NICOTINE RESEARCH ANALYSIS

Overview

Does tobacco industry involvement influence research findings? This project analyzes over 2,000 published nicotine and tobacco research papers to investigate whether financial conflicts of interest—such as tobacco company affiliations, consulting fees, or industry funding—are associated with more favorable (positive) study outcomes. Using co-authorship network analysis and statistical methods (odds ratios, chi-square tests, permutation tests with Holm-Bonferroni correction), we compare outcome patterns across three groups: papers with tobacco industry ties, papers with declared conflicts of interest (non-tobacco), and independent papers with no conflicts. The goal is to provide transparent, data-driven evidence on the relationship between industry involvement and the direction of research conclusions in nicotine science.

2,175
Research papers analyzed
7,006
Authors identified
111
Tobacco company papers (5.1%)
425
COI declared papers (19.5%)
1,639
Independent papers (75.4%)
0.791
Industry assortativity (clustering)
Co-Authorship Network

Co-authorship network of nicotine/tobacco researchers. Each dot is a researcher (minimum 3 papers); lines connect co-authors who published together (thicker = more shared papers). The network uses a force-directed layout: connected authors pull toward each other while unconnected ones repel, so clusters reveal research communities that frequently collaborate. Nodes settle into position after a few seconds. Click any node to see that author's papers. Click a legend item to hide/show that category.

Author Category

Tobacco Company
COI Declared
Independent
Loading network

    Key Findings

    1 How We Classified Outcomes

    We used a computer program to scan each paper's summary (abstract) for about 100 keyword patterns that indicate whether the research found nicotine/tobacco products to be helpful or harmful. About half the papers (1,075 of 2,175) contained clear enough language to classify. The other 1,100 could not be classified because their summaries were missing, too short, or purely about methods (e.g., study designs, review protocols) rather than results.

    Positive: 392  |  Negative: 331  |  Mixed: 291  |  Neutral: 61  |  Not classifiable: 1,100

    2 Outcome Distribution by Category

    Among the 1,075 classifiable papers, how do outcomes break down by group? The bars below show the percentage of positive, negative, mixed, and neutral findings for each category. Notice how the green (positive) segment is larger for COI-declared papers.

    3 Conflicted Research Skews Positive

    This is the central finding: papers where authors declared a financial conflict of interest were 46% more likely to report favorable results for nicotine/tobacco products than independent research. In plain terms, if you picked a random COI paper and a random independent paper, the COI paper is nearly 1.5× as likely to have a positive conclusion.

    Positive rate: 43.2% of COI papers vs 34.3% of independent papers
    Odds ratio = 1.46 — meaning COI papers have 1.46× the odds of a positive outcome. The 95% confidence interval (1.09–1.95) does not cross 1.0, confirming this is statistically significant (p = 0.012).

    4 Consulting & Advisory Ties Drive the Bias

    Not all conflicts are equal. When we break down COI papers by the type of conflict, consulting and advisory board relationships (294 papers) are the main driver of positive bias. Industry funding shows the strongest effect (5.8× the odds) but with only 11 papers, that could be due to chance.

    Consulting/Advisory (294 papers): 1.42× the odds of positive ✓ significant
    Legal (97 papers): 1.57× the odds
    Financial Interest (35 papers): 1.60× the odds
    Fees/Compensation (19 papers): 1.15× the odds
    Industry Funding (11 papers): 5.77× the odds
    ✓ = statistically significant (p < 0.05). Others may be real effects but sample too small to confirm.

    5 Tobacco Company Papers — No Clear Bias

    Surprisingly, papers written directly by tobacco company employees do not show a significant skew toward positive outcomes. Their positive rate (36.5%) is close to independent research (34.3%). This might be because tobacco-authored papers face more scrutiny, leading to more cautious framing, or because the sample (111 papers) is simply too small to detect a difference.

    Positive rate: 36.5% (Tobacco) vs 34.3% (Independent)
    Odds ratio = 1.10 (95% CI: 0.65–1.88). The confidence interval crosses 1.0, so we cannot rule out no difference (p = 0.78).

    6 Tobacco Researchers Stick Together

    The co-authorship network shows that tobacco company researchers overwhelmingly publish with each other. They form tight, insular clusters with very few connections to independent researchers. A clustering score (assortativity) of 0.791 — where 1.0 would mean complete isolation — confirms this near-total segregation.

    1,651 authors  |  13,405 co-authorships
    92 research communities detected  |  Largest connected group: 89% of all authors

    7 COI Authors Bridge Research Communities

    Unlike insular tobacco clusters, researchers with declared conflicts of interest tend to collaborate across different research groups. They act as "bridges" between communities that would otherwise be isolated. This central network position means their work reaches — and potentially influences — a wider audience.

    Bridge score (betweenness centrality):
    COI: 0.0035  |  Tobacco: 0.0011  |  Independent: 0.0006
    Higher = more connections between different groups. COI authors score 6× higher than independent ones.

    8 How Confident Are These Results?

    To make sure these findings aren't due to chance, we tested them using multiple independent methods. All key results held up across every test, including a correction (Holm-Bonferroni) that accounts for running many tests at once, which can inflate false positives.

    Pattern test across all 3 groups: p = 0.029
    Randomization test (COI vs Independent, 10,000 shuffles): p = 0.012
    Network centrality differences: p < 0.001
    ✓ = statistically significant. A p-value below 0.05 means there's less than a 5% chance the result is due to random chance alone.
    Methodology

    This analysis examines 2,175 nicotine/tobacco/e-cigarette research papers from PubMed (1964–2025). Papers are classified into three mutually exclusive categories based on author affiliations and conflict-of-interest statements:

    • Tobacco Company — Authors affiliated with Philip Morris, BAT, JUUL, R.J. Reynolds, Altria, Swedish Match, and other tobacco companies
    • COI Declared — Authors with disclosed conflicts (consulting fees, advisory boards, funding) not tied to tobacco companies
    • Independent — No conflict of interest declared

    Outcome Classification. Each paper's abstract is analyzed using an enhanced keyword classifier with approximately 100 directional patterns. The classifier assigns one of five labels:

    • Positive — Abstract contains language indicating favorable outcomes for nicotine/tobacco products: harm reduction, cessation aid efficacy, lower toxicant exposure, improved health markers, therapeutic benefits, or reduced-risk product claims
    • Negative — Abstract contains language indicating adverse outcomes: health harms, addiction, cardiovascular/respiratory toxicity, carcinogenic effects, increased mortality, gateway effects, or youth uptake concerns
    • Mixed — Abstract contains both positive and negative directional keywords, or explicitly describes mixed/conflicting results
    • Neutral — Abstract is present but contains no strong directional language in either direction; purely descriptive or methodological findings
    • Not applicable — Abstract is missing, too short, or entirely methodological (e.g., protocol papers, systematic review methods) with no classifiable outcome language

    The classifier first checks for negative patterns (e.g., "lung cancer", "nicotine addiction", "cardiovascular disease"), then positive patterns (e.g., "smoking cessation", "harm reduction", "reduced exposure"). If both are found, the paper is labeled Mixed. If neither is found, it defaults to Neutral (if abstract exists) or Not applicable. This keyword approach has known limitations—see the Limitations section below.

    Limitations
    • PubMed-only corpus. This analysis is limited to papers indexed in PubMed. Research published in non-indexed journals, conference proceedings, or industry-only reports is not captured.
    • Keyword-based outcome classification. Outcomes are classified using automated keyword matching on abstracts (~100 patterns), not manual expert review. This may misclassify nuanced or ambiguous findings. 1,100 of 2,175 papers (50.6%) could not be classified ("Not applicable") due to missing or purely methodological abstracts.
    • COI disclosure depends on self-reporting. The "Independent" category includes papers where no conflict was declared — this does not guarantee the absence of undisclosed industry ties or indirect funding.
    • Industry affiliation detection is pattern-based. Authors are flagged as tobacco-affiliated by matching against a curated list of 15+ company names. Affiliations through subsidiaries, shell organizations, or intermediary consulting firms may be missed.
    • Correlation, not causation. The association between COI declarations and positive outcomes does not prove that conflicts of interest cause biased results. Selection effects, publication bias, and legitimate scientific disagreement may also contribute.
    • Network limited to co-authorship. The network only captures co-authorship relationships. Other forms of collaboration (shared funding, advisory roles, citation networks) are not represented.
    • Temporal coverage is uneven. The corpus spans 1964–2025, but the majority of papers are from the last two decades. Earlier papers often lack structured COI statements, which may inflate the "Independent" category for older research.