The main study questionnaires administered to those with the tracer conditions had three modules regrouping eight sets of questions, namely: socio-demographic characteristics (including demographic, educational, and occupational characteristics), family composition, and income; attitudes and beliefs regarding health services; social support; health care entitlements; out-of-pocket expenses; regular source of medical care; perceived need for care and perceived health (limitations and symptoms); and utilization of health services, in terms of the number of and reasons for visits.
In Rio de Janeiro, however, the questionnaire had one family module only to avoid repeated questioning on socio-demographic characteristics of household members, family composition, household income, and family health insurance, should there be more than one individual with a tracer condition in the household. All questionnaires were pre-tested and validated prior to the main survey in enumeration areas other than those retained for the study. The fieldwork was carried out by local teams trained for the purpose or, in the case of Mexico, contracted out to a private firm. Prior to the survey, all interviewees were given a written document informing them of the main objectives of the survey and how the collected information would be used. They were also told that while there may not be any direct benefits to them, their participation in the survey would provide valuable input toward improving the health services. Respondents were also informed that data would be treated in a confidential manner respectful of their anonymity and confidentiality. Table 4 gives the response rate per stratum and city. Table 4. Total number of households and screening response (in households and percent), by city and sample stratum. | Sample stratum | Mexico City | Rio de Janeiro | Rosario | Number of households | Screening response | Number of households | Screening response | Number of households | Screening response | Household | % | House hold | % | Household | % | 50% poorest | 7 295 | 4 141 | 56.8 | 4 705 | 4 064 | 86.4 | 2 966 | 2 748 | 92.7 | 30% intermediate | 6 328 | 2 771 | 43.8 | 4 298 | 3 028 | 70.5 | 3 964 | 3 269 | 82.5 | 20% richest | 8 329 | 2 597 | 31.2 | 5 643 | 3 516 | 62.3 | 2 613 | 1 810 | 69.3 | Sum | 21 952 | 9 509 | 43.3 | 14 646 | 10 608 | 72.4 | 9 543 | 7 827 | 82.0 | Note: Nonresidential and unoccupied households have been excluded from the total number of households for all cities. In Rio de Janeiro, a further 1 616 households were excluded as their occupancy status could not be determined. |
Data entry and editing in all three surveys required: (i) exhaustive visual revision of all the questionnaires to verify the adequacy of the linkages between the screening questionnaire and the tracer questionnaires; (ii) coding the variables, according to code books prepared and updated during the work; (iii) entering the data into computers; and (iv) editing the data, including cross-checking the coded and entered data against the questionnaires, and verifying the consistency between the data pertaining to different variables. Although each team used different software for data entry and editing they followed the same algorithm of procedures. Statistical methods We used path analysis to test the theoretical underpinning and the hypothesized associations in the model. Path analysis involves the use of a series of structural equations to estimate the magnitude of the hypothesized linkages between sets of variables (Alexander and Markowitz 1986). A least squares step-wise multiple regression analysis [3] was performed for each one of the endogenous variables (enabling, need, and utilization variables), including as predictors all preceding variables in the causal model, to calculate standardized partial regression coefficients (path coefficients). The standardized regression coefficient estimates the direct effect of a predictor variable on the dependent variable, controlling for the effects of all other independent variables in the equation. Since the total effect of a variable on another is the simple correlation between these two, the difference between the total effect and the direct effect yields the total indirect effect of the independent variable, that is, its effect exerted through other variables. Path coefficients thus help determine the relative importance of direct and indirect effects that predictor variables exert on the outcome variable (Johnson and Wichern 1988). As indicated in Appendix I, composite indices were created to measure attitudes and beliefs, social support, out-of-pocket payment, and perceived need. We also had to resort to either logarithmic or square-root transformation of the variables family income and number of visits because of skewed distribution (see footnotes in Tables 6-a, -b, -c for reliability scores, and skewness and kurtosis changes). Determinants of the Use of Health Care Services
in Argentina, Brazil and Mexico In this section, we will first present the results of the utilization survey for each country separately. Thereafter, we will have a summary discussion of the evidence from the three countries together to highlight commonalities and differences. For each country, we present below in tables and figures, and per tracer condition, the final sample sizes (Tables 5-a, -b, -c); descriptive characteristics of the study populations (Tables 6-a, -b, -c); bivariate correlation coefficients (Tables 7-a, -b, -c); findings of the path analyses (Tables 8-a, -b, -c); and, finally, in Appendix 2, path diagrams for Argentina (Figures 2-a, 3-a, 4-a), Brazil (Figures 2-b, 3-b, 4-b) and Mexico (Figures 2-c, 3-c, 4-c) . Rosario, Argentina Table 5-a gives the number of cases included in the analysis. The vast majority of the individuals had at least one visit to a physician during the preceding 6 months, and many had several ( Table 6-a ).
In the hypertensive group, entitlement to health coverage was positively correlated with the level of education. There was also strong correlation between attitudes and beliefs regarding health services and the importance of out-of-pocket payments for individuals on one hand, and the perceived health needs on the other ( Table 7-a ). Some correlation was also detected between the number of physicians’ visits and perceived need, entitlement, and out-of-pocket payment valuation. For prenatal care, the level of education has an important positive correlation both with family income and with entitlement ( Table 7-a ). The latter two were also highly correlated. In addition, attitudes and beliefs regarding health services were correlated with out-of-pocket money valuation. More importantly perhaps, the number of visits had the highest correlation with the level of education and income, and with entitlement. For the diarrhea tracer, as for prenatal care, education had high correlation coefficients both with income and with entitlement ( Table 7-a ). As expected, the level of education was associated with regular source of care and with perceived need. Again, family income also correlated positively with health coverage and with perceived need. As to the number visits by the child, his or her age, perceived need-for-care, and having a regular source of care showed the highest correlations. Table 8-a and Figures 2-a, 3-a and 4-a in Appendix 2 depict, in both tabular and diagrammatic forms, the results of the path analysis. Only statistically significant regressions, path coefficients and associations are shown. The slight differences between the coefficient values in the table and figures reflect the use of both least-squares regression and LISREL for analysis (see Footnote 2). Overall, for hypertension, the model could only explain 9% of the total variation in service use. Although this was lower than expected, it was nevertheless not surprising given the relatively lower predictive power of a behavioral model for more discretionary services. Enabling factors were the main determinants of service use, indicating the sensitivity of physician utilization to health care prices and fees, and coverage. Also noteworthy was the role of gender, social support, and attitudes and beliefs in explaining perceived need-for-care.
The model proved to be more powerful in explaining service use in relation to prenatal care (R square = 0.32). The main predictors were level of education, entitlement, social support, and age, positively, and attitudes and beliefs and out-of-pocket payments, negatively. As to the factors impacting on the key enabling variables, both entitlement and out-of-pocket payment were explained almost solely by family income; social support by level of education and by income; and regular source of care by out-of-pocket money. For diarrhea, use of physician services appears to be determined directly by need-for-care, by the child’s age (negatively so), and by having a regular source of care. It is also noteworthy, as well as somewhat expected, that perceived need for care and consequent utilization of services was largely dependent on mother’s level of education. To synthesize, in Argentina, the model better explained service use in relation to prenatal care than in relation to diarrhea or, even less, hypertension. While enabling factors and need were the only determinants of service use for hypertensive individuals, predisposing factors were equally important in explaining use for prenatal care and diarrhea. Surprisingly, need-for-care had no bearing on service use during pregnancy. However, the fact that having a regular source of care, health care coverage, and out-of-pocket payments were among the key explanatory factors for all three tracers, indicates that health care services are not accessible to all in an equitable manner, despite the high level of education and awareness of the importance of seeking health care for the tracer conditions at under study. Rio de Janeiro, Brazil As in Rosario, a large majority of individuals with the tracer conditions under study visited a physician at least once during the 6-month period preceding the survey (Tables 5-b and 6-b ). We observed high degrees of correlation between income, education, and entitlement irrespective of the tracer condition ( Table 7-b ). Table 8-b gives the results of the path analyses. For hypertension, the amount of variation in service use that is explained by the model remained very low, at 6.87 percent of the total variation. Moreover, none of the predisposing variables was included in the final model which contained only regular source of care and entitlement as enabling variables, in addition to perceived need. The latter, in its turn, was mainly explained by age, sex, out-of-pocket payment, and having a regular source of care.
In the case of prenatal care, the fitted model fared much better, and explained 15.9 percent of the total variation in physician utilization, education and having regular source of care being the main explanatory factors. Income, attitudes and beliefs influenced women’s perception of need-for-care, along with out-of-pocket payment and having a regular source of care. As for diarrhea, the amount of the total variation explained was much higher, that is, 25%. Service use was, as expected, determined by the age of the child, out-of-pocket payment, and perceived need, the latter being itself influenced by age of child, out-of-pocket payment, and income. In summary, the model’s predictive power was low overall and across the three tracer conditions. Nevertheless, enabling variables, especially having a regular source of care and health coverage, proved to be the main variables influencing service use. Surprisingly, perceived need had no bearing on service use in pregnancy, but having a regular source of care had. The effect of predisposing factors was negligible, except for the age of the child in case of diarrhea, and the level of education in seeking and using prenatal care. Mexico City, Mexico The proportion of individuals who had at least one visit to a physician, and the average number of visits during the 6 months preceding the survey were, in general, slightly higher compared with the study populations in Argentina and Brazil, except for prenatal care (Tables 5-c and 6-c ). In the hypertension group, it was noteworthy that there were positive correlations between schooling and social support, and that attitudes and beliefs correlated positively with both direct out-of-pocket payment and perceived need. In addition, service use was positively correlated, albeit to a lesser extent, with income, attitudes and beliefs, social support, entitlement, and having a regular source of care. In the case of prenatal care, similar correlations were found between schooling, social support, and income, and service use, and between attitudes and beliefs and out-of-pocket payment. The variables with the highest correlation with service use were schooling, income, social support, and perceived need. As expected in the case of diarrhea, age was negatively with service use. In turn, there were positive correlations between schooling and both income and social support; between income, and entitlement and perceived need; and between attitudes and beliefs, and both direct payment and regular source of care ( Table 7-c ). The results of path analyses are shown in Table 8-c and in Figures 4-a, 4-b and 4-c in Appendix 2 . In the case of arterial hypertension, all the variables in the model were found to have a direct or indirect effect on utilization, although the proportion of the variance explained was very low, about 10%. The main factors that had a direct effect were perceived need, entitlement, direct payment, social support, and attitudes and beliefs. In addition to the direct effects, the main indirect effects included age, sex, attitudes and beliefs, and social support through perceived need. In the case of prenatal care, the model did not fare any better. Contrary to observations in the other two countries, income, education, and social support were the main explanatory variables and, to a lesser extent, out-of-pocket payment and perceived need. As for diarrhea, the main variable with direct influence was the age of the child, although more variables were found to have a significant effect on service use when LISREL was used (see Figure 4-c in Appendix 2 ), resulting in a slight increase in the proportion of the total variation explained. In summary, except for regular source of care, enabling factors were the main explanatory variables in all three models. Education, income, and attitudes and beliefs also played a secondary role in directly influencing service use, whereas age was the key factor in the case of diarrhea. Discussion According to Andersen and many others, access to health care is equitable to the extent that the "need" variables account for a large proportion of variations in service use (Andersen 1968). Service use may take different forms, depending upon its type, i.e., discretionary or non-discretionary; its purpose, i.e., preventive vs. curative; the place of service delivery, i.e., out-patient clinic vs. hospital care; or its nature, i.e., physician visits vs. dental etc. In this study we tried to cover a large spectrum of health care needs by selecting three distinct tracer conditions in terms of their nature, and the type and purpose of the services they will require, so as to be able to better understand the relative role of the enabling and need variables in times of health care reform. Obviously, such a study has its own limitations, not the least being its cross-sectional nature which does not allow for clearly differentiating causes from effects. Moreover, and more importantly perhaps, is the limitation in observing or measuring the trends, and hence the changes in the relative role of a number of key variables like income and entitlement over time, as reforms evolve or get implemented. While Andersen’s model has so far been unable to explain a large proportion of the variation in service use in developed countries, we thought that it might prove to have a higher explanatory power in the health care context of higher middle-income countries such as Argentina, Brazil and Mexico. In these countries, availability of health care services has ceased to be an issue while accessibility - or more importantly, affordability - remains a major concern. While the model did not prove to be more powerful, the findings nevertheless concur with the observations made in the previous chapters with respect to the lack of equity in access to care. A number of caveats of a methodological nature are in order. First, because the original model in Figure 1 did not fit the data, an exploratory data analysis was conducted to seek the best model capable of explaining the utilization of the health services for the three tracers across the three countries. In order to find the best model for each tracer, a heuristic model was defined based on a number of statistical considerations: (i) a path entered the model whenever it was significant at 10% level; (ii) the p-value of the Minimum Fit Function Chi Square test was maximized; and, (iii) the model needed to be considered be non-recursive and observe the sequential nature of Andersen’s original model. Second, there were measurement issues related to multi-dimensional constructs as described in Appendix I. Several constructs had either low or moderate internal consistency scores (Cronbach’s alpha). Third, several key variables, including income and the number of visits did not distribute normally, therefore requiring either logarithmic or square-root transformations. This was especially problematic, statistically speaking, for the dependent variable, number of visits. Fourth, regular source of care and entitlement, in the case of Mexico, were measured as dummy variables on a dichotomous scale. This may violate the assumptions behind the use of regression analysis, especially if the split between the two categories are not in the order of 0.25/0.75, as was the case in Mexico for entitlement. Finally, although originally intended, as evident in the sampling strategy selected (see above), we decided against presenting the findings per income strata. While sub-group analysis could have enriched our understanding of the main determinants of service use, it would have rendered our findings more unstable, statistically speaking, because of the reduced sample size, without, in our opinion, necessarily much affecting the overall predictive power of the underlying model. Hypertension In all three countries the predictive power of the model was the lowest for hypertension. This was somewhat to be expected given the nature of the condition and the consequent discretionary use of services for its management, especially in middle-income countries where it is yet to be recognized as a major health problem with severe health consequences. The results indicate that enabling variables rather than perceived need play a more prominent role in service use, which suggests lack of equity in access to health care. In addition, the effect of the key predisposing factors such as income, education, and attitudes and beliefs on utilization - which were highly correlated with one another - exerted through entitlement and out-of-pocket payment also points toward the same conclusion. In Argentina, entitlement and out-of-pocket payment directly influenced the number of visits. Entitlement had a negative effect, however, implying that a better situation of entitlement corresponds to a lower number of office visits. Since in this country survey, entitlement was measured on a scale with the highest value meaning better quality and more coverage, the direction of the association suggests inequity in access to care. In Brazil and Mexico, the interpretation is similar, although the results are positive because of the measurement of entitlement as a dummy variable, where 1 indicated the existence of health insurance and 0 its absence. In all these countries, health care coverage does not necessarily mean access to better quality and a larger range of services but, rather, easier access to a physician. In fact, the observed direct effect of the enabling variable regular source of care on the number of visits in Brazil is a case in point. The direct and negative effect of the enabling variable, out-of-pocket money, on the number of visits in Argentina and Mexico is revealing: the higher the costs associated with service use are perceived, the more the services are used. Although, this may seem at first to be contradictory, it was not unexpected. As the poor and the indigent tend to have poorer health than the rich, they are often obliged to allocate a larger portion of their disposable income to health care at the expense of other needs. Prenatal care With respect to this tracer, the three countries showed similar results, not only in terms of the proportion of service use variation explained by the model - as high as 32% in Argentina and as low as 11% in Mexico - but also in terms of the key explanatory variables. For instance, perceived need was not significantly associated with service use in Argentina and Brazil, nor was having a regular source of care significantly associated with service use in Argentina or Mexico. Moreover, in Brazil, out-of-pocket payment did not appear to have any direct or indirect effect on service use, either through perceived need or regular source of care. The same applies to entitlement, which was associated with service use only in Argentina. It seems that in Brazil, being educated and having a regular source of care are the main explanatory factors. Indeed, level of education was the only common predictor in the three countries, influencing prenatal care use directly or indirectly, through social support in Mexico and regular source of care in Brazil. Finally, the only direct effect of income was observed in Mexico, although it exerted an indirect effect through entitlement in Argentina and Mexico. These differences may be due to a somewhat similar health care financing and delivery system in Argentina and Mexico that differs from that in Brazil. In Brazil, the social network of formal and informal connections with providers, or at least with those in the public health care system, and the fact that health care is less tiered could explain why income does not appear to be a predictor of service use, especially since the advent of the unified health model (SUS). This interpretation is more likely to reflect the reality for prenatal care and delivery than for curative and less discretionary services. For instance, the average number of prenatal care visits in Brazil was about 10, higher than in Mexico and about the same as in Argentina. Diarrhea Again, there were differences between the three countries in terms of the total amount of variation in service use explained, which ranged from a high of 25% in Brazil to a low of 13% in Mexico. There were also differences in the nature of the main predictors, except for the most significant one, age. The level of education of the mother, having a regular source of care, and perceived need proved to be other key predictors in Argentina. In Brazil and Mexico, the effect of out-of-pocket payment, both directly and indirectly through perceived need, was considerable. Mexico was unique in that social support and attitudes and beliefs had a direct influence on service use. Based on these results, one could argue that, at least for this tracer, the health care system may be more equitable in Argentina than in Brazil or in Mexico. Indeed, Argentina appears to offer, in our opinion, the best primary health care of the three countries. As for the Brazil and Mexico, the findings are disconcerting because of the association between out-of-pocket payment and service use considering the characteristics of this tracer and the fact that it is relatively simple to manage. The seemingly significant dependence on the ability to pay to access to services for diarrhea can be considered indicative of inherent inequality of the health care system. Moreover, in Mexico City, out-of-pocket money is also directly influenced by social support and directly influences the existence of a regular source of care. The converse effect of out-of-pocket payment on the number of visits in Brazil could be explained by either the different nature of the measurement of this concept, which focused on the self-assessed expected costs rather than real costs, or the health seeking behavior of the poor, who might take the child to a physician in anticipation of higher costs in case of delayed diagnosis and treatment. Another important observation, in our opinion, is the high degree of correlation between education and/or income and entitlement, especially in Argentina, considering that all three countries have adopted heath care reforms with privatization of health care delivery and financing as one of their main features. We believe that the results are consistent with observations in previous chapters in that the health care systems in all three countries do not appear to be equitable for meeting all types of health care needs. However, based on these findings, we are not in a position to claim that that they have become more or less equitable as a consequence of health care reforms in the works in these countries. On the other hand, one could consider this study as a baseline for future studies to monitor changes in the relative importance of key variables such as income, entitlement, and having a regular source of care, and thus gauge the evolution and effects of the privatization of health care financing and delivery in Argentina, Brazil and Mexico.
[1] The survey was coordinated in Argentina by Stella Sanchez, with the participation of Irene Luppi; in Brazil by Lenaura Lobato, with the participation of the PRODEMAN-UERJ (University Research Program on Social Demands, Rio de Janeiro State University); and in Mexico by Silvia Tamez, with the participation of Marco Zepeda.
[2] The census enumeration areas are called AGEB in Mexico City, Setor in Rio de Janeiro, and Radio in Rosario.
[3] The stepwise selection of independent variables is a combination of backward and forward procedures. The first variable considered for entry into the equation is the one with the largest correlation with the dependent variable. This variable is first examined through application of the entry criterion, and then of the removal criterion. In the next step, variables not in the equation are examined for entry. After each step, variables already in the equation are examined for removal. Variables are removed until none remain that meet the removal criterion. Variable selection terminates when no more variables meet entry and removal criteria (Norusis 1986). The criteria for inclusion in and exclusion of the variables in the model were established on the basis of the probability associated to statistic F (0.05 and 0.10 respectively).
We also used LISREL (Bollen 1989) which allows for more flexibility in terms of assumptions than more conventional least-squares multiple regression analysis. All path diagrams in figures below display results obtained through the use of LISREL and differ only slightly from the results obtained by regression analysis.
Copyright 2000 © International Development Research Centre
info@idrc.ca |
Document(s) 3 de 14
