LAUSR

Identifying mechanisms of neurodegenerative disease causation has long seemed to be beyond the reach of traditional epidemiological tools. Elucidating a plausible mechanism for initiation of amyotrophic lateral sclerosis (ALS) has appeared particularly elusive. The impression that environmental risk factors for ALS were not providing consistent direction meant there was no sturdy epidemiologically based “handle” to grasp when trying to envisage a biological mechanism for triggering sporadic ALS. Reasons often cited for reluctance to accept clues from classical epidemiology are potential limitations inherent in some epidemiological research due to unknown confounding and reverse causality. Reality is more complex. Lack of consensus on how to sift through epidemiological data has resulted in different risk factors receiving emphasis from different investigators. Society has found it difficult to accept even conclusions driven by unambiguous epidemiological evidence, with no possibility of confounding or reverse causality. In 1986, Martijn Katan proposed a novel approach to evaluating whether the observed relation between low serum cholesterol levels and cancer was causal. He suggested exploiting the genetic differences in amino acid sequences of apolipoprotein E (ApoE), which had been shown to affect cholesterol levels; people with ApoE2 had lower low-density lipoprotein cholesterol (LDL-C) and total cholesterol (TC) levels than those with ApoE3 or ApoE4. If low cholesterol levels caused cancer, then the ApoE2 allele should be more common among cancer patients and ApoE3 or ApoE4 more common among controls. If ApoE phenotypes were distributed similarly among cancer patients and controls, this would suggest that the association was spurious. Mendelian randomization (MR) methods follow this reasoning. They apply instrumental variable (IV) methods developed in the field of economics to infer causality in the presence of unmeasured confounding. The concepts underlying IV or MR methods are illustrated in the Figure, and the underlying assumptions are listed in the Supplementary Table. They use a measurable “instrument,” for example, allelic allocation, as a surrogate for an unmeasurable (or less reliably measured, or confounded) intermediate phenotype (smoking or LDL-C) and consider the relationship of allelic allocation with disease (cancer or ALS) to represent what the relationship of the intermediate phenotype (smoking or LDL-C) would be with disease (cancer or ALS) without the effect of the unmeasured confounders. A causal relationship may be inferred if several assumptions hold (see Supplementary Table). The study proposed by Katan was performed 20 years later. Its findings, as did those of a subsequent study using similar methods, established definitively that “Low plasma LDL-C levels were robustly associated with an increased risk of cancer, but genetically decreased LDL-C was not. This finding suggests that low LDL-C levels per se do not cause cancer,“ because if low LDL-C caused cancer in general, then it would not be possible biologically for genetically mediated low LDL-C levels not to be associated with cancer. Comparison of 2 reviews by leaders in the MR field, written 15 years apart, shows that application of MR methods has become more nuanced. The key message is that findings from MR studies need to be interpreted critically, with close attention to the context from which they arise, and with utilization of internal and external comparators. Epidemiological analyses using MR methods need to