The Brazilian Unified Health System (Sistema Único de Saúde−SUS) is the universal public healthcare system of Brazil funded in the 1990s [10]. Patients enter the system through primary care; those with suspicion of cancer are referred to secondary care and confirmed cases are referred to specialized tertiary care centers such as oncology centers, oncology units, and hospitals where the diagnosis is confirmed and radiation therapy, systemic therapy (chemotherapy, hormonal therapy [androgen deprivation therapy-ADT], etc.), surgery, and palliative care is administered accordingly. Information Technology Department of SUS (DATA-SUS) maintains a nationwide database of its patients which is updated monthly and uses the International Classification of Diseases (ICD-10) code [11].

Data on general deaths as well as prostate cancer new cases, treatments, and deaths was obtained on 06/15/2023 from the following DATA-SUS warehouses: Ambulatory Information System (SIA) (sia.datasus.gov.br/principal/index.php, accessed on 25/01/2024), Hospital Information System (SIH) (sihd.datasus.gov.br/principal/index.php, accessed on 25/01/2024), and Cancer Information System (SISCAN) (www.inca.gov.br/publicacoes/manuais/manual-do-sistema-de-informacao-do-cancer-siscan-modulos-1-2-3-e-4, accessed on 25/01/2024). The information was limited to the years 2013 to 2021 and men over the age of 50. Information on population was obtained from the Brazilian Institute of Geography and Statistics (IBGE) from their population projection estimates between 2013 and 2021 (www.ibge.gov.br, accessed on 25/01/2024). Information on health expenditure was obtained from the Institute for Health Policy Studies (IEPS), a non-profit independent organization (ieps.org.br/institucional, accessed on 25/01/2024). This information was limited to the year 2021. Information on the number of SUS hospitals in the year 2023 was obtained from the database of the Brazilian Ministry of Health-The National Registry of Health Establishments–CNES (www.gov.br/saude/pt-br/acesso-a-informacao/acoes-e-programas/cebas/cnes-cadastro-nacional-de-estabelecimentos-de-saude, accessed on 25/01/2024). Information on number of robotic surgical systems per state in 2023 was obtained from www.portaldavinci.com.br(accessed on 25/01/2024). For the correlation study, these economic and healthcare indices were assumed to have been stable throughout the time of our PCa data (2013 to 2021).

Patient data was limited to the annual number of new PCa cases, PCa-related deaths, and PCa-related treatments per state, including surgeries, radiation therapies, and systemic therapies (hormonal therapy, chemotherapy, etc.). DATA-SUS maintains information on patients referred to private practices [11]. No data was acquired from practices outside of the SUS.

Yearly PCSIR and PCSMR were calculated per 10,000 men aged 50 years or older. Mortality to incidence ratio (MIR) was calculated as a proxy for cancer-specific survival [12]. The number of each PCa treatment was divided by the number of new cases to calculate their rates. Regions were compared using a two-sided nonparametric test (Wilcoxon Rank Sum), p-values were Holm’s corrected, and those smaller than 0.05 after correction were considered statistically significant.

We used Winsorized Pearson correlation as a more robust method of correlation analysis [13]. We used rank transformation on hospitals per capita to reduce the impact of one outlier that was five standard deviations higher than the average [14].

Joinpoint analysis [15] was used to detect temporal changes and its results were used as predictors in a multivariable regression model. Eventually, we incorporated year, region, and binary indicators of pre-2018 and post-2018 as well as pre-pandemic and post-pandemic (2020) as predictors of the model [5,16].

All analyses were performed on R version 4.1.2 (2021-11-01) on RStudio platform 2022.07.1 + 554 “Spotted Wakerobin” and using package tidyverse, ggpubr, and ggstatsplot [17]. The study was exempted from ethical review since it only used public deidentified data.

Figs. 1 and 2 spread the data through time and region to show trends. Based on the multivariable regression model (Table 2), the annual PCSIR for PCa was relatively stable between 2013 and 2018; however, after 2018 it increased (β = +3.66, p < 0.001) while after 2020 it decreased (β = −2.15, p = 0.002). In contrast, both the annual PCSMR and the MIR remained relatively unchanged throughout the studied period, extending into the pandemic years. Remarkably, all-cause mortality (ACM) experienced a significant surge during the pandemic (β = +58.98, p < 0.001).

PCa treatments (Fig. 2, Table 2) remained relatively stable throughout the period, however, notable shifts emerged post-2018. Any treatment rate (ATR) decreased after this year (β = −0.09, p = 0.016), SR increased (β = +0.05, p = 0.017) while RTR and STR decreased (β = −0.06 and −0.10, p = 0.005 and 0.002, respectively).

States’ economic and healthcare status were measured based on the following indices: average health expenditure (HE), state’s ranking in the number of SUS hospitals per capita (Hos-PC), the number of robotic surgical systems per million population (RSS-PM), and the percentage of hospitals under SUS (SUS-coverage). Fig. 3 shows the correlation between these indices and PCSIR, PCSMR, MIR, and treatment rates. RSS-PM and SR were positively correlated (p < 0.001). HE was negatively correlated with PCSIR and RTR while it was positively correlated with MIR and STR (p < 0.001 for all). Hos-PC was positively correlated with PCSMR and STR (p = 0.041 and 0.004, respectively). SUS-coverage was positively correlated with MIR and STR, and negatively correlated with PCSIR, SR, and RTRs (p < 0.001 for all).

PCSIR was strongly negatively correlated with STR (β = −0.60, p < 0.001), negatively correlated with ATR (p < 0.001), and positively correlated with RTR (p = 0.004). MIR was positively correlated with ATR and STR (p < 0.001 for both), and negatively correlated with RSS-PM (p = 0.003). Fig. 4 shows a multivariable regression model predicting all-cause mortality rates before the 2020 pandemic from PCa statistics. All-cause mortality was positively correlated with PCSMR (β = +9.78, p < 0.001), and negatively correlated with SR (β = −24.67, p = 0.006). Details of states’ health economy status, coefficient estimates, as well as 95% confidence intervals, and p-Values pertaining to Figs. 3 and 4 are available in supplementary Tables S1–S3.

Our analysis of regional differences (Table 1, Fig. 1) is not in concordance with a prediction study by Santos et al. [2] While we saw the highest PCSIR in the Southeast region followed by the South, Santos et al.’s predicted the highest PCSIR in the Northeast followed by the Midwest. This is possibly due to their different information source (Population-Based Cancer Registries 1987–2019).

A 2022 study by Iser et al. analyzed PCSMR trends in different regions of Brazil between 1990 and 2019 [28]. Similar to our study, they found the highest PCSMR in the Northeast region. In our study, the Midwest had the second highest PCSMR, contrary to Iser et al.’s which showed the North to have the second highest mortality. This discrepancy is possibly due to the difference in the time of our data. Similar to our study, they found high PCSMRs in the states of Bahia (BA) and Tocantins (TO).

We saw a significant difference among regions in their choice of treatment. While the Midwest region treated the majority of cases with surgery (median rate 0.67), other regions preferred systemic therapy (ranging between 0.75 in the North to 0.54 median rate in the South) and radiation therapy (from 0.26 in the South to 0.06 median rate in the North, p < 0.001). The difference could be the result of better access to robotic surgery in the Midwest (0.57 RSS-PM, Supplementary Materials) and/or limited access to radiation therapy or systemic therapy in the region. According to Fonseca et al., patients living in the Midwest between 2017 and 2018 had to commute more than 300 km to receive radiation or systemic therapy [11].

To comprehend this section of our study, it’s imperative to bear in mind three pivotal concepts:

Correlation does not establish causation; rather, it signifies an association between variables [29].

In our data, PCSIR and STR were strongly negatively correlated (β = −0.60, p < 0.001). Clinically this makes sense since PCSIR is increased with screening, which in turn reduces the age of the patients and the number of cases that require end-stage treatment [30]. The same logic explains the negative correlation between PCSIR and ATR (β = −0.37, p < 0.001), an indicator of active surveillance rate. Screening also increased the possibility of detecting cancer in the localized stage, hence the positive relationship between PCSIR and RTR (β = +0.24, p = 0.004).

Correlation between SUS-coverage and any of the rates can be the result of selection bias: low SUS coverage may mean losing patient information to private practice. If this is not the case, then the correlations may reflect the public healthcare policy of Brazil: SUS prefers systemic therapy (β = +0.38, p < 0.001) to surgery (β = −0.50, p < 0.001) and radiation therapy (β = −0.27, p < 0.0001), which increases MIR (β = +0.31, p < 0.001). Or it could mean that states with high SUS coverage are less likely to perform PCa screening, which lowers PCSIR (β = −0.34, p < 0.001) and the stage of PCa cases at diagnosis.

The negative correlation between PCSIR and MIR is expected since the two are not truly independent. However, their strong negative correlation (β = −0.87, p < 0.001) plus the absence of correlation between PCSMR and MIR (β = +0.21, p = 0.18) can reflect the impact of screening on PCa survival. Regular PCa screening leads to discovering cases that would not need any treatment, this may falsely give the impression that screening increases PCa survival. All-cause mortality in Fig. 4 was strongly correlated with PCSMR, showing the heavy impact of this cancer on men’s health. The negative correlation between SR and all-cause mortality could result from better access to healthcare or increased screening.

DATA-SUS is one of the largest and most comprehensive healthcare systems in the world. The trends observed in their data likely provide an accurate reflection of the genuine impact of public screening policies on PCa patients. The notable shift in PCSIR metrics post-2018 underscores the significance of international guidelines in shaping clinical practices in Brazil. Consequently, these changes influenced PCa indices, as indicated by our correlation studies, which in turn strongly influenced treatment decisions. This correlation may be attributed to alterations in patient demographics, such as age, and the stage of cancer at diagnosis.

A major limitation of our study is the traveling patients. The states of Roraima (RR) and Amapá (AP) from the north region show zero radiation therapy cases which is due to the absence of radiation therapy centers in their states. Fonseca et al. [11] showed that patients from these states had to travel up to 2000 km for radiation or systemic therapy. Another limitation of our study is the absence of data on age, race, and cancer stage, which could explain the correlations, and information on private practices which is important in reducing bias [26,31,32].