Correction Models of Pneumonia Mortality Rates in the Elderly in Brazil from 2005 to 2016

Objective: To estimate the burden of pneumonia (PNM) mortality in the elderly in Brazil using different methodologies for rate correction. Method: A time-series study of mortality rates from PNM in the elderly from 2005-2016. For the correction of the rates, two strategies were used: 1-correction by active search (CBA) and 2-correction by active search more redistribution (CBAR). In addition, a methodology for comparing corrected rates, calculated lower respiratory tract infection rates (ITRi) using estimates from the Global Burden Disease study. Trend was as-sessed using regression models. Result: From 2005 to 2016, 546,068 deaths were registered, which, after using the CBAR correction strategy, left 501,380, a reduction of 8.2%. There was an increasing trend of PNM mortality rates for all regions of Brazil and for the elderly population. The largest linear increase in mortality rate due to CBAR pneumonia was observed in the Northeast. Conclusion: These estimates of the magnitude and trend presented make it possible to analyze the behavior of mortality rates due to pneumonia in the elderly in Brazil and to subsidize the National Immunization Program on the need to include a specific pneu mococcal vaccine for the entire elderly population.


Introduction
Since the beginning of the use of mortality data, which occurred in England in the middle of the 16th century, the information has enabled the construction of indicators that can characterize the epidemiological profile of a population. Fact that contribute to monitor health conditions and support the implementation of public health interventions targeted at a specific population. In Brazil, the Mortality Information System (SIM), implemented in 1975, has accumulated, over time, improvements in its performance, both in increasing its coverage and in the quality of information on the cause of death [1][2][3]. This improvement is possible due to the efforts made by SIM managers at different levels of performance. However, this work has a high cost and requires qualified professionals [4]. It is observed that there are still quality problems in the registration of the basic cause of death in the Death Certificate (DO), with high percentages of little useful codes (garbage codes -GC), registered as the basic cause of death [5]. These GC are contained in the various chapters of the 10th edition of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) [6].
In addition, another problem is the lack of completeness of the information in variables present in the DO. Common examples of incorrect completion of the DO are statement not of the underlying cause but of its consequence; or just the commitment of a specific organ or system; or even the generic name of the disease, without the necessary specifications, among other examples [4,7]. Some strategies were implemented by the Ministry of Health (MS) in order to reduce the underreporting of deaths in the SIM and improve the quality of filling in the cause of death. The following stand out: a) The research "Active Search for Deaths and Births" [8], whose purpose was to assess the underreporting of deaths in the SIM; b) The project to reduce ill-defined causes (chapter XVIII of ICD-10); c) The project to reduce the less useful GC codes [8,9], linking mortality data with the hospitalization information system; and d) Implantation of verbal autopsy forms for deaths without medical assistance.
These initiatives have been successful in terms of improving SIM coverage and also in the significant reduction of GC in chapter XVIII of ICD-10, but the challenge of reducing the other CG in the other chapters of ICD-10 remains, especially the causes not specified within the chapters. Currently, an intervention project is underway with medical professionals in large hospitals in the metropolitan regions of the country with the aim of qualifying the completion of the DO, in order to reduce the percentage of GC [2].
Advances in the SIM computerized system, together with improved coverage and quality of information, have enabled the development of research using its databases. The use of linkage between databases is a good example of this evolution. This technique allows to qualify the data of the information system when relating the same individual to another information system, such as, for example, the Hospital Information System (SIH), which is responsible for processing and registering all hospital admissions financed by the Unified Health System in Brazil (SUS) [4,10,11].
A recent study that used the linkage strategy between these two  The second strategy, which is a correction factor for PNM mortality data recorded in SIM, was carried out based on the results of the study on the redistribution of the basic cause of death using data matching between SIM and SIH-SUS from 2008 to 2013 [10]. The ICD-10 codes registered in the primary diagnosis field (or secondary diagnosis, only in the case of external causes) of the AIH were used to review the basic cause of death, taking into account the diagnoses of hospitalization SIH other than PNM. Seattle (IHME). In defining this new mortality rate in the elderly, the deaths estimated by GBD were used as a numerator in the group of causes of death due to ITRi using the estimates of the GBD study [13]. The estimate by brazilian regions was possible due to the cooperation agreement between the MS of Brazil and the IHME for the production of the GBD indicators for the Federation Units (UF).
GBD methodological details are available in other publications of the group, which updates its procedures and concepts since the first publication [14].
As GBD's estimates of the number of deaths by ITRi (ICD-10: A48,1, A70, J09-J15,8, J16-J16,9, J20-J21,9 and P23,0-P23,4) are available on the IHME online page by UF in Brazil, by year and by age group, the number of deaths by region was calculated from the sum of the UF [15]. Thus, the GBD mortality rate was calculated by adopting, as a numerator, the death estimate by ITRi; and, as a denominator, the same population previously described, by region, age group and year. For the analysis of the trend of mortality due to SC pneumonia, CBA, CBAR and ITRi-GBD, dispersion diagrams between the coefficients and the years of study were made in order to visualize the function that could be expressing a relationship between them. From this relationship, it was possible to choose the simple linear regression model (Y = β0 + β1X), which, in addition to its statistical power, allows to define a forecast function [16]. In this model, β0 is the average annual coefficient and β1 is the coefficient of linear effect (speed). For modeling, the mortality coefficients were considered as a dependent variable (Y axis) and the years of the study series as an independent variable (X axis).
For the adjustments of these models, classic assumptions were verified, such as normality, independence and homoscedasticity of the residues [16]. The Durbin-Watson test was used to observe the autocorrelation between the residues. It was adopted as a measure of precision of the model the use of the coefficient of determination (R2). A significant trend was considered as one whose estimated model obtained a significant regression coefficient (p <0.05

Results
In the period from 2005 to 2016, 546,056 deaths in the elderly were registered in the SIM in Brazil, with the primary cause of PNM.
After applying the correction factor (CBAR), 501,380 records were considered as PNM, which corresponds to an 8.2% reduction in deaths with PNM registration as the basic cause in the period. Table   1 and Figures 2-4 show mortality rates due to PNM in the elderly without correction and those corrected during the study series.

Discussion
The main results of this study were a) Estimation of mortality rates due to PNM, using correction methods for underreporting deaths [8] and the basic cause of death [10]; and b) The analysis of the trend in mortality rates due to PNM in the elderly by regions and age group.
There was a progressive increase in mortality rates due to PNM in elderly people in Brazil directly proportional to age; it was also possible to describe the differences between the geographic regions of the country.
Brazil is a country of continental dimensions and important social and economic differences in each region [18,19]. In this study, the Northeast region, followed by the North, were those that showed the highest increase in the mortality rate due to CBAR pneumonia in all age groups. In this sense, research has shown that these two regions present disparities in the indicators of education, material deprivation and use of health services when compared to the others [19]. Recent national research on the trend of mortality due to PNM in Brazil also lists the Northeast and North regions as being disparate in relation to the others [20]. There is no doubt that the increase in investment in public policies aimed at vaccination is associated with the relevant reduction in general and specific mortality due to PNM and influenza, as well as the reduction of health inequities in the country [21]. An important result of this study was the identification of a tendency to increase the magnitude of mortality rates due to PNM in elderly people in Brazil in all regions of the country, in the period from 2005 to 2016.
This increase in mortality was directly proportional to the increase in the age group of the elderly. Studies have shown an increasing trend both in mortality [20] and in morbidity due to PNM in the elderly in Brazil, even after the implementation of intervention strategies [22]. With advancing age, in addition to increasing exposure to environmental factors, the body's defense mechanisms reduce the ability to fight pathogenic microorganisms, increasing susceptibility to communicable diseases (immunosensitivity or immune aging), which is aggravated by the multicomorbidities present in this age group [6,7,[23][24][25]. Such an assertion may be associated with the increase in mortality rates in elderly people aged 80 years or older observed in this research. This increase is in contrast to what has been observed in other countries that implanted conjugated pneumococcal vaccination for children longer than in Brazil, in addition to having also implemented it for adults [23,26,27]. Although in these countries the number of deaths due to ITRi in the elderly is higher than in other age groups, mortality rates have been decreasing.
The main speculation is that it is associated with the effect of herd immunization and also with vaccination in the elderly themselves [23,26]. Until the completion of this study, the 23-valent polysaccharide vaccine was available only in private clinics, with limited access to the elderly with greater purchasing power.
The vaccine is only available free of charge for special groups of hospitalized elderly and/or residents of long-term care facilities, as well as for people with chronic and immunodepressed diseases [28]. There is a need to direct efforts to increase the availability of the vaccine to all the elderly in the country. Despite the important improvement in the coverage and quality of the information contained in the SIM [2,3,29], there are still regional inequalities in terms of filling the causes of death in the DO, which makes it difficult to know the 'true' burden of PNM in the Brazil. In addition, it is important to emphasize errors in filling out and coding the causes of death in DO. These are observed, more frequently, in the elderly population due to the multiple comorbidities presented that make it difficult to define the real basic cause of death [6,7,23].
Part of the deaths registered as PNM are, in fact, an associated cause or do not represent the basic cause of death [6,30,31]. A study carried out in 2010 in Brazil reclassified 65.5% of ill-defined causes [3]. Another study redistributed 67% of the cases of neoplasms after applying linkage between databases [32]. In 2018, a study redistributed 17.4% of the cases in which PNM was considered a basic cause of death [10]. Still in that same study, after applying the strategy to redistribute the basic cause of death, the mortality rate due to PNM in Brazil reduced seven deaths per 100,000 inhabitants. In particular, the study presented here adopts CBAR strategies, correction of under enumeration and overestimation of deaths, as a more appropriate estimate of the burden of PNM in the elderly population for Brazil. The use of these two strategies allowed to know more robust estimates of mortality rates due to PNM in the elderly in Brazil, which contributed to the reduction of garbage codes.
The use of ITRi-GBD rates served as a parameter to compare two proposed correction methodologies: CBA and CBAR. Although a previous study showed an increase in both the number of deaths and mortality rates due to ITRi in the population aged 70 years or older [33], in this research the trend of ITRi-GBD rates showed a more stable behavior during the series when compared with CBAR rates.
The correction factors used by the GBD methodology to project the rates appear to be much higher at the beginning when compared to the end. This fact can be seen in the improvement in the quality of SIM data in recent years [2,3,29]. There are several actions taken The adoption of this strategy, CBA, has already been applied in other studies with the same purpose, which is to correct the underregistration and bring it closer to values closer to the real [7,35].
But in this study, we corrected in addition to the under-registration the coding errors of the basic cause. Another aspect that should be emphasized is the lesser protection of the herd conferred by infant 10-valent pneumococcal vaccination in the extreme elderly, in addition to the low effectiveness of the vaccine for disease prevention in the elderly when compared to young adults. It is also important to highlight the increase in the circulation of non-vaccine serotypes in this population [36].
This study enabled estimates closer to the reality of mortality rates due to PNM in the elderly in the five regions of Brazil, since both the under-registration correction (CBA) was adopted, which in theory would increase the number of notifications, as well as errors codification through the redistribution of deaths by PNM to other causes. The main limitations of the study were the presence of a certain degree of uncertainty in the estimated rates, even after using the proposed methodology. The additional use of basic cause redistribution research may present limitations related to possible biases that may have occurred in the database linkage process, such as different patterns of cause of death between cases that were paired and not paired [10]. Despite the limitations listed when using data from the MS information systems, it is believed that the technique proposed in this study represents a viable and adequate alternative for estimating the mortality burden of PNM in the elderly. In addition, studies that include validation and reliability strategies for improving methodologies to correct information on the basic cause of death are necessary and should be encouraged to improve the SIM.

Conclusion
This study proposed and used methodologies to correct the mortality rate due to PNM in Brazil that can be applied in analyzes for other groups of causes. The results showed that in 2016 there were more risks of mortality from PNM in the Southeast and North regions; and that the trends of increase in mortality rates due to PNM in elderly people aged 60 years or older in Brazil, with higher rates of increases, were in the Northeast and North regions of the country. The magnitude and trend estimates presented in the study make it possible to analyze the behavior of mortality rates due to PNM in the elderly in Brazil. In addition, they can subsidize the National Immunization Program on the need to include a specific pneumococcal vaccine for the entire elderly population.