Author: Bryan P. White
Original Publication: 09/20/2019
The Bradford Hill method of causation is commonly used in epidemiology to assess the relationship between a disease and its causes, and attempt to assign both a causal factor and the strength of that factor. Known as the “Hill” method, this method includes the following criteria to judge a causal relationship: 1.) Strength (how strong is the association between cause and effect); 2.) Consistency (how reproducible is the observed pattern); 3.) Specificity (are there more than one potential cause, or is this cause the only, most likely cause?); 4.) Temporality (does the effect occur after the cause, and in what timeframe?); 5.) Biological gradient (strength of the exposure should cause a worse effect); 6.) Plausibility (is there a plausible mechanism between the cause and effect, given current knowledge?); 7.) Coherence (is the effect supported by established laboratory data that reflect epidemiological occurrences and disease knowledge?); 8.) Experimental data (is there experimental evidence for the relationship?); 9.) Analogy (has this type of relationship been observed in other scenarios, eg. a chemical causing cancer, etc.?); 10.) Reversibility (if the cause is removed, is the disease prevented from occurring?) (Hill, 1965).
One example of the application of Hill’s method of causation is between the potential effects of anti-psychotic medications on diabetes (Holt and Peveler, 2006). A large percentage of people diagnosed with diabetes are also diagnosed with psychiatric conditions, so the question frequently arises does one cause the other - can antipsychotics cause diabetes? One of the challenges is that many of the risk factors that lead to diabetes are similar to those that lead to severe mental illness (SMI), which requires the administration of antipsychotics. Another challenge is identifying proper control groups to test this causation problem, and many studies designed to test the efficacy of antipsychotic pharmaceuticals were not designed to test long-term side-effects (e.g., weight gain/diabetes).
Looking at strength of association, a relative risk (RR) of less than 2 might be considered weak, with more than 3 considered strong. In many cases, those receiving antipsychotic drugs can have up to 7 times RR for developing diabetes compared to control populations. However, using the general population as a control can be problematic, and many published papers do this, and a few papers report reduced risks (depending on the drug). Overall, with an average RR less than 2, the probable strength of this association is relatively week.
Given a low risk rate, a high consistency rate might have increased evidence for a causal link between antipsychotics and diabetes, but since the development of diabetes is a negative result, pharmaceutical-company funded studies have the risk of only reporting positive outcomes (no diabetes development for a drug). Looking at long-term, longitudinal studies, the use of antipsychotics is consistently not associated with the development of diabetes.
In terms of specificity, biological gradient, coherence, and experimental evidence: the association is also week. In terms of temporality, plausibility, and analogy: there is some evidence that the association might exist. Overall, the conclusion is that if there is an association between SMI and antipsychotic drugs, that it is a weak association and those with SMI will have a low risk of developing diabetes due to their condition and/or treatment.
Hill, A. B. (1965). The environment and disease: association or causation?
Holt, R. I. G., & Peveler, R. C. (2006). Antipsychotic drugs and diabetes—an application of the Austin Bradford Hill criteria. Diabetologia, 49(7), 1467-1476.