The first goal of this article was to assess the levels of metabolic control among elderly diabetic population in Costa Rica, using the guidelines of the ADA as a point of comparison . Mean levels of HbA1C, HDL, and LDL, and the proportion of patients having elevated levels in the same biomarkers, seem high in a first look. However, these are similar or even lower than those reported in high-income populations in the U.S. [20, 27–31], Germany , Sweden , and France . Elevated levels of DBP and SBP are considerably higher than in the U.S., while mean levels of triglycerides are lower . When compared to another developing country, prevalence of poor levels of HbA1C, DBP, triglycerides, and LDL are lower in this Costa Rican population than in a slightly younger population in Medellin, Colombia; only the proportion of HDL ≤ 40 mg/dl was higher in Costa Rica than in Medellin . Preventive health behaviors are broadly practiced among the Costa Rican older population with DM: only 40% is not having aspirin therapy regularly, and 46% report not to have an influenza vaccine during the last 12 months. Additionally, only 10% report not to have had a diabetes examination by a physician during the last year.
If less strict endpoints of HbA1C are used (8% or 9% instead of 7%), the proportion of diabetic older patients with poor glycemic control decreases considerably, and this change strengthens the argument that, among Costa Rican elderly, levels of metabolic control seem favorable.
Elevated BP levels among elderly Costa Ricans with DM found in this study signals an important public health problem in the country. The combination of hypertension and DM yields very high total costs for the public health care services . Further research is needed to learn the reasons for this poor hypertension control in Costa Rica, especially because there are no differences across health care settings. Preventive behaviors linked more strongly to hypertension than to DM or to dyslipidemias should be studied. Prevalence of high blood pressure in the general elderly Costa Rican population is extremely high: 64% among females and 55% among males . DM patients do not escape to this pattern. Given that cardiovascular disease mortality is not high in this population, one wonders whether blood pressure distribution of other populations is applicable to Costa Rica . High prevalence of elevated blood pressure persists even if cutoffs for defining elevated SBP and DBP are raised to 150 mmHg and 90 mmHg respectively.
Aside from hypertension, Costa Rican elderly with DM do not seem to perform worse at metabolic control than populations in the industrialized world. Why are metabolic control levels so favorable among this population? One possible explanation is that Costa Ricans are commonly inclined to preventive health behaviors. Palmer  documents that Costa Rica was, in 1914, one of the few Latin American countries in which the goals of the Rockefeller Foundation anti-hookworm campaign were achieved. According to Palmer, the success of this campaign can be explained by the active involvement of Government employees and rural teachers, and the willingness of the Costa Rican population to trust in these public workers. Regarding more recent data, with a non-representative sample of diabetic patients in San Jose, Costa Rica's capital, Firestone et al. found that the levels of DM-specific knowledge were greater than in a sample of Spanish-speaking U.S. Latinos in Starr County, Texas .
However, in general, achievements in health status in Costa Rica have been linked to effective primary health care services, provided mainly by public institutions and with a strong commitment with equity [12–14]. Although the Costa Rican government had a long tradition of providing those services (which was in part possible because of the non-existence of military expenditures since the 1949 Constitution abolished the armed forces), in the 1970s there was a breakthrough with the opening of hundreds of Rural Health Posts . More recently, a health sector reform launched in 1995 fostered the opening of community health centers or EBAIS to improve the access to primary health care services in underserved areas. It is hard to tell whether this "Costa Rican model" can be exported to other Latin American countries with much larger or less integrated populations, weaker institutions and fewer resources available for public health.
These free primary health care services have been key mechanisms in the control of communicable diseases, maternal mortality, and children malnutrition. Nonetheless, the epidemiologic transition accompanied with the process of population aging [2, 3] can bring changes that occur faster than the institutional ability to respond to them. This possible lag in institutional response may undermine the efficacy of this kind of services, especially in chronic disease management.
The article's first goal-the assessment of metabolic control among the older population in Costa Rica-serves as an introduction for the second goal: to study whether metabolic control among EBAIS patients is different than metabolic control among patients at other health care settings. The working hypothesis was that EBAIS older patients with DM had metabolic control levels equal or better than diabetic elderly patients at other clinical settings because EBAIS are aimed to provide only primary health care services, and patients who need specialists are then referred to physicians at hospitals or clinics. The present article shows that, among the elderly population with DM in Costa Rica, there are no major differences in poor metabolic control between patients from public hospitals and clinics on one side, and patients from community health centers on the other. However, community health center patients are more likely to have elevated levels of LDL and triglycerides. Since worse levels of metabolic control are not found for the other studied biomarkers (HbA1C, HDL, SBP, and DBP), these results may mean that lipid control for elderly diabetic patients at community health care centers is not as thoroughly examined as at public hospitals and clinics, or at private clinical settings, perhaps because many EBAIS lack laboratory facilities. No differences in lipid-control medication intake or frequency of cholesterol examination were found across clinical settings; besides, differences in triglycerides and LDL levels are not explained by differences in obesity or caloric intake, given that the multivariate models control for the effect of these variables. These differences may also be related to particularities in DM education at community health centers.
It is important for the public health care system to address the high levels of blood pressure as well as the poorer lipid control among patients that go to EBAIS because these community health centers aim to improve the access to primary health care to underserved and poor populations, and primary health care is an effective way of managing chronic diseases such as DM by decreasing the need of secondary and tertiary health care services [40, 41]. Even in industrialized countries such as the U.S.A., community health centers have been seen as an inexpensive way of providing preventive services to underserved and poor populations [31, 42, 43]. However, in terms of DM control, community health centers in other countries have not always been successful [30, 31, 43].
The advanced stage of the epidemiological transition in Costa Rica is happening in the context of a health sector reform. The community health centers, a key component of the reform, are increasingly providing primary health care services to elderly patients with chronic disease. If the system of community health centers in Costa Rica turns to be successful in reducing the burden of both communicable and chronic diseases, the Costa Rican health sector reform can serve as an experiment for other health sector reforms that international organisms are fostering in developing and transitional countries .
It is also worth noting that there are no differences in metabolic control between patients treated at public services and patients treated at private services, except in the frequency of DM examination by physicians. Diabetic elderly patients that visited a private outpatient service or that were treated at their own house are more likely to lack physician examination at least once a year. It is not clear whether this difference is due to financial barriers or medical practices, but the results should call the attention to users and providers of private health care in Costa Rica.
The analysis in this paper is performed using data from a nationally representative survey of elderly in Costa Rica that gathers self-reported information, as well as biomarkers. This kind of dataset is rare in developing countries. However, there are limitations in the data. An important one is that the study relies on self-reported information which might be affected by reporting bias. Diabetic respondents in the survey were determined based on self-reported answers to the question "Has a doctor or medical personnel ever told you that you have diabetes or high blood sugar levels?" This approach might overestimate the size of the actual diabetic population, and introduce biases in summarizing the outcome variables. If a physician ever told a person that she had impaired glucose tolerance or a one-time elevation in glucose, it is possible that this non-diabetic person will answer "yes" to the question in the survey instrument. We consider that it is inconvenient to validate the self-reported diagnosis with metabolic data because metabolic data is the outcome of this article. Therefore, any correction based on the biomarkers will bias upwards the prevalence of poor metabolic control. A way to explore the bias due to self-report is by using complementary information. The survey inquires about insulin use and other prescribed medication that respondents have in their house. Interviewers ask to see all prescribed drugs and enumerate them. Thirteen percent of the sample (weighted estimate) report not to be taking insulin or oral medication and also lack diabetes-related medication in their houses. If excluding this group, results are roughly the same although the mean levels of the biomarkers and the prevalence of poor metabolic control are slightly higher than with the original sample, but the differences are not statistically significant (α = 0.10). The most important differences between the two sets of results are that the difference in prevalence of LDL ≥ 100 mg/dl between community health center patients and patients at other health centers widens, especially among patients with statin medication, and that the percentage of elderly diabetic people without physician examination during the last 12 months drops from 10% to 4%. We decided not to exclude respondents without DM medication from the analyses because, even though some of them might be persons that do not have the disease, others might be diabetic patients that do not care about the treatment for their illness. Therefore, the analysis would be based on a selected population of people that do care about controlling their disease. We do not report the results based on this subsample that excludes these potential non-diabetic respondents, but we can provide them upon request.
The limitation due to self-reported data is also true for the main explanatory variable, place of last outpatient visit, as well as for most of the control variables. Reporting bias may arise from people that identify community health centers as clinics. Associated with this limitation, there is the problem that the survey questionnaire asks for the place of the last outpatient visit, but the questionnaire inquires neither whether the person visited another health care setting before the last outpatient visit, nor the reason of the last outpatient visit. It is possible that an undetermined proportion of the diabetic older population that reported last outpatient visits might refer to consultations about health problems other than DM-related. We believe this lack of information would not introduce a strong bias in the estimates, insofar as physicians are supposed to be examining patients' DM control, regardless of the reason for the visit. However, this kind of bias can not be discarded.
Another data problem is its limited statistical power to detect some differences, because the sample size of all DM patients is just about 500. Given that the subsample size of EBAIS patients is around 180, and the levels of metabolic control at the EBAIS range from 38% to 84% (depending on the indicator of metabolic control), the logistic regressions estimated for the final analyses have a statistical power of 0.80 to detect only odds ratios that are larger than 1.75 (or smaller than 0.55) for most of the biomarkers, and larger than 2.50 (or smaller than 0.30) for annual physician examination. Statistical power decreases even more when subsamples based on medication therapy are analyzed rather than the whole diabetic subsample. Power calculations are based on formulas for logistic regression analysis .