Prospective Evaluation of Few Homeopathic Rubrics of Kent’s Repertory From Bayesian Perspective

Abstract

Absolute grading system of homeopathic repertories poses substantial threat to reliability; however, it may be resolved by evaluating rubrics prospectively using likelihood ratio (LR). The authors evaluated few “physical general” rubrics from Kent’s repertory—“chilly,” “hot,” “ambithermal,” “preference for hot/cold food,” “desire/aversion for fish/egg/meat/sour/pungent/salt/sweet/bitter”—prospectively in West Bengal, India, for 1.5 years using the Outcome Related to Impact on Daily Living scale. Per symptom/rubric, LRs < 1.5 were discarded. A total of 2039 encounters were analyzed for thermal relations and 4715 for desires/aversions for specific food items. Comparison with Kent’s repertory revealed discrepancies. One new rubric with corresponding medicines was suggested to be introduced, new entries of medicines were recommended, and some seemed to maintain their ascribed importance. The authors refrained from converting LRs into typefaces prematurely; still they propose introducing LR to repertories for a structural update, changing its use, and enabling homeopaths to make more reliable predictions.

Homeopathic medicines are shortlisted in repertories for a given symptom/condition (rubric), based and graded (typeface) on occurrence in proving and casual clinical experience. Kent used 3 grades in his repertory ¹ :

Bold (3 marks)/first grade: Found frequently in “all or the majority” of provers, confirmed by reproving, and verified clinically on the sick

Italics (2 marks)/second grade: symptoms found in few provers, confirmed by reproving, but occasionally verified clinically on the sick

Plain Roman (1 mark)/third grade: symptoms experienced “now and then” in proving, not yet confirmed by reproving but verified by curing patients—so accepted as clinical symptom only

This method of constructing repertories has so far remained qualitative, and this absolute grading in place of relative ones without any consistent rule or quantification poses a substantial threat to their reliability. However, this structural shortcoming may be resolved by evaluating rubrics prospectively and by systematic analysis using Bayes’ theorem and likelihood ratio (LR) that generate reasonable certainty that the given medicine shall work in the given condition. The grading should depend on the difference between medicine-population and rest of the population. Bayes’ theorem predicts that the chances of success with a medicine increase if a symptom is frequently present in patients who are cured by that medicine, more frequently than in other patients. ² And LR is the modern epidemiological tool for determining the characteristic and keynote symptoms of medicine. ^3,4 LR measures the prior and posterior probability of work of a medicine when a certain symptom is present. The gold standard of homeopathy is “cure.” ⁴ LR is the frequency (prevalence) of a symptom in the population “cured” (improved) by a certain medicine divided by the frequency of the same symptom in the remainder of the whole treated population. ³ Positive LR (LR+) is used to calculate changes in odds (or chances) if the symptom is present. Odds become greater when LR+ >1; higher is better to include. Negative LR (LR−) is used to calculate changes in odds (or chances) if the symptom is absent. Odds become smaller when LR− <1 (between 0 and 1); lower is better to exclude.

Expressing results using LR and statistical probabilities (confidence intervals) may not eliminate but can still minimize bias to a considerable extent and may validate repertorial rubrics and medicines. This requirement has already been felt and initiative is ongoing since the past decade. ⁵ In this multicentered project, the authors intend to evaluate prospectively few “physical general” rubrics from the most frequently used Kent’s repertory and thereby validate the rubrics as well.

Materials and Methods

The symptoms of homeopathic medicines were assessed prospectively by 6 doctors. The symptoms were “chilly,” “hot,” “ambithermal,” “preference for hot/cold food,” “desire/aversion for fish,” “desire/aversion for egg,” “desire/aversion for meat,” “desire/aversion for sour,” “desire/aversion for pungent things,” “desire/aversion for salt,” “desire/aversion for sweet,” and “desire/aversion for bitter.” These symptoms were selected as being most frequently encountered keynotes (physical general symptoms) for prescribing different homeopathic medicines. The symptoms were recorded in all consecutive appointments at the 6 different homeopathic settings in West Bengal, India: the medicine and otorhinolaryngology (ear-nose-throat [ENT]) outpatients of 3 government homeopathic hospitals, 1 homeopathic outpatient of a health center, and 1 private practice setting. The assessment was made from October 2013 until February 2015.

As a rule of thumb, per symptoms or rubrics, LRs <1.5 with corresponding medicines were discarded from pick-listing (also because LR values between 1.0 and 1.5 hardly change posterior probability). ² The rest of the LRs, that is, 1.5 or more with their corresponding medicines were planned to be reported along with their 95% confidence intervals (CI) and prevalence of each symptom in the general population. While calculating LRs from the 2 × 2 contingency tables, the cells “a” to “d” were constructed as follows:

The number of patients with a certain symptom improved by the medicine under question

“Remainder of the population” presenting with the symptom (not responding to the medicine under question and other medicines as well, plus population responding to other medicines)

Number of patients with absence of the symptom, still improved by the medicine under question

“Remainder of the population” not having the symptom (not responding to the medicine under question and other medicines as well, plus population responding to other medicines)

A minimum of 2 in fields “a” and “c” and more than 5 for “a + c” were considered for analysis. The analyzed symptoms were recorded from every new patient older than 2 years of age in first visit and then medicines were prescribed according to symptom similarity. In every follow-up visit, patients’ responses were assessed using the 9-point (−4 to +4) Outcome Related to Impact on Daily Living scale ⁶ and were documented for analysis. Thorough case taking was done prior to every change of medicine. If new medicines were indicated, new entries were made considering the cases as new ones. In acute cases, the outcomes recorded at the last follow-up were taken for analysis. In chronic cases, we started to record outcomes after a minimum of 3-month follow-up. The outcomes were adjusted at each follow-up consultation until the last follow-up. All the data were taken from routine practice and the patients were intervened as per presenting symptoms and the physicians’ perception.

At the end of data collection, a total of 2039 prescriptions were analyzed. We evaluated 2039 encounters for thermal relations and 4715 for desires/aversions for specific food items. As a single patient might have desire/aversion for multiple food items, redundant entries generated 4715 data in total for these items. A specially designed spreadsheet was provided for systematic data collection to all the participating homeopathic doctors. We did not ask for ethical clearance. According to the Helsinki agreement, this was not necessary if there was no influence on daily practice from the research.

The presence of a symptom, that is, the preference, desire, or aversion for a specific food item (cold, warm, fish, egg, meat, sour, pungent, salt, sweet, and bitter) was distinguished on the basis of 3 degrees: 0 = absent, 1 = mild/less intense, and 2 = strong. Grade 2 (strong) was considered as an intensity that could not be controlled by will and was spontaneously mentioned. Grade 1 (mild) was the intensity that was less, but confirmed on questioning. Grade 0 indicated that symptoms were absent or occurring seldom or never or was not stronger than in the average population. In cases of thermal relations, patients were asked which season they felt more acutely. A patient was considered as hot (“warm” blooded) if he/she preferred winter and/or could not tolerate summer. “Chilly” (“cold” blooded) patients were defined “a priori” vice versa.

Results

Thermal Relations

“GENERALITIES—Heat, sensation of”: Only 8 medicines having LR >1.5 could be enlisted: Kali-s, Lach, Med, Nat-m, Pic-ac, Puls, Ruta, and Sulph; LR ranging from 1.53 (95% CI = 1.30-1.82) to 1.84 (1.60-2.12). (Supplemental Table 1; Figure 1; all supplemental tables are available at http://chp.sagepub.com/supplemental)

“GENERALITIES—aggravation, heat and cold”: Fifteen medicines with LR >1.5 were reported: Anac, Bell, Carb-v, Carc, Caust, Graph, Hyper, Kali-c, Merc, Med, Nux-v, Phos, Phyt, Sil, and Symph; LR ranging from 1.52 (95% CI = 0.42-5.48) to 5.04 (95% CI = 2.82-9.01). (Supplemental Table 2; Figure 1)

“GENERALITIES—Heat, vital, lack of”: Only a single medicine was picked: Psor (LR 1.61; 95% CI = 1.07-2.43) was picked. (Supplemental Table 3; Figure 1)

Figure 1.

Thermal relations.

Desire for Food Items

Cold food: Seventeen medicines having LR >1.5 were enlisted: Alum, Am-c, Ant-c, Arg-n, Ars, Ars-i, Carc, Cinch, Cond, Cupr, Hamam, Hydras, Kali-c, Lac-c, Mur-ac, Nat-c, and Phos; LR ranging from 1.61 (95% CI = 0.44-5.95) to 6.87 (95% CI = 2.56-18.39). (Supplemental Table 4; Figure 2)

Warm food: Fourteen medicines showing LR >1.5 were enlisted: Aesc, Aloe, Am-m, Apoc, Ars-i, Calc-s, Dulc, Ign, Kali-ar, Kali-bi, Kali-i, Mag-p, Petr, and Staph; LR ranging from 1.50 (95% CI = 0.47-4.86) to 3.52 (95% CI = 1.99-6.21). (Supplemental Table 5; Figure 2)

Fish: Eighteen medicines were found to have LR >1.5: Alum, Arg-n, Carc, Caps, Carb-an, Cond, Con, Iod, Kali-bi, Mag-p, Merc-i-r, Petr, Phos, Ran-b, Sang, Spig, Thyr, and Zinc; LR ranging from 1.59 (95% CI = 0.79-3.22) to 3.44 (95% CI = 1.54-7.69). (Supplemental Table 6; Figure 2)

Egg: Seventeen medicines having LR >1.5 were enlisted: Agar, Bacil, Calc-c, Calc-p, Chel, Cist, Cup, Hydr, Hyper, Ign, Merc-c, Nat-c, Phos, Puls, Ran-b, Rat, and Teuc; LR ranging from 1.50 (95% CI = 0.41-5.53) to 5.11 (95% CI = 1.74-15.00). (Supplemental Table 7; Figure 2)

Meat: Sixteen medicines with LR >1.5 were enlisted: Acon, Anac, Bacil, Bell, Carc, Caps, Gnaph, Graph, Iod, Ipec, Merc, Nit-ac, Pic-ac, Rat, Spig, and Tub; LR ranging from 1.64 (95% CI = 0.94-2.87) to 5.43 (95% CI = 2.60-11.35). (Supplemental Table 8; Figure 2)

Sour, acids, and so on: Twenty medicines were identified having LR >1.5: Act, Aloe, Anac, Ant-c, Apis, Aur, Bacil, Bar-c, Carb-v, Chel, Cond, Gels, Graph, Lach, Med, Mer-c, Pic-ac, Puls, Rat, and Sep; LR ranging from 1.50 (95% CI = 0.41-5.54) to 6.40 (95% CI = 2.39-17.14). (Supplemental Table 9; Figure 2)

Pungent things: Fifteen medicines with LR >1.5 were shortlisted: Acon, Apis, Ars, Bell, Calc-f, Cist, Hep, Iod, Mur-ac, Nit-ac, Petr, Rat, Symph, Syph, and Thuj; LR ranging from 1.51 to 6.27. (Supplemental Table 10; Figure 2)

Salt things: Eleven medicines with LR >1.5 were enlisted: Acon, Apis, Arn, Kali-c, Mur-ac, Nat-s, Nit-ac, Phyt, Sep, Syph, and Zinc; LR ranging from 1.52 (95% CI = 0.76-3.06) to 4.84 (95% CI = 2.46-9.51). (Supplemental Table 11; Figure 2)

Sweets: Twenty-four medicines having LR >1.5 were found: Alum, Am-c, Canth, Caps, Carb-v, Con, Dulc, Ign, Ipec, Kali-s, Lac-c, Lach, Led, Merc-c, Phos-ac, Pic-ac, Plb, Ran-b, Sil, Spig, Staph, Teuc, Thyr, and Tub; LR ranging from 1.53 (95% CI = 0.44-5.37) to 3.37 (95% CI = 1.15-9.89). (Supplemental Table 12; Figure 2)

Bitter: Fourteen medicines with LR >1.5were shortlisted: Acon, Aloe, Apis, Calc-f, Hep, Kali-c, Lac-c, Led, Medo, Merc-i-r, Plb, Spig, Syph, and Zinc; LR ranging from 1.55 (95% CI = 0.41-5.86) to 4.49 (95% CI = 1.67-12.04). (Supplemental Table 13; Figure 2)

Figure 2.

Desires for food items.

Discussion and Conclusion

In this article, we tried to introduce statistical reasoning to reevaluate entries of Kent’s repertory. Kent’s methodology was fundamentally problematic being dependent on the frequency of use of a medicine. If a medicine is seldom used, even its most important symptoms will not be mentioned in bold type in his repertory. As influence of chance was not addressed, discrepancies were elicited while comparing with our newly developing repertory. The differences can also be explained by statistical uncertainty caused by small samples and recall or expectation bias in expert experience. In our method, it does not really matter if the medicine is seldom or frequently prescribed. We therefore believe that our data may be more correct than the existing entries in the repertory. And large amounts of data could be collected for evaluating LRs of the medicines with minimal interference of daily practice.

The results should be interpreted cautiously. The research was carried out by several independent and experienced homeopaths, well familiar with the Materia Medica of the investigated symptom. Choice of the medicine did not influence the scoring of outcomes. Relative grading using absolute numbers apparently gives the impression of some absolute truth, without reflecting statistical uncertainties and measurement bias. To overcome the problem, LRs are accompanied here by confidence intervals. This will enable other users to compare their assessments and interpretations of the symptoms and their own population. But one of the obvious tasks that remained unaccomplished is translation of the LR cutoff values to reflect the conventional typefaces of Kent’s repertory. Earlier classifications of cutoffs ^2,5 were based on some theoretical considerations merely meant to indicate the necessity for future discussions about this, and especially about the difficult point of discarding the existing entries in the repertory. It was stressed that these cutoff values were arbitrary. Suitable cutoff values depend on prior chance and thus may be subject to change in different populations. Thus, not being certain about translating the LR values into typefaces and to avoid fruitless disputes, the authors opted for reporting the LRs only and without typefaces along with their 95% confidence intervals and prevalence of each symptom in the general population. The LR values obtained from this research need to be subjected for a larger set of assessments on data collected from across the globe for arriving at unquestionable cutoff values, thus helping uniform interpretation. However, the issues related to confirmation bias may have been resolved due to longer follow-up and prospective gathering of data in this project.

The study reproducibility and generalizability remained questionable. There is large variance of interpretation of symptoms, in age, gender, culture, profession, and so on. Significant bias and variance may exist within our research group also. However, in this multicentered research, achievement of consensus was not possible; hence, interrater variance was not reported. Expectation bias, qualitative vagueness, and causality attribution were not addressed as well. This article is also subjected to the inherent limitations of its study design, for example, the placebo effect, the therapeutic relationship with the clinician (empathy, compassion, social desirability, etc), the regression effect toward the mean especially in acute cases, and the use of undisclosed concomitant treatments, if any.

Many differences were elicited while comparing ours with Kent’s repertory. One new rubric needs to be added—“GENERALITIES: aggravation, heat and cold” with their corresponding medicines. Surprisingly, contrary to our common practice, only a single remedy Psor. (LR 1.61) could qualify under the rubric “Heat, vital, lack of” in Kent’s repertory; and Nat-m mentioned in Italics for bitter food desire even could not qualify. Similarly, Phos. received a bold typeface in Kent’s repertory under “DESIRES, cold food,” while it scored only an LR of 1.81. Cupr. strikingly received an LR 5.49. Phos. (LR 2.00) and Calc. (LR 2.54) received similar importance in desire for fish and egg, respectively. Graph (LR 1.74) and Merc (LR 1.94) seem to maintain their importance in meat desire; Ant-c (LR 2.56), Apis (LR 2.56), Carb-v (LR 2.26), and Chel (1.50) in desire for sour and acids; Ars (LR 1.87) for desire for pungent things; Am-c, Kali-s, and Plb (LR 2.11 each) and Tub (2.25) for sweet desire. Many new entries seemed probable under most of the rubrics evaluated. Thus, many frequently prescribed medicines were found overrated; seldom used medicines remained underrated.

The results are well documented and give indications for a gradually improving repertory. But the study results should not be treated as the ultimate and it needs replication. The authors are still gathering data for further verification and reconfirmation of the elicited LRs. Since the past decade, considerable initiatives have already been taken for refining and standardizing the homeopathic literature, in terms of introducing LR and polarity analysis for modifying Kent’s and Boenninghausen’s repertories. ^7,8 We should continue to update repertories consistently using LR instead of casual observations to indicate which medicines are indicated if a symptom is present, and how strong the indication is. Bias in different forms already exist and shall continue to exist, but may be curtailed by systematic research.