Flogging toothpaste may be a dull business, but for once eyes must surely have shone brighter than teeth in the marketing department at Colgate this week. A gift of a study, published in the BMJ last Thursday, linked poor toothbrushing to heart disease. The media predictably flipped the message, with headlines certain to fix a smile on even the most jaded of Colgate lips. Auntie exhorted us to ‘Brush teeth to halt heart disease’, while the Daily Mail directed ‘Clean your teeth twice a day to keep a heart attack at bay’. The ping was at last back in the Colgate ring of confidence, for who needs advertising, when sparkling headlines (351 of them, according to google) say it all?
The research is not without interest. It kicks off with what we might call intuition bias, by echoing the old saw about an apple a day. And what could be more likely than scabby teeth means dodgy tickers, as plaque ridden teeth mirror plaque filled arteries (Dr No kids you not – one Mail commenter equates dental plaque with arterial plaque – and advises using floss to remove it!). But what about the science?
On the face of it, the study appears sound enough, and ticks most of the right boxes – apart from, as we shall see, a crucial one. It is population based, and involved large numbers of subjects. It is ‘prospective’, in that the analysis starts with the risk factors, and moves forward in time to see who develops disease, even if, somewhat confusingly, the study was done retrospectively – in other words, after the event – making it in fact a so-called retrospective, or historical, cohort study. Such studies differ from true prospective studies in that, instead of actively following up subjects over time, they passively make use of data that has already been collected; and so are reliant not on their own fastidious data collection and follow-up, but on the past exertions of others, which may – or may not – have been sufficiently rigorous.
The subjects came from the Scottish Health Survey, a comprehensive occasional survey that collects a wide variety of demographic, risk factor and health related McData from a cross-section of Scots people. The outcomes – ‘cardiovascular disease events’ – were identified using the lugubriously named Information Services Division, a Scottish outfit that collects, amongst other things, identifiable patient data including main diagnosis on hospital admissions, and official records of deaths. Having identifiable data allowed the researchers to ‘link’ the SHS data to the outcome data – and so analyse how risk factors affected outcome. So far, so (reasonably) good (ISD hospital data is considered to be reasonably robust).
The analysis was done using Cox’s proportional hazards model, a widely used but – we are entering dark territory here – poorly understood statistical method that allows analysis of survival data even when the final outcome is not known for all subjects, based on multiple risk factors. We may think of it as a specialised form of multiple regression, that allows us to tease out the extent to which numerous risk factors act and interact to affect survival. The study’s headline result can be summed up simply enough: compared to those who brushed their teeth twice a day, those who brushed less than once a day were 1.7 times (95% CI 1.3 to 2.3) more likely to experience a cardiovascular disease event, even when allowing for the influence of a large number of other risk factors (age, sex, socio-economic group, smoking, physical activity, and visits to dentist, BMI, family history of cardiovascular disease, hypertension and diabetes) – and the 1.7 (compared to 1) gives rise to the ‘70% increased chance of heart disease’ widely reported in the press.
Now, at this point, we could embark on a detailed assessment of the rigour of the study and its methods. We could ask, for example, do we know that self-reported tooth-brushing behaviour – a so-called proxy – actually relates accurately to oral hygiene (in fact, research suggests it does). We could ask about outcome identification (recall it relied on inpatient statistics and deaths – what about patients solely in primary care?). We could even assess the viability of the Cox proportional hazards model (and probably end up with a thumping migraine) – but we need do none of this, because the study contains one big black hole – the mystery of what happened to the toothless hags. As long as that black hole remains, the study is flawed.
It goes like this. The SHS collects data on tens of thousands of people. This particular study, reasonably enough (cardiovascular disease mainly affects older people), decided to limit subjects to those aged 35 and over, giving a starting pool (calculated from figures in the paper) of 16144. From that pool, 204 were excluded because of missing demographic data (fair enough) and 386 because of pre-existing heart disease (fair enough). And then – a staggering, whopping 3685 – nearly one quarter of subjects – were excluded because they were edentulous (that’s no teeth to you and me). One in four subjects, so to speak, bit the dust, or at least would have done, had they any teeth with which to bite the dust.
This creates a black hole that we know very little about at the heart of the study – apart from, that is, a throwaway line that describes this group as more likely to be older, female, and smokers – a bunch, in other words, of toothless hags.
Now the problem is this: we simply do not know what happened to the toothless hags, or what the effect of their exclusion from the analysis had on the results. Let us suppose, for example (not unreasonable, because they are older, and female) that they are in fact survivors who did not suffer cardiovascular ‘events’. We might also suppose that a good number of them did not major on oral hygiene (presumably they were edentulous for a reason) and so many – we may suppose – belong to the poor oral hygiene group. Yet they have been excluded from the analysis, in effect weighting those who were analysed, in that poor oral hygiene group, towards a more male group, who are more likely to suffer a cardiovascular event. We are in effect selecting subjects for inclusion in the poor hygiene group who are more likely to have an adverse event – a form of error called selection bias, which – surprise surprise – can falsely elevate, and so over-estimate, the calculated risk.
All of this is conjecture, but reasonable conjecture; the problem is we simply do not know. The researchers provide no rationale (and Dr No cannot realistically see one) for excluding the edentulous from the analysis; and yet they were excluded. Dr No wonders why. And so it is that this apparently whiter than white study has a black hole at its heart – the mystery of the toothless – and quite possibly bearded, we just don’t know – hag.