availability of a health center, presence of a shop that serves the local community, presence of a home-based shop, whether the community received government assistance, presence of a formal or informal credit institution, presence of a high school, average rent for a house, presence of community irrigation, and mean monthly wages for men and women.

As mentioned above, for inclusion into the model, we focused principally on data obtained in the baseline survey from 1997. However, because the 1997 data for some community characteristics were unavailable, we used 2003 information as necessary.

We estimated the impact of Oportunidades on adult health, focusing on 3 primary sets of outcomes: 1) BMI, overweight, and obesity; 2) blood pressure and hypertension; and 3) self-reported health and fitness. The 2 methods of analysis were ordinary least squares (OLS) regression adjusted for survey design and nonparametric matching techniques. All analyses were conducted by using STATA version 9.0 for Windows (StataCorp LP, College Station, Texas).

For BMI and blood pressure, we identified and excluded implausible values and outliers more than 3 standard deviations from the mean (<5% of values removed). Participants were excluded if they were younger than 30 years. Descriptive statistics were generated by community (intervention and control) and by households within those communities (intervention and control). Between-group comparisons were made with 1-way analyses of variance for continuous variables and analyses of proportions for noncontinuous variables.

We first examined the association between participation in Oportunidades and various adult health outcomes by conducting simple OLS regressions (linear and logistic) with program participation, age, and sex as independent variables. We then repeated these analyses with the 47 demographic, household-level, and community-level control variables described above. Standard errors were adjusted for intercluster correlation at the community level.

In order to examine the robustness and sensitivity of the OLS estimations, we also used nonparametric matching methods to assess the effect of Oportunidades on adult health. These methods are nonparametric techniques that control for observable differences between intervention and control communities. The methods can be more flexible than multiple regression methods, which rely strongly on assumptions of linearity (28). Because of limited space and the similarity of the outcomes with these 2 approaches, results reported are based on the standard OLS regressions.

The Oportunidades evaluation was approved by the research committee at the National Institute of Public Heath in Mexico and the Committee for the Protection of Human Subjects at the University of California at Berkeley. Participants were invited to participate in the evaluation after receiving a detailed explanation of the survey procedures and were asked to sign an informed consent declaration at that time.

Age- and sex-adjusted BMI was significantly lower in the intervention group than in the comparison group (26.57 kg/m² vs 27.16 kg/m², P < .001), and the significance was attenuated with the inclusion of individual-, household-, and community-level covariates (Table 3). The prevalence of age- and sex-adjusted obesity was also significantly lower in the intervention group (20.28% vs 25.31%, P < .001) as was the prevalence of overweight (59.24% vs 63.04%, P = .03); the difference in obesity but not overweight was retained after adjusting for covariates.

Oportunidades participants had lower SBP and DBP than did adults in comparison areas, when controlling for all covariates. Uncontrolled hypertension was present in 33.80% of the Oportunidades participants and 34.52% of the comparison group. Participation in the program was associated with increased likelihood of a participant's having had a blood pressure test in the 5 years before the survey, which was part of the Oportunidades program requirements. Program participation was also associated with a reduced number of self-reported symptoms relating to hypertension, such as headaches, dizziness, or buzzing in the ears.

The Oportunidades group was better able to participate in medium-effort ADL than was the control group, and the results were sustained after including covariates. Adults living in Oportunidades communities had fewer sick days in the 4 weeks before the survey and fewer days in which they were unable to do household chores. No significant difference was seen between the groups in self-reported ability to perform heavy-effort ADL.