To avoid potential bias in our study design, coinvestigators independently selected articles for inclusion, analyzed article conclusions, and examined financial sponsors. Article conclusions were classified as “favorable,” “neutral,” or “unfavorable” by two coinvestigators who had no knowledge of financial sponsors. Funding source was classified as “all industry,” “no industry,” “mixed,” or “not stated” by another coinvestigator who had no knowledge of article conclusions. In addition, relationship of sponsors to the beverage being studied was characterized according to whether a favorable finding would have “benefit,” “no relationship,” or “antagonism” to apparent financial interests. Then, associations between funding source and article conclusion were calculated, with adjustment for relevant covariates in some models.
Selection of Articles
We aimed to take the broadest view of the literature within the area of beverages and health, and therefore included a range of article types in the categories of interventional studies, observational studies, and scientific reviews. Scientific reviews—but not commentaries, editorials, or letters—were included because these articles are abundant, often cited, and potentially influential, and generally presumed to be objective. (However, we found very few scientific reviews that declared a source of funding; therefore, the numerical contribution of these articles to analyses of potential bias was small. We also conducted an analysis with scientific reviews excluded, focusing on only interventional studies.)
Articles included in this study were initially identified by the study coordinator (LIL) using OVID-Medline literature searches. The literature searches were designed in consultation with a medical librarian to have high sensitivity and intermediate specificity for inclusion criteria using the following terms to identify articles focusing on the beverages of interest: soft drinks, carbonated beverages; fruit juice, apple juice, orange juice, prune juice, cranberry juice, grapefruit juice, grape juice, guava juice, pear juice, pineapple juice, vegetable juice, carrot juice, tomato juice; milk. Additional terms for health and disease states of interest were included in the searches.
We used six inclusion criteria for study articles: (1) The topic relates directly to soft drinks, juices, or milk, or an inherent component of one of these beverages (e.g., calcium in milk). (2) At least one main endpoint relates directly to health, disease, or a disease marker. For example, an article demonstrating a health benefit of antioxidants in juice would be included, whereas an article describing manufacturing techniques to maximize antioxidant concentrations in juice would be excluded. (3) The article involves or considers research with humans or clinical materials derived from humans. (4) Conclusions relate directly to the beverage under study. For example, an article examining the effect of dietary calcium on bone mineral density would be included only if implications to the health effects of milk consumption are stated explicitly. (5) The article is classified as an interventional study, an observational study, or a scientific review according to standardized criteria listed below (see Assessment of Covariates). Articles in the categories of commentaries, editorials, letters, and miscellaneous were excluded. (6) The article was published in the 5-y period between 1 January 1999 and 31 December 2003.
Articles identified by literature search were then examined individually by the study coordinator and removed if any inclusion criteria were not met. Several additional articles were removed from the study for the following reasons: article was coauthored by one of the coinvestigators involved in this study (to avoid potential bias); one of the coinvestigators involved in classifying article conclusions had previous knowledge of the article's sponsorship (also to avoid potential bias): or both coinvestigators involved in classifying article conclusions determined that all inclusion criteria were not met. (A list of included articles is available from the authors upon request.)
Classification of Article Conclusions
The study coordinator provided two coinvestigators (CBE and DSL) with each article's abstract and discussion/conclusion section (as available). The coinvestigators were given no information relating to the identity of the article (e.g., journal, title, authors) or to financial sponsorship. When electronic documents were available, a simple text file was utilized. For articles without electronic versions, photocopies were made that excluded or obscured any identifying information. The coinvestigators classified article conclusions independently and then met to resolve discrepancies, using the categories outlined below.
Favorable—if both coinvestigators agreed that: (1) the conclusions suggested beneficial health effects or absence of expected adverse health effects, and (2) no statements were made that cast the product in a negative light.
Unfavorable—if both coinvestigators agreed that: (1) the conclusions suggested adverse health effects or absence of expected beneficial health effects, and (2) no statements were made that cast the product in a positive light.
Neutral—if the coinvestigators agreed that the conclusions were neither favorable nor unfavorable, or if the coinvestigators could not agree on classification.
Characterization of Financial Sponsorship
The study coordinator examined each article (and supplemental material, if relevant) in its entirety for information about financial sponsorship. A coinvestigator (MG) was given a list of all sponsors of each article linked to the type of beverage under study (soft drinks, juice, or milk). The coinvestigator was given no further information relating to the identity of the article (e.g., journal, title, authors) or to its methods or results.
The coinvestigator used generally available information, obtained in part by Internet searches, to characterize each sponsor as: (1) industry—including for profit and nonprofit affiliations (e.g., US National Dairy Council), (2) industry-associated—including governmental agencies that work with industry to promote consumption of specific foods or commodities (e.g., US Department of Agriculture), (3) nonindustry—including governmental agencies with no industry association (e.g., US National Institutes of Health), university, and independent foundations, philanthropies, and other nonprofit organizations, and (4) unknown. Funding source was then classified for each article as outlined below.
All industry—if all sponsors were classified as category (1) above.
No industry—if all sponsors were classified as category (3) above.
Mixed—if any sponsor was classified as category (2) or (4) above, or if the article had sponsors that were classified into more than one category.
We considered the possibility that an industry sponsor might fund a study or scientific review examining a competitor's product. For example, milk and soft drink consumption are reciprocally related among children [6
]; thus, a negative conclusion relating to soft drinks would arguably be advantageous to a dairy-affiliated organization. For this reason, the relationship of a financial sponsor to the beverage under study was characterized as outlined below.
Benefit—if a positive finding appeared to be in its commercial interest.
Antagonism—if a negative finding appeared to be in its commercial interest.
No relationship—if the sponsor appeared to have no commercial interest at stake.
Unknown—if commercial interest could not be determined.
Sponsors with no association or affiliation with the food industry (e.g., government, university, independent nonprofit) were characterized as “no relationship.” No articles in the “all industry” category had sponsorship that was characterized as “no relationship” or as “unknown,” nor did any have multiple sponsors in different categories. Thus, each “all industry” article could be subcategorized as “benefit” or “antagonism.”
Assessment of Covariates
Three covariates were examined: publication year (available from Medline), article type, and potential author conflict.
Article type was classified according to the following definitions: interventional study—if humans consumed, or if human tissue was exposed to, a food or food component with the intention of measuring a biological response; observational study—if data were collected on participants without the intervention of the investigators; and scientific review—if no original data were reported and if published research was analyzed in a systematic fashion.
Potential author conflicts for each article were identified if an explicit statement to this effect was made in the article about any author; or if a coinvestigator (MG) determined that the declared affiliation for any author might benefit from a positive conclusion relating to the beverage under consideration.
To evaluate changes over time in the percentage of articles with declared funding, we used the Mantel-Haenszel chi-square test for trend and exact binomial 95% CIs.
For analyses of the relationship between conclusion and funding source, we focused on the most discrete categories of funding: all industry—benefit, no industry, and all industry—antagonism. Studies with mixed funding were excluded because they represent a heterogeneous group, with different proportions of industry funding, potentially obscuring underlying relationships. Studies with no listed funding were also excluded from these analyses.
We evaluated the association between article conclusion (favorable, neutral, and unfavorable) and funding source using an exact linear-by-linear association test, pooling all article types. When evaluating this association for only interventional studies, we used Fisher's exact test, collapsing articles with favorable and neutral conclusions.
Using logistic regression analysis, we calculated ORs of conclusions for all industry compared to no industry funding. We computed two sets of ORs, one collapsing articles with a favorable or neutral conclusion and the other eliminating those with a neutral conclusion. Adjusted analyses controlled for relevant covariates, including publication year, beverage type, and potential author conflict of interests. One all industry—antagonism article was categorized as unfavorable, a situation in which the sponsor was perceived to benefit from a negative conclusion about a competitor's product (see above). Therefore, we considered this article as favorable to the sponsor's interests, and reclassified it as such for the purpose of calculating ORs, per a priori hypothesis.
Role of the Funding Source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.