The Impact of Inequalities and Health Expenditure on Mortality Due to Oral and Oropharyngeal Cancer: A 20-Year Trend Analysis

Objectives: To assess the magnitude and trend of mortality rates due to oral (OC) and oropharyngeal cancer (OPC) in the 133 Intermediate Geographic Regions (IGR) of Brazil between 1996 and 2018 and to analyze its association with sociodemographic variables and provision of health services. It also aims to compare the trend of mortality from neoplasms that have been reported as associated with HPV (OPC) with the trend of neoplasms that have been reported as not associated with HPV (OC). Methods: We obtained mortality data from the Mortality Information System in Brazil and analyzed the trends using the Prais-Winsten method. Then, we assessed the relationship between mortality trends and socioeconomic, health spending, and health services provision variables. Results: The median of the APC of the country’s mortality rates was 0.63% for OC and 0.83% for OPC. Trends in mortality in the IGRs correlated signicantly with the Human Development Index and government expenditure on ambulatory health care and hospitalizations. Conclusions: Mortality from both types of cancer decreased in those IGR in which the government spent more on health and in the more socioeconomically developed ones. This study found no epidemiological evidence that HPV plays the leading etiological factor in OPC in Brazil.

and enormous social inequality, the lack of knowledge about the dynamics of the disease and their relationship with the characteristics of the territory are obstacles to the planning of actions tailored to the needs of the population.
The description of the magnitude and trend of cancer incidence and mortality rates is crucial for understanding its evolution in the population context. There is a signi cant variation in the distribution and temporal trends of oral and oropharyngeal neoplasms worldwide, justi ed by the different exposure patterns to risk factors. Analyzing the trend in the incidence of these diseases, Chaturvedi et al. [8] veri ed a variation in the etiological pro le of OC and OPC between high and low/middle-income countries. The authors identi ed an increasing pattern in OPC incidence, contrasting with a decreasing (or stable) trend in OC incidence in high-income countries. This difference was attributed to the growing role of the HPV virus in oropharyngeal cancer's etiology -which is currently considered the leading risk factor for OPC in some countries, like the USA [9] -, concomitant with the reduction in the prevalence of tobacco consumption in these countries -the most classic risk factor for OC [10]. Concerning low and middle-income countries, tobacco still seems to be the leading risk factor for both types of cancer [8], but this assumption still needs evidence. Like Brazil, most countries do not have cancer databases that inform, for example, the HPV serology of tumors. Knowing the epidemiological behavior of this disease, from a perspective that can elucidate aspects related to its etiology, is necessary for the timely prevention of this condition, as it allows identifying changes in its patterns.
In 2017, the Brazilian Institute of Geography and Statistics (IBGE) divided the Brazilian territory into 133 Intermediate Geographic Regions (IGR). The IGR corresponds to an intermediate geographical strati cation between states and cities. These territorial delimitations always include large urban centers as a reference for urban functions of greater complexity to the neighboring towns [11].
Among other characteristics, IGR tends to cover the path through the health system that a cancer patient, resident of these, goes through; larger cities act as a reference (of different natures, including health) for smaller cities. This characteristic makes IGR an opportune territorial division to study the determinants of mortality from OC and OPC, identifying which factors are associated with more favorable regional outcomes. Nevertheless, to this date, no study examined mortality due to OPC and OC varies by IGR.
The present study aims to assess the magnitude and trend of mortality rates due to OC and OPC in the IGR of Brazil between 1996 and 2018 and to analyze its association with variables of a sociodemographic nature and provision of health services. Additionally, it aims to compare the trend of mortality from neoplasms that have been reported as associated with HPV (OPC) with the trend of mortality from neoplasms that have been reported as not associated with HPV (OC) [12], respecting current knowledge, that considers these two diseases as distinct clinical and epidemiological entities. By detailing this analysis by IGR and understanding how these rates behave in different pro le territories, we expect to counterbalance the over-generalization present in most existing studies.

Methods
This ecological study used data from the Mortality Information System (SIM, for the acronym in Portuguese) of Brazil from 1996 to 2018. The study period begins in 1996, when the SIM adopted the tenth revision of the International Classi cation of Diseases (ICD-10) [13] and ends in 2018, which is the most recent year with available data in the SIM. The database encompasses all deaths whose underlying cause was oral cancer (C02.0 -C02.3, C02.8, C02.9, C03 -C06) and oropharyngeal cancer (C01, C02.4, C09, C10 e C14.2).
The selection of which ICD-10 codes make up each type of cancer was made based on the proposition of Chaturvedi et al. [14]; it is described in Supplementary Chart S1. We also collected information about sex, age, and residence local of each case. The 5570 Brazilian towns were gathered into 133 Intermediate Geographic Regions (IGR) of the country. The IBGE created this territorial division as an intermediary administrative hierarchical level encompassing nearby towns within the same state [11]. Death rates due to oral and oropharyngeal cancer were calculated per 100,000 inhabitants for each IGR. The rates were adjusted by sex and age group ( ve-year intervals), using the direct method and the population distribution proposed by the World Health Organization [15].
Deaths with absent information on sex, age, or local of residence were redistributed based on the proportion of deaths with existing information. Deaths by ill-de ned and unspeci ed causes were also redistributed proportionally, according to the method of reclassifying garbage codes proposed by the Global Burden of Disease Study 2010 [16].
The outcome on cancer mortality and all covariates were calculated for each IGR. The HDI of each IGR was obtained by a weighted average of the index in each town, using the number of inhabitants of the town as the weighting factor. The FHS coverage was calculated using an analogous scheme. The number of FHT was divided by the population in the respective year, multiplied by 3000, and, nally, multiplied by 100 to obtain the percentual coverage for each town. According to the Ministry of Health, this number (3000) is the ideal number of inhabitants for each FHT [17]. Government expenses related to outpatient procedures and hospitalizations per capita were calculated directly for each IGR: the total annual amount, in Reals, was divided by the population in the same year.
For the trend analysis of mortality between 1996 and 2018, we applied a generalized linear regression model using the Prais-Winsten method, in which the dependent variable was the log transformed death rate, and the independent variable was the year of death. This method allows assessing the trend with correction for rst-order autocorrelation. The assessment of the annual percent change (APC) and its respective 95% con dence interval (CI 95% ) used the formula recommended by Antunes and Waldaman [18]: In which b1 is the regression coe cient, and b1lower/b1upper are the limits of its CI 95% . This procedure allows to classify the rates as ascending (APC and CI 95% positive), declining (APC and CI 95% negative), or stationary (CI 95% that includes the zero). The APCs were assessed for each IGR and thus aggregated by the Brazilian macro-regions: North, Northeast, Southeast, South, and Midwest.
We used Pearson's correlation to analyze and quantify the association between the APCs and the covariates HDI, FHS coverage, and public expenditure on health. The IGRs were classi ed into quartiles for each covariate. All analyses were performed using the Stata 14.0 software.

Results
From 1996 to 2018, 73,563 people died from OC, and 63,970 died from OPC in Brazil. The median of the APC of the country's mortality rates, considering the trend of mortality in the 133 IGRs, was 0.63% for OC and 0.83% for OPC. In general, median APCs were higher in the North and Northeast for both types of cancer (Table 1).  Tables S1-S5). When comparing the sexes, the median APC was higher for men than for women in all regions, except for the South. The most expressive discrepancy between the sexes occurred in the North, and the most discreet in the Southeast (Fig. 1).
Regarding trends of OPC mortality, the biggest median APC occurred in the Northeast (2.66%), whereas the lowest was in the Southeast  Tables S1-S5). When comparing the sexes, the median APC was higher in men than in women in all regions (Fig. 1).
Trends in mortality from OC and OPC in Brazil's IGRs correlated signi cantly with the HDI, FHS coverage, and government expenditure on ambulatory health care and hospitalizations. IGRs in the South and Southeast regions had higher HDI and per capita government expenditure on outpatient procedures and hospitalizations than the remaining regions. In contrast, IGRs in the Northeast region had higher FHS coverage by (Fig. 2). IGRs with low HDI tended to present with steeper increasing trends of mortality, i.e., the APC for OC and OPC was inversely associated with the HDI. The correlation between the APC and government expenditure on outpatient procedures and hospitalizations followed the same pattern, except for the mortality of women. The correlation between mortality trends and FHS coverage, however, was positive and statistically signi cant ( Table 2), depicting that the health program (the FHS) provided a higher coverage in the IGRs with higher increase of mortality. When it comes to compare the magnitude of mortality, the richer regions of the country, the Southeast and South, had an overall reduction of death rates from the beginning (1996-1998) to the end (2016-2018) of the study period. However, the opposite occurred in the poorer regions, the North and Northeast, where the rates were higher in the end (Fig. 3).
The median APC of mortality countrywide was higher for OPC than for OC (0.83% and 0.63%, respectively). The same occurred when speci cally assessing mortality in men (1.43% and 0.88%). However, this pattern was not homogeneous across the macro-regions.
When comparing the two types of cancer -for both sexes -only the Midwest had discrepant trends: the median APC was negative for OC and positive for OPC (Table 1).

Discussion
This study assessed the trend of mortality from OC and OPC in the last two decades in Brazil and the relationship between these trends and the region's human development, provision of health services, and governmental health expenditures. We found that mortality from both types of cancer decreased in those IGR in which the government spent more on outpatient procedures and hospitalizations and in the more socioeconomically developed ones. The ndings suggest that these elements may be effective in reducing mortality from OC and OPC in the country. These ndings are the most important result of the study.
In the analysis of association with health service provision, we found evidence pointing to contrasting scenarios. in ve years [20]. The ndings related to outpatient expenditures follow the same logic. IGR that have this level of care more productive may have provided more opportunity for early diagnosis, which is a predictive factor for survival in head and neck cancers [21].
The ndings regarding FHS coverage do not follow the same reasoning. Coverage by these teams represents access to primary health care, which is the main entryway to the Brazilian Uni ed Health System (Sistema Único de Saúde). A study carried out in Brazil indicated that coverage by Oral Health Teams -which are part of the FHS -was inversely associated with death rates in men from OC and OPC. The authors justi ed the results explaining that the FHS promoted the expansion of access to health in unassisted areas [22].
Our results indicated that IGRs with more expressive coverage by FHS have greater APC in mortality. We suggest that they are re ecting the pro-equity character of the implementation of the FHS, which prioritizes more vulnerable regions. Studies that evaluated the impact on the population's oral health have rati ed this [23]. Thus, coverage by FHS would be acting as a proxy for the IGR's social development in the analysis. Another possibility is that the IGRs with the broader coverages are those that historically had a more insu cient health provision. In this perspective, the FHS compensates for a historical de ciency not because it prioritizes the most deprived regions but because it is the only health care service available -as it is a more nancially feasible solution. Our results suggest that the structuring and expansion of health services in Brazil must occur networked and sync with intersectoral policies for social development to compensate for the inequities in illness and death due to OC and OPC.
We identi ed that the HDI showed a negative association with the mortality trend, for both types of cancer, considering both sexes and each separately: the more developed the IGR, the greater the decrease in mortality due to OC and OPC. The HDI, which summarizes the dimensions of longevity, education, and income, is lower in the North and Northeast. Explaining this association is complex, as there are several dimensions related to the social determinants of health [24] involved in this relationship. Socioeconomic factors appear to be involved in OC and OPC mortality since exposure to risk factors. A study carried out in Scotland found that, for individuals living in disadvantaged communities, smoking is a mechanism to deal with the stress generated by personal struggles, including economic de ciencies. The obstacles of living in an environment with few opportunities and few resources would intensify these di culties [25].
Food pattern, inputs and protective or harmful environments [26], occupational exposures [27], and access to health services and health information [5] are all aspects potentially related to OC and OPC and which may be in uenced by socioeconomic characteristics. The negative association between mortality and HDI reported here is compatible with previous studies [28][29][30][31].
A substantial number of studies on the incidence of OC and OPC in high-income countries have shown trends of increase in the rates of incidence due to OPC in the last two decades, contrasting with a decreasing or stable trends for OC [8,[32][33][34]. This epidemiological pattern has been associated with the rising role of the Human Papillomavirus (HPV) in the etiology of OPC [35], which is not observed in the OC. In the present study, the similarity in mortality trends for both types of cancer suggests that the predominant risk factors are common and are operating with similar intensity in both anatomical regions, which is not compatible with the course of HPV. This conclusion ts with results of Anantharaman et al. [36]. In a study of 1,420 cases of head and neck cancer, they investigated the percentage of oropharyngeal tumors positive for HPV-16, which was 60% in the USA, 31% in Europe, and only 4% in Brazil. From this perspective, the most classic risk factor (tobacco use) would be operating as a leading etiological factor for OPC incidence and mortality.
We identi ed higher APCs in mortality rates due to OPC compared to OC in all macro-regions, except for the Southeast. We would expect an opposite situation in a country where HPV is the protagonist of oropharyngeal carcinogenesis. HPV-related head and neck neoplasms are less lethal than non-related ones [37]. A recent multicenter study in Europe found almost 50% of reduction in the risk of death in HPV-16 positive cases of OPC when compared to negative ones [38]. Analyzing a cohort of 235 patients in Scotland, Wakeham et al. observed that patients with high-risk HPV positive OPC had 89% less risk of death than the negative ones [39].
This study used the Brazilian IGR as a unit of analysis. This construct is the product of a new territorial division, which considers the existence of a hub with more complex urban functions for each region. This division respects the concept of network-territory, which re ects the relationship of social subjects with spaces, incorporating their ows and their connections with different territories [11]. Oral cancer and oropharyngeal cancer are severe diseases, which require diagnostic and therapeutic resources of high complexity, usually present in large urban centers. The use of the territorial division in IGR was the strategy that was used to encompass the entire path of the sick individual, without the burden of over-generalization.
This study's main limitation is its data source, which is an information system -the Mortality Information System. In the last two decades, death records in Brazil have improved substantially [40]. However, its completeness and quality in identifying the underline cause of death vary between regions of the country. An international comparative study of cancer mortality excluded Brazil due to problems with registering the underline cause of death [41]. To counteract this limitation, we corrected the number of deaths of each IGR using the methodology of the Global Burden of Disease Study 2010 [16]. This approach, in addition to considering ill-de ned deaths, corrects the "garbage codes": the deaths registered with codes that do not reproduce the underlying cause of death (for example, "senility"). Besides, concerning primary health care coverage, we only measure FHS coverage, which is the principal strategy for organizing this level of care in Brazil, but not the only one.
The knowledge that factors of socioeconomic origin in uence mortality from oral cancer is well established in the literature. The results of this study indicate that the availability of this knowledge was not su cient to reduce the inequities related to this outcome in Brazil: the less developed regions had the greatest APC of mortality due to OC and OPC. This study innovates in addressing the IGRs as a population aggregate, which tends to encompass the path of patients with cancer in the health care network. It was possible to identify that mortality rates are decreasing in regions that spent more at the outpatient and hospital levels. This suggests that the investment in health care network is effective for this outcome. Finally, when comparing trends in mortality from oral cancer and oropharyngeal cancer, this study found no evidence that, in Brazil, HPV plays the leading etiological factor in oropharyngeal cancer.
However, this conclusion requires studies that measure the virus (and other signs of its role as an etiological factor) in these tumors and the incidence of these diseases.

Declarations Funding
This study was nanced in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES, Brazil) -Finance Code 001.

Autorship Contribution
ARC: substantial contributions to conception and design, acquisition of data, analysis, and interpretation of data; drafting the article; nal approval of the version to be published. AB: substantial contributions to conception and design, acquisition of data, and analysis of data; drafting the article; revising the article critically for important intellectual content; nal approval of the version to be published. JLFA: substantial contributions to conception and design, analysis and interpretation of data; revising the article critically for important intellectual content; nal approval of the version to be published. FNH: substantial contributions to conception and design, and interpretation of data; revising the article critically for important intellectual content; nal approval of the version to be published.