Estimating the incidence of cocaine use and mortality with music lyrics about cocaine

Cocaine use and mortality have rebounded in the United States after years of decline. From 2012 to 2017, cocaine overdose deaths increased by about 200%, and initiation also rose, with first-time use up 57% and 6.2% of those aged 18–25 reporting past-year cocaine use in 2017. Historically, cocaine has been used in social settings and has been portrayed as a marker of status and affluence, with popular media shaping public perceptions by glamorizing its effects and downplaying risks. Contemporary popular music frequently references cocaine and other substances, often in contexts of wealth and sociability, potentially reflecting or influencing real-world behaviors. Prior work documents extensive substance portrayals in popular media and music, but studies have often been descriptive, limited in scope or time, and not specific to cocaine nor quantifying links to epidemiologic outcomes. This study examines whether trends in cocaine mentions in song lyrics are associated with national trends in incidence of cocaine use and cocaine overdose mortality, aiming to assess whether popular music can signal or predict epidemiologic patterns relevant to public health.

Prior research shows that substance use is prominently portrayed in popular media and music. Content analyses report drugs as a dominant theme in music, with one study estimating 33.3% of top-charting songs portray substance use and averaging 32.5 drug references per hour. Given that most Americans listen to music regularly (average 32.1 hours per week), exposure to drug references is substantial. Studies link media exposure with youth initiation and use of tobacco, alcohol, and cannabis. Qualitative work found certain music subgenres (e.g., "screw") reinforce codeine syrup use through media modeling, and analyses of country music lyrics indicate increasing associations between alcohol, sex, and portrayals of women. However, many studies have been descriptive, covered few songs per year, focused on specific charts (e.g., Billboard Top-100), and have not specifically investigated cocaine or quantified effect sizes linking drug mentions to population-level drug use or mortality. This study addresses those gaps by focusing on cocaine-specific lyrics over two decades and modeling associations with national incidence and mortality.

Design: Ecological time-series analysis relating annual prevalence of cocaine mentions in song lyrics to national incidence of cocaine use and cocaine overdose mortality in the United States. Time frame: 2000–2018 for lyrics; 2002–2017 for incidence; 2000–2017 for mortality. Analyses emphasized contemporary trends given the digital transformation of the music industry and recent resurgence of cocaine outcomes in the 2010s. Data sources: (1) Incidence of cocaine use (ages 12+) from the 2017 National Survey on Drug Use and Health (Table 7.28A), converted to incidence per population using US Census midyear estimates. (2) Cocaine-related overdose deaths from CDC Multiple Causes of Death WONDER database (ICD-10 T40.5), normalized by midyear population. (3) Cocaine street price (purity- and inflation-adjusted) from UNODC (2000–2016). (4) Lyrics data from Lyrics.com. Cocaine mentions in lyrics: A comprehensive list of cocaine slang terms was compiled using sources such as Urban Dictionary and the DEA Slang Terms and Code Words guide. Ambiguous terms were removed after dual independent coding of a 1000-item sample; excluded terms included "snow", "blow", and "dust" due to ambiguity. The final list included 18 terms: cocaine, coke, coca, coco, powder, baking soda, bakin soda, arm and hammer, kilo, 8 ball, eight ball, llello, yayo, yeyo, nose candy, white horse, white line, and doing lines. Programmatic queries to Lyrics.com retrieved annual counts per term; counts were summed to annual totals of unique slang mentions. To adjust for variability in the total volume of lyrics over time, counts were normalized by the annual count of the word "love", yielding a relative prevalence ratio of cocaine mentions to love mentions. Stationarity and preprocessing: Augmented Dickey-Fuller tests indicated nonstationarity across lyrics, price, incidence, and mortality series. First-order differencing was applied to achieve stationarity, verified via autocorrelation and partial autocorrelation inspections. Modeling: First-order finite distributed lag (DL) models with unrestricted coefficients were fit using ordinary least squares to regress first differences of ln(outcomes) on first differences of lyrics (contemporaneous and lagged) and price (as a proxy confounder). Stepwise selection began with contemporaneous lyrics and added lag terms consistent with epidemiologic plausibility (1–3 years) until further additions were non-significant. Final equations included:

• Mortality: Δln(deaths_t) = α + β0Δlyrics_t + β1Δlyrics_{t−2} + β3Δlyrics_{t−3} + β4Δprice_t + ε_t
• Incidence: Δln(use_t) = α + β0Δlyrics_t + β1Δlyrics_{t−1} + β2Δprice_t + ε_t The response was log-transformed to interpret coefficients as percent changes. Sensitivity analyses excluded the ambiguous term "8-ball" and used negative controls with codeine-related and heroin-related slang lists to test specificity. Cross-correlation analyses evaluated lead–lag relationships between lyrics and outcomes in both directions across lags. Statsmodels (Python) was used for all analyses. Models were used to project 2018–2020 incidence and mortality under an assumed linear growth in cocaine-mention prevalence in lyrics.

• Data set: 6923 unique cocaine mentions from 5955 unique songs by 2418 artists; mean 364 mentions/year (SD 73).
• Trend in lyrics: Cocaine mentions were stable from 2000–2010, then increased by 190% from 2010 to 2017.
• Epidemiology trends: Cocaine overdose deaths rose every year starting in 2012, increasing 200% from 2012 to 2017.
• Top slang terms in lyrics (2000–2018): "coke" (2358 songs; 34%), "cocaine" (1469; 21%), "powder" (810; 12%), "coca" (579; 8.4%), "kilo" (470; 6.8%). Greatest growth: "kilo" +1950%, "yayo" +750%, "coco" +640%, "baking soda" +300%.
• Distributed lag models: A 0.01 increase in the relative prevalence of cocaine mentions in lyrics (vs. "love") was associated with an estimated 11.8% increase in incidence of cocaine use in the same year and a 15.5% increase in cocaine mortality 2 years later. Associations were robust to inclusion of street price; price decreases were associated with increased mortality at a 2-year lag.
• Table 1 parameter highlights: Use model contemporaneous lyrics β≈11.16 (p=0.039). Mortality model lyrics at 2-year lag β≈14.39 (p=0.037). Other lags largely non-significant.
• Cross-correlation: Significant correlation between lyrics and incidence in the same year (r=0.5713, p<0.05) and between lyrics and mortality at a 2-year lead (lyrics leading deaths; r=0.5747, p<0.05). No evidence that deaths later predict lyrics.
• Sensitivity and specificity: Results robust after removing "8-ball" (death model p=0.04; use model p=0.03). Negative control models with codeine- and heroin-related lyrics showed no significant lag relationships with cocaine epidemiology.
• Projections: Incidence projected at 344 per 100,000 (2018), 385 (2019), 416 (2020). Cocaine overdose mortality projected at 5.16 per 100,000 (2018), 5.66 (2019), 7.12 (2020).

The study demonstrates that increases in cocaine mentions in song lyrics are temporally associated with increases in cocaine use incidence within the same year and with overdose mortality two years later. Cross-correlation corroborates these lead–lag relationships, suggesting that popular music may both reflect and potentially influence evolving patterns of cocaine use. The 2-year lag between lyric trends and mortality aligns with epidemiologic evidence indicating approximately 1–3 years between initiation and seeking treatment, implying a plausible incubation period from initiation to fatal outcomes. These findings highlight a potential dual role of media—both as an early measurement signal of shifting social norms and as an exposure that may shape attitudes and behaviors toward cocaine use. The observed shift toward newer slang terms suggests changing demographics and settings for cocaine use, potentially indicating broader accessibility beyond elite venues and into younger and lower-income populations. Specificity analyses, showing no similar relationships for codeine or heroin lyrics with cocaine outcomes, strengthen the argument that cocaine-specific media signals align with cocaine epidemiology. Monitoring lyrical content could thus provide timely insights for public health surveillance and inform prevention and intervention timing, particularly within the identified 2-year window before mortality risk peaks.

Monitoring popular music lyrics offers an early signal of cocaine-related epidemiologic trends. Cocaine mentions in lyrics are associated with contemporaneous increases in cocaine use and with increased overdose mortality after approximately two years, and evolving slang usage may reflect emerging user populations and contexts. These results suggest value in integrating media analytics into public health surveillance. Future research should examine causal pathways, explore other substances, investigate more granular temporal and geographic resolutions, and assess how lyrical content shapes perceptions and behaviors. Given music’s wide reach, artists and industry stakeholders might consider the potential influence of drug-related messaging on youth and public health.

• Causality cannot be inferred; analyses identify associations despite efforts to reduce spurious correlations through first-differencing and inclusion of cocaine street price as a proxy confounder.
• Potential incomplete capture of cocaine slang; ambiguous or evolving terminology may lead to under- or misclassification, although sensitivity analyses (e.g., removal of "8-ball") did not materially change results and inclusion of additional terms like "blow" increased associations.
• Temporal aggregation at annual level limits granularity; both lyrics and epidemiologic data were available only yearly, yielding 18 effective observations for lyrics (2000–2017).
• Mortality definition using ICD-10 code T40.5 may include varying underlying causes; improvements in overdose testing over time could influence observed death trends.
• Immediate mortality following initiation might occur but was not observed; addiction progression typically involves repeated use over time, making near-term mortality less likely on average.
• Model specification choices (e.g., lag structure, normalization by "love" mentions, assumption of linear growth for projections) may influence estimates.