County-level determinants of cancer incidence in the United States, and the drivers of Iowa's rising rate

Introduction

A prior version of this project, despite an extensive analytical apparatus, was structured to confirm a single hypothesis: that agricultural pesticides cause cancer. Its competing-risk analyses were framed to defend that signal rather than to weigh alternatives on equal terms, it did not examine Iowa, and a key environmental covariate (radon) was never obtained. This re-analysis reopens the two questions the prior framing could not answer without bias: which factors actually predict cancer across U.S. counties when none is privileged, and why Iowa's age-adjusted incidence is rising while most states decline.

We adopt three disciplines throughout. No exposure is favoured a priori; effect sizes are calibrated against established carcinogens; and association is kept distinct from causation, hazard from risk, and detection from true incidence. The honest answers below overturn the original conclusion, which is the point of an unbiased design.

What predicts cancer across counties

Forty-two candidate factors were scored against all 26 cancer sites on equal footing. Smoking is the strongest correlate in the dataset, within a tight cluster of socioeconomic and metabolic factors. Pesticides rank in the bottom third (31st of 42; glyphosate 39th), a placement robust to resampling. In a common regression the herbicide to kidney association is real but small (about +1.6 percent per standard deviation), whereas the smoking to lung association is approximately +17.6 percent, an order of magnitude larger.

Why Iowa is rising

Iowa's age-adjusted all-cancer incidence is approximately flat since 1999 (+0.06 percent per year), but the national rate fell by about 5 percent over the same period. Iowa therefore diverged upward by failing to decline. Decomposition attributes most of the gap to a slower fall in prostate and lung cancer (a smoking legacy sustained by weak tobacco-control policy), while the fastest-rising sites are ultraviolet related (melanoma) and alcohol related (liver, oral cavity, oesophagus). For kidney, thyroid, and melanoma, incidence rises far faster than mortality, the established signature of screening and incidental detection rather than rising lethality. A natural experiment reinforces the point: neighbouring farm states that apply the same pesticides declined like the nation, and within Iowa no single measured factor explains the county pattern.

Candidate chemical exposures

We assessed four chemicals proposed to act through the gut microbiome or as direct toxicants, scored by an independent three-rater panel for mechanistic plausibility, human evidence, dose realism at Iowa exposure, local relevance, and testability. None reaches a high score as a driver of Iowa's burden, and all sit below the established causes (smoking, alcohol, obesity, detection). Drinking-water nitrate is the most defensible: Iowa's median exposure lies within the range where prospective cohorts report effects, yet below a standard set in 1962 for an infant condition rather than cancer.

Limitations

This is an ecological design: county-level associations need not hold at the individual level (the ecological fallacy), and a known carcinogen, radon, even reverses sign against lung cancer here because high-radon rural counties smoke less. The age-stratified data are national, so Iowa's early-onset signal cannot be localised in this dataset. The chemical assessments draw on the published literature and, for nitrate and PFAS, real exposure data, but the ecological layer cannot establish causation. The work is hypothesis-generating; the decisive tests are individual-level, described in the linked sections.