The unit of observation for this article is the clinical trial. The cohort of clinical trials included any informative human study (pharmacokinetic, safety, efficacy, etc.) that we identified through the “Written Request” mechanism of the pediatric exclusivity program. The pediatric exclusivity program was designed to provide economic incentives to companies to conduct pharmacokinetic, safety, and efficacy studies in children. The Written Request, generally issued by FDA prior to initiation of pediatric exclusivity studies, contains required elements of the requested studies, including indication, number of studies, sample sizes, trials design, and age ranges to be studied. The results of the studies comprise much of the final study report that is submitted to the FDA. It is noteworthy that marketing exclusivity is extended regardless of the outcome of the trials requested by the FDA in the Written Request, provided that the companies fairly respond to the terms of the Written Request. From the Written Requests, final study reports, and FDA’s regulatory actions on the submissions, we recorded the basic elements of design and subsequent labeling changes for each product. All studies completed in response to a Written Request and submitted to FDA are evaluated by the Agency. Thus, this cohort captures 100% of the studies evaluated as part of the pediatric exclusivity program within our time frame.
The primary independent variable was based on the labeling change that resulted from the exclusivity submission. Labeling change categories included positive labeling changes, negative labeling changes, submission completed but no labeling change, submission/application withdrawn, or product withdrawn from the market. Examples of positive labeling changes included “safety and effectiveness established”, and “approved for use in children”. Examples of negative labeling changes included “no meaningful clinical activity”, “black box warning”, and “increased mortality reported in the product compared to placebo”. We collapsed the categorical labeling change variable into the primary independent variable. If the labeling change was positive, we categorized the outcome as positive; we otherwise categorized the outcome as negative. The categorization of the labeling changes was adjudicated by the authors affiliated with FDA. Two authors (DKB and JL) first categorized the labeling changes. Label changes were then categorized by three authors (LM, DA, and RR). Differences were then discussed and adjudicated (DA, DKB, DM, JL, LM, and RR). Reconciliation was required for two products.
Drug labeling changes also were classified with respect to public health impact. Labeling changes were classified as “key labeling changes” if the studies resulted in substantive dosing changes, new safety information, or lack of efficacy in phase III testing. Study design was classified as follows: efficacy (randomized trials, either placebo or active comparator), dose-ranging or multi-dose pharmacokinetics, single-dose pharmacokinetics, or safety. Race and ethnicity were not uniformly provided for each study; but most of the trials conducted have submitted the fraction of children enrolled who were white. These data are reported as a marker of racial and ethnic diversity of enrolled children. We recorded the annual sales (http://www.drugtopics.com/drugtopics/
) for each product for the most recent available year that the product had sole marketing protection in the US. The number of centers at which each trial was conducted was transformed into a categorical variable—<7 centers, 7–17 centers, 18–34 centers, and > 34 centers.
The primary outcome was publication of the main study results in a peer-reviewed journal. If the results of two or more studies were combined into one publication, (4
) then all studies with data reported in the paper were counted as published. If data from a study was combined with data from non-exclusivity trials (e.g., a population pharmacokinetic study), then the study was counted as published. If studies of two products were performed in the same trial and the trial results were published (5
) then both studies were counted as published. We only counted as published those articles that could be found by Medline search.
We used four separate search strategies to obtain publication status. One author searched Medline (JL, 6/2006), entered the generic name of the product, and limited the search strategy to “all child (0–18 years)”, “1998–2005”, and “English language”. She read each abstract and compared the abstract with the Written Request or final study report. If the abstract was a potential match, the article was obtained and read. A second author (DKB, 6/2006) entered the product’s generic name, and limited the search to “1998–2005”, “English Language” and ages of trial participants from the Written Request and/or final study report. A third author (PBS) used key words from the study design provided by the Written Request and the generic name. This final strategy allowed for the capture of manuscripts prior to 1998. Each search strategy had high sensitivity; 6 articles were found by only one of the three strategies. When a paper was found to be a match, we recorded the 2004 journal impact factor (http://isiknowledge.com/jcr
). Finally, a study coordinator at Duke University requested publications (including those published, in press, or submitted) from studies conducted for Pediatric Exclusivity from each company. This resulted in no additional publications. We did not include publications limited to abstract or publications in trade journals because it was our primary aim to assess ready availability of results to the practicing physician. The analysis was limited to studies for which the data were submitted to FDA’s Pediatric Exclusivity Board by December 1st
, 2004 because of the delay between completion of study, manuscript preparation, and publication.
Reported p-values and 95% confidence intervals (CI) are two-tailed. Odds Ratios (OR) and 95% CI were derived from logistic regression modeling (STATA 8.2). We used forward selection logistic regression and the maximum number of variables in the model did not exceed the variables presented in the final model. We limited the number of variables in accordance with previously published regression methods. (6
) We repeated the multi-variable analysis using three models: a parsimonious model that included only those variables that were associated (p< 0.05) with publication in forward selection, a model that included variables associated with publication and was adjusted for year of submission, and a model that included variables associated with publication and was adjusted for year of submission and log-sample size (of children enrolled) (7
). Variables that were analyzed in forward regression but were not retained included number of centers (analyzed as a continuous and as a categorical variable outlined above), fraction of children enrolled who were categorized as white race, and annual sales of drug in the latest year of patent protection. The Institutional Review Boards and Ethics Committees of the enrolling sites approved the protocols included in this analysis. Dr. Benjamin had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, and approval of the manuscript are independent of any funding organization.