Disease and Economic Burden of Hospitalizations Attributable to Diabetes Mellitus and Its Complications: A Nationwide Study in Brazil

Diabetes is associated with a significant burden globally. The costs of diabetes-related hospitalizations are unknown in most developing countries. The aim of this study was to estimate the total number and economic burden of hospitalizations attributable to diabetes mellitus (DM) and its complications in adults from the perspective of the Brazilian Public Health System in 2014. Data sources included the National Health Survey (NHS) and National database of Hospitalizations (SIH). We considered diabetes, its microvascular (retinopathy, nephropathy, and neuropathy) and macrovascular complications (coronary heart disease, cerebrovascular disease, and peripheral arterial disease), respiratory and urinary tract infections, as well as selected cancers. Assuming that DM patients are hospitalized for these conditions more frequently that non-DM individuals, we estimated the etiological fraction of each condition related to DM, using the attributable risk methodology. We present number, average cost per case, and overall costs of hospitalizations attributable to DM in Brazil in 2014, stratified by condition, state of the country, gender and age group. In 2014, a total of 313,273 hospitalizations due to diabetes in adults were reported in Brazil (4.6% of total adult hospitalization), totaling (international dollar) Int$264.9 million. The average cost of an adult hospitalization due to diabetes was Int$845, 19% higher than hospitalization without DM. Hospitalizations due to cardiovascular diseases related to diabetes accounted for the higher proportion of costs (47.9%), followed by microvascular complications (25.4%) and DM per se (18.1%). Understanding the costs of diabetes and its major complications is crucial to raise awareness and to support the decision-making process on policy implementation, also allowing the assessment of prevention and control strategies.


Introduction
Non-communicable diseases (NCD) are the leading cause of disability and mortality globally, being responsible for 39.5 million deaths in 2015 [1]. Diabetes mellitus (DM) is one of the four major NCDs, together with cardiovascular diseases, cancer, and chronic respiratory diseases [2]. Diabetes prevalence is rising, representing a growing challenge to public health. A total of 415 million people

Ethics Approval
The Ethics Committee of Federal University of Goias in Goiania, Brazil, granted ethical approval for this investigation in October 2014 (# 852808). Considering we used secondary publicly available data, with no personal identifiers, the Institutional Research Board (IRB) waived written individual consent.

Diabetes Prevalence
The prevalence of self-reported diabetes was obtained from the 2013 NHS [5], stratified by gender, age groups and state within the country. The original NHS database was analyzed and estimates of self-reported diabetes prevalence were generated considering a positive response of surveyed individuals to question Q030 "Has any doctor ever given a diagnosis of diabetes?", excluding individuals reporting diagnosis of gestational DM (as responded in a different question of the survey). Evidence suggests that self-report of a physician's diagnosis of diabetes is a good estimate of diagnosed diabetes [20].
As individuals with diabetes who are unaware of the disease may also be hospitalized due to diabetes or its complications, we considered the prevalence of undiagnosed diabetes for this study. To account for undiagnosed diabetes, the prevalence of self-reported diabetes was multiplied by a factor of 2, based on recent evidence from the Brazilian literature, indicating that half of the individuals with diabetes diagnosis by laboratory confirmation were unaware of their disease [7]. This strategy has been applied by other authors for the estimation of diabetes disease burden [21].

Hospitalization and Cost Data
All hospitalizations occurring nationwide in SUS are recorded in a National Hospitalization Information System (SIH), which includes information on hospital admissions and discharges and its costs to the SUS. We obtained raw hospitalization data from SIH-SUS, without personal identification information, which are publicly available online [22].
A standardized hospital admission form (AIH) reports the main hospitalization diagnosis. We considered two types of AIH: AIH-1 (conventional hospitalization authorization) for the estimates of diabetes hospitalizations, and AIH-5 (long-term hospitalization authorization) considered in addition to AIH-1 for economic burden estimates.
International statistical classification of diseases and related health problems, 10th revision (ICD10) codes [23] assigned for admission diagnoses of all hospitalized patients are recorded in the SIH. The databases were extracted in October/2016, and the following variables were considered: type of AIH, state of residence, sex, age group, date of admission, admission diagnosis, and cost of hospitalization. The data were extracted and analyzed in Microsoft Excel (R) Office Excel (R) 2007 (12.0.4518.1014) MSO (12.0.4518.1014) spreadsheets.

Population Data
Considering 2014 as the base year for this cost analysis, population estimates for 2014, by age group, and gender for each state were obtained from the National Institute of Geography and Statistics. The total adult population (20 years and older) in 2014 was estimated at 137,640,060 inhabitants [24]. The age groups were divided in 5-year strata from the age of 20. Standardized hospitalization rates adjusted by gender and age group were calculated using direct standardization method.

Outcomes of Interest
We considered hospitalizations due to diabetes and its complications. Complications included microvascular (retinopathy, nephropathy, and neuropathy), macrovascular (coronary heart disease, cerebrovascular disease, and peripheral arterial disease), respiratory and urinary tract infections, as well as selected cancers. Hospitalizations for DM in pregnancy (ICD-10 O24) were excluded.
Hospitalizations were thus categorized into two groups: (1) those in which the main diagnosis was reported as diabetes and coded as ICD10 E10 to E14; (2) those in which the main diagnosis was reported as any chronic complication of DM and related diseases, including infectious and neoplastic diseases for which DM is considered to be an important risk factor. The list of diagnosis included in this second group was adapted from the one considered by the American Diabetes Association's study of economic costs of diabetes in the US [25], and included 66 diagnosis as coded by ICD-10 three-digit codes (Supplementary Table S1).

Relative Risks
For each and all chronic complications of DM and related conditions considered, we obtained individual relative risks of hospitalization for people with diabetes compared to those without the disease. Relative risk estimates for each diagnosis were obtained through systematic literature reviews (Supplementary Table S1).

Burden of Hospitalizations Attributed to Diabetes and Its Complications
Except for hospitalizations in which the main diagnosis was reported as DM, the proportion of hospitalizations due to DM was estimated using the attributable risk methodology. This method considers that diabetic patients use healthcare services more than non-diabetics and that a portion of the care associated with such medical care can be attributed to diabetes. The risk of presenting a particular medical condition, according to the presence or absence of DM, and the proportion of the population with the disease are combined to calculate the etiological fraction.
The etiological fraction for each of the 66 conditions considered were calculated using the following formula: where RAPI is the fraction of risk attributable to the population for the medical condition "i" due to diabetes, P represents the prevalence rate of diabetes in each state by gender and age group, and iRR is the relative risk of hospitalization for people with diabetes compared to those without the disease. A total of 702 demographic strata were generated, resulting from a combination of 2 sex categories, 13 age categories (i.e., 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, 65-69, 70-74, 75-79 and 80+ years), and 27 states. The number of hospitalizations for each of the conditions considered, by each of these 702 demographic strata (sex, age group, and state) was obtained.
Specific etiological fractions were then applied to the number of hospitalizations obtained for each of the 702 demographic strata, resulting in 46,332 estimates (66 ICD-10 codes multiplied by 702 demographic strata) of the proportion of hospitalizations attributable to DM.
Results were further grouped in four age groups (20-44, 45-64, 65-74 and 75+ years old), and are reported by sex, age group and main diagnosis. Diagnosis groups considered for reporting results are DM, cardiovascular disease, kidney disease, eye disease, neurological disease, infectious disease, and neoplasms.
Proportion of hospitalization and population hospitalization rates (per 10,000 population aged 20 years and older) are reported, comparing overall hospitalizations and those due to diabetes.

Direct Medical Costs Attributed to Diabetes and Its Complications
Economic burden analysis considered the SUS perspective as payer. A top-down costing methodology was used considering on the combination of prevalence and relative risks [26]. This methodology allocates to diabetes a portion of the total expenditures of hospitalizations (for several conditions) that could be due to the diabetes, based on the estimate of the proportion of total services consumed by individuals with the disease [27,28] as described above.
Hospitalization charges are based on Diagnostic Related Groups (DRG), with addition of values resulting from on intensive care unit (ICU) stay, certain special medications, prostheses and other selected materials. In addition to these direct medical costs, which include hospital stay, staff, diagnostic and therapeutic procedures, materials and drugs, non-medical costs for hospital stay of a parent or caregiver accompanying the hospitalized individual is also included. Reimbursed values by cost items are standardized nationwide based on SUS own price list [29].
Monetary values were obtained in Brazilian reais (R$) and then converted to international dollars (Int$) considering the purchasing power parity (PPP) (conversion factor 1.748) [30].
Total hospitalization costs and average cost per hospitalization, by diagnostic groups, and by specific hospitalization cause are presented, comparing all hospitalizations and those attributed to diabetes. Costs are further presented stratified by gender and age group.

Results
We considered the national prevalence of undiagnosed diabetes as 12.4%, varying by age group and state (Supplementary Table S2). As such, we estimated that 17,320,339 adult individuals in the country would have diabetes.
A total of 11.3 million hospitalizations were registered in 2014 in the SIH/SUS, of which 8,629,004 million (76.2%) were adults (20 or more years). Of these adult hospitalizations, 284,675 received an authorization for prolonged stay (AIH-5).
In 2014, an estimated 313,273 hospitalizations due to diabetes occurred in Brazil, corresponding to 3.6% of total hospitalizations and representing a hospitalization rate of 22.8/10,000 adults. Excluding hospitalizations for pregnancy, childbirth and the puerperium, hospitalizations attributable to diabetes represent 4.6% of total adult hospitalization in Brazil in 2014.
Among these, DM per se (ICD-10 codes E10-E14) accounted for 41.9% of hospitalizations, followed by cardiovascular diseases attributable to diabetes (26.5%) ( Table 1). The population hospitalization rates increased from 3.5 and 3.8/10,000 adults for men and women, respectively, aged 20-44 years to 146.0 and 133.3 for the age group of 75 and over. Women were hospitalized more than men when considering both absolute number and crude hospitalization rate. However, when considering age-standardized rates, these are higher for men (23.9/10,000 population) when compared to women (21.9/10,000 population). While the average cost of a hospitalization of an adult individual was Int$709 in 2014, the average cost of a hospitalization due to diabetes and related diseases was 19% higher, reaching Int$845. Among the hospitalizations due to diabetes, hospitalizations due to kidney (Int$1602) and cardiovascular (Int$1529) diseases were the ones with higher average cost, and hospitalizations due to diabetes had the lower average cost (Int$364). Average hospitalization cost was significantly higher in men in all age groups and for all diagnosis groups, except for selected neoplasms, probably because of breast cancer costs included in this group (Table 2).
Total costs for adult hospitalization in the SUS in 2014 were approximately Int$6.1 billion. Admissions due to diabetes and related conditions reached Int$264.9 million, representing 4.3% of total hospitalization costs. After excluding hospitalizations for pregnancy, childbirth and the puerperium, this proportion increased to 4.8%. Diabetes mellitus per se accounted for only 18.1% of total costs attributable to hospitalization due to diabetes and related conditions, with cardiovascular diseases attributable to diabetes (47.9%) accounting for the higher proportion of overall costs. Total hospitalization costs were significantly higher in men from 20-74 years. The reverse was observed in the age group of 75 years and older (Table 3).
Among the hospitalizations with the main diagnosis reported as DM, the number of hospitalizations (52.5%), and total costs (46.2%) related to unspecified DM ICD-10 code E14) were the most observed, despite presenting the lower average hospitalization cost. The second most relevant cause of hospitalization in this group was insulin-dependent hospitalizations (ICD-10 E10), with the higher average hospitalization cost (Table 4).
Cardiovascular diseases due to diabetes accounted for 13.1% (n = 82,958) of admissions and 14.3% (Int$126,849,817) of costs of all hospitalizations for cardiovascular diseases in SUS. In hospitalizations due to diabetes, average hospitalization costs due to cardiovascular disease were 10.4% higher than non-diabetes hospitalizations. Among all hospitalizations due to cerebral infarction (ICD-10 code I63) and transient ischemic stroke and related syndromes (ICD-10 code G45), 25% of hospitalizations and costs could be attributed to diabetes (Table 5).
Microvascular diseases due to diabetes (kidney, eye and neurologic diseases) accounted for a greater share of total hospitalizations (29.1%) and associated costs (24.5%). Of worth noting is the high number of hospitalization and overall costs with diabetes hospitalization due to renal diseases, in particular due to chronic kidney disease (ICD-10 code N18) ( Table 6).
Hospitalizations for respiratory and urinary infections, for which diabetes was considered a risk factor, represent a small percentage (5.3%) when compared to the cardiovascular (13.1%) and microvascular (29.1%) groups in relation to total SUS and accounted for 6.5% of hospitalizations due to DM. Even so, this percentage was reached due to the large participation of respiratory infections (96.5%) in this group, with emphasis on pneumonia per unspecified organism (ICD-10 J18) (69.9%) ( Table 7).

Discussion
Brazil is one of the most populated countries in the world, with an estimated population of 137.6 million adults in 2014 [31]. Based on recent prevalence estimates, we have estimated that 17.3 million individuals aged 20 years and older had diabetes in Brazil (Supplementary Table S2). Despite increasing trends in diabetes prevalence in the country, mortality due to diabetes declined 1.7% per year (from 40.6/100 thousand population to 33.7/100 thousand population) from 2000 to 2011, probably as a result of better access to healthcare, thus reducing mortality due to acute events [32]. However, when diabetes was analyzed as an associated cause of death due to other causes, there was an increase of 8% between 2000 and 2007 [33], most likely representing deaths due to chronic diabetes complications and related conditions.
Hospitalizations represent an important part of the consumption of health resources in different health systems and countries around the world and patients with type 2 diabetes had higher rates of hospitalization than the general population [34]. In the United States in 2012, diabetes hospitalization costs were the most significant cost component (43%) of direct medical costs ($176 billion) associated with diabetes, which added to $245 billion when considering both direct and indirect costs [25].
The estimated costs of hospitalizations due to diabetes and related conditions estimated in this study (Int$264.9 million) represent 4.6% of all hospitalizations and 0.45% of all expenditures for actions and public services of health provided by the Ministry of Health in 2014 (Int$58.3 billion) [35]. In this same year, total health expenditures in Brazil were 8% of its Gross Domestic Product of which 46% was associated with public health expenditures (Int$606 per capita) [36]. This spending is equivalent to Int$1.92 per adult by the federal government only with hospitalizations for DM and its complications. The average value of an adult hospitalization due to diabetes was 19% higher than a hospitalization without diabetes, and hospitalizations due to kidney and cardiovascular diseases were the ones with higher average cost.
Most countries in Latin America have adopted public health systems with universal coverage in the last few decades. Nonetheless, disparities in per capita government health expenditure can be observed in the region [37] and a wide difference can be identified between countries that share historic similarities, with Venezuela with the lowest (Int$270.88), and Cuba (Int$2366.06) the highest per capita government health expenditure [36]. When contrasting with high-income developed countries in other regions, disparities are more pronounced, with United States (Int$4541.17), United Kingdom (Int$2807.62), and Japan (Int$3115.08) among the highest per capita expenditures [36].
Our results demonstrated that the population aged 65 years and older used a much larger portion of hospital resources, both in number of hospitalization and costs, similar to results demonstrated in the United States in 2012 [25]. Cardiovascular complications attributable to diabetes also represented the largest share of all hospitalizations, both in number of hospitalizations and costs.
Although when considering crude rates, women were more likely to be hospitalized than men, when adjusting for age taking into consideration the different age structure between men and women, men are more likely to be hospitalized than women. Although men had relatively higher hospitalization costs than women from age group 20-74 years, except for the ≥75 years age group, this may be due to the relatively longer life expectancy in women, compared with men [38]. Hospitalizations reported as having diabetes as the main diagnosis were the most frequent (41.9%), although with lower costs. They currently represent a small proportion of all hospitalization expenses for the Brazilian National Health System, but are expected to increase considerably as the population ages. Moreover, hospitalization costs related to diabetes, but not captured by a first listed diabetes diagnosis, must be integrated with these costs to give a more comprehensive picture of the overall disease burden attributable to diabetes.
The total number of hospitalizations due to DM-related conditions was 2.4 times that of hospitalizations for first-listed DM; however, spending was almost 5.5 times higher. Microvascular diseases due to diabetes (kidney, eye and neurologic diseases) accounted for a greater share of total hospitalizations (29.1%) and associated costs (24.5%), part of which could be prevented with a better metabolic control. These results were in accordance with others that show hospitalizations for diabetes complications had a higher average cost than those for diabetes itself [25,39,40].
The hospitalization costs due to infectious diseases and selected neoplasms in adults with DM were 4.8% and 3.9%, respectively, of the total hospitalizations due to DM. Although it represents a small percentage compared with vascular diseases, currently, this was the first Brazilian study to consider DM as an important contributor to such hospitalizations and costs.
Comparisons with Brazilian studies for 1999-2001 [41] and 2008-2010 [42], that used the same attributable risk methodology to estimate hospitalizations for DM in the Brazilian public network, should be performed with caution. In both previous studies, the results encompassed all age groups while in the current, only adults. In addition, in the two previous studies, hospitalizations were also estimated for the general medical conditions group, i.e., all other ICD-10 diseases that are not attributed to diabetes or its complications but for which individuals with DM were hospitalized more frequently. In contrast, in the current study, some of these conditions, such as certain neoplasms and lower respiratory and urinary tract infections, were computed as diabetes complications. It is also recognizable that the more recent literature has brought lower relative risks from international studies for the calculation of etiological fractions, although it was partially offset by the double of self-reported prevalence.
Our study has several limitations which are worth noting. The source of the data (SIH/SUS) was initially developed for administrative-financial functions for the purpose of collection and may not be free of coding errors, intentional or not. This is reflected by the high number of individuals hospitalized for whom the main diagnosis reported as DM was "unspecified DM-E14" (n = 68,987), and "insulin-dependent hospitalizations-E10" (n = 38,883), significantly higher than those reported as "non-insulin dependent diabetes-E11" (12,707). We believe that most of these cases reported as E14 are indeed individuals with type 2 DM. Also, many of the cases reported as E10 may be individuals with type 2 diabetes using insulin.
In addition, the SUS covers 75% of the population in Brazil, which means about one quarter of population with diabetes was not included in the data analysis, and thus our estimates may be underestimated. Moreover, the diabetes hospitalization rates may be different among those not covered by SUS. As SIH data do not incorporate critical variables with explanatory potential, such as body mass index, race, schooling, severity of the clinical condition at the time of hospitalization, degree of use of the services, readmissions and other, we are not able to identify the role of these possible factors in diabetes hospitalizations.
Another limitation is we considered only the adult population, as the focus of the study was on type 2 DM, which is more amenable to prevention strategies. As such, although hospitalizations due to type 1 DM may have been inadvertently included in our estimation, on the other hand, we might have underestimated cases of type 2 DM in those younger than 20 years of age.
Finally, as the more recent prevalence estimate available was that for self-reported diabetes [5], to account for undiagnosed diabetes, and as done by other authors [21], we applied a factor of 2, considering high quality recent evidence from Brazil [7]. The resulting diabetes prevalence of 12.4% considered in our analyses is consistent with sub-national studies in Brazil which a decade ago showed 2 digit prevalence figures in selected regions of the country, being 12.1% in the city of Ribeirão Preto [43], 12.4% in Porto Alegre [44], and 13.5% in São Carlos [45]. It is also consistent with prevalence estimates considered for global disease burden estimate studies [21].

Conclusions
This study presented a detailed overview of the hospitalizations attributable to DM in the Brazilian public network. It is a study that deals with the epidemiological and economic aspects of the public expenses with a disease. They portrayed an "epidemiological iceberg" present in developing societies such as Brazil. By increasing the incidence and severity of other diseases, diabetes increases the chances of hospitalization of patients and the use of more aggressive therapies. We believe that improving the quality of life of these patients depends, among other measures, on the expansion of preventive activities in order to reduce the need for hospitalization, minimize complications and reduce the severity of other more general medical conditions. Our estimate is part of the monitoring and analysis of the health situation for the necessary interventions. Understanding the costs of diabetes and its major complications is crucial to raise awareness and allow the assessment of strategies to reduce its prevalence and their impact.