|Year : 2017 | Volume
| Issue : 4 | Page : 262-268
Confirmation and confirmation bias: The role of prognostic factor research
Independent Researcher, The Netherlands
|Date of Web Publication||29-Dec-2017|
Dr. Lex Rutten
Aard 10, 4813 NN Breda
Source of Support: None, Conflict of Interest: None
Background: Some homoeopathic practitioners fear that prognostic factor research (PFR) of homoeopathic symptoms neglects the context of these symptoms. Materials and Methods: The function of repertory and materia medica in the homoeopathic method is discussed. Previous research shows how the repertory can be improved. Outcome and Discussion: The homoeopathic procedure can be divided into two stages: an open mind stage to make an inventory of all possible medicines and a confirmatory stage to select the medicine that fits the totality. Closing the open mind stage too soon will cause confirmation bias. PFR is meant to improve the repertory in several aspects. Bayes' theorem provides an algorithm for homoeopathy and the inherent likelihood ratio (LR) a generalisation of Hahnemann's aphorism 153 to a broad differentiation of importance of symptoms arranged by peculiarity and by prevalence in respective medicine populations. Conclusion: PFR does not alter the way we select homoeopathic medicines, but it improves the repertory. LR generalises Hahnemann's aphorism 153 to an algorithm that distinguishes a large range of symptoms, peculiar and less peculiar, according to their importance for the choice of a medicine.
Keywords: Aphorism 153, Bayes' theorem, Confirmation bias, Homoeopathic symptom, Repertory
|How to cite this article:|
Rutten L. Confirmation and confirmation bias: The role of prognostic factor research. Indian J Res Homoeopathy 2017;11:262-8
|How to cite this URL:|
Rutten L. Confirmation and confirmation bias: The role of prognostic factor research. Indian J Res Homoeopathy [serial online] 2017 [cited 2019 Nov 22];11:262-8. Available from: http://www.ijrh.org/text.asp?2017/11/4/262/221960
| Introduction|| |
Prognostic factor research (PFR) is new in homoeopathy; it assesses the prognostic relationship between symptoms and personal characteristics and outcome of treatment. New techniques and developments still have to prove themselves, and it is only natural that practitioners regard them critically. On the other hand, the instruments of homoeopathy – materia medica and repertory – should also be reviewed. Better materia medica and better repertory render better homoeopathic treatment. These two instruments are used alongside each other and complement each other. Critics of PFR state that the choice of a homoeopathic medicine cannot be improved by assessing symptoms disconnected from the whole context of the patient. That is partly true, but we all know that the repertory also considers symptoms disconnected from their context. We know that the repertory has considerable flaws, but we can still use it. PFR is meant to repair these flaws, but that does not essentially change the use of the repertory. This paper considers the role of the repertory; the discussion about PFR is, in fact, a discussion about the right use of the repertory, about confirmation and confirmation bias.
Prescribing the right homoeopathic medicine is not a process of mere calculating. At the end, it is based on recognising the complete picture: different symptoms confirm each other, and the totality is more than the sum of individual symptoms. At the start of the consultation, however, the practitioner must have an open mind. The first symptoms elicited in this process can still indicate a larger number of medicines, some well-known, some less known to the prescriber. A prematurely restricted choice neglects the richness in nuances of homoeopathic pictures but may also leave essential personal characteristics of the patient unrevealed that may come up after in-depth interrogation. This in-depth interrogation explores the fit of the totality of a limited number of medicines; medicines that are excluded at this stage will not be explored in depth. Roughly speaking, choosing the best homoeopathic medicine consists of two stages: the first open mind stage to make a preliminary selection of possible medicines and the second 'confirmatory' stage to explore which medicine fits the totality the best.
| Materials and Methods|| |
Analysing homoeopathic practice functions of homoeopathic materia medica and repertory can be distinguished. Applying Bayes' theorem shows the structural shortcomings of the repertory. A previous PFR project in the Netherlands shows how the repertory can be improved., Recent pilot studies to assess the prevalence of homoeopathic symptoms show some considerations about systematic assessment of homoeopathic symptoms.
Open mind and repertory
In the open mind stage, the practitioner will use the homoeopathic repertory to scan all possible medicines. In this stage, the context of the whole personality is consciously disregarded and this should be filled in later. In this stage, practitioners check modalities, food aversions/desires and a limited number of mental symptoms, among other symptoms in all patients. Many patients are focussed on their complaint and not aware of the fact that they are more chilly than others or use more salt than others because they like it so much. Such symptoms might not be brought forward if not systematically checked in every patient. The systematic enquiry about such symptoms makes a good start for thinking about a limited number of medicines, and then, the confirmation stage can begin by asking more direct questions for confirmation of each medicine. Of course, this is a simplified description of the medicine selection process; if we are not satisfied with our initial selection of medicines, we will re-open our mind for new possibilities, and every practitioner has his own methods in this respect.
The instrument that suits the open mind attitude best is the homoeopathic repertory that arranges medicines according to symptoms, complaints and characteristics. The repertory disregards the complete picture of the medicines, except for a few subrubrics that specify symptoms that are very specific for single medicines. Especially, the disconnectedness of the whole picture helps us to keep an open mind.
If our open mind closes too soon, we risk confirmation bias. It is only human that practitioners like to see their pre-existing ideas confirmed. That is what confirmation bias is about: confirmation of pre-existing ideas in new observations. If we enter the confirmation stage too soon, we risk confirmation bias. During the consultation, the practitioner develops ideas about possible medicines guided by the complaints of the patient and the answer to various questions, but these questions are partly guided by previous answers.
Our expectations of a chance of a curative effect of various medicines change with every new symptom or characteristic that comes up during the consultation: a symptom can confirm or contradict specific medicines. The belief before a new symptom comes up is called the prior belief and the belief after the symptom is called the posterior belief. This posterior belief becomes the prior belief before the next symptom and so on. At a certain point, the practitioner starts asking confirmatory questions guided by the then preferred medicine(s). This selection of questions can lead to confirmation bias, limiting the choice of possible medicines in the wrong direction. You may notice the loquacity of the patient more easily when the patient tells you that tight clothing around the neck is intolerable, indicating medicines such as Lachesis. Then, you wonder if the patient is jealous but do you trust a negative answer to that question? People do not easily admit that they are jealous. Possibly you disregard the denial of jealousy and persist on your choice for Lachesis but did you consider Tarentula? Tarentula has both intolerance of clothing around the neck and loquacity. In the original Kent repertory, Tarentula is present in the rubric 'loquacity' only in plain type, but in the Dutch prospective assessment of this symptom, loquacity was present in 2 out of 8 patients responding well to Tarentula, resulting in a likelihood ratio (LR) = 3.85 (95% confidence interval (CI) 1.16–12.91). This was not much lower than LR of loquacity for Lachesis: LR = 5.34 (95% CI 3.43–8.35). Hence, your confirmatory questions should not only consider Lachesis but also Tarentula.
Intensity of symptoms
It is also possible that confirmation bias influences cut-off values for symptoms: the symptom is regarded as positive if present in a lower intensity. An example: if you consider the medicine Natrum muriaticum (Nat-m) because of the symptoms, you have so far you might consider to ask for the symptom 'recurrent herpes of the lips'. Probably, you will accept a frequency of once in every 2 years as a confirmation for Nat-m while a frequency of more than several times a year would be required to start thinking of Nat-m in the beginning of the consultation. Hence, confirmation is a normal process in the homoeopathic consultation, but we must be aware that it can lead us in the wrong direction.
Confirmation, materia medica and repertory
When we ask questions to confirm specific medicines, we apply our knowledge of materia medica. This knowledge is partly in books, partly in our heads. Sources for this knowledge are intoxication data, proving data and data from experience in daily practice. There is, however, variation in the occurrence of symptoms in provings and in the practice experience of every individual homoeopathic practitioner. This leads to different opinions about the importance of symptoms regarding homoeopathic medicines.
An example of variation in personal experience of different doctors as a source of confirmation bias. In the Netherlands, a group of homoeopathic doctors attended scheduled consensus meetings regarding best cases of specific medicines. These cases were restricted to long-lasting chronic complaints; the improvement was longer than 1 year and very likely caused by the medicine. This project was called 'materia medica validation'. A meeting concerning best Sulphur cases was attended by 20 doctors and they presented a total of 23 best Sulphur cases. One doctor presented two Sulphur cases and one of them had the symptom 'fear of death'. None of the other patients had a fear of death. According to the experience of this doctor, 'fear of death' was an indication for Sulphur because 50% of his best Sulphur patients had that symptom. For the 19 other doctors, 'fear of death' had no relationship with Sulphur. The actual prevalence of the symptom 'fear of death' was 4% in 23 patients. Later prospective PFR in 4094 in the Netherlands patients assessing this symptom with 88 Sulphur patients showed only one patient with 'fear of death' (1%).
Another example from the Dutch materia medica validation project illustrates the role of confirmation bias in homoeopathic prescribing. Another validated medicine was Stramonium. Out of 12 cases, 5 (42%) had the symptom 'fear of the dark'. Faced with that result several other colleagues stated that they never prescribed Stramonium if there was no fear the dark in the patient. This leads to clear confirmation bias: those colleagues will never see a Stramonium case without fear of the dark and their opinion will never change. This confirmation bias becomes more problematic if these doctors teach materia medica. Their students will learn not to prescribe Stramonium if there is no fear of the dark. Practitioners who do not prescribe Stramonium if there is no fear of dark will also introduce this confirmation bias in consensus meetings.
Preventing confirmation bias with the repertory
Above is stated that closing the open mind too soon will increase confirmation bias. We must gather as much information as possible without preference for specific medicines and not go to the materia medica too soon to try to confirm specific medicines.
The function of the repertory is that it broadens your scope of possible homoeopathic medicines. It is the opposite of materia medica because it tears all symptoms apart as separate items. Why is this necessary? Suppose the patient tells you that he has a fear of death. Of course, with basic homoeopathic materia medica knowledge, the homoeopathic medicine Aconitum comes to mind, but do you also think about, say, Latrodectus mactans, or Veratrum album? If you had time to read all the 'mind' sections of the materia medica, you would encounter these medicines among many different others. Reading further about these medicines, you might then realise that the patient has chest pain extending to his left axilla or that he has much thirst for cold drinks, leading to Latrodectus mactans or Veratrum album respectively, but for this, you have to read hundreds of pages of materia medica from A to Z.
Even if we had time enough to read all this materia medica, it would still be hard to keep an open mind about all possible combinations of symptoms leading to numerous different medicines. At this point, the repertorisation of a limited number of symptoms helps us to keep an overview of all symptom-medicine relations. After a complete inventory of these relations, you can start thinking about the whole picture: which medicine fits best the data, your experience and your intuition?
Prognostic factor research and repertory
Of course, the repertory should be correct, but that is not the case. In Kent's original repertory, Latrodectus mactans is not mentioned in the rubric 'fear of death', despite the fact that the symptom is very prominent in the materia medica of Latrodectus mactans. The medicines Natrum muriaticum and Sulphur, however, are present in this rubric. These medicines are used very frequently and we might wonder if 'fear of death' is really more frequently present in patients responding well to these medicines than in the remainder of the population. The materia medica does not give us that information because the materia medica does not compare all symptoms with other medicines.
The comparison of each symptom with other medicines can be provided by the repertory. At least, it should do so, but – as explained elsewhere – many repertory rubrics are seriously flawed because medicine entries are based on absolute occurrence while they should be based on relative occurrence (prevalence). A symptom is an indication for a specific medicine only if it occurs more frequently in patient responding well to that medicine than in other patients. In statistical terms, LR should be greater than one. To assess the LR for each medicine for the symptom, we have to count prevalence of the symptom in populations responding well to specific medicines and in the whole population. This research process is called PFR.
From aphorism 153 to likelihood ratio
Hahnemann already recognised the importance of individual symptoms. His aphorism 153 is the most important principle in homoeopathy: more peculiar symptoms are more important. Peculiar means that the prevalence of the symptom in the general population is low. Now look at the formula for LR:
LR = (prevalence in the target (medicine) population)/(prevalence in the remainder of the population)
In homoeopathy, the population responding well to any specific medicine is just a small part of the whole population, so the remainder of the population is nearly the whole population. The smaller the denominator in this formula (prevalence in the remainder of the population), the higher the LR. In other words, a peculiar symptom is identical with high LR: aphorism 153 is identical with a small denominator in the LR formula, i.e., a low prevalence of the symptom in the whole population. The fact that a peculiar symptom, resulting in high LR, increases the chance that the corresponding medicine will work considerably can be understood by looking at Bayes' theorem:
Posterior odds = LR × prior odds (odds = chance/[1-chance]; chance = odds/[1+odds])
Bayes' formula is, after two centuries of struggle, accepted all over the world and present in many computer programs. It describes how we learn from practical experience and is the scientific rectification of Hahnemann's aphorism 153: a more peculiar the symptom renders a higher LR.
An example: the symptom 'nausea from hearing organ music' is very peculiar with a prevalence of probably <1 in thousand. Only one medicine, Physostigma, is known for this symptom and the few patients with this symptom will be much more likely to respond to Physostigma than to other medicines. Hence, the prevalence of this symptom is much higher in the Physostigma population than in the remainder of the population and LR is probably higher than 1000. Suppose that the prior probability that Physostigma works is 1%. After the symptom 'nausea from hearing organ music', the posterior probability becomes 91% if the LR = 1000.
LR covers aphorism 153 completely, but aphorism 153 only covers the denominator in the LR formula. LR is more generally applicable because it also comprises the nominator, the prevalence of the symptom in the population responding well to a specific medicine. Peculiar symptoms in homoeopathy have prevalence below, say, 1 in 100. In that case, high LR is predominantly caused by the low prevalence in the denominator, but what if the prevalence in the denominator – in fact in the whole population – is not so low, say, more than 1%? In that case, we do not have a really peculiar symptom and aphorism 153 does not apply, but we know that some medicines are more related to a specific medicine than others. In statistical terms, this means that the prevalence of the symptom is higher in the population responding to medicine A than to medicine B.
An example: in the Dutch assessment of six homoeopathic symptoms, the symptom 'recurrent herpes of the lips' was present in 2 out of 19 Rhustoxicodendron (Rhus-t) patients (10.5%) and in 24 out of 156 Natrum muriaticum (Nat-m) patients (15%). This results in LR = 2.1 for Rhus-t and LR = 3.4 for Nat-m. In Kent's repertory, both medicines are in bold type, so no difference can be concluded from the repertory. This difference can only be assessed by PFR showing the differences of the prevalence in the nominator of the LR formula.
Applying LR, we also understand the most important shortcoming of the repertory: large symptom rubrics are unreliable, and especially, frequently used medicines are over represented in those rubrics. This is illustrated by [Table 1], based on the above-mentioned Dutch assessment of the symptom 'fear of death' in 4094 patients. Hitherto repertory entries were based on absolute occurrence and we see that there are patients with fear of death responding well to Natrum muriaticum (Nat-m) and Sulphur (Sulph), rectifying an entry in this rubric in the old system. Intuitively, we understand that there is a difference between one in 88 patients for Sulph and one in four patients for Cenchris (Cench), indicating that fear of death is much more important for Cench than for Sulph. This difference is shown by the prevalence of the symptom in both populations and the resulting LR by comparing this with the remainder of the population. For clarity, not all data for LR calculations are shown (see [Box 1]) in [Table 1], just the prevalence of the symptom in the medicine population and LR. This shows the relationship between prevalence and LR population.
|Table 1: Outcome of assessment of the homeopathic symptom ‘Fear of death’. The number of patients responding well to each medicine with the symptom and the total number of patients responding well to each medicine. This renders the prevalence in each population. The prevalence of the symptom in the whole population (n=4094) was 3.9% (158 patients). Hence the LR can be calculated|
Click here to view
Another systematic shortcoming of the repertory is unreliable polar rubrics: symptoms with opposite expressions, such as aversion/desire, aggravation/amelioration and chilly/hot. Arsenicum album (Ars) is generally known for chilly patients, but, due to variation, some Ars patients are hot. Therefore, we see Ars in the rubric 'chilly' and in the rubric 'sensation of heat', albeit in different degrees; in Italics in 'chilly' and in plain type in 'sensation of heat'. The problem with this is that we look at one rubric only, so you look at the rubric 'sensation of heat' and accept this symptom as a confirmation for Ars. Probably, the prevalence of 'sensation of heat', however, is lower in the population responding well to Ars than in the remainder of the population and LR <1, not indicating Ars. This problem is present in most rubrics of polar symptoms.
There is a solution for the problem of polar symptoms called polarity analysis (PA) in a computer repertory program. This program subtracts the entries of the opposite symptoms from the entries of the chosen symptom. Hence, if you enter 'sensation of heat', the computer subtracts 2 (entry in Italics in 'chilly') from 1 (plain type entry in 'sensation of heat') resulting in minus one and therefore a relative contraindication for Ars. This is equal to LR <1, but the actual LR value allows us to calculate how strong this contraindication is.
PA has proven to be successful I the randomised controlled trial on ADHD by the Swiss paediatrician Frei et al. In this research, the success of the first prescription mounted from 28% to 48% using PA.
| Discussion|| |
Some people fear that new research methods such as PFR are a threat to the core-requisite of homoeopathy; finding the total picture, individual symptoms should confirm the totality of the patient. Such fear is rectified, but we must realise that we already have an instrument disregarding the context of the patient: the homoeopathic repertory. Most practitioners use the repertory to keep an open mind about all possible medicines. For most practitioners, it is impossible to have an oversight in memory of all homoeopathic medicines. If we concentrate ourselves on a limited number of medicines in the beginning of the consultation we risk confirmation bias, we look for and recognise easily the symptoms that fit our pre-existing ideas.
PFR does not change our methods; it merely improves the repertory. There is much to improve regarding the existing homoeopathic repertory because there are several serious and systematic mistakes: entries based on absolute occurrence instead of prevalence and confusing polar symptoms. PFR involves the use of LR as the main constituent of Bayes' theorem. Bayes' theorem provides homoeopathy with a scientific algorithm and LR is a generalisation of Hahnemann's aphorism 153. Aphorism 153 only considers a low prevalence of a symptom in the general population, the denominator of the LR formula. LR also includes different prevalences of symptoms in respective medicine populations, the nominator of the LR formula. At present, the typeface of medicines in symptom rubrics gives an inaccurate idea of the prevalence of the symptom in respective medicine populations. In PFR, we count symptoms in the whole population and in respective medicine populations enabling us to calculate prevalence and LR.
PFR is still in its infancy, and we have to newly develop this method. Checking a symptom in each patient in PFR is quite different from eliciting symptoms in our usual consultations. We must use our clinical judgement to interpret the answers of the patient. We also have to use cut-off values for the intensity of symptoms more consciously. Then, there is the problem of constituting medicine populations; what should be the result and is this result really caused by the medicine?
Every research raises new questions and new doubts, but we have to realise that we hitherto avoided these questions and doubts. We have to improve our instruments to obtain a more effective method that can be more easily understood by other physicians. This will result in better outcome in randomised controlled trials to prove the efficacy homoeopathy. This has already been demonstrated for Polar Analysis.
| Conclusion|| |
PFR does not alter the homoeopathic method. The homoeopathic medicine should fit the whole picture of the patient, and individual symptoms must fit in this context. Nevertheless, most practitioners use the homoeopathic repertory indicating medicines by individual symptom to keep an open mind about all possible medicines. The repertory has serious systematic shortcomings, and PFR helps to mend these.
The most important aspect of PFR is assessing the prevalence of symptoms, in the whole population and in populations responding well to specific medicines. This way it is possible to generalise the idea of Hahnemann's aphorism 153 about the importance of peculiar symptoms to an algorithm that distinguishes the importance of both peculiar and less peculiar symptoms for specific homoeopathic medicines.
Financial support and sponsorship
Conflicts of interest
| References|| |
Rutten L. Prognostic factor research in homoeopathy. Indian J Res Homoeopath 2016;10:59-65.
Rutten AL, Stolper CF, Lugten RF, Barthels RW. Statistical analysis of six repertory rubrics after prospective assessment applying Bayes' theorem. Homeopathy 2009;98:26-34.
Rutten L. Data collection: Treat every variable as a treasure. Homeopathy 2015;104:190-6.
Rutten L, Muraleedharan KC, Shinde VH, Manchanda R. What is a homoeopathic symptom, in daily practice and research? Indian J Med Res 2017;11:11-20.
Klein JG. Five pitfalls in decisions about diagnosis and prescribing. BMJ 2005;330:781-3.
Frei H, von Ammon K, Thurneysen A. Treatment of hyperactive children: Increased efficiency through modifications of homeopathic diagnostic procedure. Homeopathy 2006;95:163-70.