Classification Cascades: ICD-10, Trade Codes, and Cannabis Policy

Classification systems—standardized taxonomies such as the International Classification of Diseases (ICD), the Harmonized System (HS) for trade, and the Classification of Functions of Government (COFOG)—constitute foundational infrastructures of modern governance and economic measurement. This paper proposes the concept of classification cascades: the hypothesis that revisions to classification systems generate predictable, multi-stage disruptions that propagate from administrative coding through institutional processing to measurable behavioral and economic outcomes. I test this hypothesis using two complementary empirical strategies. First, I estimate event-study models exploiting staggered ICD adoption across 15 OECD countries (2005–2020), documenting discontinuities in pharmaceutical consumption consistent with cascade-induced shifts in recording and utilization at the point of classification revision. Second, I implement a difference-in-differences (DiD) design using U.S. Medicare claims data surrounding the ICD-9 to ICD-10 transition (October 1, 2015), comparing psychotherapy codes (treatment group) against evaluation and management codes (control group). The DiD estimates reveal a statistically significant increase in claim denial rates of +3.36 percentage points in the transition year (p = 0.004) and +2.99 percentage points in the post-transition year (p = 0.014). Descriptive evidence further documents a 6.9% decline in service volume and a 7.6% decline in Medicare payments by 2016 relative to the 2014 baseline. These findings are externally validated by Stewart et al. (2019), who document parallel disruptions in diagnosis recording across ten U.S. health systems. These findings suggest that classification cascades may represent a general class of administrative policy shock with implications extending beyond healthcare to international trade, public finance, and regulatory governance.