With the exception of some rare familial forms of the disease, the cause of amyotrophic lateral sclerosis (ALS; motor neuron disease) is unknown (Bruijn et al., 2004). Although 90–95% of ALS cases are apparently sporadic, SOD1 gene mutations with low penetrance (Cudkowicz et al., 1997; Anderson et al., 1997; Shaw et al., 1998; Al-Chalabi et al., 1999; Orrell et al., 1999; Andersen et al., 2003; Majoor-Krakauer et al., 2003; Nogales-Gadea et al., 2004) and other gene variants or mutations may contribute to the population risk (Al-Chalabi et al., 1999; Mitchell, 2000; Lambrechts et al., 2003; Majoor-Krakauer et al., 2003).
Advances in understanding the molecular basis of neurodegenerative diseases point clearly to the importance of genetic factors in conditions that until recently we took to be determined mainly, if not solely, by environmental factors, for example, Alzheimer's disease and Parkinson's disease. Genetic predisposing factors may, of course, require an environmental trigger to activate disease processes. There has been no shortage of epidemiological research into ALS attempting to identify a link between ALS and trauma, exposure to toxins, physical activity and a variety of other possible risk factors (Mitchell, 2000; Worms, 2001; Armon, 2003a, b). However, the only really consistent epidemiological risk factors for ALS are increasing age, male sex and a family history of ALS (Kurtzke, 1991; Mitchell, 2000; Worms, 2001; Armon, 2003a), although cigarette smoking may also be a risk factor (Nelson et al., 2000; Armon, 2003a; Weisskopf et al., 2004). Even in Guam, where a restricted population with high risk of developing ALS has been under intense (albeit intermittent) scrutiny for over 50 years, controversy persists and no environmental (or genetic) causes have been identified beyond doubt (Kurland and Mulder, 1954; Cox and Sachs, 2002; Armon, 2003b; Banack and Cox, 2003; Plato et al., 2003; Waring et al., 2004).
In this context, the study reported in Brain (page) raises important issues of methodology and interpretation that need to be resolved if epidemiology is to contribute decisively to future studies on the pathogenesis of ALS and other neurodegenerative disorders. Identical issues arise from recent reports that service in the Gulf war of 1990–91 might increase the risk of developing ALS (Haley, 2003; Horner et al., 2003; Rose, 2003; Armon, 2004).
Chio and colleagues were prompted to examine the relationship between ALS and football following indications that there might be an excess of deaths owing to ALS among Italian professional football players. A survey on the causes of death among Italian footballers, organized by an Italian prosecutor with an interest in health and environmental issues, was triggered by concerns about illegal drug use among the players. There were 375 deaths in a cohort of about 24000 players. Eight deaths were attributable to ALS, yielding a standardized proportional mortality ratio (SPMR) of 11.6%, with a 95% confidence interval (CI) of 6.7–20. These data have not been published, and because Chio and colleagues identified weaknesses in the methodology of this survey, they decided to carry out a more rigorous analysis of the relationship between football and ALS. To do so, they identified a cohort of (male) professional football players who had played at least one official match in the Italian series A or B divisions between 1970 and 2001. Players who were not Italian-born were excluded. Identification of the cohort and ascertainment of the diagnosis of ALS were probably as complete as could be achieved in a retrospective study extending over so many years. Information on players' medical history, family history, level of activity and playing position in the field, occupations other than football, physical trauma and exposure to drugs or environmental toxins were recorded—one presumes mainly from family informants, although the authors do not state exactly how many affected individuals they were able to examine.
In total, 18 players developed ALS during the study period, but three were excluded because they were not born in Italy, and 10 were excluded because they did not play during the study period. This left five ALS cases in a cohort of 7325 players. In the 137078 person-years of follow-up that this represents, the expected number of ALS cases was 0.77 based on population data from two Italian ALS registries that between them cover a population of 8.5 million (about 15% of the Italian population). This yields a standardized mortality ratio (SMR) of 6.5 with 95% CI 2.1–15.1. Increased risk of developing ALS (SMR 12.2; 95% CI 3.3–31.2) was associated with a midfield playing position, and with a longer duration of playing (SMR for >5 years of playing versus ≤5 years was 15.2; 95% CI 3.1–44.4).
It is important to consider whether the observation of five cases (an excess of about four in 7000) is sufficient for us to accept that these are significant findings. Chio and colleagues have made some assumptions in coming to this conclusion of significance, and we should assess whether these are valid. The study does not compare the football players with a matched control group in a prospective design. This would have been highly impractical, and so the assumption that the general population risk can be used to estimate the expected risk to players is made. Although this is, of course, reasonable given the rarity of ALS, it may not be appropriate, because football players are a selected population, chosen on the basis of great neuromuscular ability. Complex genetic factors and environmental factors may influence physical stamina and fitness (Holden, 2004; Le Galliard et al., 2004). If, for example, risk genes for ALS also conferred some advantage in sport, such as increased strength or stamina, any association seen might not be due to anything specific to football itself, but merely the result of selection of genes for sporting ability for sporting ability. This would still be of interest, but it would affect the interpretation of the association and the direction of future research. Comparison with another group of sportsmen would allow this possible explanation to be examined.
Is a significance level of 0.05 acceptable for epidemiological research? In the early days of genetic association studies, 95% confidence was accepted as sufficient statistical proof, but it soon became obvious that because there are thousands of genes and thousands of diseases, multiple testing was occurring, even though each test was performed by different researchers on a different sample set. As a consequence, many published genetic associations have not been replicated. Performing 100000 tests will result in 5000 significant results at a 95% confidence level even if there is no real difference between groups, and because of publication bias, most of the 100000 tests will not even be available to the scientific community. Although Chio and colleagues have examined the relationship between football and ALS, this is just one of many hundreds or thousands of potential environmental risk factors that have been and will be examined. Current thinking is that a P-value of less than 10−6 is required for genetic association because all potential studies need to be taken into account. Should we expect a similarly stringent level of statistical significance for epidemiological research? If not, many published associations are likely to be difficult to replicate.
It is striking that the phenotypes of the affected football players in this study are quite different from what might be expected given their age and sex. Chio and colleagues comment that the mean age of onset is about 10 years lower than expected for sporadic ALS, and very similar to the age of onset of familial ALS (Kurtzke, 1991). Although none of the cases studied had a family history of ALS, there are striking differences in penetrance among different forms of familial ALS (Ruddy et al., 2003), so we can speculate (as do Chio and colleagues) that these cases might have had a genetic predisposition to ALS. In addition, three of the five affected subjects had bulbar onset, which is unusual. About a quarter of people with ALS can be expected to present with bulbar symptoms. Of these only about a seventh are men below the age of 50, compared with two-fifths who are women over the age of 50 (unpublished data from the King's database).
We know of no evidence that football players elsewhere in the world have an increased risk of developing ALS. The association noted by Chio and colleagues was first identified more by chance than by design, so this association is unlikely to have been detected in other epidemiological studies. It will be important to carry out similar studies in cohorts of football players from other countries. A strength of the present report is the quality of the population data, which allows reasonably robust estimates of SMRs for ALS to be derived from a large population sample. It is desirable, in addition, to design robust case–control studies comparing football players with other athletes and with people form sedentary occupations. This will be time-consuming and expensive, but necessary if we are to know for certain that a career as a professional football player is indeed associated with increased risk of developing ALS.
Assuming that professional football (or something associated with professional football) is a true risk factor, what explanations might be offered? Chio and colleagues offer four possibilities. First, could the increased risk be related simply to participation in the sport and to the degree of physical exercise demanded of players? Second, could trauma (e.g. heading the ball) be involved? Third, could exposure to illegal drugs or toxins be responsible? And finally, does professional football expose players to high levels of toxic herbicides or fertilizers that may be used to maintain football grounds? Unfortunately, none of these possibilities can be proved at present. Although physical activity has been put forward as a risk factor for ALS, the epidemiological data are weak (Longstreth et al., 1998; Scarmeas et al., 2002; Armon, 2003a,b). Increased physical activity has been reported to be either beneficial or harmful in transgenic SOD1 ALS mice (Veldink et al., 2003; Mahoney et al., 2004). Could the effect be due to performance-enhancing or recreational drugs? This seems unlikely, but we have no way of knowing without further case–control and prospective studies.
Whatever the explanation, it is essential to debate further the appropriate methodology and level of stringency for statistical analyses in epidemiological studies of rare conditions such as ALS. The Guam ALS–Parkinson's disease–dementia saga, the recent studies on ALS risk in veterans of military service and the study by Chio and colleagues raise crucial questions, including the validity of the associations claimed. If these lifestyles and occupations really are associated with an increased risk of developing ALS, we must determine whether the explanation lies in genetics, in as yet undefined exogenous factors or (as seems most likely) in complex interactions between genetics and environment. One way forward is to forge international collaborations based on rigorously defined population databases and a common epidemiological methodology. It is encouraging that the European ALS Consortium has established just such a structure (EURALS). One of EURALS's early challenges might be to grapple with the issues raised in the study by Chio and colleagues. A prospective case–control study would be a long-haul project, but worth the effort.