Causation in health sciences is best understood in the vector model of mutual manifestation and nonlinear interaction rather than within the mono-causal neuron model.
While almost all causation in medicine is multifactorial, that feature alone does not explain genuine complexity and nonlinear interaction. A model of interaction must be added, allowing for context-sensitive effects and explaining how each individual can be medically unique.
Health and illness are to be understood as thoroughly holistic phenomena – in the philosophical sense of holism – rather than caused by the mere addition of many discrete factors. The bio-psychosocial model suggests more the latter rather than a genuine interaction.
The evidence pyramid and its hierarchical structure reveals a commitment to an unsustainable and largely unquestioned orthodoxy that singular causal truths are derived from general causal laws, which prevents progress on many complex medical phenomena.
Existing research methods in medicine presuppose that same intervention will have the same effect in different patients, which is already known to be dubious. A mutual manifestation model of causation, instead, explains how the recipient of an intervention very much contributes to the effect produced.
The direct transfer of probabilistic results from population-level studies to individual patients is invalid and may put them at risk. A propensity theory of probability, consistent with a theory of causal dispositionalism, explains individual variation better and allows more accurate patient risk assessment and prediction.
Subtractive and same-level interference with symptoms is a better alternative to an additive and reductionist approach. Subtractive interference aims at removal of an aggravating cause rather than addition of a counteracting treatment. The former has the benefit of being cheaper for the health service but also avoids the danger of ‘allostatic’, cumulative stress effects, if the aggravating cause is allowed to persist.
Individual variation within a population is to be expected, even within a narrowly defined sub-group. RCTs fail to include heterogeneity, overdetermination, marginals and propensities. Causal dispositionalism suggests the positive ways in which the methods of RCT can be supplemented.
Causation is one, single thing; but it cannot be reduced to something else that could then be used as an infallible method for its discovery. Hence, we have to accept a position of methodological pluralism in which we use different methods and look for consistent and coherent results. While correlation data and comparative studies can indicate causation, they should also be supported by theoretical or mechanistic knowledge if we are to claim with confidence that we have discovered causes.
The external validity of population studies to the individual clinical situation should be given a sound ontological basis, which should then also inform public health policy.
Causality, Complexity and Evidence in Health Sciences 