1 INTRODUCTION

Oral cancer (OC) remains one of the most common cancers worldwide,1 especially in developing countries.2 These tumours are responsible for approximately 145 328 deaths worldwide per year.1 In Brazil, the 5-year survival rate for patients with OC varied from 40%–60% depending on the study location,3, 4 and there is no evidence of any improvement.3 The Brazilian OC survival rate is lower than in other countries, as Índia (59%–79%),5 Korea (63.5%)6 or USA (62.9%–75%).7, 8 This may be due to late diagnosis and delayed treatment start.9, 10 Thus, despite the universal access that is constituent to the Brazilian Unified Health System (acronym in Portuguese, SUS), some challenges still need to be overcomed in establishing an effective OC care network.11

Prior studies carried out in Brazil indicated that a good service structure is associated with better work processes in primary health care (PHC).12 Also, greater coverage of oral health services in PHC13 and Dental Specialties Centers (acronym in Portuguese, CEO)12 were associated with lower OC mortality rates. In Brazil, the National Oral Health Policy (acronym in Portuguese, PNSB) and the National Oncological Care Policy (acronym in Portuguese, PNAO) were implemented in 2004 and 2005, respectively, to articulate primary, specialized, and hospital care, organize a care line and reduce OC burden. Within this context, CEO are public health services that are responsible for providing dental care in endodontics, periodontics, minor oral surgery, care for people with disabilities and diagnosis of oral mucosa lesions, with an emphasis on OC.12 CEO receive and treat patients by PHC facilities, improving the interface14 between primary and secondary dental care, playing an essential role in OC prevention and diagnosis, and contributing to comprehensiveness of dental care.15 However, evaluation of OC services’ quality, especially CEO, are scarce.16-18

As part of the efforts to improve quality of SUS services, the Brazilian Ministry of Health implemented the Program for the Improvement of Access and Quality of Dental Specialties Centers (acronym in Portuguese, PMAQ-CEO). PMAQ is one of the world's largest payment-for-performance (P4P) programmes globally,19 and uses a framework that evaluates aspects of the structure, work processes, and impacts of SUS’s services.20 Services that opt to take part of PMAQ complete a self-evaluation and receive a external evaluation from a pool of Universities that partnered with the Brazilian Ministry of Health to implement the programme. The evaluation of work processes comprise aspects that include governance, service organization, and planning.21 Quality improvement programmes based on P4P are uncommon in Dentistry, but are somewhat popular in medicine, especially in PHC.19, 22, 23 They may contribute importantly to improve services and providers’ performance in critical indicators related to OC diagnosis and management, and can potentially reduce disease burden. Notwithstanding, we have not found previous studies evaluating the findings of large-scale programmes based on P4P in OC care work processes. This study aims to analyse quality indicators related to OC care in CEO services, comparing changes over two PMAQ-CEO external evaluation cycles (2014 and 2018) using latent transition analysis (LTA).

2 METHODS 2.1 Study design

This is a panel ecological study,24 that analysed secondary data from the first (2014) and second (2018) cycles of the external evaluation phase of PMAQ-CEO.20 Aggregate variables for the CEO level were used as units of analysis. The study followed STROBE guidelines.

2.2 Study context and period

PMAQ-CEO was a P4P program implemented by the Brazilian Ministry of Health. CEO services could opt to take part voluntarily and the programme had as one of its goals promoting a culture of evaluating public services. The programme consists of successive cycles. Each cycle consists of three phases (admission, external evaluation, and certification/re-contractualization) and a cross-sectional centre line (development of actions).20 Performance evaluations were carried out during the programme's external evaluation phase in 2014 (first cycle) and 2018 (second cycle).

2.3 Setting

Brazil is the largest country of Latin America, with a population nearing 210 million inhabitants in 2018, distributed in 26 states and the Federal District, and 5570 municipalities. Brazil ranks 75th in the Human Development Index, with a value of 0.752. Brazil's Gross Domestic Product (GDP) per capita is 5521 billion Reais, and the Gini Index of household income per capita is 0.6086.

Brazil has five regions (North, Northeast, Midwest, South, and Southeast), with the North being the largest in territorial area, the least populated (3.77 inhabitants/km²), and the one with the smallest GDP. In contrast, the Southeast region, the second smallest in Brazil in terms of area, is the most populous and developed, with a population density of 92.0 inhabitants/km² and the country's highest GDP.

2.4 Participants

In 2014, 988 CEO were accredited by the Ministry of Health and 932 (94.3%) were evaluated in the first cycle of PMAQ-CEO. In the second cycle (2018), 1097 CEO were accredited and 1042 (95%) were evaluated. Losses were because they had recently lost their accreditation, or were not operating, due to the impossibility of accessing the CEO location, or due to refusal. In this analysis, only CEO that participated in both cycles of the programme were included.

2.5 Data collection process

In each cycle of the PMAQ-CEO, the external evaluation phase was carried out by a pool of universities that partnered with the Brazilian Ministry of Health. The questionnaires used in data collection were organized into three modules, and different data collection strategies were adopted.20 The first module evaluated the specialized dental care centre's infrastructure using a script for in loco observation. The second module evaluated the oral health care work processes by interviewing the manager and one dentist of the CEO, and also examined documents in the services. The third module focused on the quality of care, access, and service utilization and comprised the interview of ten patients who were waiting for a consultation in the CEO waiting area.

2.6 Variables

All the questions pertaining to OC care included in the external evaluation of PMAQ-CEO were analysed, as follows: five variables from the ‘Biopsy and referral to anatomohistopathological analysis’ block; two variables from the ‘Care for OC’ block; and other three variables from other parts of the questionnaire (related to the training of professionals, adherence to protocols, and waiting time for OC care). However, four variables were excluded from the final model due to the high collinearity with other variables, high number of missing data, or for interfering with the model's adjustment parameters, sometimes making the analysis impossible and not creating latent classes. After exclusions, six variables were used in the final model, as described below.

Two variables indicated the articulation of CEO with PHC (‘Training with PHC professionals to detect OC’ and ‘Adherence to clinical protocols for referring PHC patients to the CEO’). Two other variables pointed to the CEO’s ability to diagnose OC (‘Performs biopsies’ and ‘Has a laboratory for referrals of histopathological specimens’). There was also a variable for monitoring/recording of OC cases (‘Has records of users diagnosed with OC’), and an indicator of the articulation of the CEO with the hospital network (‘Referral of confirmed OC cases to the hospital network is available’). All variables had ‘Yes’ or ‘No’ answers. A supporting document was also requested for some of these variables (availability of a clinical referral protocol and of a register of users with cancer).

2.7 Statistical analysis

We estimated each variable's absolute frequencies and percentages, as well as the proportional difference between cycles 1 and 2. LTA was used to identify classes of adequacy for the work processes of the CEO related to OC care and to model the transition between these classes over time. LTA is a longitudinal version of latent class analysis (LCA).25 LTA is a robust statistical approach used to identify underlying subgroups in a population sharing similar characteristics, based on some observed categorical variables.25 The LTA considers that the units of analysis can change latent class over time; therefore, in LTA, the terminology ‘latent status (LS)’ is adopted instead of ‘latent class’.

We estimated three groups of parameters in LTA: (1) probabilities of the characteristics of units belonging to an LS; for example, when modelling the features of the work process of the CEO over time, these probabilities reflect the proportion of CEO with specific traits that belong to each LS; (2) prevalence of observations in each LS at each time; and (3) transition probabilities between LS over each time25 (we calculated the proportions in relation to the total CEO assessed and, additionally, in relation to the number of CEO that could improve or get worse). In the LTA, multiple aspects of a phenomenon can be accessed over time to jointly indicate the LS in a multivariable way.25

The first step of LTA was selecting the number of LS. Thus, we prepared models with different LS numbers, and we selected the ones with the best conceptual interpretability and adequate goodness of fit according to the statistical likelihood ratio test, p-values, degrees of freedom (df), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and entropy. We tested models with three to five LS. As our models had less than 99 degrees of freedom, we also considered the likelihood ratio Chi-Square test and the respective p-value for choosing the model. These parameters compare the tested empirical model with an ideal hypothetical model. The lower the test value and the higher the p-value, the more similar the empirical model will be to the ideal hypothetical one.25, 26 Low AIC and BIC values reflect a better balance between model fit and parsimony. Entropy close to 1 indicates a straightforward design of LS.26 We performed LTA with MPlus software, version 7.4 (Muthen & Muthen, California, United States).27

We prepared choropleth maps describing the distribution of latent transition patterns for the work processes in Brazilian CEO. They were categorized as ‘Better work process in both cycles’ when they were LS 2 in both cycles; ‘Improved’ when they changed from the LS 1 to any other LS, or when they changed from the LS 3 or LS 4 to the LS 2; ‘Average work process in both cycles’ if the CEO were LS 3 or LS 4 in both cycles; ‘Worse work process in both cycles’ when they were LS 1 in both cycles; or ‘Worsened’ when they changed from LS 2 to any other LS, or if they changed from LS 3 or LS 4 to LS 1. We conducted spatial aggregations by the municipality, that is, the maps reflect the proportion of CEO in the municipality with each of the transitions indicated above, classified on a 5-level scale: 0%, between 0% and 25%, above 25% and below 50%, above 50% and below 75%, and above 75%. We also performed hotspot analysis for each of the transition situations to identify nearby areas with a concentration of statistically significant ‘hot’ spots (areas with a high concentration of the event) and ‘cold’ spots (areas with small concentration of the event). The maps were designed by the ArcGIS program, version 14.0 (California, United States).28

2.8 Ethical considerations

The Ethics Committee of the Federal University of Pernambuco (UFPE) approved the project under process CAAE 23458213.0.0000.5208, on January 30, 2018, per the National Health Council's ethics resolutions. All participants provided consent to take part.

3 RESULTS

Nine hundred and sixteen CEO were included in the study. They were located in 766 (13.7%) Brazilian municipalities distributed in the five macro-regions of the country – 59 in the North (6.4%), 349 in the Northeast (38.1%), 332 in the Southeast (36.2%), 115 in the South (12.5%), and 61 in the Midwest (6.7%).

We observed an improvement for four of the six variables analysed referring to the work process in OC care, especially for the number of CEO providing continued education for PHC professionals to detect OC cases (increasing by 55% between cycles). However, we identified a 17.8% decline in the number of CEO that reported performing biopsies and 18.3% in those that had hospitals for referring confirmed OC cases (Table 1).

TABLE 1. Descriptive statistics of the variables used in latent transition analysis (N = 916) Latent status indicators (Oral health teams´ oral cancer work process) Time 1 (2014) Time 2 (2018) Variation N % N % % Providing training to PHCa professionals to detect oral cancer 353 38.5 547 59.7 55.0 Agreed clinical protocols for referring PHC users to the CEOb are in place 531 57.9 619 67.6 16.6 Performs biopsies 753 82.2 619 67.6 −17.8 Has reference for referral of histopathological specimens 696 76.0 729 79.6 4.7 Has records of patients diagnosed with oral cancer 500 54.6 665 72.6 33.0 Has a hospital service to reference patients who received an oral cancer diagnosis 734 80.1 600 65.5 −18.3

Table 2 shows the model fit information used to select the final model in this study. We chose the four LS model due to its better conceptual interpretability, and goodness of fit – less likelihood ratio Chi-Square, p-value >.05, entropy closer to 1, and high values of AIC/BIC (very close to those verified in the other models).

TABLE 2. Model fit information used in selecting the LTA model Number of Latent Status Likelihood Ratio Chi-Squarea Degrees of Freedom p-valueb AICc BICd Entropye Oral health teams´ work process 3 −5352.3 26 <.050 10756.6 10881.9 0.885 4f −5129.9 39 .831 10337.7 10525.7 0.934 5 −5436.7 54 .793 10981.4 11252.1 0.915

The top panel of Table 3 exhibits the probabilities of response to the item, depending on the LS association. The combination of probabilities of response to the item provides a notion of what characterizes the four different LS. CEO classified in LS 1 had an inadequate provision of most of the services listed (the worst situation). The CEO of LS 2 were those with the highest proportions of adequacy for the six items evaluated (the best situation). The CEO of LS 3 and 4 were mostly those with shortcomings in the organization of the OC care network (considered average situation). Those from LS 3 had an inadequate articulation with PHC (they did not perform continued education in OC care with PHC professionals and did not have clinical protocols agreed for referring PHC users to the CEO). Those in LS 4 were not well articulated with hospital care (there was no reference for referral of confirmed OC cases); the registration of users diagnosed with OC was also deficient in both LS. Table 3 also presents the prevalence of each LS, by cycle, and the transition probabilities. In the first cycle of PMAQ-CEO (2014), most CEO belonged to status 3 (N = 459, 50.1%), followed by LS 2 (N = 294, 32.1%) and LS 1 (N = 163, 17.8%). In the second cycle (2018), most CEO were classified in status 2 (N = 498, 54.4%). Comparing the cycles, 419 (45.7%) CEO did not change the LS (1→1, 2→2, 3→4), 269 (29.4%) improved (1→2, 1→4, 3→2), and 228 (24.9%) reduced their quality (2→1, 2→4, 3→1). If the proportion of changes are compared accounting to the impossibility of worsening for CEO that were in LS 1 in the first cycle 1, and for improving for CEO in LS2 in the first cycle, the proportions were as follows: 43.2% improved (269 of the 622 that could increase – excluding 294 CEO that were already in LS 2); and 30.3% decreased their quality (228 of the 753 that could decrease – excluding 163 CEO that were already in LS 1) (Table 3).

TABLE 3. Four-status model of quality and transition probabilities for the oral health teams´ work process in Brazilian CEO Latent Status 1 2 3 4 Worse work process Better work process Deficiency in referral and communication between CEO and PHC Deficiency in referral and communication between CEO and hospitals Probability of response to indicatora PHCb professionals’ training to detect oral cancer Yes 31.1 73.7 28.7 51.1 No 68.9 26.3 71.3 48.9 Agreed clinical protocols for referring PHC users to CEOc Yes 9.1 100.0 47.7 100.0 No 90.9 0.0 52.3 0.0 Performs biopsies Yes 0.0 100.0 100.0 100.0 No 100.0 0.0 0.0 0.0 Reference for referral of histopathological specimens Yes 34.8 96.3 89.4 78.0 No 65.2 3.7 10.6 22.0 Records of users diagnosed with oral cancer Yes 37.0 100.0 44.6 17.8 No 63.0 0.0 55.4 82.2 Reference of confirmed oral cancer cases to hospital services Yes 43.7 97.8 79.1 1.1 No 56.3 2.2 20.9 98.9 Prevalence of Latent Statusd Time 1 (Cycle 1 of PMAQ-CEOe, 2014) 17.8 32.1 50.1 0.0 Time 2 (Cycle 2 of PMAQ-CEO, 2018) 32.4 54.4 0.0 13.2 Transition probabilitiesf LSg 1. Worse work process 52.1 28.8 0.0 19.0 LS 2. Better work process 6.1 88.4 0.0 5.4 LS 3. Poor linkage: CEO and PHC 42.3 41.6 0.0 16.1 LS 4. Poor linkage: CEO and hospital 0.0 0.0 0.0 0.0

The choropleth maps (Figure 1) evidenced that practically all Brazilian States had CEO that improved work processes between the first and second cycles, that is, they moved to a class with better work processes for OC care. The Southeast and South regions had the highest percentage of CEO with the status ‘Better work process in both cycles’, emphasizing the states of Mato Grosso do Sul (MS) and Goiás (GO) in the Midwest region, and Ceará (CE), Piauí (PI), and Pernambuco (PE) in the Northeast region. The States of Acre (AC), Amapá (AP), Amazonas (AM), Pará (PA), Roraima (RR) and Tocantins (TO), all in the North region, and Maranhão (MA), Piauí (PI), and Bahia (BA), in the Northeast region, and Goiás (GO), in the Midwest region, in turn, had higher percentages of CEO had a bad performance in both cycles or deteriorated from the first to the second cycle in both cycles.

Choropleth maps of the transition of the work process for oral cancer care in the CEO, cycles 1 (2014) and 2 (2018) of the PMAQ-CEO, Brazil 

The hotspots (Figure 2) revealed the presence of areas concentrating improvements in the Northeast, with the exception of Maranhão (MA) and Piauí (PI), where the CEO worsened or remained in the worst LS. In contrast, the Southeast and South regions stand out for maintaining their status of ‘Better work process’ in both cycles. On the other hand, some hotspots in all the Northeastern states showed areas where the work processes deteriorated or remained bad. The same characteristic was shown in Amazonas (AM), Pará (PA), and Tocantins (TO) in the North region of Brazil.

Hotspots showing the spatial correlations of latent status transitions in CEO, cycles 1 (2014) and 2 (2018) of the PMAQ-CEO, Brazil 

4 DISCUSSION

In this ecological study involving all the CEO that participated in first and second cycles of PMAQ-CEO, services were classified in LS, according to the quality of the work processes for OC care. Changes between cycles were identified using LTA to test models with 3, 4 and 5 LS. The model with the best fit and most adequate model from a theoretical viewpoint had four categories, as follow: LS 1 – deficiency in the provision of services; LS 2 – a higher proportion of adequacy of the services offered; LS 3 – limited integration between CEO and PHC; and LS 4 – limited integration between CEO and hospital services. Comparing the cycles, 419 CEO remained in an LS with similar indicators of adequacy of the work process, 269 improved the adequacy status of the services, and 228 had decreased their quality. The indicators that most impacted the worsening of quality were the reduction in the number of CEO that performed biopsies and that had a hospital referral for confirmed cases of OC. In addition, there were inequalities in the provision of care for OC among Brazilian regions. The Southeast, South and Midwest concentrate areas with services that had the best work processes in both cycles, the North revealed a pattern of worsening or maintenance of worse work processes, and the Northeast had a mixed pattern – with some areas that improved and others that deteriorated.

This study has some limitations. The use of secondary data limits the analysis to existing data. It is possible that the positive findings have been overestimated, as they are based on answers provided by the CEO´s professionals. Not all CEO in Brazil were evaluated in both cycles, which made it impossible to include 100% of these facilities in the present study. However, more than 90% were included. The PMAQ-CEO data are from 2014 and 2018. After this period, political (such as changes in the federal government),29 economic (economic crisis and changes in SUS financing),30, 31 and health (such as the COVID-19 pandemic)32 events may have had repercussions on the work processes of the CEO, so that the extrapolation of our findings for today must be done with caution. On the other hand, strengths of this study include the assessment of specialized public oral health services that provide care for OC in all Brazilian States, the use of standardized assessments, the use of innovative analytical methods, and the longitudinal design. LTA allowed us to identify changes in the CEO’s work process, while choropleth and hotspots maps allowed us to identify areas with the better or worse provision of specialized OC services.

Our findings revealed advances in the implementation of the PNSB and the potential of the PMAQ-CEO in inducing quality in OC care; however, some weaknesses are still worrisome, especially the referral and counter-referral processes15, 17 of oral health services for OC prevention, diagnosis, and treatment.11 Map analysis also suggests that the programme may have widened the gap between the wealthiest (South and Southeast) and the most impoverished regions (North and Northeast), despite having fulfilled its objective of inducing quality. These difficulties, in addition to problems in accessing APS and CEO services, possibly contribute to the overall stationary OC mortality rates in Brazil33, 34 and point to the need to improve the work process better and organize the OC Care Network.

In the first cycle of PMAQ-CEO, most CEO belonged to LS 3, while the proportion of CEO belonging to LS 2 were predominant in cycle 2, indicating the overall quality of OC care improved in Brazil. Parameters that involve training for performing biopsies and the PHC protocol for referrals to the CEO have improved when comparing the first and second cycles. Screening protocols and better continued education strategies contribute to an early OC diagnosis.10, 35, 36 These findings reflect the advances of the Oral Health Care Network concerning the comprehensiveness of care by CEO and PHC services. This finding is in accordance with an ecological study that evaluated the structure and work process of primary dental care services using data from the PMAQ of PHC (1st and 2nd Cycles), that showed that the number of primary care teams that made appointments every day of the week increased by 8.7% and that those that provided home visits increased from 44.1% to 52.9%. Also, there was a reduction in the number of teams that reported guaranteeing the agenda for accommodating spontaneous demand, varying from 62.4% to 60.1% and in the continuity of treatment, varying from 63.4% to 61.1%.23

The expansion in the coverage by oral health services in recent years and the implementation of Decree 7.508/2011, which defined PHC as the main entrance to SUS and as the core element in the organization of health services network, is supposed to contribute to explain this finding. Data from the Brazilian Primary Care Information System (acronym in Portuguese, SISAB) revealed an expansion of almost 10% in primary dental care services during the period analysed (from 38.5% in 2014 to 42.2% in 2018). The bill 12.372/2012 established that the SUS has up to 60 days, starting on the histopathological diagnosis, to start cancer treatment. Therefore, a remarkable improvement in the dialogue between oral health services in primary and secondary care has been identified.37 However, the coverage of SUS specialized oral health services is still low (15% of the Brazilian municipalities),11 which can hinder timely diagnosis and treatment, decreases the quality and survival time of OC patients, and increases the treatment costs. On the other hand, the increase in the number of CEO is associated with a decrease in the number of hospitalized cases of OC, reducing the number of stage IV cases, and reduction of mortality rates due to OC in Brazil.38 The process of organizing referral and counter-referral flows from the health region perspective, as proposed by the legal grounds, is complex and continuous. Therefore, it is possible that the impact has been occurring slowly and gradually, being more visible over time.

We observed a decrease in the proportion of CEO that performed biopsies between the first and second cycles, which may reflect the lack of a specific production target for these procedures. This finding is particularly relevant since biopsy is the appropriate diagnostic method for OC, and the CEO is the preferred service for performing this procedure in SUS. In a prior study that analysed data from the first cycle of PMAQ-CEO, findings showed that the availability of specialized oral medicine care is one of the least prevalent in in CEO, being present in only 65.5% of CEO evaluated.18 Another study that analysed data from the I and II Cycles of the PMAQ-AB (the equivalent of PMAQ-CEO to primary care services), revealed that only 58.9% of the PHC teams in the Northeast region of Brazil carried out campaigns to detect oral lesions and refer suspected cases of OC. In addition, 43.9% of them register and follow up suspected/confirmed cases, but only 20.7% have documents that prove these records.22 The lack of a goal in OC care leads managers to direct services to other priorities.11, 39 Thus, although OC diagnosis represents a priority for CEO, a relevant proportion of primary dental care teams do not actively search and identify patients at risk for OC and potentially malignant oral mucosa lesions.39 However, although the number of CEO that undergo biopsies has decreased, there is evidence that the rates of oral biopsies were fairly stable in Brazil between 2016 and 2019,40 suggesting that either the PHC is referring biopsies to other services (or that the PHC itself is performing them), or that the CEO itself is referring patients who need biopsies to other services. However, more empirical evidence is needed.

While several services improved the quality of care for OC between the first and second cycles of PMAQ-CEO, there was still a large number of CEO where the programme was not successful in inducing it (e.g., 32.4% of the CEO that remained in the worst LS in both cycles). Some of these were due to reduced adequacy of OC care work processes, especially those that belonged to LS 3 and migrated to LS 1 (21.18%). This was more frequent in the North and Northeast regions of Brazil. The findings in the state of Piauí (PI) can be considered paradoxical because they were polarized. The state stood out both in the group of the best and worst results. It is important to determine individual service and contextual factors of the municipalities that can better explain this polarization. The distribution of latent transition patterns for these Brazilian regions agreed with the findings of other studies, where places with the worst socioeconomic indicators had more limitations in the provision of health services.41

Prior studies reported that financial resources tend to be channelled