Correlation and causation dermatology

Correlation does not imply causation is a common mantra in all clinical research. But then, when do we know when correlation is indeed causation? Double-blinded randomized controlled trials are, famously, the Gold standard to conclude a causality in medicine. As a clinical trial, the factor tested is always prior to its consequences, thus assuring temporality (the precedence of the cause to it's consequence), and thanks to randomization and blind the only difference between the groups is the treatment tested. However, randomized controlled trials are not always an option, for ethical or practical reasons. How can we then infer causality using observational studies such as cohorts? This age-old question dates back to the beginning of randomized controlled trials when Braford Hill after the famous first randomized controlled trial 'Streptomycin treatment of pulmonary tuberculosis' went on to study the effect of smoking on lung cancer. He then developed the criteria that bear his name, notably the strength of the correlation, its consistency, its specificity, temporality, dose–response relationship and (physiopathological) plausibility.1 These criteria are the most common tool epidemiologists use when evaluating causality. However, in chronic inflammatory dermatological conditions, they can be very difficult to judge. Indeed, Zhou et al.2 in their article remind us aptly that although psoriasis and dyslipidemia are correlated it is very difficult to know which came first, and, therefore, assess temporality. Unfortunately, a physiopathological explanation can be thought of both ways with either abnormalities in lipid metabolism leading to skin inflammation or cytokine network from psoriasis affecting lipoproteins. In a similar manner, when evaluating the link between atopic dermatitis and skin cancer, Luo and al point out the multiple, almost impossible to consider factors that might bias this relationship of atopic dermatitis and skin cancer, and probably explain why this relationship has been so inconsistently found in various epidemiological studies. Mendelian randomization is not a randomization nor was it invented by Mendel. Instead, mendelian randomization mimics the randomization of clinical trials using genetic variants as instrumental variables. Instrumental variables are variables that have a causal link with the exposure factor but not with the outcome or with potential bias. In a Mendelian population, genetic variants are distributed at random, thus mimicking the randomization of clinical trials. We can then select genetic variants that are linked to the exposure, and assess whether they are linked with the outcome. Because genetic variants always come before the phenotypes they are associated with, it solves the question of temporality to. With the increased availability of genome-wide association studies (GWAS), mendelian randomization studies became more common. This also led to strong methodological strengthening of Mendelian randomization and reporting of such studies making them possible in dermatology.3 In one such study, Luo et al.4 showed that atopic dermatitis can indeed cause cutaneous squamous cell carcinoma and basal cell carcinoma but not cutaneous malignant melanoma or actinic keratosis. This has important consequences for clinical practice as it implies that atopic dermatitis patients should have a regular check for these skin cancers. This also will likely lead to more physiopathological examination as to why atopic dermatitis causes squamous cell carcinoma but not its common premalignant precursor. We can even go further with Mendelian randomization and perform two-way Mendelian randomization to assess which of the two conditions is the cause and which is the effect. If the genetic variants associated with lipid metabolism anomaly are linked with psoriasis but the genetic variants associated with psoriasis are not linked with lipid metabolism anomaly then lipid metabolism anomaly causes psoriasis and not the contrary. Zhou et al. showed using this method that indeed lipid metabolism anomalies cause psoriasis. This opens the door for potential adjunct treatment for psoriasis based on common cholesterol-lowering treatment. However, for this indication, these will still have to be evaluated in randomized control trials. We declare no competing interests. Data sharing is not applicable to this article as no new data were created or analyzed in this study.