Will this medicine work for me? Towards a scientific answer

Which answer would you prefer: 1. “This medicine works better than a placebo”, or: 2. “I estimate the chance that this medicine will work in your case to be 60%”?

These two answers reflect a two and a half century lasting dispute between two statistical methods, ‘classical’ (frequentist) and Bayesian. The first is regarded to be more scientific, the latter played a major part in solving many of history's most important problems, like deciphering coded messages in WO II and predicting disasters. Nowadays many computer programs incorporate Bayes’ theorem to handle experiential knowledge.

Because RCT evidence does not allows other conclusions, the patient can only expect a yes-or-no statement about efficacy of conventional medicines. But this answer might still be false in his particular case. Other factors, like correct diagnosis, genetic susceptibility and comorbidity, also determine if the medicine works. The diagnostic process preceding the prescription is Bayesian and renders the probability of a specific diagnosis.

Bayesian philosophy is about learning from past experience, e.g. about characteristics of patients responding well to specific medicines. Like a medical diagnosis, the choice of a homeopathic medicine is a Bayesian process. Different personal characteristics add up, stepwise increasing the chance that a specific homeopathic medicine will work.

Hitherto homeopathic doctors have been using Bayesian statistics implicitly: characteristic symptoms pointing towards a specific medicine occur more frequently in patients responding well to that medicine than in patients responding to other medicines. It is a small step to make this rule explicit in various types of research and data collection. All we need to know is the prevalence of a symptom in the population responding well to a specific symptom and in the remainder of the population. The ratio between these two is called the Likelihood Ratio (LR).

The research we need is accepted in conventional diagnosis research. Like all kinds of research we will have to deal with possible bias; like our reference standard: what is a good result? Symptoms should be defined more accurately, etcetera. These problems have been neglected in the past. We must realise that this research is meant to improve homeopathy, not to prove it. However, improved homeopathy will render better proof.

Several methods for Bayesian assessment of symptoms are demonstrated. The most valid and time-consuming method is prospective research of a small set of symptoms, but even with this method we can achieve a tremendous scientific improvement of homeopathy within a limited amount of time. Within ten years we could know LRs of characteristic symptoms for our most frequently prescribed homeopathic medicines. Applying the formula that goes with Bayesian theory we will be able to tell the patient: “Based on the symptoms you gave me I expect the chance that medicine A works for you to be x%”.

Keywords:

Clinical research - research methodology