Global Co-Occurrence of COPD and Peripheral Artery Disease in Older Adults: A Machine Learning Analysis of Spatial Patterns and Population-Attributable Risks.

Global median incidence rates for COPD and PAD in adults aged 70 years and older were 1150.16 and 695.4 per 100,000, respectively.

• Data were drawn from Global Burden of Disease (GBD) 2021 estimates across 202 countries and territories.
• The analysis focused specifically on adults aged ≥ 70 years.
• Disease-specific incidence quartiles were used to characterize spatial co-occurrence.

High-incidence regions for COPD and PAD exhibited partial but not complete geographic overlap globally.

• Spatial co-occurrence was characterized using disease-specific incidence quartiles across 202 countries and territories.
• A quartile-based composite risk index was constructed to summarize cumulative exposure patterns.
• The partial overlap suggested both shared and distinct regional risk drivers.

Smoking, high LDL cholesterol, and kidney dysfunction were identified as shared exposure correlates associated with incidence of both COPD and PAD.

• These shared exposures were identified using random forest models with SHAP (Shapley Additive Explanations) to rank exposure contributions.
• Negative binomial regression with population offsets was used to estimate associations and population-attributable fractions.
• These exposures were distinguished from disease-specific exposures that correlated with only one condition.

Elevated LDL cholesterol had the largest positive association with COPD incidence, with a relative risk of 1.149.

• This finding was identified through negative binomial regression models.
• High LDL cholesterol was also a shared risk correlate for PAD, making it relevant to both conditions.
• The association is ecological in nature and should not be interpreted as an individual-level causal effect.

High sugar-sweetened beverage intake exhibited the largest positive association among PAD-specific exposures, with a relative risk of 1.059.

• Sugar-sweetened beverage intake was categorized as a disease-specific exposure for PAD rather than a shared exposure with COPD.
• This was identified through negative binomial regression with population offsets.
• Several additional exposures were also found to be disease-specific for either COPD or PAD.

Model-derived cumulative population-attributable fractions (PAFs) were 29.7% for COPD and 23.7% for PAD.

• PAFs were estimated using negative binomial regression with population offsets.
• These figures represent the proportion of disease incidence potentially attributable to the modeled exposures at the population level.
• The authors explicitly cautioned that these PAFs should not be interpreted as individual-level causal effects due to the cross-sectional and ecological study design.

Random forest models with SHAP values were used to rank the relative contributions of exposures to COPD and PAD incidence.

• SHAP (Shapley Additive Explanations) provided interpretable rankings of variable importance from the machine learning models.
• This approach was combined with negative binomial regression to estimate both rankings and quantitative associations.
• The analytical framework included a quartile-based composite risk index to summarize cumulative exposure burden.