Glen Clack, Chief Medical Officer at TheraCryf, shared on LinkedIn:
“… or, how a caveat dies on its way to Parliament.
In July the Department of Health and Social Care published an impact assessment supporting the extension of plain packaging to cigars, pipe tobacco and rather a lot else besides. Buried in the evidence base is a sentence to the effect that cigar and pipe smoking carry increased all-cause mortality, increased rates of smoking-related cancers, cardiovascular disease and respiratory conditions relative to never-smokers. Beneath it sits a footnote: Shaper and others, 2003.
I should declare at once that I hold no interest of any kind in the tobacco trade, and that I am not about to tell you cigars are harmless. They are not. But the sentence above is a summary of a summary of a paper that said something rather more careful, and the manner of its unravelling is, I think, the most useful thing a clinician will read this month. It is certainly more useful than anything I have to say about mahogany boxes.
Two populations wearing one label
The trouble begins with the phrase ‘cigar smoker’, which names not one population but two.
The first has never smoked cigarettes. The second smoked them for years, stopped, and took up cigars instead. The US Food and Drug Administration’s systematic review of this literature calls them primary and secondary cigar smokers, and the distinction is not cosmetic. Secondary smokers inhale roughly twice as often as primary smokers, and they arrive at their first cigar with lungs that have already been comprehensively worked over.
Separate them and the picture changes. In the American Cancer Society’s first Cancer Prevention Study, lung cancer mortality ratios were 2.1 in primary cigar smokers and 6.29 in secondary. For chronic obstructive pulmonary disease the primary figure was 1.42 and not statistically significant; the secondary figure was 4.39. All-cause mortality among primary cigar smokers came out at 1.08, and coronary heart disease at 1.05, which is to say hovering at the threshold of detectability.
Before anyone reaches for a corona in triumph, note what survives the separation. Among primary cigar smokers who reported no inhalation whatever, oral cancer mortality was 6.98, laryngeal 10.6 and oesophageal 3.40. In the EPIC cohort, exclusive current cigar smokers showed upper aerodigestive cancer at a hazard ratio of 3.5. A pooled analysis published this year, restricted to 518 people who had smoked pipes or cigars and never cigarettes, found all-cause mortality at 1.24 and COPD at 2.02.
So the cancer signal is real, substantial, and largely indifferent to whether you inhale. It is the cardiovascular and respiratory limbs, in the older literature, that turn out to belong in significant part to the cigarettes these men used to smoke. The exposure was labelled ‘cigars’. The exposure was partly Woodbines.
The same error, held up to a mirror
Every clinician of a certain vintage knows the J-shaped curve: the finding, reproduced for decades, that moderate drinkers outlive abstainers. It launched a thousand newspaper columns and a good deal of well-upholstered self-justification at dinner parties.
The difficulty is that the abstainer category contains two populations. There are people who have never drunk, and there are people who stopped because they had become ill. Put the sick in the reference group and everybody else looks marvellous by comparison. This is the sick-quitter effect, and it is the same structural error as the cigar problem, simply pointed the other way: contaminate the exposed group and you overstate harm, contaminate the reference group and you manufacture benefit.
The paper that identified it was a prospective study of 7,735 middle-aged British men, 504 of whom died over seven and a half years. It found the U-shaped curve and then, rather than celebrating it, took it apart as systematic misclassification. There is a detail in the abstract that I find quietly remarkable: the highest mortality of all was among ex-smoking non-drinkers. The authors had already noticed, in 1988, that people carry their old exposures into their new categories like luggage.
The paper is Shaper, Wannamethee and Walker, in the Lancet. The cohort is the British Regional Heart Study.
The same three names. The same cohort. Fifteen years before the cigar paper.
This is the part that ought to give one pause. The investigator who taught the field to distrust a contaminated reference group went on to publish a study whose exposed group was contaminated in precisely the same way; noted the fact in his own text, as he was bound to; and it made no difference at all to what happened next. The error is not a failure of intelligence or of care. It is structural, and it survives the people best equipped to catch it. When the abstainer bias was finally removed, through meta-analyses restricted to studies that had handled it properly and through Mendelian randomisation in very large cohorts, the cardioprotective effect of moderate drinking largely evaporated. It had been luggage all along.
A third form, for which we have no excuse
There is a version of this that requires no contamination whatever, only aggregation, and it is the one I meet most often in my own work.
‘HRT’ is not an exposure. It is a class label covering distinct molecules, given by distinct routes, at distinct points in a life. Oral conjugated equine oestrogen with medroxyprogesterone acetate, started a decade past the menopause, is not transdermal oestradiol with micronised progesterone started at it; the first-pass hepatic effect alone is sufficient to separate the thrombotic risk profiles. Yet an estimate generated in the first population has been quoted, for twenty years and with great confidence, to women in the second.
Molecule, route, timing. Whenever a class label is doing duty as an exposure variable, the only respectable question is what precisely is being averaged, and whether the person in front of you is anywhere near the average.
Why nobody checked
Here is what I think is the useful lesson, and it is not really about cigars at all.
The J-curve was interrogated relentlessly for thirty years. Every methodologist with a pulse had a go at it. Why? Because ‘alcohol is good for you’ is a surprising claim, and surprising claims attract scrutiny the way a wet dog attracts a clean carpet.
Nobody interrogated the pooled cigar estimate, because cigars are harmful, and so the answer looked right. A right-looking answer stops enquiry dead.
We tell ourselves that scepticism is a disposition, evenly applied. It is nothing of the kind. It is a finite resource, and we spend it almost entirely on findings we dislike. The estimates least likely ever to be examined are precisely the ones that confirm what we already believe, which is an uncomfortable thought for anyone who has ever cited a paper because its conclusion was congenial and its methods section looked long enough to be respectable.
The three-document exercise
If you teach, this is worth ten minutes with the documents on screen.
Shaper 2003 reports estimates for pipe and cigar smokers, primary and secondary combined, and says so plainly.
Chang and colleagues, in the FDA review of 2015, separate the two groups, quantify the gap, and name prior cigarette smoking as the principal confounder requiring exclusion.
The impact assessment of July 2026 cites Shaper for the proposition that cigar smokers have elevated cardiovascular and respiratory risk relative to never-smokers, unqualified, in support of regulations to be laid before Parliament.
The caveat is present in the primary source, sharpened in the review, and gone by the time it reaches the instrument that changes the law. No one lied. Each step is a defensible summary of the one before it. That is what makes it worth teaching: citation decay is not a scandal, it is entropy, and it requires active work against it rather than indignation.
A note from the day job
As some of you will know, I spend my working life designing trials in oncology. Prior therapy is not something one discovers in the analysis. It is a stratification factor, specified in the protocol before the first patient is consented, because no regulator on earth would accept a single hazard ratio from an arm that pooled treatment-naive and heavily pretreated patients. We would be laughed out of the room, and rightly.
Observational estimates that commit exactly that error, meanwhile, proceed unstratified into secondary legislation. I do not think this reflects badly on epidemiologists, who have known about it since at least 1988 and said so. I think it reflects on what happens to a qualification as it travels.
Four questions
Ask them of any observational estimate you are about to act on, or cite, or put in front of a patient.
- What is in the exposed group besides the exposure?
- What is in the reference group besides its absence?
- What is the label averaging over?
- And the difficult one: does this estimate point the way I expected it to, and if so, when did I last actually check it?
I have no commercial interest in the tobacco industry. The interview I performed with Christopher Foxley-Norris from FOREST in my second year at medical school, and subsequence defence of the right to smoke tobacco in debate, was only to wind up my peers… I realise that this belligerent behaviour will come as a shock to many of you…References available on request, and I would be glad of the argument.”
FDA Authorizes Marketing of Four New Nicotine Pouch Products