Abstract
The latest estimates of the Global Burden of Disease (GBD) study are now published (1). Headache disorders once again rank among the top causes of health loss quantified as years lived with disability (YLDs) (2). And, once again, the scientific, clinical and public-health communities can expect a flood of papers re-publishing the data.
GBD 2023 (1) is the first iteration since GBD 2016 (3) for which those who actually collated and modelled the data have produced separate analyses for headache (2). However, between these analyses, 31 known other papers have republished GBD headache data: two from GBD 2017 (4) and nine from GBD 2019 (5), while GBD 2021 (6) opened the floodgates to at least 20 (there may be others we have not identified). The number across all neurological disorders is much larger, and this is itself greatly exceeded across medicine, but here we address the headache community.
Four key methodological issues need to be understood. First is that, from GBD 2017 onwards, population-based studies have been the only source of data contributing to GBD headache estimates. Second is that no new headache data have been entered since GBD 2017 (the last literature review informing headache estimates was conducted in September 2017); changes in these estimates since then are due only to refinements in data modelling. Third, where few or none exist, GBD imputes data from other timepoints or countries (“borrowing strength” over time and space) to produce best estimates for all populations across the time series. Fourth is that, at disease-level four (individual headache disorders), GBD generates estimates only for migraine and tension-type headache (TTH) (although each includes a proportionate contribution from medication-overuse headache (MOH)) (2); overall estimates for headache disorders (disease-level three) are the sum of these: they do not include other headache types.
Errors and misinterpretations
One paper extracted child and adolescent data from GBD 2017 (7). The authors partly misunderstood the methodology behind these data: evidence includes reference to “different case definitions”, a related, mistaken belief regarding data sources (“registries based on medical data versus patients’ reports”) and the perception that GBD includes other headache disorders (“We decided to focus on TTH and migraine …”). Nonetheless, this paper presents pertinent messages not evident from the capstone paper (4).
One paper carries the endorsement of the “GBD 2021 headache collaborators” (8). This is not what it might suggest: no-one involved in modelling the data published in GBD 2021 (6) is among the 277 listed co-authors. Aside from reporting nothing new, the paper has many errors (Table 1). These, it should be noted, survived feedback within the partly decentralized publication process managed by the Institute for Health Metrics and Evaluation (IHME), which leaves the lead-authors responsible for their correction.
Errors and misinterpretations identified in reference 8.
DW = disability weight; GBD = Global Burden of Disease study; MOH = medication-overuse headache; SDI = sociodemographic index; TTH = tension-type headache; UI = uncertainty interval; YLD = year lived with disability
The overwhelming concern is the 29 papers in Table 2. Most bear the hallmarks of artificial intelligence (AI)-generation. Large language models degenerate, and ultimately collapse, as they consume data from the internet recursively generated by themselves (9). Authors with a poor understanding of GBD, or of headache (both in evidence in these papers), may not recognise this, even when it produces nonsense, as Table 2 shows. They should, however, be struck by the widely overlapping uncertainty intervals, such that claimed per-person increases in global headache burden are illusory, except that the focus on numbers of cases rather than age-standardized rates is a common obfuscation. Certainly the total headache burden has increased hugely since 1990: so has the world's population, by almost 3 billion! (10) (Table 2)
Errors and misinterpretations in 29 papers republishing headache data from GBD 2017 onwards.
CGRP = calcitonin gene-related peptide; GBD = Global Burden of Disease study; PMID = PubMed Identifier; UI = uncertainty interval
Intellectual property, publication ethics and integrity of the literature
A clear example of academic misconduct was submission of a manuscript – not, so far as we know, accepted anywhere – citing data from GBD 2023 while these were still under strict embargo. But many of the papers copy, republish and effectively claim ownership of vast quantities of GBD data (sometimes in tables extending over more than 100 pages) that the authors did nothing to collect. None gives credit, more than acknowledging “thanks” to “GBD collaborators”, to those who actually did so.
These papers disrespect intellectual property, and they disturb the integrity of the headache epidemiological literature. While none of those in Table 2 was an authorized product of GBD, they may be perceived as such, undermining the credibility of GBD itself. GBD is an enormously valuable resource, created with huge and sustained effort (1). Public-health policies often depend upon it, particularly in low- and lower-middle-income countries without the resources to gather health data locally. Its trustworthiness must not be compromised by papers that deem quality of no importance in pursuit of publication.
Failure of peer review
How have these papers, offering no new insights and scientifically unworthy, passed peer review? We have reviewed many such manuscripts, always recommending rejection, which to our knowledge has always been the editors’ decision. They are simply submitted elsewhere. It is easy to blame peer reviewers, who, certainly, are sometimes at fault, but highly productive paper mills flood the system. Many peer reviewers are recruited to the task without themselves knowing how headache disorders are modelled in GBD, and they cannot fairly be held to account for their inability to recognise charlatanism in what might appear to be an authorized GBD manuscript. Neither can editors, if they, too, lack this particular expertise but wish to accept manuscripts likely to gather many citations: if they don’t, it appears that other journals will.
And because these papers do gather citations, the paper mills keep turning, rapidly escalating the problem as Table 2 shows. We fear what will happen in the wake of GBD 2023 (1,2).
Confronting the problem
Confronted it must be. Our purpose here is to raise awareness, encouraging editors of headache journals and peer reviewers to push back.
But we doubt that peer review is equipped to provide the complete solution. IHME, sadly, is being forced to restrict access to GBD data, which it has hitherto steadfastly maintained, both for transparency and as a free good to provide governments and other stakeholders access to best-quality data for their countries or regions. They did not do so for others to exploit in order to embellish their own publication records. This is an abuse, which, ideally, should be stopped at source. Many of the papers in Table 2 acknowledge government grant support. This should cease.
Footnotes
Author contributions
All authors contributed equally to this editorial.
Declaration of conflicting interests
All three authors are GBD headache collaborators within the GBD Collaborator Network.
Funding
The authors received no financial support for the research, authorship and/or publication of this article.
