From epidemiology to polypharmacy, Matthew Edwards provides a selection of highlights from the IFoA’s forthcoming Longevity Bulletin
Researching research: the ups and downs of epidemiology
Matthew Edwards and Dan Ryan
Readers of articles summarising medical studies might assume that the combination of rigorous study design, impartial analysis and independent peer review would make the output 'cast iron'. Why, then, do we read statements such as:
- 'Much of the scientific literature, perhaps half, may simply be untrue' (Richard Horton, editor, The Lancet, April 11, 2015)
- 'It is simply no longer possible to believe much of the clinical research that is published, or to rely on the judgment of trusted physicians or authoritative medical guidelines' (Dr Marcia Angell, former editor of the New England Journal of Medicine).
What has driven these concerns? And if they are well-founded, how can we assess the validity of any published research? How did epidemiology come to be in this position?
Epidemiology became a promising field of medicine in its own right in the 1950s, with well-known studies relating to smoking and the discovery of the harm caused by tobacco, as well as hypertension. How, in the field's infancy, would one interpret or judge epidemiological results?
In 1965, Bradford Hill (whose work with Richard Doll helped determine the link between smoking and lung cancer - see Figure 1) published 'The Environment and Disease: Association or Causation?' in the Proceedings of the Royal Society of Medicine. In this paper, he introduced what have been known since as 'the Bradford Hill criteria' (see box). They have remained the standard criteria for distinguishing association and causation ever since, with a much wider scope than just medicine.
The rise and fall of epidemiology?
For some time, there has been a 'diminishing returns' problem: each decade has felt less illuminating than the previous decade. Far from the aforementioned ground-breaking studies on smoking and hypertension, modern epidemiology seems more concerned with continually conflicting claims about aspirin, broccoli, coffee and so on (the list could be extended alphabetically to Z without much difficulty). And quantification of the reputed harms or benefits - with typical risk factor effects shown of the order of 10-20% for particular causes of death - shows life expectancy impacts of the order of a few weeks, compared with the massive life expectancy impact of five-10 years (for a 50-year old) implied by the smoking studies.
Hierarchy of evidence
The hierarchy of medical evidence is often expressed in the pyramid shown in Figure 2. Higher layers represent increasing internal validity as the risk of bias is reduced (if not eliminated). The efforts taken to allow for confounding factors increase the likelihood that the phenomena observed are likely to result from the causative mechanism of interest.
However, the dividing lines are not always clear cut. For example, while a cohort study tracks the experience of many individuals over time to assess the longitudinal impact of risk factors or treatment, carefully selected pairwise comparisons in a case-control study may provide stronger explanatory power. One problem with very carefully designed studies, though, is that the results will hold true only if the selected population from which participants are drawn is representative of the population to which the treatment or intervention is applied (this issue is explored further in the article on polypharmacy).
The importance of sense checking
While epidemiology remains a vital field of exploration, all claims need to be properly 'sense checked' (given the unreliability of the peer review system). The Bradford Hill criteria are an excellent reference point, but there are other points for actuaries to consider:
- Impact - what is the all-cause mortality impact if the results are correct? For many findings published in recent years, the impact on life expectancy is negligible
- Data bias - how might bias be present in data selection, or in the operation of confounding factors?
- Commercial bias - given the penalties of $38bn applied to pharma firms since 2000, it is clear that questionable approaches have often been adopted
- Association or causation - few study types allow us to safely infer causation, although association may be sufficient in a typical insurance underwriting context
- Biological plausibility - are the results plausible regarding the underlying biological or real-world process involved and, similarly, are they consistent with clinical evidence?
Proper appreciation of models requires us to give attention to the real world underlying the models, as Bradford Hill warned us many years ago. Epidemiology is more likely to be misleading when it becomes a purely statistical enterprise, divorced from its underlying reality.
Case study: Bias risk in observational studies
An interesting question is the degree of bias present in observational studies, where studies are constructed 'retrospectively' through the analysis of historical datasets rather than set up on a prospective basis. The study would typically split records according to patients' apparent use of a particular medication or treatment - but adhering to that medication is likely to be a sign of a 'doing what I can to be healthy' attitude, and hence associated with other healthy behaviours.
The most famous (or rather, infamous) example of this bias relates to advice recommending hormone replacement therapy (HRT) following the results of the 1985 Nurses' Health Study. This was an observational study that found a 42% reduction in cardiovascular risk associated with HRT, while an equivalent randomised controlled trial conducted some years later showed a 29% increase in cardiovascular risk from HRT. The difference was seen as largely attributable to other healthy behaviours of those women who had opted for HRT.
Other studies have looked at this issue from the perspective of adherence within the placebo groups of randomised controlled trials - ie what is the mortality of 'placebo taking' versus 'placebo neglecting' individuals? Such studies have shown very striking differentials of the order of 50% risk reduction in respect of placebo adherence - which is acting here purely as a marker for general healthy behaviour, independent of the treatment being studied.
The Bradford Hill criteria
Strength (effect size):The larger the association between 'dose' and response, the more likely that it is causal.
Consistency (reproducibility): Consistent findings observed by different persons in different places with different samples strengthens the likelihood of an effect.
Specificity: Causation is likely if there is a very specific population at a specific site and disease with no other likely explanation.
Temporality: The effect has to occur after the cause.
Biological gradient(ie dose-response relationship): Greater exposure should generally lead to greater incidence of the effect.
Plausibility: A plausible mechanism between cause and effect is helpful (although knowledge of the mechanism is limited by current knowledge).
Consistency between epidemiological and clinical findings increases the likelihood of an effect.
Experiment: Do preventive actions taken on the basis of an assumed causal association alter the outcomes?
Analogy: What are the effects of similar factors?
Polypharmacy: is too much medicine reducing life expectancy?
Dr Malcolm Kendrick
Polypharmacy refers to the prescription of multiple medications. An increasing number of people feel that we are in danger of doing more harm than good, with an increasing number of medical interventions.
Part of the problem is the continual lowering of the 'normal versus high' thresholds for measures such as blood pressure and cholesterol. Along with the many millions of people 'diagnosed' with hypertension, more than 60% of the adult population is now considered to have a raised cholesterol level.
The labelling of more and more conditions as diseases that require medication has led to an increasing number of drugs being prescribed - and this can be seen in many countries. However, in the UK, another factor has exacerbated the polypharmacy problem. This is the Quality Outcomes Framework (QOF), which regulates payments to GPs.
QOF incentivises GPs to prescribe in accordance with the guidelines. However, these guidelines are based on evidence relating to drugs in isolation. There have been limited attempts to establish whether combining medications creates an additive benefit - but prescription drugs are powerful agents that should be used with caution, as this 2014 study from Harvard University makes clear:
About 128,000 people [in the US] die from drugs prescribed to them. This makes prescription drugs a major health risk, ranking 4th with stroke as a leading cause of death. The European Commission estimates that adverse reactions from prescription drugs cause 200,000 deaths; so together, about 328,000 patients in the US and Europe die from prescription drugs each year.
The conclusion of a study in Israel which looked at reducing the number of medications given to elderly and disabled patients showed reduced costs, improved quality of life and a reduction in overall mortality. It could be said that stopping drugs is the single most effective drug treatment currently available.
Other subjects covered in the Bulletin
The forthcoming issue of the Longevity Bulletin also contains articles on:
Diabetes and pharma: How pharma is both helping diabetes (metformin) and exacerbating it (statins),by Nicola Oliver
Opioids in the UK:A medical analysis by Dr Chris Martin
Opioids in the US: A statistical overview by Dr Magali Barbieri
Juvenescence: How pharma applies in the anti-ageing space, by Simon Pavitt