Lung cancer incidence decreases with elevation: evidence for oxygen as an inhaled carcinogen

Daniel Himmelstein

Kamen Simeonov

March 16, 2015

iPQB Journal Club

Video Summary by SciShow

Video Summary by DNews

Lung Cancer Rates, 2007–2011

SciShow: Oxygen is Killing You

oxygen-driven tumorigenesis

  • Decreasing ambient oxygen inspired by cancer-prone mice doubled time till tumorigenesis and decreased tumor bulk and genomic instability (Sung et al., 2011)
  • Increased cancer in cases of neonatal oxygen supplementation (Spector et al., 2005)
  • oxygen toxicity appears most profound in the lung (Jackson, 1985)

ODT progenitors:

radiation hormesis skeptics

Clarice Weinberg

Wesley Van Pelt

Study Design

  • outcomes: lung, breast, colorectal, and prostate
  • cancer incidence rather than mortality
  • counties were the smallest feasible unit
  • why not barometric pressure?
    • 100% oxygen at sea level
    • 88.7% at 1,000 m
    • 78.5% at 2,000 m
    • 69.2% at 3,000 m 
  • population-weighted elevation
    • Census blockgroup
    • 66.5 blockgroups per county
    • 1348.3 persons per blockgroup
  • excluded counties with fewer than 10,000 individuals
  • observations: counties
  • outcome: all-site cancer incidence
  • predictors: metro, white, black, education, income, obesity, percent male, and smoking

Choosing QC Thresholds using Residuals

Additional Predictors

  • minimize confounding effects by adjusting for additional factors
  • six general covariates to account for unknown or immeasurable risk factors or biases
  • included major cancer-specific risk factors
  • omitted redundant variables to reduce collinearity

best subset

  • evaluates all possible predictor combinations
  • advantages:
    • forced elevation inclusion
    • exhaustive
    • range of models (optimal = min BIC)
    • defined statistics
  • disadvantages:
    • overfitting
    • collinearity


  • coefficient shrinkage
  • variable selection
  • regularization parameter optimized with cross-validation
  • advantages:
    • prevents overfitting
  • disadvantages
    • limited statistics

weighted regression by county population (squareroot maxed to 250k)

Best Subset Regression

Were the entire United States situated at the elevation of San Juan County, CO (3,473 m), we estimate 65,496 99% CI [46,855–84,136] fewer new lung cancer cases would arise per year (ceteris paribus and assuming 2000-census county-populations).

Lung Cancer versus Elevation

County Stratifications

Population Subgroupings

Elevation Replacements


  • most variables covary with elevation
    • omitted-variable bias
    • aka uncontrolled confounding

Finally, following previous analyses, we used lung cancer as the indicator of accumulated population exposure to smoking. This adjustment for lung cancer in multivariable regressions may have overadjusted, if altitude has a beneficial effect on lung cancer. (Ezzati et al. 2012)

Lung cancer as a smoking proxy

spurious associations

Ed Yong Tweet

  • publicly available data
  • PeerJ free publication promotion
  • CC-BY (attribution only)

broader impacts

Elevation and Lung Cancer

By Daniel Himmelstein

Elevation and Lung Cancer

Lung cancer incidence decreases with elevation: evidence for oxygen as an inhaled carcinogen

  • 2,372