Introduction

Changes in technologies, treatments, medical knowledge, and cultural norms provide cause to review and change disease definitions and diagnostic thresholds, a task that is commonly undertaken by expert panels, consensus meetings, or influential workgroups who publish findings as statements, special reports, or as part of clinical practice guidelines. While such changes can be beneficial, there is an increasing recognition that widening of disease definitions may be one factor contributing to the problem of over-diagnosis, occurring across a range of conditions including pulmonary embolism, breast and prostate cancers [1],[2]. The concern expressed by some researchers is that for some people with milder symptoms, at lower risks, or in earlier stages of possible disease, the harms of a diagnostic label and treatment may outweigh benefits [3],[4].

At the same time there is accumulating evidence about pervasive financial ties between pharmaceutical companies and health professionals [5], including those writing guidelines [6] and disease definitions [7]. While noting the value of professional–industry collaborations, a 2009 Institute of Medicine (IOM) report found “widespread relationships with industry have created significant risks that individual and institutional financial interests may unduly influence professionals' judgments,” and that these “conflicts of interest” threaten the integrity of research, the objectivity of education, the quality of patient care, and public trust in medicine [5].

The 2009 report recommended professional societies and other organisations drafting clinical practice guidelines should “generally exclude as panel members individuals with conflicts of interest.” A subsequent 2011 IOM report on how to produce trustworthy guidelines included recommendations that “whenever possible,” guideline developers “should not have” conflicts of interest, that only a minority should have conflicts, and that chairs should be free of conflicts [8].

As both reports make clear, in addition to financial ties there are non-financial or intellectual conflicts such as academic advancement, and there should be no assumption that having a conflict is unethical, or “that any particular professional will necessarily let financial gain influence his or her judgment” [5].

A 2011 systematic review found many clinical guideline panels have failed to disclose financial ties, and those that did disclose had a “high percentage” of individuals with financial conflicts of interest [6]. Studies analysing ties of working groups for the Diagnostic and Statistical Manual of Mental Disorders (DSM), which set definitions and diagnostic criteria, have also found a majority of members with ties [7]. Kung and colleagues recently found two-thirds of individuals chairing guideline committees had conflicts of interest [9].

Few studies [7] have examined the financial ties of members of panels reviewing and changing definitions of common conditions, whether as part of practice guideline development or other processes. Our aim was to identify guideline panels in the US setting that have most recently made decisions about definitions or diagnostic thresholds for common conditions, and to report on any proposed changes and their industry ties.

Methods

List of Conditions

On the basis of the method previously used by Choudhry and colleagues [10], we derived a list of common conditions in the United States, drawing from a list of the ten most costly adult diseases [11], the top 20 therapeutic classes of drugs, and the top 25 individual drugs by expenditure [12]. Consistent with that method, drugs used to treat many non-specific conditions were excluded (e.g., pain killers). For situations in which a drug was approved for a number of conditions, we identified the most common condition for inclusion (e.g., etanercept ultimately mapped to rheumatoid arthritis, not psoriatic arthritis). If a condition in the top ten costly disease list was too broad or diffuse, or covered many specific conditions, it was excluded (e.g., back problems). A flowchart of the method appears in Figure 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flowchart identifying study conditions and panels reviewing definitions.

Note: bipolar/depression was one panel.

https://doi.org/10.1371/journal.pmed.1001500.g001

Results

After analyzing source documents [11],[12], the following drug classes, individual drugs, and conditions were excluded when identifying study conditions, as they were too non-specific or too broad, and did not map to specific conditions enabling analysis: oncologics; autoimmune diseases; narcotic analgesics; anti-epileptics; vaccines; hormonal contraceptives; immunostimulating agents; bevacizumab; oxycodone; pegfilgrastim; cancer; trauma-related disorders; and back problems.

From an initial list of 16 included common conditions, for two—osteoarthritis, HIV—we could identify no panel that made decisions about definitions or diagnostic thresholds since 2000 in the US context specifically. For the remaining 14 conditions, we identified the most recent panels that deliberated and made decisions about disease definitions, all of which proposed changes. For asthma and high cholesterol we identified two panels each, one constituted under the government funded NIH [13],[14] and one by professional societies [15],[16], reflecting the two main types of panels identified in this study. A single panel, the DSM-V Mood Disorders working group, proposed changes to two different conditions, bipolar and depression, in two separate web-based publications [17]. A full list of the final 14 conditions, 15 panels and 16 publications, key changes and rationale, analysis of panel decisions, and disclosed ties appears in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Conditions and characteristics of the panels and publications included in the study.

https://doi.org/10.1371/journal.pmed.1001500.t001

Among 16 publications, all authors in our study agreed that proposals in ten publications would tend to widen definitions (Table 2) and for one, narrow the definition. For the remaining five publications the impact was unclear. Rationales for the benefits of widening definitions or expanding diagnostic categories included: evidence about the risk of future adverse events for people previously considered normal (pre-hypertension); simplification (gastroesophageal reflux disease [GERD]); standardisation for research (rheumatoid arthritis); and the emergence of new evidence about biomarkers, tests, or treatments (Alzheimer disease). Among 15 panels, six included mention of possible harms of proposed changes (Table 3), albeit briefly, with three of those including citations in that mention [17]–[19], two citing primary studies [18],[19], and one of those citing a review of primary studies as well [18]. One publication referred to the potential negative consequences for those who would be labelled by the expanded definition [20], and only one referred to overdiagnosis [21].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Different ways to expand disease definitions.

https://doi.org/10.1371/journal.pmed.1001500.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Mention of possible harms of proposed changes to definitions.

https://doi.org/10.1371/journal.pmed.1001500.t003

The average number of panel members was 21 (range, five to 52). Among 15 panels, 12 included members disclosing financial ties to multiple companies, one panel disclosed ties to a single company only (GERD) [21], one stated that members had no relevant conflicts of interest (diabetes II) [22], and one had no disclosure section (high cholesterol 2002) [14], also the oldest panel. For a total of 2,081 individual ties across all categories recorded in the study, there were 55 discrepancies, 2.6%, arising from the independent extraction, mainly involving one or other extractor accidentally missing or adding a specific tie, or making errors by entering a specific tie into an adjacent column or row in a spreadsheet. All were resolved by discussion.

Among 14 panels with disclosure sections, the average proportion of members with industry ties was 75% (range 0%–100%) (Table 4). For members with ties, the median number of pharmaceutical or device companies to which they had declared ties to was seven (Table 4). For the nine panel publications disclosing multiple separate categories of tie, on average, members with industry ties were a consultant/adviser for four companies, received speaker fees/honoraria from two companies, and they or their institutions received research support from three. Twelve panels were chaired or publications led by authors with industry ties, most commonly to multiple companies. Among panels disclosing any ties to government agencies, non-government organisations, or publishers, on average around one-third of panel members disclosed these ties.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Nature and extent of disclosed ties, by panel.

https://doi.org/10.1371/journal.pmed.1001500.t004

For the 12 panels for which ties were disclosed to more than one company, almost all companies with ties to the three highest proportions of panel members were also active in the market for that panel's condition, with at least one drug on the market or in the research pipeline (Table 5). For example, with the chronic obstructive pulmonary disease [COPD] publication, Astra Zeneca, Boehringer-Ingelheim, and GSK—all companies with drugs for the condition—each had financial ties to 11 of 12 members, including the chair [18]. With the DSM-V Mood Disorders work group, Pfizer and Lilly—with drugs for depression and bipolar—had ties to five of the 12 members [17]. Similarly, companies marketing hypertension drugs—Bristol-Myers Squibb, Merck, Novartis—each had financial ties to eight of the 11 members of the panel which created the new diagnostic category “pre-hypertension” [23].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Companies with highest proportions of ties, and drugs in therapeutic area.

https://doi.org/10.1371/journal.pmed.1001500.t005

To evaluate any potential impact of the IOM recommendations regarding industry ties, we compared the panel publications released in 2012—after both IOM reports [5],[8]—to those released earlier. We found similar proportions of members disclosing industry ties (76% was the average across 2012 panels; 74% was the average across pre-2012 panels); a small reduction in the median number of companies to which those members disclosed ties in the 2012 panels (four in 2012 panels; seven pre-2012 panels); and similar proportions of panel publications widening definitions (four of six of 2012 publications; six of ten of pre-2012 publications).

Discussion

In this cross-sectional analysis of panels making recent decisions on definitions of common conditions in the US context, we found most panels proposed widening definitions and most had a majority of members with multiple ties to pharmaceutical companies. Proposals to widen fell into three inter-related categories: creating new categories of “pre-disease”; lowering diagnostic thresholds; and proposing earlier diagnosis or different diagnostic methods (Table 2). In some cases a clear rationale was offered for these changes—as when the hypertension panel cited evidence from original studies and meta-analysis linking normal blood pressure with elevated risks as the reason to create “pre-hypertension” [23]. In other publications, including the 2007 panel proposing changes to the diagnosis of asthma [13], the rational was less clear, more complex and diffuse.

Notwithstanding the problem of under-diagnosis, a growing body of evidence suggests over-diagnosis may be occurring across a range of common conditions, including hypertension [24], asthma [25], attention deficit hyperactivity disorder (ADHD) [26], and COPD [27]. Yet less than half of the study publications mentioned potential harms of proposed changes to definitions, and none included a rigorous evidence-informed discussion of those risks or how they might be mitigated.

In a three-part publication in 2011 [28]–[30] proposing new categories of “pre-clinical” Alzheimer disease (for research only at this stage) and “predementia”—which would clearly expand the population labelled—there was one short reference to the need to study the “ethical and practical implications” of diagnosing people at a “preclinical” stage [30]. The panel proposing changes to assessment and classification of COPD briefly mentioned that diagnostic methods “may lead to more frequent diagnosis of COPD in older adults with mild COPD as the normal process of aging affects lung volumes and flows” [18], but did not explicitly refer to the risk of “over-diagnosis” as it had done in a previous version of its report [31]. Proposing changes to ADHD diagnostic criteria—in part to make the condition more amenable to being a “lifespan” disorder involving adults as well as children—the DSM-V panel mentioned potential increases in prevalence but suggested they would be “negligible” (Table 3) [32].

Among panels disclosing ties, almost all chairs had financial ties to industry, and an average of three-quarters of members had ties to a median of seven companies, commonly working as consultants, advisers and/or speakers, as well as receiving research support. Companies with financial relationships with the greatest proportion of panel members were marketing or developing drugs for the same conditions about which those members were making critical judgements. GSK for example, marketing top-selling products for asthma, had financial ties to 20 of the 24 members of the 2009 asthma panel, and all 20 were consultant/advisers and/or declared speaker/honoraria ties to GSK [15].

This study has several important limitations. First, the lack of a comparison group means it is impossible to draw any inference of association between frequency of industry ties and proposals to change disease definitions. The exclusion of common conditions too broad to enable a focussed analysis of single panel publications (e.g., back problems) means it may have missed potentially important examples of changing disease definitions and limits generalizability of findings. The focus on the United States—chosen explicitly because of its globally influential panels such as DSM-V workgroups—also limits generalizability. A fourth limitation is reliance solely on disclosed ties, likely leading to an underestimate of their extent. Finally, we note that while we tried to ensure an exhaustive and multi-layered search strategy, we are unaware of any established method for identifying panel publications that review or propose changes to disease definitions.

Notwithstanding these limitations, the study has strong clinical, research, and policy relevance. Its novel focus on panels reviewing and proposing changes to common disease definitions or diagnostic criteria will help deepen understanding of the nature of what's been described as the “modern epidemic” of over-diagnosis [2]. Moreover, the group of 16 publications includes influential articles affecting the definition of 14 common conditions and impacting directly on medical policy and practice around the world.

The study findings are consistent with and help augment the evidence-base about industry ties of influential medical professionals. The 2011 systematic review found 56%–87% of clinical guideline writers had conflicts of interest [6], similar to our finding of 75% across disease-defining panels. Kung and colleagues found 71% of guideline committee chairs had conflicts [9], again similar to our findings. While these proportions may reflect the level of ties among medical specialists more generally, they are in stark contrast to IOM 2009 and 2011 reports calling for panels to generally exclude people with conflicts of interest [5],[8]. As reported above, we found no change in the proportion of members disclosing ties in the 2012 publications, after release of both IOM reports.

At least two publications [20],[21] made reference to members believing industry ties did not influence their decision-making, and we make no suggestion to the contrary. Indeed our data do not support any inference industry ties are associated with widening definitions or failure to rigorously assess potential harms of that widening. With anemia in chronic kidney disease, a panel with a high proportion of ties raised thresholds, effectively narrowing the definition [33]. There will doubtless be other cases where diseases have been widened by panels of medical specialists without industry ties. Moreover, as Lurie and colleagues found in the context of drug regulation, the financial conflicts of expert advisory committees did not correlate significantly with their voting outcomes [34]. Medicalization and over-diagnosis are driven by many factors—technological, professional, commercial, legal, and cultural [3].

While inferences of association or causation between industry ties and expanding disease definitions cannot be drawn, our findings can be considered in the context of broader evidence about potentially distorting biases associated with widespread industry sponsorship and financial ties in medical research [35]–[37], education [38], and practice [5], and in relation to “key opinion leaders” who speak and consult for industry [39].

In 1999 Schwartz and Woloshin found changes to definitions of high blood pressure, high cholesterol, diabetes, and overweight would “dramatically inflate disease prevalence” and “ultimately label 75% of the adult U.S. population as diseased” [40]. They concluded the “extent to which new ‘patients’ would ultimately benefit from early detection and treatment of these conditions is unknown. Whether they would experience important physical or psychological harm is an open question.” To what extent newly created “patients” produced by widening disease definitions will experience important harms remains a largely unanswered question, over a decade later.

This study did not investigate the merits of the proposed changes to the conditions identified. However, findings that diagnostic thresholds are being lowered by panels dominated by those with financial ties to multiple companies that may benefit directly from those decisions raise questions about current processes of disease definition. While it may be more difficult to locate senior specialists without industry ties, two recent IOM reports have encouraged such a change [5],[8], and models already exist for panels free of such conflicts, including the NIH consensus development program [41].

Several unanswered questions arise from this study, which could benefit from further investigation. Researchers might fruitfully examine how definitions are changing over time, what dollar amounts are being received from industry by panel members and organisations that auspice them, and how panel proposals impact on potential markets of sponsors. Most importantly enhanced research and policy attention might be directed at designing new processes for reviewing disease definitions, free of financial conflicts of interest and informed by rigorous analysis of benefits and harms.

Supporting Information

Acknowledgments

We would like to thank Elaine Beller and Rae Thomas at the Centre for Research in Evidence-Based Practice, Faculty of Health Sciences and Medicine, Bond University for assistance in design and analysis; Sarah Thorning, from the same centre, for assistance with development of the search strategy; and research assistant Peter Coxeter from the same centre for independent extraction of disclosure material. We thank David Henry for his comments on an early draft of this manuscript.