A Conflict of Interest Tool
Adrian Wilkins, Matt Cutting, Helga Perry, Chris Martin, Chris Dankwa
This is Carl’s original posting for the problem. • Conflict of interest in the academic literature, develop an automated measure of conflictedness for particular drug/field e.g thromboprophylaxis
When using evidence to support clinical decision-making, it is important that the evidence used is as free from bias as possible, and where that is not possible, that some kind of allowance is made for bias. An important source of bias in the literature is commercial bias, where it is in the interests of commercial organisations to weight the literature in favour of their products. This does not necessarily involve deliberate manipulation, but may be a result of investment of resources in ensuring publication of favourable results and neglect of unfavourable results, generating a publication bias.
Conflict of Interest statements The International Committee of Medical Journal Editors states that the authors should make a statement of conflict of interest (COI) on a separate page on the page following the title page, and they suggest the use of a standard form (http://www.icmje.org/coi_disclosure.pdf).
Pubmed provides access to well-structured meta-data on a very large body of published biomedical literature. However, there is no inclusion of COI in the meta data. In order to access the conflict of interest, then access is needed to the full-text versions of publications. The body of free access is increasing, but is still not the majority of articles. Where access is available, the structure of the conflict of interest statement is inconsistent, and sometimes vague. The COI statement is not coded or structured to a well-defined format that lends itself to parsing. The pubmed header data does include well-structured authorship data, and tags for affiliation – the institution(s) primarily connected with the work.
FDA drug database
The US Food and Drug Agency makes a database of currently approved drugs publicly available. (http://www.fda.gov/Drugs/InformationOnDrugs/ucm135821.htm) This includes the drug names and the associated sponsor of any license application.
Pharmaceutical industry statements
In 2010 the US Patient Protection and Affordable Care Act required drug companies to declare payments to physicians. The slipping deadline is some time in 2013. Many organisations are now compliant and publishing the payments list online or in submissions to regulators. Examples include the Lily registry (http://www.lillyphysicianpaymentregistry.com/Registry). Other information on grants might be made available from drug companies like this example from Pfizer. (http://www.pfizer.com/responsibility/grants_contributions/transparency_in_grants.jsp)
Clinical trials databases
There are various international databases of clinical trials that oindicate a sponor and applicants. Examples include the US www.clinicaltrials.gov, International Standard Randomised Controlled Trial Number (ISRCTN) Register (http://www.controlled-trials.com/isrctn/), and the WHO maintain a database (http://apps.who.int/trialsearch/).
There are databases of patents on the web that could provide further information of potential relevance such as the Intellectual Property Office (http://www.ipo.gov.uk/p-find.htm) and the US Patents and Trademarks Office site (http://patft.uspto.gov/).
Recently announced database assigning a unique identifier to all researchers. Not yet really off-the-ground, but perhaps an application like this could develop a good synergy with such a database. (http://www.nature.com/news/scientists-your-number-is-up-1.10740)
We have considered various approaches to the problem.
Scraping COI statements
This gives rich and specific information on individuals, but is difficult to access and is highly unstructured. Statements are also prepared by the authors themselves and may be deficient. Data linkage from various sources Various databases could be linked and used to track individuals, specific papers and organisations to rate potential conflict of interest. This is attractive because of the large amount of data available, but it may be incomplete in that some companies but not all will publish data and there are no standards to simplify this process making the compilation labour intensive and expensive to maintain.
Traffic light rating
It might be possible to generate some kind of rating score of reliability for an individual or a particular paper. This is attractive in usability terms but would be difficult to achieve in a reliable manner meaning many ‘clean’ papers or authors being labelled as vulnerable to COI unjustly.
Social networking approach
It may be possible to harness the power of crowdsourcing to rate reliability or highlight potential conflict of interest for papers and authors in the same way that Diggit, Facebook and Twitter have links by papers to invite public participation on discussing a piece of work.
We have chosen combination of linked datasets and social networking. We have constructed a rudimentary test set of articles, organisations and papers that users can view and edit.
The data consists of a sample of paper headings from Pubmed with the associated meta data including the authors and affiliations. This is linked with the FDA drug data base including drugs with the generic and commercial names, and license sponsor. A separate table of conflicts exist which provide links between organisations and a particular paper (and therefore the associated authors) along with a ‘conflict type’ and space for notes. The notes can be used to enter explanatory text, or possibly the COI statement from the paper if available. The conflict types are currently ‘funding’ and ‘employments’.
This is a web-based tool (on a laptop!). The user arrives at the database page either by a search link, entering the URL or by clicking an icon by a paper. On landing at the page the relevant article entry is loaded. The number of links to ‘authors’, ‘products’, ‘organisations’ and the number of entered conflicts. The user can create a new conflict by linking a paper, organisation and a conflict type with free text notes.
The site could be further developed by building in links to pharmaceutical company data on payments to physicians, patents databases and other relevant sources. Whilst an SQL database sits behind the site, it may be possible to use API links to databases elsewhere. One disadvantage of linking to web based data is dependence on material that may change or disappear with or without warning, so it may be better to scrape relevant information and store it locally.