Reliability of self-reported constitutional questionnaires in Ayurveda diagnosis.

BACKGROUND: Ayurveda is one of the most ancient and widely practiced forms of medicine today, along with Traditional Chinese Medicine. It consists of determining an individual's constitution or Prakriti and current imbalance(s) through the use of multimodal approaches. Ayurveda practitioners may choose to include either a self-reported or structured interview constitutional questionnaire as part of the Prakriti assessment. Currently, there is no standardized or validated self-reported constitutional questionnaire tool employed by Ayurveda physicians or western Ayurveda educational institutions.
OBJECTIVES: To examine test-retest reliability of three self-administered constitutional questionnaires at a one month interval and internal consistency of items pertaining to a single constitution.
MATERIALS AND METHODS: Three constitutional questionnaires were administered online. 19 participants completed three questionnaires at two time points, one month apart. Age range was 21-62 years old with a mean age of 34. Of the 19, 5 were male and 14 female. Vata, Pitta, and Kapha scores obtained from each questionnaire were standardized to give a vector of three relative percentages, summing to 100. These percentages were further translated from numerical values to one of ten possible dosha diagnoses.
RESULTS: Analysis indicated that the three questionnaires had moderately good test-retest reliability according to numerical scores, but highly variable reliability according to discrete Ayurveda diagnosis. Internal consistency pertaining to individual constitutions within one questionnaire was poor for all three primary doshas, but especially for Kapha.
CONCLUSION: Further research is necessary to develop a reliable and standardized constitutional questionnaire.

Introduction

The traditional Indian medical model known as Ayurveda is one of the most ancient and widely practiced forms of medicine today, along with Traditional Chinese Medicine. As demand in the western world for traditional medicine increases, there is a growing interest to ensure quality in training, research, and practice [1]. Treatment efficacy is the most prolific type of research in Ayurveda [1]; however there is little research examining reliability of the various diagnostic techniques upon which treatment prescription and efficacy depend. The Ayurveda diagnosis consists of determining an individual's constitution and current imbalance(s) through the use of a multimodal approach including observation, physical exam, pulse diagnosis, and health history. Many Ayurveda physicians and western Ayurveda schools also employ some version of a constitutional questionnaire during the initial patient intake. These questionnaires are often made available online and are popular tools for self-diagnosis amongst the general public. They are not however, standardized or have evidence of validity. If included in the overall assessment, these questionnaires may impact diagnosis and long-term treatment recommendations. Research on their reliability and validity is therefore imperative.

According to the philosophy of Ayurveda, humans have physical and behavioral differences that are classified into one or more of three metabolic forces, or doshas [2]. These doshas, known as vata, pitta, and kapha, are the vital bioenergies responsible for promoting and sustaining the health of each individual. Each dosha comprises of five elements: earth, air, fire, water, and space. Vata is the combination of air and space, pitta- of fire and water, and kapha- of earth and fire.

An individual's specific Prakriti or constitution refers to the physical and behavioral qualities that remain stable throughout one's life [3]. Ayurveda considers seven Prakriti classifications, however, there are ten possible combinations depending on relative predominance of dosha: vata, pitta, kapha, vata-pitta, pitta-vata, vata-kapha, kapha-vata, pitta-kapha, kapha-pitta, or Tridosha; the dosha listed first being the more dominant of the two for an individual who is Dvidoshic. Each Prakriti classification describes the predominant dosha(s) that is likely to overpower the others, producing a certain set of characteristic physiologic imbalances. From an Ayurvedic point of view, knowing one's Prakriti enables a person to make educated lifestyle choices in order to minimize the effects of such inherent tendencies [4].

Vikriti on the other hand, is a term used to describe the changed condition of body, mind, and consciousness [4]. While Prakriti remains stable throughout one's life, Vikriti is a temporary state of imbalance of the doshas, constantly changing depending on one's lifestyle habits. It is imperative for the success of an Ayurveda treatment plan that the physician correctly diagnoses an individual's Vikriti; but, it is very helpful for the individual and the physician to know the underlying Prakriti or constitution as well as it can inform potential future imbalances, disease susceptibility, and long-term treatment plans [5], [6]. When evaluating effectiveness of an Ayurveda prescription, for example, it is important to consider Prakriti. An appropriate constitutionally based prescription can enhance therapeutic effects and minimize adverse effects.

Traditionally, there are four methods of determining or diagnosing an individual's Prakriti: observation, physical exam, pulse diagnosis, and health history [4]. These methods used together are preferable to any used alone as individually, they may lead to bias in Prakriti diagnosis [7]. Several studies have examined reliability and validity of Prakriti diagnosis by way of various methods: pulse taking alone [1], [6], interview-based or self-reporting questionnaires in combination and alone [8], [9], [10], and incorporating several of the diagnostic methods together [1], [8], [9]. To date, results of these studies have varied greatly, from low to moderate levels of reliability, and there is not a validated standard by which to compare one method to another.

Ayurveda practitioners may choose to include either a self-reported or structured interview constitutional questionnaire as part of the Prakriti assessment. At present, there is no standardized or validated self-reported constitutional questionnaire tool employed by Ayurveda physicians or western Ayurveda educational institutions. Rather, there are a wide variety of questionnaires, many of which are publicly available. It is our intention to contribute to the existing literature on examination of the reliability of self-reported constitutional questionnaires by investigating three publicly available questionnaires developed by two of the most well known western Ayurveda educational institutions and by a private international Ayurveda products company. We are unaware of any prior research investigating the reliability or validity of these specific self-reporting constitutional questionnaires. We chose these questionnaires as our starting point because these sources are easily available to the general public and influence western trained Ayurveda practitioners, leading many to employ these or very similar questionnaires in private medical practice. For purpose of this research, we have decided to focus solely on the subject of reliability of self-reported questionnaires without incorporating any other diagnostic methods (i.e. pulse, interview-style questionnaires) due to not having a validated standard by which to compare results.

Reliability refers to consistency and repeatability of outcome measures [1], [11]. Test–retest reliability is used to assess consistency of measures between two points set apart by a length of time. It is necessary to separate the two measures by an adequate amount of time so that results are not influenced by the observer's memory. Based on methods used to assess test–retest reliability for various other self-reporting diagnostic questionnaires [10], [14], [15], [16], we chose a 1-month time period to ensure an adequate length of time before retesting.

Internal consistency reliability refers to consistency of responses across individual items within a test that intend to measure the same construct [1]. Here, it is a way of assessing the amount of agreement amongst questions that examine the vata, pitta, and kapha diagnoses. If, for example, a survey uses a number of questions to assess vata dosha – i.e., a high score on these questions is supposed to identify presence of vata dosha in the respondent – then we would expect there to be high levels of agreement between answers to these questions.

It is our hope that, by exploring the structure and reliability of various self-reporting questionnaires used by western Ayurveda institutions, we can contribute to development of a reliable and validated self-reporting constitutional questionnaire tool that can be shared widely by Ayurveda educational institutions and practitioners.

Materials and methods

Participants were recruited by two methods: 1) an email sent to the local naturopathic school's student body and 2) flyers posted on the school's campus and throughout the greater Portland area. Eligible participants (Table 1) were administered three Ayurveda constitutional questionnaires at two separate time points, one month apart, using REDCap [17], an online survey tool. Participants were able to access the questionnaires online from any computer with Internet access. They were given exactly one week to complete the questionnaires at each time point. All participants provided informed consent.

Table 1

Inclusion/Exclusion criteria.

Participants inclusion criteria

• Adults 18+ years of age • Willingness to complete the three questionnaires online at two separate time points, one month apart • Access to a computer and internet • Ability to read and write in English

Participants exclusion criteria

• Self-reported history or current diagnosis of cognitive impairment that would reduce his or her ability to complete the questionnaires • Anyone who has previously completed an Ayurveda constitutional questionnaire and can recall his or her constitutional type. • Anyone who has previously been given an Ayurveda constitution (such as: vata, pitta, kapha, or any combination thereof) by a healthcare professional and can recall his or her constitutional type.

We studied three constitutional questionnaires used at prominent western Ayurveda educational institutions and made publicly available on their websites. Two questionnaires were developed by the institutions, while the other originated with an established private international Ayurveda products company.

Questionnaire 1 answer options were a degree of agreement on a scale from 0 to 6 (0 = does not apply, 3 = applies somewhat, 6 = applies most); items were grouped into sections labeled as relating to vata, pitta, or kapha dosha. Questionnaires 2 and 3 gave three distinct answers to each item, corresponding to the three doshas and labeled as V, P, and K (and therefore also not disguised).

None of the questionnaires assign a numerical scoring key nor is there an established guideline provided for in previous research. For the purpose of this study, we used the largest of V,P,K scores to determine primary dosha. If both remaining (V,P,K) scores were 25% of the total score, the participant was considered to have a single dosha. If one of the remaining V,P,K scores was ≥25% of the total score, the participant was considered to have Dvidosha, with the second largest score determining secondary dosha. If all three scores were ≥25%, the participant was considered to have Tridosha. If a participant scored equally Dvidoshic at one timepoint (i.e. vata-pitta/pitta-vata) and single dosha at 2nd time point (i.e. vata-pitta) that was at least ½ matching, this was considered dosha agreement.

All three questionnaires instructed participants to answer based on what was true for them in general, not based on situations that may have come up in the prior few weeks. This instruction is meant to differentiate between Prakriti, which remains stable throughout one's life, and Vikriti, which is a temporary state of imbalance. Each questionnaire used its own version of this instruction. The questionnaires were duplicated in REDCap as closely to printed version as possible.

Sample size was estimated by the number of participants needed to detect a mean change of 0.2 (20%) in a single standardized constitution measure (e.g., vata score), ranging from 0 to 1. Note that this is less than what would be associated with any change in discrete diagnosis [18]. Since no prior data on variability or test–retest correlations were available, we used very conservative estimates for the variability between individuals (SD = 0.2) and correlation between tests in the same individual (r = 0.3). Using the program G*Power (v.3.1.9.2) [19] with these estimates and a paired t-test design, we calculated a necessary sample size of 17, in order to obtain 90% power to detect a mean change of 0.2 in the constitution score, with a customary alpha level of 0.05. Statistical software, SPSS v.22, was used for statistical analysis.

Test–retest reliability

In order to determine test–retest reliability of the three individual questionnaires at one month interval, the results for each questionnaire were first standardized as a vector of three numerical values (V, P, K), representing proportions of vata, pitta, and kapha in the dosha, summing to 100%. Reliability was tested through a distance measurement (measure of absolute deviation between Test 1 and Test 2 scores), by agreement in diagnosis, and by intra-class correlation co-efficients [20].

Each questionnaire gives separate, summed vata, pitta, and kapha scores. For each questionnaire, standardized proportions were determined by dividing each separate score by the sum so that the total V + P + K was equal to 1 [8]. Results of each questionnaire were compared at two different time points. If the time 1 score is (v1, p1, k1) and the time 2 score is (v2, p2, k2), then the distance D is calculated according to the formula

The composite distance D between Test 1 and Test 2 (V, P, K) vectors varies between 0 and , with 0 meaning the two vectors are the same and meaning that the diagnoses are completely different [6], [8]. In addition, to assess distance, intra-class correlations of the test and retest V and P/(P + K) scores were also computed. Note that the V, P, K space is two-dimensional, and that V, P, and K are inversely correlated, by design. P/(P + K) gives a second generator for the space, not a priori correlated with V.

As a secondary analysis, we also assigned an Ayurveda constitutional type to each participant. Numerical V,P,K scores were converted to 10 possible Ayurveda diagnoses (or classes). Dosha was defined based on a weighting of the three dosha types: vata, pitta, and kapha using the rule that any score equal to or above 0.25 was included in the constitution and order would determine primary and secondary dosha. None of the questionnaires assigned a numerical scoring key nor was there an established guideline provided in previous research, and thus this weighting system was arbitrarily assigned for this study. For example, suppose an individual's V, P, K scores were 0.6,0.3,0.1. Since two scores- V and P were equal to or above 0.25, a Dvidosha constitution was assigned. Furthermore, since V (0.6) was larger than P (0.3), the constitution was vata-pitta. Once discrete types were defined, we calculated the proportion of participants for whom the questionnaire produces the same constitutional diagnosis, at both tests.

Internal consistency

Internal consistency could be tested for Questionnaire 1, because it used designated lists of items for assessing each of the vata, pitta, and kapha constitutional types. A Cronbach's alpha statistic was used to calculate internal consistency from pairwise correlations between items that aim to assess the same dosha [12], [13].

As an exploratory analysis, we also computed intra-class correlation co-efficients to assess the agreement between the three questionnaires, in assessing prominence of the three doshas at baseline.

Methods by comparison to other self-reporting questionnaire studies

Kurande et al. assessed inter-rater reliability of pulse, tongue, and Prakriti assessment through Ayurveda practitioner inspection, history taking, and palpation, and by a researcher developed self-reporting questionnaire comprised of 75 items, 25 items relating to each of the three dosha types – vata, pitta, and kapha [8]. Each question of the self-reporting questionnaire required the subject to choose a level of agreement between 1 of 3 options. Study subjects were second-year Ayurveda college students, and none were asked if they had previously been given a Prakriti diagnosis. In our study, we excluded participants who had any prior knowledge of their Prakriti diagnosis in order to minimize subjective bias. The Kurande et al. questionnaire was administered at 1 time point and diagnosis was compared to other Prakriti assessment tools, unlike in our study in which we measured test–retest reliability of each questionnaire to itself over 2 time points, 1-month apart. Statistical analysis for our study was carried out similarly to Kurande et al., weighting each class of diagnosis type and using distance measurements to define level of agreement. In our study, Cronbach's alpha statistic was used for Questionnaire 1 to assess internal consistency.

Rastogi et al. developed a prototype Prakriti analysis tool assessing inter-rater reliability by comparing Prakriti diagnosis of a self-reporting questionnaire from 1 time point to that of Ayurveda physician diagnosis [9]. As mentioned earlier, we avoided using physician assessment comparison due to not having a validated physician assessment standard by which to compare.

The Rastogi and Kurande et al. studies attempted to minimize environmental or disease factors that may influence external expression of Prakriti by selecting healthy volunteers of a young age [8], [9]. While our study did not exclude participants with disease conditions, our hope was that measurement of test–retest reliability of each questionnaire from 2 time points would minimize this influence.

Similar to our study, Shilpa et al. developed a Tridosha, self-reporting questionnaire tested for reliability over 2 time points, approximately 1 month apart [10]. We chose to test the prior mentioned self-reporting questionnaires vs building upon the work of Shilpa et al., in order to assess reliability of what is currently used by western trained Ayurveda practitioners.

Results

Baseline socio-demographic characteristics of study participants were calculated (Table 2). Of the 26 participants recruited, 19 completed three questionnaires at two time points, one month apart (Fig. 1). Age range was 21–62 years old with a mean age of 34. Of the 19, 5 were male and 14 female. 16 participants were from Portland, Oregon and surrounding areas, and 3 were from out-of-state. 19 participants were included in final analysis of test–retest reliability.

Table 2

Baseline characteristics of study participants.

Characteristic	EnrolledN = 26	CompletedN = 19
Sociodemographic
Age, year, mean (SD)	34 (10)	35 (11)
Gender, no. (%)
Female	18 (69)	14 (74)
Male	8 (31)	5 (26)
Location, no. (%)
Portland, OR & Surrounding Area	23 (88)	16 (84)
Out-of-State	3 (12)	3 (16)

Fig. 1

Study diagram.

For all questionnaires, the mean (SD) distance between test and retest (V,P,K) was found (Table 3). Intra-class correlation co-efficients-ICC (1,1) were calculated for vata and the proportion of pitta among pitta and kapha P/(P + K). ICC (1,1) results indicate what proportion of variation in outcomes is due to real differences between individual participants, as opposed to variability of diagnosis in the same individual.

Table 3

Test–retest agreement results (Distance and intraclass correlation coefficients ICC [1,1]).

Questionnaire (Q)	Mean (SD) Distance	Vata ICC (1,1)	95% CI	P/P + K ICC (1,1)	95% CI
Q1	0.05 (.03)	0.69	(0.37, 0.87)	0.69	(0.37, 0.87)
Q2	0.11 (.07)	0.86	(0.67, 0.94)	0.81	(0.58, 0.92)
Q3	0.10 (.07)	0.89	(0.76, 0.96)	0.76	(0.51, 0.91)

Percentage of test/retest agreement for discrete diagnoses was also calculated using the rule that any score equal to or above 0.25 was included in the constitution and order would determine primary and secondary dosha (Table 4). The 3 most common diagnoses across both time points listed in descending order from most to least common was Tridosha, vata-pitta, and pitta-vata tied with pitta-kapha (Table 4). The least common diagnosis across both time points was kapha tied with kapha-vata. Percentage of agreement between time points for each individual questionnaire was also calculated resulting in Q1 with highest level of agreement (Table 4).

Table 4

Test–retest discreet diagnosis agreement results and total.

	Time Point 1 (n)			Time Point 2 (n)			Total
	Q1	Q2a	Q3	Q1	Q2	Q3b
Vata	0	2	0	0	3	0	5
Pitta	0	1	1	0	0	0	2
Kapha	0	0	1	0	0	0	1
Vata-Pitta	2	2.5	7	1	3	6.5	22
Vata-Kapha	0	0	0	0	2	0	2
Pitta-Vata	0	1.5	4	0	3	1.5	10
Pitta-Kapha	0	2	2	0	2	5	11
Kapha-Vata	0	1	0	0	0	0	1
Kapha-Pitta	0	0	0	0	2	1	3
Tridosha	17	9	4	18	4	5	57
Total	19	19	19	19	19	19	114
% agreement	Q1: 95% Q2: 42% Q3: 63%

Q2 at Time Point 1 resulted in 1 participant scoring equally Dvidoshic as both vata-pitta/pitta-vata, counted as 0.5 for each dosha.

Q3 at Time Point 2 resulted in 1 participant scoring equally Dvidoshic as both vata-pitta/pitta-vata, counted as 0.5 for each dosha.

In order to determine a discrete diagnosis from questionnaire scores, it is necessary to choose some minimum cutoff for inclusion of a type in the diagnosis. Since the 0.25 standard was arbitrarily set (the questionnaires themselves provide no guidance), and since this resulted in clinically counterintuitive findings of high numbers of Tridosha diagnoses for Q1, we performed a post-hoc re-analysis of discrete diagnoses using a 0.30 standard for constitutional inclusion. The three most common diagnoses across both time points in re-analysis were vata-pitta, pitta-vata, and Tridosha, with kapha the least common (Table 5). Percentage of agreement between time points for each individual questionnaire was calculated again in re-analysis, resulting in a comparable degree of agreement for all three questionnaires (Table 5).

Table 5

Test–retest discreet diagnosis agreement results and total post-hoc re-analysis with 0.30 standard.

	Time Point 1 (n)			Time Point 2 (n)			Total
	Q1a	Q2b	Q3	Q1	Q2c	Q3d
Vata	0	3	1	0	5	2	11
Pitta	0	1	4	0	2	3	10
Kapha	0	1	1	0	0	0	2
Vata-Pitta	4.5	3.5	6	4	3	5.5	26.5
Vata-Kapha	1	0	0	2	1	0	4
Pitta-Vata	2.5	4.5	4	4	2	4.5	21.5
Pitta-Kapha	2	2.5	1	2	2.5	3	13
Kapha-Vata	2	1	0	1	0	0	4
Kapha-Pitta	0	1.5	1	0	2.5	1	6
Tridosha	7	1	1	6	1	0	16
Total	19	19	19	19	19	19	114
% agreement	Q1: 58% Q2: 63% Q3: 63%

Q1 at Time Point 1 resulted in 1 participant scoring equally Dvidoshic as both vata-pitta/pitta-vata, counted as 0.5 for each dosha.

Q2 at Time Point 1 resulted in 2 participants scoring equally Dvidoshic as both vata-pitta/pitta-vata and pitta-kapha/kapha-pitta, counted as .5 for dosha.

Q2 at Time Point 2 resulted in 1 participant scoring equally Dvidoshic as both pitta-kapha/kapha-pitta, counted as .5 for dosha.

Q3 at Time Point 2 resulted in 1 participant scoring equally Dvidoshic as both vata-pitta/pitta-vata, counted as .5 for dosha.

Calculation of Cronbach's α, for baseline responses to Questionnaire 1, produced estimates of moderate to low internal consistency (Table 6), with especially poor consistency in the assessment of kapha.

Table 6

Calculation of Cronbach's α, for baseline responses to Questionnaire 1.

	α	Interitem correlations for 14 questions (range)	Internal consistency
Vata	0.523 (N = 26)	−0.33 to 0.62	poor
Pitta	0.604 (N = 25)a	−0.45 to 0.76	questionable
Kapha	0.184 (N = 23)b	−0.46 to 0.60	unacceptable

When imputing possible values for one missing answer in one participant, α varies from 0.583 to 0.604.

One participant was missing many scores. When imputing possible values for one missing answer in each of two other participants, α varies from 0.206 to 0.373 (N = 25).

As an exploratory analysis, we calculated intra-class correlation co-efficients for absolute agreement between the three questionnaires at baseline (Table 7). Scores show that there is meaningful lack of agreement between the three scales.

Table 7

Intra-class correlation co-efficients-ICC (2,1) for absolute agreement between questionnaires at baseline.

	ICC (2,1)
Vata	0.520
Pitta	0.185
Kapha	0.376

Discussion

Many experienced practitioners use some form of Ayurvedic constitutional questionnaires to guide the initial intake, while also taking into account, other tools such as pulse diagnosis, health history, and physical exam (including tongue assessment). Nevertheless, many experienced Ayurveda practitioners do not consider self-reporting constitutional questionnaires to be very reliable as a stand-alone diagnostic tool. This may explain the tolerance of wide variation in questionnaire type and format.

Several studies have examined reliability of Prakriti diagnosis by way of pulse taking alone [1], [6], [18], and by incorporating several of the other diagnostic methods together (pulse + history taking + observation) [1], [8], [9]. To date, results of these studies have varied greatly from low to moderate levels of reliability. As a result, validity was not considered an objective of this study, due to lack of any validated standard to which questionnaire results can be compared. We recognize this limitation and realize that physician assessment may need to be included for future studies regardless of lacking a validated standard.

None of the questionnaires disguised dosha categories in question. Because of this, it was important in the exclusion criteria that participants have not previously received an Ayurveda diagnosis. By the very nature of reading the responses within each labeled category, however, participant answers may have become gradually biased throughout the process of taking each questionnaire, so that by the last one taken, they begin to identify with a specific dosha. It would be preferable, in a standardized questionnaire, for items and/or responses corresponding to different doshas to be randomly ordered, without identifiable labels. Furthermore, it would have been beneficial to exclude participants with any known pathology in order reduce influence on external expression of Prakriti.

Questionnaire 1 is uniquely different than Questionnaires 2 and 3. The answer options within each section of Questionnaire 1 are a degree of agreement on a scale from 0 to 6 (0 = does not apply, 3 = applies somewhat, 6 = applies most) with each statement pertaining to each particular dosha-vata, pitta, and kapha. Questionnaires 2 and 3 ask participants to choose 1 of 3 possible responses for each question, with each answer corresponding to one dosha. Research in scale construction for social and personality psychology tells us that different scale types may produce different responses, even when the question is the same [21]. Response type may influence the accuracy of assessment and therefore, the accuracy of Ayurveda diagnosis. According to Questionnaire 1, but not to Questionnaires 2 and 3, most participants are Tridoshic. A Tridoshic Prakriti is generally one that is in perfect balance, not requiring treatment, and is infrequently diagnosed [8], [18]. This finding could be the function of the 25% cut-off that was arbitrarily assigned for this study in order to assign discrete diagnoses. Although this was not stated as an aim, it is notable in assessing the quality of this questionnaire. Interestingly, post-hoc re-analysis with a 30% cut-off for discrete diagnoses corrected for this high degree of Tridoshic results in Q1, but also lowered the rate of agreement between time points for Q1 from 95% to 58%. Clinical application of self-reported questionnaires ultimately relies on diagnosis, and variation in results due to the cut-off standard may impact clinical decision-making associated with treatment.

Cronbach's α statistics showed variable internal consistency in measuring the three dosha types, with especially poor consistency in the assessment of kapha. The discrepancy in consistency brings up two important questions. 1) Is this questionnaire better adept in evaluating those with vata and pitta primary dosha types and 2) Might inconsistency of kapha responses relate to a western cultural bias against the identifiable cluster of kapha characteristics?

A kapha dosha body type is structurally bigger and of a slower metabolism. Although not portrayed negatively in Ayurveda theory, kapha dosha types may be perceived negatively in western culture where thin-framed body types are generally more desirable [22]. Overall, kapha was the dosha diagnosis least represented amongst participants in this study (Table 6, Table 4). This may be due small sample size (N = 19), difficulty assessing kapha through a self-reported questionnaire format, psychologically influenced based on desirability of dosha traits, or due to administering the survey mainly to naturopathic students who may not represent a typical distribution of body types.

It is important to note that Prakriti features described in questionnaires were originally observed within the context of an Indian population, and thus may not be verbatim transferable to other ethnic groups and geographies. In particular, our sample consisted primarily of white American graduate students, a highly specialized population. Although details have not been published, it is certainly possible that there is better evidence for reliability and validity of the questionnaires, in other populations. While we do not know the background development of specific questionnaires used for this study, we need to take this limitation into consideration for development of any future self-reporting Ayurveda questionnaire.

Within each set of items in Questionnaire 1, we identified a wide range of inter-item correlations, including negative relationships. Future research may include factor analytical techniques aimed at identifying a set of items with high inter-correlation that may be more reasonably identified as markers of the underlying constitutional type. This type of analysis could also be employed with questions having designated vata, pitta, and kapha responses, in order to identify sets of items for which there are high rates of response agreement. Development of such internally consistent item lists will be an important future step in producing a valid questionnaire.

It is interesting to note that some questionnaires demonstrate small mean changes over time (using the distance formula) and have relatively good intra-class correlations; but that extracted discrete diagnoses still change for most participants. Depending upon how the scales are used, this might greatly impact our assessment of test–retest reliability. Moreover, we should note that this small study was only powered to find meaningful changes in the distance metric [6], [8]; our estimates of Cronbach's α, intra-class correlation, and rate of discrete test–retest agreement in dosha are relatively imprecise.

Conclusion

Analysis performed in this study indicates that three prominent self-reporting Ayurveda questionnaires have moderately good test–retest reliability according to numerical scores, but highly variable reliability according to discrete Ayurveda diagnosis. Internal consistency pertaining to individual constitutions within Questionnaire 1 is at best questionable for all three primary doshas, and poor for kapha. Further research is necessary to develop a reliable and standardized constitutional questionnaire.

Sources of funding

This work was supported by grant R25AT002878 from the National Institute of Health/National Center for Complementary and Integrative Health, NIH/NCCIH).

Conflict of interest

None.