The Trust Behind the Table
On the doctor’s desk lies a recommendation. Behind it is a research review, and behind the review are a number of studies. The patient may see none of these documents. Yet they enter the consultation and help determine what will happen to his body.
When the doctor says that a treatment has demonstrated benefits, the statement rests on work performed by many others. Some have examined patients; others have analysed the data, assessed the manuscript and brought the findings together. Trust passes through many hands before reaching the person who needs help.
I think about this chain of people as I read two reports by Richard Van Noorden in Nature. The first, from 2023, describes clinical trials so seriously flawed that their findings may be unreliable. The second, in the January 2025 issue, follows efforts to develop better ways of checking them.
What do researchers, editors and authors of research reviews owe someone who must rely on knowledge he cannot verify for himself? They owe him evidence that can be examined, honest accounts of uncertainty and corrections that lead to action. At the same time, they owe the researchers under scrutiny fair treatment.
This responsibility begins long before the recommendation reaches the doctor’s desk.
Knowing Through Others
In “Epistemic Dependence”, the philosopher John Hardwig examines how we depend on the knowledge of others. We can have good reasons to accept a claim without ourselves possessing the expertise needed to assess the evidence behind it. Experts, too, must rely on other experts. The division of intellectual labour makes it impossible for any individual to verify everything.
The patient’s dependence is therefore part of a wider pattern. The doctor also depends on others. So does the researcher using a method developed by someone else, and the editor who entrusts parts of the assessment to peer reviewers.
As a former researcher and teacher, I recognise this dependence. Every reference in a text points towards work I have not carried out myself. I must assess what I can build upon and how far the evidence reaches. A reference does not relieve me of responsibility for the use I make of it.
Hardwig helps us understand why trust is necessary. But that necessity also makes the question of trustworthiness more demanding. When others must build on my work, my accuracy matters beyond the text I am writing.
Studies with the Outward Form of Knowledge
The anaesthetist and journal editor John Carlisle examined more than five hundred manuscripts describing randomised controlled trials. For more than a hundred and fifty, he obtained access to anonymised data from individual participants.
He found errors in 44 per cent. In 26 per cent, the problems were so extensive that, in his judgement, the studies could not be trusted. The cause could have been inadequate research practices or fabricated data. Carlisle called them “zombie trials”: they had the outward form of research, but their foundations failed under closer examination.
These figures concern a particular selection of manuscripts submitted to one journal. They do not measure how much medical research as a whole is unreliable.
Nevertheless, the difference between what Carlisle could see with and without access to the underlying data is striking. When he had only the summary tables, he identified far fewer serious problems. A table can look orderly even when the material behind it does not stand up to scrutiny.
As a reader, I usually encounter the finished account: the description of the methods, the figures and the conclusion. I can examine the argument. But an orderly argument offers no guarantee that the investigation took place as described.
When Many Studies Produce Too Much Certainty
A research review is intended to help us see what several investigations show collectively. If unreliable studies are included, the synthesis can create an impression of greater certainty than the evidence warrants.
The 2023 report describes research into tranexamic acid given preventively during caesarean births. A 2021 meta-analysis of 36 trials estimated a reduction of around 60 per cent in the risk of severe bleeding. A large American trial involving approximately eleven thousand participants found only a small effect in 2023, which was not statistically significant.
The researcher Ben Mol believes that problems with some of the smaller trials could explain the difference. This is his assessment, and the discrepancy does not itself prove that the studies were fabricated. The example nevertheless shows how much can be at stake when findings are combined.
The question here concerned preventive use during caesarean births. It must be distinguished from treating bleeding that has already begun. Otherwise, our own discussion of unreliable research may create another misunderstanding.
For me, the philosophical question lies in the movement from findings to certainty. How confidently can we speak? What must be examined before a conclusion is given greater weight? When the evidence weakens, the confidence of the claim must weaken with it.
The Absent Person
Central to Løgstrup’s thinking is the idea that we hold something of another person’s life in our hands. This is easy to see in the encounter between doctor and patient. The patient comes with his body, his anxiety and his hope for help.
In research, the connection is less visible. The future patient is usually absent when data are recorded, a manuscript is assessed and a research review is written. Yet something of his life may be affected by the decisions being made.
Here, I draw on Løgstrup’s thought to reflect on a responsibility that passes through institutions and documents. The researcher is responsible for the data. The editor is responsible for the assessment before publication. The review author is responsible for what is carried forward. Their tasks differ, but dependence connects them.
A deadline can feel closer and more pressing than the patient who will one day live with the result. For that very reason, the work must be organised so that this absent person also matters in the decisions being made.
Openness That Makes Inquiry Possible
In A Question of Trust, Onora O’Neill examines why more information and greater openness do not automatically create better trust. Information must be traceable to sources and open to examination. A flood of information can also make judgement more difficult.
This sharpens the question of what research owes those who depend on it. Making documents available is insufficient if no one can establish the connections between them, ask questions or receive answers.
In this context, I therefore understand openness as a practical opportunity for scrutiny: clear methods, traceable information and responsible access to the material. Participants’ privacy must also be protected. They have entrusted personal information to researchers, and that trust must be honoured too.
The more recent Nature report shows how demanding scrutiny can be. In the INSPECT-SR project, more than sixty researchers examined clinical trials in fifty Cochrane reviews. They reported “some concerns” about 25 per cent and “serious concerns” about 6 per cent of the trials.
These were not figures for proven fraud. The sample was not representative, and the project was also intended to test the assessment methods. Project leader Jack Wilkinson pointed out that the proportion might have been overestimated because some checks were subjective or difficult to apply. At the time, the study was available as a preprint.
The scrutiny itself must therefore remain open to correction. In the 2023 report, Alexandros Sotiriadis explains that a protocol allowed studies that failed a trustworthiness assessment to be excluded without declaring them fraudulent. That distinction protects both the evidence base and the person under scrutiny.
When a Correction Fails to Reach Its Destination
The first report also describes the consequences of Yoshihiro Sato’s fabricated research into bone health. Retracted studies had been used in numerous research reviews and guidelines. The researchers examining their impact repeatedly alerted others to this. At the time the report was written, only 11 of 88 reviews had been updated.
The error had been discovered, but it continued to live in texts that others might use.
Here, Løgstrup’s thought becomes concrete once again. Responsibility for another person includes the painstaking work that follows discovery: examining which conclusions change, correcting the review and notifying those who have built upon it.
This requires time and resources. The newer report discusses assessment tools that cost money and researchers who lack access to them. In the first report, Ben Mol points to promotion systems that, in his view, place too much weight on publication counts. Taken together, these accounts lead me to ask what institutions actually enable people to do thoroughly.
If scrutiny and correction are to be part of research’s responsibility, they must also be part of its working conditions.
Back to the Desk
The recommendation still lies on the doctor’s desk. The patient cannot trace every figure back to each participant, each analysis and each assessment. Nor can the doctor repeat the entire investigation.
They must be able to trust that others have done their part, and that someone will intervene when the evidence fails.
Behind the table, then, is a responsibility that continues after publication. It involves making the evidence open to examination, describing uncertainty honestly, responding fairly to objections and ensuring that corrections reach those who need them.
The recommendation will then also bear traces of the questions that were asked, the errors that were corrected and the uncertainty that was allowed to remain visible.
On the other side of the desk sits a person who must live with the consequences of what we say we know.
References
Hardwig, J. (1985). Epistemic dependence. The Journal of Philosophy, 82(7), 335–349. https://doi.org/10.2307/2026523
Løgstrup, K. E. (1956). Den etiske fordring. Gyldendal.
O’Neill, O. (2002). A question of trust: The BBC Reith Lectures 2002. Cambridge University Press.
Van Noorden, R. (2023). How many clinical trials can’t be trusted? Nature, 619, 454–458. https://doi.org/10.1038/d41586-023-02299-w
Van Noorden, R. (2025). Flawed trials pollute gold-standard medical reviews. Nature, 637, 256–257. https://doi.org/10.1038/d41586-024-04206-3
On the other side of the desk sits a person
who must live with the consequences of what we say we know.
This essay was written in a conversation with Claude/Anthropic and OpenAI/ChatGPT