A liability, science and media briefing based on consideration of the recent high profile research publication, reference:
N Engl J Med (2019) Vol. 381 p 1801-1808. Daniel F. Mackay et al.
Neurodegenerative Disease Mortality among Former Professional Soccer Players
Science research and popular narratives both have influence with the decider-of-fact at common law. It follows that both have an influence on liability exposure. This report has two main aims:
1) To present a brief analysis of the science research report (referenced above).
2) To report and assess the popular themes chosen by the press in response to the publication of this research in Nov 2019.
Emerging liability risk.
If some aspect of football playing was found to be causal of neurodegenerative disease (NDD) then similar scenarios could expose a range of liability insurance policies. Scenarios could include school sports, amateur sports, other professional sports and some occupations.
In the UK, severe dementia affects around 4% of people aged over 65. It is a care-intensive condition . It does not usually affect people of working age . A new liability exposure could be a significant one, based on general damages and care costs.
The study was of causes of death. Compared with a non-footballer reference group, former professional footballers, born before 1977, had a higher rate of death being attributed to neurodegenerative disease. The statistical analysis is probably correct. The finding is in need of explanation.
There was no new data on football-related risk factors, such as concussions, and there was no data on the many relatively protective factors that may be present in the non-footballer reference group.
The media has not been much engaged in a critical analysis of the study, preferring instead to praise it and assume it means what the researchers say it means. The media then go on to explain the results in terms of preconceived models.
There are several reasons to curb any enthusiasm for interpreting the findings of this study in terms of liability insurance exposure. Without further work, the study offers no new insights into the causes of neurodegenerative disease in former professional footballers, or anyone else, and has nothing to say about legal causation.
Several of the findings point to study-design factors, as opposed to biological factors, being the origin of the measured risk increase. But this cannot be definitively decided without access to the data.
If the media view is a persuasive one then this study confirms a high degree of risk to the health of football players, but, the media view is overstated. While liability exposure is influenced by media fashions the traditional tests of breach, generic causation and specific causation (including material contribution and acceleration) hold sway in rational jurisdictions. If it relied solely upon the findings of this study, an ‘emerging risks reserve’ for indemnifying sport-related neurodegenerative disease would not pass the ‘balance of probabilities’ test.
Study Design Factors
Interpretation of science reports begins with an assessment of study quality. This may be achieved by an analysis of its strengths and weaknesses and from quantifying the degree to which the main findings may be different under different design assumptions, and, different by chance. Many reviewers attempt to substitute a bias analysis for a quality analysis, but while there is some overlap in these concepts they are not the same. A design may well be unbiased and yet, completely wrong.
From an informed knowledge of dementia and of sport, studies aimed at identifying dementia risk associated with sports participation, or sportsmen more generally, would need to account for: age , aerobic fitness effects, early life exertion patterns, CVD risk factors, genetics, epigenetics (especially early life/prenatal nicotine exposure), childhood stressors, use of recreational substances (smoking), alcohol abuse (possibly binge drinking and alcoholism), diet, hydration habits, common diseases (especially inflammatory ones), economic status (at all ages), access to health care, well-being motivation, IQ, education level, cognitive reserve etc. This means that any true control group would need to be matched in several important ways, not just by age. The alternative is that statistical correction be made for any un-matched variables.
Given the available understandings, a fairly fixed proportion of people would develop degenerative disease if there was no intervening event (e.g. death from other causes). Rather than initiate a disease process, the observed risk factors may accelerate the biological and cognitive processes leading to manifest degenerative dementia.
There is a theory that education and a fulfilling life can increase resilience both by direct biological effects in the brain (a ‘wiring effect’) and in coping behaviours . If this is true, brain development in childhood/teenage would be most important. Study reference groups which do not match for childhood/teenage factors would be of limited value in understanding initiation or acceleration of NDD.
Probably positive design factors
Former footballers (FF) were identified from a profession database – no other selection criteria were applied. 80% of registered FF were probabilistically matched with health records. This is a reassuringly high rate. The missing 20% could introduce bias into the results, but without knowing the exact reasons for database mismatches nothing further can be said other than that neurodegenerative disease bias would need to be fairly strong if the given results depended on it.
A large number of age matched subjects were included. 7,676 FF and 23,028 in the reference population. This provides a good chance of statistical power and should largely remove the age=risk effect. Fine statistical adjustment for age would still be needed.
Minimum age at study baseline was 40, so there was a high chance of large numbers of neurodegenerative disease diagnoses and deaths during an average follow-up window of 18 years. Again, this provides an opportunity for statistical power .
Outcomes were as stated on death certificates as either primary cause of death or, contributory cause of death. FF status per se was unlikely to influence the records made and so there should be no differential bias introduced. This is a good thing.
However, the actual cause of death in neurodegenerative disease (NDD) varies and is in the vast majority of cases not the NDD itself. A very high proportion die of pneumonia, infections or heart failure. Choosing to record neurodegenerative disease as the primary cause of death in such cases is unfortunately subject to reporting fashions, which may have changed over the study period, and especially since the Astle coroner’s verdict . Compared with autopsy (not a perfect method, but regarded as the gold standard) the rate of agreement in dementia cases can be 50%.
Aspirational pneumonia is usually and probably rightly attributed to dementia. This, and interventions to prevent it, would also usually trigger dementia as the recorded cause of death.
It would therefore be very helpful to the interpretation of this study if there was data on the reasons why NDD was recorded as the primary cause of death.
Such is the nature of dementia that people with a lower cognitive reserve are more likely to be medicated for neurodegenerative disease during their later years and the presence of a written history of dementia medication could influence the diagnosis on the death certificate.
The reference group was matched on the basis of affluence as determined by post code at around the time of death. Affluence has an effect on health care seeking and exposure to some risk factors e.g. diet. Socio-economic risk factors are well known in some diseases of ageing.
The accuracy of death certificates was not assessed. If the accuracy was different from FF to reference group this could introduce bias. This potential factor cannot be assessed from the report.
This would be especially problematic for the assessment of contributory causes of death. Dementia can affect all kinds of mortality processes but the significance attached would be a matter of judgement and intention. For this reason primary and contributory causes of death should not be combined in the analysis unless the separate results are also presented.
Post-code based matching applies only to the period around the time of death. FF usually originate from the less financially advantaged social strata. Social mobility before 1977 was low. A difference in childhood/teenage risk factors may not be accounted for by the chosen matching method.
Standard statistical tools for medical research are designed on a precautionary basis. This creates the impression of significant differences where any normal person would say there were none.
The importance of these potentially positive and negative factors can only be judged from the results and analyses.
Primary cause of death
Neurodegenerative disease was listed as the primary diagnosis on the death certificate at a rate of 1.7% (n = 134) among former soccer players and 0.5% (n=120) among the reference group. Comparison gives a hazard ratio (HR) of 4.10 (95% CI = 2.88 to 5.83). The HR value is both high and precise; chance alone is not a likely explanation. Adjustment was made for the risk of death from heart disease, which was less common in FF, and for “any cancer”, and this brought the HR value down to 3.45 (2.11 to 5.62) . This remains a high and precise HR value, not likely due to chance.
The reason for co-adjustment was not explained and the separate effects of adjustment for each were not reported. The reason for not adjusting for lung cancer alone was also not explained yet the risk of lung cancer in FF was very much below that of the reference group [HR = 0.53 (0.40 to 0.70)]. The average effect for “any cancer” was null. A fuller account of the effects of adjustment is needed if it is to be interpreted.
Given that age is a strong risk factor for developing NDD, if footballers lived for longer then more of them would be at risk of NDD. It would be expected that the observed lower risk of lung cancer and heart disease would lead to a relatively longer life on average. In fact there was no significant difference in average age at death for the 15% that died during the study period. This puzzle could be explained in part by an asymmetric acceleration of NDD. Details of the age profile would be needed to make an assessment of this.
There was no difference in risk of death from stroke, yet in most studies, high aerobic fitness is a protective factor against stroke. FF certainly have periods in their lives when they have high aerobic fitness and for most this is a lasting condition. Lack of protective effect in stroke is a puzzle.
Unfortunately the authors do not say if the relative risk of NDD increased with age. If the authors had found an age-dependant HR value it is assumed that they would have said so given that they several times pointed out when the Cox proportional hazards assumption was not met. That they didn’t say this would then mean that the HR value was the same at all ages. In other words, no asymmetric acceleration of NDD. Acceleration would often be needed to evidence and argument for material contribution.
The authors also didn’t say whether the HR values for NDD mortality varied by post code classification. Given the effort to match the reference group by post code the authors clearly thought this would be an important factor. Checking to see if it was an important factor would have been consistent with that hypothesis. There was no report of this.
Joint cause of death
For NDD as both primary and contributory causes of death (n= 222 in FF (2.9%) and 228 in the reference group (1.0%)), HR = 3.53 (2.72 to 4.57). Again this is a high and precise HR value and is unlikely to be caused by chance alone. The value did not vary with field position (outfield vs goalkeeper). HR values for this combined outcome appeared to vary by disease type. Alzheimer’s, motor-neurone and Parkinson’s disease all showed high and precise HR values. The values are reported as being different, however, with the exception of Parkinson’s disease, there is a less than 50% chance that they are actually different degrees of risk. These individual disease HR values also did not vary with field position. HR values for these diseases as the primary cause of death were not separately reported yet these would in principle be the more reliable death certificate reports.
There were also high HR values for non-Alzheimer’s dementias (e.g. vascular dementia) and dementia ‘not otherwise specified’.
Alzheimer’s, motor-neurone, Parkinson’s and vascular dementia are distinct in terms of anatomical location, cell mortality processes and biochemistry. There are some very general commonalities but it would be remarkable if some common factor acted to increase the biological risk of all these forms of dementia. A non-biological explanation would be the more natural subject of investigation, even if only to eliminate its effect from further analysis.
The number of cases where two or more diagnoses were present was not reported. It would be useful to compare the rates of multiple diagnoses with those predicted.
Once initiated, NDD is progressive and would probably be described in law as an indivisible injury. Material contributions to initiation would in that case be exposed to a risk of 100% of the liability. The current research report provides no specific evidence of contribution to initiation. However, the very high HR values obtained, if valid, would tend to be persuasive of specific causation if a valid biological rationale linking NDD to FF could be found. No such rationale is as yet established. It may be that legal contribution to initiation of NDD is not provable. Initiation of NDD has thus far eluded explanation in medical science.
The mortality risk for dementia was higher for former professional footballers than for a reference group which was chosen on the basis of post code at age of death. For many reasons, it remains uncertain whether or not this reference group is the right one. How this reference group compares with other reference groups would be of great interest. An analysis of excess NDD risk by post code classification would shine light on this.
From a medical causation view, it is unlikely that five biologically different classes of dementia would all be at increased risk for same reason, once attained-age effects have been eliminated. It is therefore a matter of high priority to discover whether or not the measured raised risks were caused by the design of the reference group. Whatever the case, there was no difference in raised risk by field position, so it is perhaps unlikely that footballing activities per se were the cause of the observed raised risk. The uncertainty in this part of the conclusion arises from the (understandably) relatively low proportion of goal keepers.
Several potentially helpful analyses were not provided. The question of diagnostic accuracy would be illuminated by separate analyses of primary and contributory mortality attributions per disease type and by decade of death (given that reporting fashions do vary in time). Age dependence of NDD risk rates would help in assessing the presence or absence of acceleration of what is a cumulative disease process. Lack of acceleration would tend to point to a non-causal association.
A great many biological (e.g. genetic) and circumstantial risk factors (e.g. propensity for health care seeking) were not studied. This means that the potential relationship between the raised risk, if not an artefact of the study design, and the fact of playing football cannot be assessed from this research. There is no basis here to conclude that playing football increases the risk of NDD.
The scale of uncertainty in diagnosis and the effects of disease correction would probably not be sufficient to explain away all of the apparently excess risk that was observed in this ~18 year window of observation. The potential scale of the effect of choosing the wrong reference group cannot be assessed from the data provided, but could be substantial . When combined, the effect of design factors would be worth a thorough assessment. Only the authors of the report, with full access to the data, could do this.
When comparing this group of former professional footballers with a reference male population matched by age and post code at time of death, a raised risk of reported death from neurodegenerative disease was observed.
Bias in the press/ news media.
In response to this research publication, several very common themes were championed by the lay press:
• Playing football is the cause of this research observation i.e. the raised risk of NDD.
• Heading a ball and concussions, and “sub-concussions” may be the relevant risk factors.
• Amateur football may not present the same risk as professional football.
• Playing sport brings general health benefits. [banning sports would be a shame]
• The football authorities (FA) are actively researching this phenomenon.
In short: the tone is that it was heroic of former professional footballers to put their health at risk, the FA are behaving (reasonably) well, sport must go on, care (and compensation) should be provided for former professional footballers.
A minority of the press offer that the FA knew there was a problem long ago and did nothing or did not do enough. In the UK the notion of corporate conspiracy as an explanation of unwanted events is not an easy win for the popular press, the public are not immediately swayed by such stories. This is very different in the USA where a conspiratorial explanation converts a possibility into a certainty.
It is often proposed that the press will only say what they think the public will believe or will persuade themselves to believe given the right kind of half-truths and appeals to common beliefs. A press which opposes common prejudices will lose its readership. It is likely therefore that deciders of fact, if not given enough scope to adopt an intellectually rigorous approach, would readily align with the view of the press. The press is therefore a bell weather of ‘readiness to believe’.
The research report described here actually provides no support for any of the themes championed by the lay press, and some of the medical press, in response to it. There is clearly a predetermined media view that the reader is thought likely to accept.
To jump from what is an uncertain fact to a decided fact because the fact can be explained by a preferred model is an example of causation bias .
In defence of the media, the research report also doesn’t say that the chosen media themes are clearly wrong. It may be offered that absence of explicit opposition to a theme may be sufficient justification for promoting it. Uncertainty provides opportunity for inventive narrative.
Exactly why the media chose this narrative line could not be guessed from the text of the reports that were assessed. It is proposed here that if the reason is a lasting one e.g. the pursuit of justice, then the media themes should be stable over time. Other reasons, such as falling out with the FA, would not be as long-lived.