[Listened as audiobook]
https://www.penguinrandomhouse.com/books/205067/overdiagnosed-by-h-gilbert-welch
This is a book about overdiagnosis of medical conditions, it digs in to the problems of screening and some implications of discovery of “incidental-oma”s, growths seen on a scan but unrelated to the reason for the scan, and other incidental findings.
This book has put into perspective some of the things that were worrying me about screening – I knew screening was often detrimental, but I hadn’t understood quite how bad it is. There are some situations where screening is worthwhile, and likewise where medical control of proxies is worthwhile, breast cancer screening and high blood pressure being good examples. Many other screening exercises lead to anxiety and risk from intervention without improving overall outcomes.
There was also some discussion of the extension and redefinition of disease to include more people in the condition – if a condition is measured, adjusting the cutoff point for diagnosis so mild versions of the disease are included. When we do this, more people undergo more invasive diagnosis and treatment, even when the condition is not interfering in their lives.
There are also some psychological and cultural aspects to the perpetuation of screening, where people believe that they have been saved by a doctor who found something they were totally unaware of and treated it. These people then become advocates for screening, “catch it early”, “it will save your life”.
The thing to look out for is the number needed to treat, if you need to treat three or four people to save one, as is the case with high blood pressure, it is clearly worthy. If you need to treat a thousand to save one, the value is much less clear cut – indeed, if those thousand then go through a procedure or take a medicine which carries risk, as most do, you don’t need to have many side-effects or bad outcomes to cancel out that one saved per thousand.
Another concern considered in the book is about the use and abuse of statistics, looking at things like ten year survival rates from initial diagnosis doesn’t hold any value in those cases where diagnosis is accelerated by screening.
This book is grounded in the American system, where there are more financial incentives to overdiagnose otherwise well people to turn them into patients, especially where the only way to know if you are getting better is to have another scan and have the radiologist tell you it looks better – there is no symptom to be relieved.
I’m not sure how well it translates to the British health system – I would like to think that the NHS is better able to understand, and therefore deal with, the problems of overdiagnosis. The NICE guidelines should incorporate mitigations for the worst overdiagnoses, but I am also sure that there will be significant pressure from the public, and some well meaning but not so well (statistically) educated medical professionals, to perform more screening, more finding things early.
Postscript: While I was writing this review the BBC published an article about How AI can spot diseases that doctors aren’t looking for. This article is an example of the perpetuation of the myths of early diagnosis – the protagonist in the piece was ‘identified’ as having osteoporosis based on machine readings of his bone density, which is claimed human doctors are biased against looking for in middle age men – it is considered more prevalent in old thin white women. The machine was looking at a scan taken for GI related issues. This is a classic example described in the book, with a symptom free vertebra collapse and low measured bone density without any symptoms like frequent fractures. Oddly, it is only at the end of the article that there is a nod toward the harms of this kind of incidental screening, and even then it concentrates more on the harm to the NHS than the people diagnosed.
The only bit of sense in the whole article is from Prof Javaid: “We want to build the evidence to use it across the NHS”. Building this evidence in a reliable way is non-trivial, and probably involves longitudinal studies of thousands and carefully designed RCTs. My expectation would be that there are a few things where the results indicate significant improvement in overall outcomes without undue anxiety or risk, but most screening will show that it either doesn’t improve outcomes, or worse, actually causes harm. But I don’t think we have the data yet to draw robust conclusions. And I doubt there will be much funding for these kinds of studies – pharma and medical tech won’t want to go looking for results that say you should use less of their products.