Report · Phase 2b

Esophageal cancer in Minas Gerais: what else goes along with the excess

Seven further characteristics, including cachaça, maize, water in 1991 and 2000 and access to diagnosis. Versão em português.

Author: Clériston Almeida Capistrano (ORCID 0009-0008-5280-1869) · independent researcher with no formal training in health sciences · cledon.org

Abstract

In Phase 2, the excess of esophageal cancer mortality in Minas Gerais followed rurality, lower municipal GDP per capita and lower altitude, with no production variable positively associated. Phase 2b tested seven further characteristics, with an analysis plan recorded in git before the data (commit 946f282), and incorporates an AI-assisted technical review (not independent peer review).

• Cachaça. Where rural agro-industry produced more sugarcane spirit (cachaça) in 2006, esophageal cancer mortality in 2015–2024 was higher (RR 1.26 (1.18–1.35) per standard deviation, 543 municipalities). The association holds with the 2017 Agricultural Census (1.27 (1.17–1.37)) and after adjustment for rurality, GDP and altitude (1.11 (1.04–1.19)). Comparing producing with non-producing municipalities, the RR is 1.30 (1.16–1.46) unadjusted and 1.11 (1.02–1.22) adjusted. For lung and other cancers, the direction is opposite. The study measured municipal production, not individual consumption, type of drink or chemical composition. The link between alcohol and esophageal squamous cell carcinoma is already established; the contribution here is a geographic pattern that may guide further research.
• Maize. We found no positive association with the share of municipal area under maize (the association is inverse, also after adjustment). This indicator does not allow a direct assessment of mycotoxin exposure.
• Well or spring water in 1991 and 2000, and rural population in 1991. Associated in the single-variable analysis, but they add no information beyond current rurality (correlations of 0.75 to 0.90 with it).
• Access to diagnosis. Unadjusted, the esophageal excess is higher where there are fewer endoscopes and the oncology hospital is farther; for other cancers, the opposite. These crude associations are incompatible with the simple explanation of higher recorded mortality due only to greater diagnostic availability. After adjustment, there was no statistical evidence of association with these indicators.

1. Data security

Only public, aggregated data: by municipality (IBGE) or by health facility (CNES, no patients). CNES files include fields such as the facility's CPF/CNPJ (taxpayer number); the script reads only the facility code, municipality, equipment, quantity and accreditation, discards the rest on reading and deletes the original file after extraction, keeping only its SHA-256 for reproduction. Published results are counts by municipality.

2. Methods

Outcome and model: those of Phase 2 (esophageal cancer deaths 2015–2024, state-based expected deaths as offset, Poisson regression with standard errors clustered by immediate region). In Minas Gerais (853 municipalities), continuous variables are standardized by the Minas Gerais standard deviation; for Brazil, by the national one; the RR is per standard deviation. Production and endoscopy enter as log(1 + x). The cachaça production indicator enters as 0/1, without standardization: its RR compares producing with non-producing municipalities. Immediate regions are identified by the official IBGE code (some names repeat across states, such as Itabaiana/PB and Itabaiana/SE).

Multiple-testing correction. Benjamini–Hochberg [3] (q) across the 7 variables of the main Minas Gerais analysis (Table 1) and, separately, in the difference tests against each comparison outcome. No correction in the adjusted models (section 4), the Brazil analysis (section 6), the sensitivity analyses, or across project phases; those p-values are descriptive.

Variables. Cachaça: sugarcane spirit from rural agro-industry (2006 Agricultural Census; 2017 as sensitivity), in litres per 1,000 inhabitants. Maize: harvested area as % of municipal area (PAM, 2005–2010). Well or spring water (1991 and 2000 Censuses; in 2000, only "on the property"). Rural population in 1991. Digestive endoscopes (CNES, equipment 0333) per 100,000 inhabitants in the immediate region, December 2015 (2020 as sensitivity). Centroid distance to the nearest municipality with a hospital accredited as CACON or UNACON (oncology), December 2015.

Missing data. Same rules as Phase 2: “-” = 0; “...”, “X” and absence from the response = missing. In the Agricultural Census, “X” means suppression when there are few producers: it indicates production, not zero. In the first run, total cachaça added own and purchased raw-material components and treated “X” + “-” as zero; this was corrected (the total is missing when any component is suppressed), and a production indicator, unaffected by suppression, was added as a non-pre-specified sensitivity analysis.

3. Minas Gerais: associations and comparison outcomes

RR per standard deviation of each variable in Minas Gerais, with 95% CI: esophagus (blue) and the comparison outcomes lung (orange) and other cancers except esophagus (green).

Table 1. Associations in Minas Gerais and formal test of the difference from the comparison outcomes (lung and other cancers; not negative controls in the formal sense [4]).

4. Adding to the Phase 2 model

Each new variable was added, one at a time, to the model with rurality (2022), municipal GDP per capita (2010) and altitude, and then also with the % of ill-defined causes.

Continuous variables: RR per standard deviation. Producers × non-producers: RR of the direct comparison.

Cachaça is the only new variable that remains positively associated after adjustment, in all three versions. Maize remains inverse. Older water, 1991 rurality and access are no longer associated: they are highly correlated with rurality and GDP (1991 water and current rurality: ρ = 0.77), and the model cannot separate them.

5. Access to diagnosis

Per the plan, an excess created by more diagnosis would appear where there are more endoscopes and oncology hospitals nearby. In Minas Gerais, esophageal cancer does the opposite (RR 0.87 for endoscopy and 1.15 for distance), while lung and other cancers follow the direction expected from an access effect. The crude associations are incompatible with the simple explanation of higher recorded mortality due only to greater diagnostic availability. However, after adjustment, there was no statistical evidence of association with these indicators. The results cannot separate differences in incidence, diagnosis, treatment and death recording; esophageal cancer prognosis depends on stage and treatment options [6].

6. Brazil (secondary)

In Brazil, cachaça is also associated with the esophageal excess (with state fixed effects: 1.09 (1.06–1.13)), as are older water and 1991 rurality; maize is not.

7. Municipalities with a signal and Minas Gerais cluster areas

Medians of the variables among municipalities with published values; in parentheses, how many municipalities enter each median. For cachaça production, only 543 of the 853 Minas Gerais municipalities have a quantitative value; the rest are suppressed or missing, and this absence may select municipalities with different characteristics. Among the 9 municipalities with a signal, the median uses 7 (Cipotânea and Ervália are suppressed). By the production indicator, which suppression does not affect, 9 of the 9 municipalities with a signal produced cachaça in 2006, against 508 of the 796 Minas Gerais municipalities with a response (64%).

9. Limitations

• Ecological study: municipalities producing more cachaça do not show that patients drank cachaça [5].
• Rural agro-industry production covers agricultural establishments and does not capture all consumption or drinks bought from other municipalities; suppression hides values in some municipalities.
• Centroid distance is not travel time; CNES records existing equipment, not examinations performed.
• Variables highly correlated with rurality and income; residual confounding is likely.
• Nitrate in water (SISAGUA) was left out: the Ministry of Health portal was not accessible from this environment.
• Analyses by an independent researcher with no formal training in health sciences, with AI assistance (Anthropic's Claude) for code and text, under human review. No machine learning model was trained or evaluated.

10. References

1. International Agency for Research on Cancer. Personal habits and indoor combustions. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, vol. 100E. Lyon: IARC; 2012.
2. International Agency for Research on Cancer. Drinking coffee, mate, and very hot beverages. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, vol. 116. Lyon: IARC; 2018.
3. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B. 1995;57(1):289–300.
4. Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology. 2010;21(3):383–388.
5. Morgenstern H. Ecologic studies in epidemiology: concepts, principles, and methods. Annu Rev Public Health. 1995;16:61–81.
6. National Cancer Institute. Esophageal Cancer Treatment (PDQ®) – Patient Version. Bethesda: NCI.
7. Instituto Brasileiro de Geografia e Estatística. Censo Agropecuário 2006 e 2017; Produção Agrícola Municipal; Censos Demográficos 1991, 2000, 2010 e 2022. Sistema IBGE de Recuperação Automática (SIDRA).
8. Ministério da Saúde. Cadastro Nacional de Estabelecimentos de Saúde (CNES) e Sistema de Informações sobre Mortalidade (SIM). DATASUS.

Code at https://cledon.org/codigo: exposicoes_2b.py, hipoteses_2b.py, gerar_fase2b.py. Plan: PLANO-fase2b.md (commit 946f282). Results: fase2b_*.csv on the map page and in the Zenodo deposit.

In memory of Francisco Cledio Capistrano (F.C.C.).