Towards standardization of Prakriti Evaluation: A scoping review of modern assessment tools and their psychometric properties in Ayurvedic medicine

Abstract

Background

Ayurveda is an ancient Indian medical system that emphasizes individualized care based on the concept of Prakriti, which represents an individual's relative proportion of three doshas (humour) at the time of conception. Prakriti is believed to remain unchanged throughout one's life span, and based on the relative preponderance of doshas, it has been classified into ten types. Researchers and practitioners have developed various assessment tools, such as questionnaires, algorithmic and machine learning-based methods, and devices to bring objectivity and replicability to the Prakriti evaluation procedure. However, a systematic evaluation of their effectiveness and psychometric properties is currently lacking in the literature.

Objective

To identify Prakriti, classical Ayurveda texts suggest various techniques, but subjective variations and bias in interpretation have been reported among practitioners.

Methods

This scoping review aims to identify modern Prakriti assessment tools available in scientific literature and describe their psychometric properties to contribute to the development of more effective and standardized methods for Prakriti evaluation. Employing Arksey and O'Malley's five-stage methodological framework, the review critically assesses Prakriti evaluation tools, such as questionnaires, machine learning algorithms, and devices, to provide insights into the robustness and accuracy of these assessment methods.

Results

Thirty-two studies meeting the inclusion criteria were included in the review. Sixteen studies utilized questionnaires for Prakriti assessment, of which three were validated, six had established reliability, and three had both reliability and validity confirmed. Five studies employed algorithm-based methods, with three of these using validated machine learning models. Eleven studies utilized devices to assess Prakriti types, with only one device validated and providing access to data for reproducibility. Questionnaires were the most commonly used tools for Prakriti evaluation, followed by algorithm-based methods and devices.

Conclusion

Questionnaires are the most validated tools presently for Prakriti assessment despite their limited psychometric robustness. Machine learning based models show potential but suffers challenges in accuracy, replicability and finding the ground truth in a robust manner, and devices are the most objective assessment tools though none of them have been appropriately validated. Despite these limitations, Prakriti diagnosis is a promising field that requires robust procedures to establish standardized, replicable and reliable measures for its evaluation.

1. Introduction

Ayurveda is an ancient Indian medical system dating back over 5000 years [1]. Based on the principles of five elements (ether, air, fire, water, earth) and three doshas (Vata, Pitta, Kapha), Ayurveda emphasizes individualized care with the concept of Prakriti being central to personalized and preventive medicine [2]. The term Prakriti has been conceptualized as a consequence of the relative proportion of three doshas – Vata, pitta, and Kapha in an individual at conception [3]. Prakriti is believed to remain unchanged throughout one's lifespan [4]. Based on the relative preponderance of doshas, Prakriti has been classified into seven types - three extreme types based on single dominant dosha (Vata, Pitta, and Kapha), three dual Prakriti types based on the dominance of two doshas (Vata-Pitta, Pitta-Kapha, and Kapha-Vata) and one balanced dosha (Vata-Pitta-Kapha) Prakriti types [5]. Contemporary Ayurveda researchers further subdivide the three dual Prakriti into six types based on active dosha (Vata-Pitta, Pitta-Vata, Vata-Kapha, Kapha-Vata, Pitta-Kapha, Kapha-Pitta), leading to total ten prakriti types [[6], [7], [8], [9]].

Five primary classical textbooks are a compendium of Ayurveda – Charaka Samhta [10], Susrutha Samhita [11], Ashtanga Hirdaya [12], Harita Samhita [13], and Sharangdhar Samhita [14]. While Charka Samhita described Prakriti based on characteristic attributes of dosha [10], Sushruta [11], Vaagbhat'a [12], Haarita [13], and Shaarangdhara [14] described Prakriti in terms of anatomical (samrachnatmaka), physiological (shaaririk), and psychological (maanshik) characteristics. To identify Prakriti based on such observable attributes, different methods of Prakriti assessment are available in classical Ayurveda texts and modern Ayurveda scientific literature.

Classical Ayurveda entails using various techniques by the Ayurvedic practitioner, such as physical examination, pulse diagnosis, and inference based on information gathered through indirect means, such as asking questions [15]. However, this approach leads to subjective variations and bias in interpreting Prakriti features among Ayurveda practitioners [16]. To bring objectivity and replicability to the Prakriti evaluation procedure, efforts in the form of the development of questionnaires, algorithmic and machine learning algorithms, and devices have been made by researchers and Ayurveda practitioners across the globe [17]. Psychometric measures such as reliability and validity (in questionnaires), sensitivity and specificity (in algorithm and machine learning tools), and replicability (in devices) have been used as the criteria to standardize the Prakriti evaluation process in those assessment tools [18]. However, despite the growing number of assessment tools available, a systematic evaluation of their effectiveness and psychometric properties is currently lacking in the literature. Therefore, there is a need for a scoping review that can critically examine the available evidence on the reliability, validity, sensitivity, specificity, and replicability of these assessment tools and their potential for use in clinical practice and research. With this background, we undertook the present review to identify the modern Prakriti assessment tools available in scientific literature and describe their psychometric properties.

This review also identifies literature gaps in the existing research and highlights future research areas, ultimately contributing to developing more effective and standardized methods for Prakriti evaluation.

It's worth noting that this review specifically focuses on the Prakriti assessment and does not address Vikriti, which represents the current state of doshas and can fluctuate due to dietary and environmental factors [19].

2. Methods

The five-stage methodology framework for conducting the review has been followed: (1) identifying the research question; (2) identifying relevant studies; (3) study selection; (4) charting the data; (5) collating, summarising, and reporting results [20]. A scoping review protocol has been developed and is available on request.

• Stage 1: Identifying the Research Question

To the best of our knowledge, there are no systematic or scoping reviews that may be used as a first-hand resource to identify a reliable/validated tool among the available Prakriti assessment tools; therefore, the research question has been kept as broad as possible to include questionnaires, scales, algorithm-based tools, and devices to assess the Prakriti body types. With this background, the research question formulated as “What are the available questionnaires, devices, algorithmic and mathematical models for objectively assessing Prakriti? What are the effectiveness and psychometric properties of each of these instruments for assessment of Prakriti?”

• Stage 2: Identifying Relevant Studies

The review followed specific inclusion and exclusion criteria based on the research question. Studies assessing Prakriti through questionnaires, algorithms, and devices published between 2016 and 2022 were included. Studies published in languages other than English and those in predatory journals were excluded. The databases searched were Google Scholar, PubMed, and Scopus. Only the first ten pages of Google Scholar results were considered [21]. Boolean operators were used to develop search queries, which included keywords like Prakriti, dosha, questionnaires, algorithms, and devices. Detailed search queries for each database are provided in the Supplementary File.

• Stage 3: Study Selection

Based on the search query, a total of 899 studies were identified from 2016 to 2022: 205 from PubMed, 175 from Scopus, and 519 from Google Scholar. After removing duplicates, 425 studies were identified. Title and abstract screening further reduced the number to 125 studies. Full texts of these 125 articles were retrieved and screened using the inclusion and exclusion criteria, including checks against Beall's list of predatory journals [22] and the UGC Care list of journals [23]. This screening excluded 39 studies published in predatory journals, one non-English study, 33 studies not relevant to the research question, and four studies inaccessible through interlibrary loan, resulting in 48 eligible studies.

Authors of studies without accessible Prakriti assessment instruments were contacted. Non-responsive authors led to the exclusion of 24 studies. Five additional studies were excluded for using previously developed questionnaires. Consequently, 19 studies were included in the review. An additional backward citation search identified 13 relevant studies, bringing the total to 32 studies. The detailed search and selection process is illustrated in the PRISMA flow diagram (Fig. 1).

• Stage 4: Charting the data.

Fig. 1.

Represents the PRISMA Flow Diagram for the scoping review process.

Relevant data from the included 32 studies were thus extracted and put into a presentation format representing the studies' relevant information using the methodological guidelines of the Joanna Briggs Institute [24]. A table was created to compare the included texts.

• Stage 5: Collating, Summarising, and Reporting Results

Following the suggestion of Arksey and O'Malley [20], the charted information of the studies has been presented in Fig. 1, and Table 1, Table 2, Table 3, Table 4, Table 5, Table 6. Table 1, Table 2 presents prakriti tools based on questionnaires and scales, Table 3, Table 4 presents prakriti instruments based on software and algorithms, and Table 5, Table 6 presents prakriti instruments based on devices and equipment.

Table 1.

Summary of studies on Questionnaire-based Prakriti assessment instruments.

Study ID	Author (year)	Name of the tool	Type of Tool [I/S]	Instructions to identify Prakriti based on lifespan	Type of Prakriti identified (No. of Prakriti Identified)
SQ1	Bhalerao, Deshpande, Thatte (2012)	TNMC Prakriti 2004 Questionnaire	I	N	EP (3)
SQ2	Dunlap et al. (2017)	N N N	S S S	Y Y Y	EP (3), DP (6), BP (1) EP (3), DP (6), BP (1) EP (3), DP (6), BP (1)
SQ3	Shirolkar et al. (2018)	PAPC	I	N	DP (3)
SQ4	Prasher et al. (2008)	Questionnaire for prakriti evaluation	I	N	EP (3)
SQ5	Rotti et al. (2014)	AyuSoft	I	Y	EP (3), DP (6), BP (1)
SQ6	Sivapuram et al. (2020)	Questionnaire for Prakriti (Ayurveda Personality) Analysis	I	N	EP (3)
SQ7	Bell et al. (2017)	Ayurvedic Constitution Questionnaire	S	N	EP (3)
SQ8	Meenakshi et al. (2021)	Prakriti assessment proforma	S	Y	EP (3)
SQ9	Joshi (2004)	Questionnaire for Statistical Modelling of Ayurvedic Diagnostic Factors	B	N	DP (6)
SQ10	Tripathi et al. (2011)	The self-assessment questionnaire to assess Prakriti	S	N	EP (3), DP (3)
SQ11	Shilpa and Murthy (2011)	Mysore Psychological Tridosha Scale	S	Y	EP (3)
SQ12	Rastogi (2012)	Prototype Prakriti Analysis Tool (PPAT)	I	N	EP (3)
SQ13	Kurande et al. (2013)	ABC Questionnaire	S	N	EP (3)
SQ14	Suchitra et al. (2014)	Ayurveda Child Prakriti Inventory	S	N	EP (3)
SQ15	Fave et al. (2015)	QDAV QDAV-R	S S	N N	EP (3), DP (2) EP (3), DP (2)
SQ16	Edwards and Streiner (2022)	New Dosha Self-Assessment Questionnaire	S	Y	EP (3)

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SQ1, SQ2, SQ3, etc.) for cross-referencing purposes. Name of tool: PAPC: Proforma for assessment of Prakruti (Psychosomatic constitution); QDAV: Questionnaire on Dosa Prakriti Ayurveda; Type of tool: I - Interview styled questionnaire; S - Self-administered; B – Both may be used in an interview or in self-administered mode); Instructions to identify Prakriti available: Y/N – Yes or No; Types of Prakriti identified: EP – Extreme Prakriti;DP–Dual Prakriti;BP: Balanced Prakriti; (x): The number, x, in parenthesis indicates the number of Prakruti classes identified within each type.

Table 2.

Detailed assessment of Questionnaire-based Prakriti assessment instruments.

Study ID	No. of items	School of thought	Scoring pattern available [(Y/N)	Validated (Y/N)	Type of reliability and/or validity	Validation sample size	Licensing for use [O/P]
SQ1	37	CS	Y	Y	CSV	30	O
SQ2	Q1 (41) Q2 (42) Q3 (32)	NA NA NA	Y Y Y	Y Y Y	TRR, IC TRR TRR	19	O O O
SQ3	58	NA	N	Y	CSV	NA	O
SQ4	73	CS, SS	N	Y	IRR	320	O
SQ5	85	CS, SS, AH	Y	Y	IRR	3416	P
SQ6	30	NA	Y	N	N	NA	O
SQ7	10	NA	N	N	N	NA	O
SQ8	34	NA	N	N	N	NA	O
SQ9	28	NA	N	N	N	NA	O
SQ10	62	CS	Y	Y	CRV	50	O
SQ11	157	NA	N	Y	CTV IC, SHR TRR	18 & 10 1548 60	P
SQ12	54	CS	Y	Y	IRR, CTV, CSV	26	O
SQ13	75	CS, SS	N	N	N	NA	O
SQ14	135	NA	Y	Y	IC, SHR, CTV, CSV	230	O
SQ15	20 15	NA NA	Y Y	N N	N N	NA NA	O NA
SQ16	39	NA	N	Y	TRR IC	76 267	O

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SQ1, SQ2, SQ3, etc.) for cross-referencing purposes. School of thought: CS - Charaka Samhita; SS - Sushruta Samhita; AH - Ashtanga Hridaya; Scoring method available: Y/N - Yes or No; Type of reliability and/or validity: TRR – Test-Retest Reliability; SHR - Split Half Reliability; IRR: Inter-Rater Reliability; IC: Internal Consistency; CRV: Criterion Validity; CNV: Convergent Validity; CTV: Content Validity; CSV: Construct Validity; Licensing for use: O - Open source; P - Proprietary; Miscellaneous: NA - Not available; Q1 - Questionnaire 1; Q2 - Questionnaire 2; Q3 - Questionnaire 3.

Table 3.

Summary of algorithm-based studies, data characteristics, and Prakriti identified.

Study ID	Author (year)	Data size	Data labelling method	Features used for classification	Types of Prakriti identified (No. of Prakriti identified)
SA1	Farooque et al. (2016)	67	AP∗	37 items questionnaire	EP (3)
SA2	Tiwari et al. (2017)	147	AP#	73 items Questionnaire	EP (3)
SA3	Gadre (2019)	330	AP$	Seven features from facial Images	EP (3), DP (3), BP (1)
SA4	Madaan & Goyal (2020)	405	Q##	Questionnaire's responses	EP (3)
SA5	Joshi Manisha et al. (2020)	107	AP$	Tongue image	EP (3)

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SA1, SA2, SA3, etc.) for cross-referencing purposes. Data labelling method: AP∗: Data labelled by Ayurveda Physician using Questionnaire and Physical Observation; AP^#: Data labelled by Ayurveda Physician using Questionnaire only; Q^##: Data labelled by authors using Questionnaire; AP^$: Data labelled by Ayurveda Physician using images only; Types of Prakriti identified: EP – Extreme Prakriti;DP–Dual Prakriti; BP – Balanced Prakriti; (x): The number, x, in parenthesis, indicates the number of Prakruti classes identified within each type.

Table 4.

Assessment techniques and validation details of algorithm-based studies.

Study Id	Technique used	Algorithms described in detail [y/n]	Validation method	Validation metric used	Availability of data and code [NN/YN/NY/YY]	Licensing for use [O/P]
SA1	Decision Tree J48 (Using Weka)	Y	NA	Kp, MAE, RMSE	NN	NA
SA2	Unsupervised, LASSO, Elastic Net, Random Forest	Y	CV-10F, OoC	Se, Sp, Ac	YNa	NA
SA3	DL, CV	N	N	Ac	NN	NA
SA4	KNN, ANN, SVM, NB, XGB, CaB	Y	HoV (80:20)	RMSE, Pr, Rc, F-score, Ac	NN	NA
SA5	KNN, ANN, DT	Y	HoV	Se, Sp, Pr, NPR	NN	NA

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SA1, SA2, SA3, etc.) for cross-referencing purposes. Techniques used: DL: Deep Learning; CV: Computer Vision; KNN: K-Nearest Neighbour algorithm; ANN: Artificial neural network; SVM: Support Vector Machine; NB: Naïve Bayes; XGB: XGBoost; CaB: CatBoost; DT: Decision Tree; Algorithms described in detail [y/n]: Y/N- Yes or No; Validation method: NA: Not available; CV-10F: Ten Fold cross validation; OoC: Out of cohort validation method (a cross validation method devised by us wherein one dataset has been used for testing and another dataset has been used for training); HoV: Hold out validation; Data size: NA: Not available; Validation metric used: Kp: Kappa statistic; MAE: Mean Absolute Error; RMSE: Root Mean Square Error; Se: Sensitivity; Sp: Specificity; Ac: Accuracy; Pr: Precision; Rc: Recall; NPR: Negative Predictive Value; Availability of data and code [NN/YN/NY/YY]: NN – Neither dataset available Nor Code; YN: Dataset available but Not code; NY: Dataset NOT available but Code is available; YY: Dataset and code, both are available; Licensing for use [O/P]: NA: Not Available; O: Open source; P: Proprietary.

^aThe paper indicates that the dataset is accessible to researchers upon request. We have sent multiple requests to the authors to obtain the dataset for our review, however, we have not yet received a suitable response or access to the dataset.

Table 5.

Devices-based Prakriti assessment instruments.

Study ID	Author (year)	Name of tool	Type of tool
SD1	Chaudhari & Mudhalwadkar (2017)	NA	Three PPG sensor-based
SD2	Goyal & Agarwal (2017)	NA	Single Pr sensor based
SD3	Khandai & Jain (2017)	NA	Three Pe and Pr sensor based
SD4	Kadarmandalgi & Asaithambi (2019)	NA	Three PPG sensor based
SD5	Joshi et al. (2007)	Nadi Tarangini	Three Pr Sensor based
SD6	Sareen et al. (2009)	Nadi Yantra	Three Pe sensor based
SD7	Selvum and Begum (2011)	Nadi Aridhal	Three Pe sensor based
SD8	Kalange et al. (2012)	Nadi Pariksha Yantra	Three Pe sensor based
SD9	Bawankar et al. (2021)	Nadi Pariksha	IoT-based pulse examination system
SD10	Joshi & Bajaj (2021)	NA	Electrical sensor-based system using ADS1293
SD11	Rao, Ravishankar, and Rao (2022)	Wrist pulse acquisition system	Three Pe sensor-based

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SD1, SD2, SD3, etc.) for cross-referencing purposes. Type of tool: PPG: Photoplethysmography;Pe: Piezoelectric; Pr: Piezoresistive; IoT: Internet of Things.

Table 6.

Detailed characteristics and validation of devices-based Prakriti assessment instruments.

Study ID	Scoring algorithm described in detail [A/NA]	Type of validation	Availability of validated data [Y/N]	Availability of equipment or device [O, P, NA]
SD1	NA	NA	N	N
SD2	NA	NA	N	N
SD3	NA	NA	N	N
SD4	NA	NA	N	N
SD5	A	NA	N	P
SD6	NA	NA	N	N
SD7	NA	NA	N	NA
SD8	NA	NA	N	N
SD9	NA	NA	N	NA
SD10	NA	NA	N	NA
SD11	A	IRR	Y (available on request)	N

Abbreviations used: Study ID: Unique identifier assigned to each study (e.g., SD1, SD2, SD3, etc.) for cross-referencing purposes. Scoring algorithm described in detail [A/NA]: A: Available; NA: Not Available; Type of validation: NA: Not Available; PV: Predictive Validity; IRR: Inter-Rater Reliability; CrV: Criterion Validity; Availability of validated data [Y/N]: Y: Yes; N: No; Availability of equipment or devices: O: Open source; P: Proprietary; N: No.

Screening was not performed in duplicates due to time constraints. However, Google Sheets were used for tabulating and organizing the retrieved studies, and Mendeley software was employed during the screening process.

2.1. Psychometric evaluation, machine learning validation, and device reproducibility

We evaluated the psychometric properties of the questionnaire-based Prakriti assessment tools using three reliability measures: internal consistency, test-retest reliability, and inter-rater reliability [[25], [26], [27], [28], [29]]. Additionally, three validity measures were used: content validity, criterion validity, and construct validity [26,[30], [31], [32], [33], [34], [35], [36]].

For the assessment of machine learning models, we used reproducibility, cross validation methodology, and standard performance metrics such as accuracy, precision, recall, F-measure, specificity, sensitivity [[37], [38], [39], [40], [41]]. The ground truth labelling method was also carefully analysed and assessed.

Furthermore, research reproducibility was used to assess the devices and equipment [42,43]. Detailed descriptions of these assessments are provided in the Supplementary File.

3. Results

Based on the methodology of Prakriti assessment, the finalized thirty-two studies were classified into three categories: questionnaire-based, algorithm-based, and device-based tools. The details of these studies are presented in Table 1, Table 2, Table 3, Table 4, Table 5, Table 6, Table 7, each providing an overview of the study characteristics and their validation properties. Specifically, Table 1, Table 3, Table 5 present the study details, such as Study ID, Author (year), type of tool, and the number of Prakriti identified. And Table 2, Table 4, Table 6 extend these details with validation characteristics, including the type of validation method, availability of data, and validation metrics. Table 1, Table 2 describes sixteen studies that used questionnaire-based Prakriti assessment tools. Table 3, Table 4 present five studies employing algorithm-based tools. Lastly, Table 5, Table 6 described studies that used device-based tools. Additionally, Table 7 presents a list of Prakriti assessment questionnaires along with their established psychometric properties, thus, offering a consolidated view of tools that have undergone rigorous validation.

Table 7.

Reliability and Validity Metrics for the validated Prakriti Assessment Questionnaires.

Study title	Prakriti assessment questionnaire	Reliability metrics	Validity metrics
Bhalerao, deshpande, Thatte (2012)	TNMC Prakriti 2004 Questionnaire	NA	CSV
Dunlap et al. (2017)	NA NA NA	IC, TRR TRR TRR	NA NA NA
Shirolkar et al. (2018)	Proforma for assessment of Prakruti (Psychosomatic constitution)	NA	CSV
Prasher et al. (2008)	Questionnaire for Prakriti evaluation	IRR	NA
Rotti et al. (2014)	AyuSoft	IRR	NA
Tripathi et al. (2011)	The self-assessment questionnaire to assess Prakriti	NA	CRV
Shilpa and Murthy (2011)	Mysore Psychological Tridosha Scale	IC, SHR, TRR	CTV
Rastogi (2012)	Prototype Prakriti Analysis Tool (PPAT)	IRR	CTV, CSV
Suchitra et al. (2014)	Ayurveda Child Pakriti Inventory	IC, SHR	CTV, CSV
Edwards and Streiner (2022)	New Dosha Self-Assessment Questionnaire	IC, TRR	NA

Abbreviations used: Reliability metrics: TRR – Test-Retest Reliability; SHR - Split Half Reliability; IRR: Inter-Rater Reliability; IC: Internal Consistency; Validity metrics: CRV: Criterion Validity; CNV: Convergent Validity; CTV: Content Validity; CSV: Construct Validity.

3.1. Details of included studies

3.1.1. Questionnaire-based Prakriti assessment instruments

Among the questionnaires reviewed, eleven used the self-assessment format, six used the interview-styled format, and two used both. The questionnaires evaluate extreme, dual, and balanced Prakriti types. As discussed above, reliability, validity, and their different types have been used to evaluate the robustness of the questionnaires. While the questionnaire contains items in the range of twenty to one hundred fifty-seven, only eight of them explicitly asked the respondents to fill in their responses, which persisted over a long period of time to ensure that the respondents were asked about their Prakriti and not the current state of doshas, i.e., vikriti. Likewise, out of the total questionnaires reviewed, sixteen are open source, two are proprietary, and the remaining one is not reproducible; hence, it cannot be described whether it is open source or proprietary. Additionally, questionnaires such as TNMC Prakriti 2004 Questionnaire [44], Questionnaire for Prakriti Evaluation [48], AyuSoft [8], Questionnaire for Statistical Modelling of Ayurvedic Diagnostic Factors [53], the self-assessment questionnaire to assess Prakriti [55], Mysore Tridosha Scale [57], Prototype Prakriti Analysis Tool (PPAT) [58], are comparatively better cited in the available literature. In the following part of this subsection, we discuss each Questionnaire in more detail.

Bhalerao, Deshpande, & Thatte [44] used a thirty-seven-item, interview-styled questionnaire. Based on the Prakriti verses described in the eighth chapter of Vimana Sthana, Charak Samhita [45], the questionnaire comprises items assessing an individual's physical, psychological, and physiological characteristics. The questionnaire included items with three options corresponding to attributes associated with Vata (V), Pitta (P), or Kapha (K). Scores were calculated by summing up the responses in the V, P, and K domains, allowing individuals to be classified into specific prakriti types based on their scores. A participant scoring ≥50 % in a particular dosha was considered to have the predominant dosha, while a score between 25 % and 35 % categorized the dosha as the secondary dosha in the Prakriti. “The validation process of the questionnaire involved pre-testing, where the agreement between the questionnaire results and the clinical assessment of Prakriti conducted independently by two Ayurveda physicians was examined in a group of 30 participants.¹” The findings revealed a high level of agreement, with more than 90 % concordance observed in the independent assessment of Prakriti by the two clinicians and the questionnaire's evaluation, thus suggesting good construct (convergent) validity. This indicates that the questionnaire's results align closely with the clinical assessment, providing evidence that they are measuring the same underlying construct of Prakriti. However, the authors did not specify how two Ayurveda expert's evaluations were combined into a final assessment of Prakriti. The instrument is available through open access.

Dunlap et al. [46] established the psychometric properties of three publicly available (but unpublished) self-reported questionnaires to evaluate Prakriti. Questionnaire 1 consisted of 41 items, questionnaire 2 consisted of 42 items, and Questionnaire 3 consisted of 32 items. Questionnaire 1 asked respondents to provide their responses on a scale of 0–6 (0 = does not apply, 3 = applies somewhat, 6 = applies most), and items were categorized according to their relationship to Vata, Pitta, and Kapha dosha. Questionnaires 2 and 3 yielded three distinct responses to each item, corresponding to the three doshas and labelled V, P, and K. In general, on the three questionnaires, participants were instructed to respond based on what was true “over their lifespan” rather than recent events. Though the authors reported no specific numerical scoring key for any of the three questionnaires, they employed the following criteria to classify the respondents into Prakriti types: the largest score among Vata (V), Pitta (P), and Kapha (K) domains were used to identify the primary dosha. If the remaining V, P, and K scores were 25 % of the total score, the participant was categorized as having a single dosha. In cases where one of the remaining V, P, or K scores was ≥25 % of the total score, the participant was classified as having dual dosha, with the second largest score determining the secondary dosha. If all three scores were ≥25 %, the participant was considered to have Tridosha. In instances where a participant scored equally for dual dosha at a one-time point (e.g., Vata-Pitta/Pitta-Vata) and had a single dosha at a second-time point (e.g., Vata-Pitta) with at least a 50 % match, this was regarded as dosha agreement. The authors established the reliability using the test-retest and internal consistency measures for the three questionnaires. Based on the responses of nineteen participants to the three questionnaires, the authors calculated Cronbach's alpha as the statistic to measure the internal consistency of questionnaire 1 (Q1) and intra correlation coefficient (ICC) as the test statistic for the three questionnaires to measure the stability of scores (test-retest reliability) over the period of one month time interval. For Questionnaire 1, the values of Cronbach's alpha were 0.523, 0.604, and 0.184 for Vata, Pitta, and Kapha, respectively. Likewise, the authors found intra-class correlation (ICC) values for questionnaire 1 as 0.69 for Vata dosha alone and 0.69 for the proportion of Pitta among Pitta and Kapha dosha combined P/(P + K); for questionnaire 2 as 0.86 for Vata and 0.81 for Pitta; and for Questionnaire 3 as 0.89 for Vata and 0.76 for Pitta, over the time span of one month. The three questionnaires are available through open access.

Shirolkar et al. [47] used a fifty-eight-item, interview-styled questionnaire for Prakriti evaluation. The instrument assesses study participants' physical, physiological, and psychological characteristics. The authors did not specify the method of scoring. While the questionnaire underwent validation through pre-testing and received confirmation from independent clinical assessments of Prakriti by two Ayurveda physicians, they did not clarify the process by which the assessments of these two experts were merged to create the ultimate Prakriti evaluation. The substantial agreement (exceeding 90 %) observed between the clinicians' assessments and the questionnaire demonstrates good construct (convergent) validity in measuring Prakriti. However, the authors did not provide additional information about the sample size used during the questionnaire's validation process. The instrument is available through open access.

Prasher et al. [48] used an interview-based questionnaire comprising seventy-three items spread over five sections – anatomical, physiological, physical, endurance, and psychological. Each section contains subsections that contain items and their corresponding responses pertaining to Vata (V), Pitta (P), and Kapha (K) dosha. The questionnaire asked respondents about their ethnicity, prior medical history, and anthropometric characteristics. Each question has multiple answer options, which refer to a V, P, or K property. The authors used the framework of Charaka and Sushruta Samhita to formulate the items of the questionnaire. The authors used a scoring of “0” and “1” to indicate the absence and presence of a dosha-specific property in an individual. The cumulative scoring method has been used in calculating an individual's V, P, and K scores. The reliability of the questionnaire has been established, and a concordance value of 80 % was found between the Prakriti assessments of two independent Ayurveda physicians using the questionnaire, thus suggesting inter-rater reliability. However, it hasn't been explained how the evaluations from two different Ayurveda physicians were combined into one Prakriti assessment. The questionnaire is available through open access.

Rotti et al. [8] used an eighty-five-item, interview-styled questionnaire embedded in software named, AyuSoft. Originally developed by Centre for Development of Advanced Computing (C-DAC Pune, India), the questionnaire comprises items pertaining to an individual's physical, physiological, and psychological characteristics. Based on Charak and Sushruta Samhita [4,5], and Ashtanga Hridaya [12], the instrument uses weightage configuration and assumes that the physical characteristics of an individual remain stable throughout one's life relative to their physiological and psychological characteristics, thus, assigning greater weightage to the physical characteristics than the physiological and psychological ones. This way, the dosha dominance is determined in cumulative percentage format in each physical, physiological, and psychological domain. Using the validation sample of 3416 healthy subjects, the Kappa coefficient was calculated as the instrument's reliability measure (inter-rater). A concordance value of 0.778 (p < 0.001) was found between the AyuSoft and the Ayurveda Physician. It is sold as proprietary software by C-DAC Pune.

Sivapuram et al. [49] used a 30-item, physician-administered questionnaire based on the three traditional methods of diagnosis to evaluate Prakriti – physical inspection (darshana), touch-based (sparshana), and history taking (prashna). Each method was assigned ten items with three choices corresponding to three doshas - Vata, Pitta, and Kapha. Final scoring was calculated using cumulative scores for each dosha from the three methods of diagnosis (pareeksha). This led to the identification of six distinct constitutional types: Vata-Pitta, Vata-Kapha, Pitta-Kapha, Pitta-Vata, Kapha-Vata, and Kapha-Pitta. The total scores for Vata, Pitta, and Kapha were derived from inspection, palpation, and history taking. Participants were classified into single dominant dosha (Vata, Pitta, or Kapha) constitutional type if a cumulative score of one dosha exceeds the score of 20 resulting from the three methods of diagnosis – inspection, palpation, and history taking. If the total score for any dosha did not exceed 20, the individual was classified as belonging to one of the combined personalities based on the order of the scores in the Vata, Pitta, and Kapha domains. No psychometric properties were established for the questionnaire, though it is available as an open access.

Likewise, Bell et al. [50] created a self-assessment questionnaire comprising ten items, each offering three alternatives corresponding to Vata, Pitta, and Kapha dosha characteristics. The method for scoring and determining Prakriti is not elucidated, and no efforts were undertaken to establish the questionnaire's psychometric properties. Nevertheless, this instrument is accessible to the public.

Meenakshi et al. [51] used a thirty-four-item, self-reported questionnaire. The questionnaire used three alternatives corresponding to three doshas, Vata, Pitta, and Kapha, against each item. The questionnaire asked the respondents to select the alternative that seemed to present over a longer period in the study participant. The study neither mentions the scoring pattern of the instrument nor the psychometric properties, though the instrument is available as supplementary material.

Joshi [52] developed a 28-item physician-administered questionnaire. Though the author claimed to establish the statistical validity of the questionnaire, neither the scoring pattern nor the validation process has been described succinctly by the authors. The authors provided the items of the questionnaires, which are available in the appendix of the paper.

Tripathi, Patwardhan, and Singh [53] modified a sixty-two-item self-assessment questionnaire originally developed by Patwardhan and Sharma [54]. The items are based on the characteristic features produced by the attributes (Gunas) of the three dosa, Vata, Pitta, and Kapha, and asked responses in a dichotomous style to indicate agreement or disagreement with the given item. According to Charaka Samhita, each dosha has specific attributes (Gunas). Vata has eight, Pitta has five, and Kapha has twelve attributes (Gunas). To measure a specific attribute (Guna), multiple items may be assigned. Each item is assigned a weight depending upon the number of items measuring the corresponding attribute (Guna) and the weight assigned to the attribute. The dosha dominance was calculated in terms of the percentage scores by dividing the total scores obtained by a participant for a dosha with the total scores allocated to that dosha and multiplying the divided score with 100 to get percentage scores. The authors established the concurrent validity of the questionnaire by determining the Pearson correlation between the scores obtained on their self-assessment questionnaire and the previously developed self-reported Prakriti assessment questionnaire [54] among a sample of fifty Ayurveda students. No published material is available to establish the reliability and validity of the previously developed instrument. Though the authors found correlation values of 0.015, 0.097, and 0.031 for the three dosha Prakriti types – Vata, Pitta, and Kapha, respectively, significance was achieved only in Vata and Kapha correlation values. The instrument is available through open access [53] and evaluates extreme and dual Prakriti types [53,55].

Shilpa and Murthy [56] developed a 157-item questionnaire to assess the psychological aspects of Tridosha-based Prakriti. The authors did not specify the scoring pattern used in the instrument. The questionnaire's psychometric properties were established using the measures of content validity, Cronbach's alpha, split-half reliability, and test-retest reliability among a sample of one thousand five hundred forty-eight healthy individuals. While content validity of the questionnaire was established by reviewing the scale with eighteen Ayurveda Vaidya, the split-half reliability analysis was established by Cronbach's alpha value of 0.976, and the test-retest reliability was established by calculating the values of Pearson correlation as 0.693 for Vata related items, 0.802 for Pitta related items, 0.718 for Kapha related items at p = 0.01 as the level of significance. The instrument is proprietary and is available to use under commercial terms.²

Rastogi [57] developed a fifty-four-item, interviewer-administered questionnaire to assess the Prakriti of healthy individuals based on the attributes (Gunas) of three doshas. The author assigned proportional weightage to each attribute (Guna) and their features in three doshas, and each dosha was assigned a total score of 960, which was then distributed among the number of classical attributes (Gunas) available to each dosha. Each Guna in a specific dosha category had an equal score. An individual Guna score in a dosha category was further subdivided based on the available features against each Guna. The author used the measures of inter-rater reliability and content and construct validity to standardize the questionnaire on a sample of twenty-six healthy Ayurveda students (consisting of sixteen males and ten females). The correlation coefficient was calculated between the rating scores of the Vata, Pitta, and Kapha subscales in the questionnaire by two independent Ayurveda practitioners. It was found to be 0.80, 0.52, and 0.40 for the three subscales. However, only the Vata subscale showed a strong significant correlation (p < 0.001), followed by the Pitta subscale (p < 0.01), and the Kapha subscale showed the least significance in the correlation (p < 0.02). The agreement between Ayurveda Physicians and the method used to merge their assessments into a single evaluation has not been explained. Likewise, the content validity was calculated after consulting the questionnaire items with six independent Ayurveda experts. Each expert was asked to rate the applicability of individual items of the questionnaire against the four levels of inference – strongly applicable, applicable, not applicable, and strictly not applicable. Only the responses against the first two levels – ‘strongly applicable’, and ‘applicable’, were used for the analysis of content validity. Construct validity was established by having the same Ayurveda experts examine the questionnaire items through traditional methods of questioning and inspection. However, the outcomes of this analysis have not been documented or disclosed. Feature expressions of Madhur Guna in the Kapha subscale and Katu and Amla in the Pitta subscale pose difficulty in objective assessment by the experts; hence, they were excluded from the questionnaire. The instrument is available through open access.

Kurande et al. [6] used three different methods – (a) a seventy-five item, self-reported ABC questionnaire; (b) an interview administered AyuSoft; and (c) pulse diagnosis by fifteen registered Ayurvedic practitioners to assess the Prakriti of individuals. The authors have not described the scoring method to evaluate the Prakriti using the ABC self-reported questionnaire. While the self-reported questionnaire was based on Charaka and Sushruta Samhita, AyuSoft was based on Charaka, Sushruta, and Ashtanga Hridaya, as discussed earlier. Though the authors found a fair amount of inter-rater reliability (Kappa value of 0.336) between the self-reported ABC questionnaire and AyuSoft assessments and a moderate amount of inter-rater reliability between physician and software (kappa value of 0.497), the reliability and validity measures of the ABC questionnaire developed by the authors were not established. The ABC questionnaire is open source.

Suchitra, Jagan, and Nagendra [58] developed a one-hundred-and-thirty-five-item Ayurveda Child Personality self-reported inventory to assess the Prakriti of children aged 6–12. The inventory used a dichotomous style of scoring as 0 and 1, indicating the absence and presence of a characteristic feature, respectively. The authors used Cronbach's alpha and split half to establish the reliability of the questionnaire, principal component analysis to establish construct validity, and review by ten Ayurveda experts to establish content validity among a sample of two hundred and thirty subjects. While Cronbach's alpha values of 0.77, 055, and 0.84 were found for the Vata, Pitta, and Kapha scales, split-half reliability values for the three scales were found to be 0.65, 0.34, and 0.84, respectively. Additionally, each item in the Vata, Pitta, and Kapha scales showed a factor analytic coefficient exceeding 0.5, with a range between 0.51 and 0.79. The instrument is available through open access.

Delle Fave et al. [59] developed a twenty and fifteen-item self-reported questionnaire on dosa Prakriti. The twenty-item instrument has nineteen items that require responses against a five-point scale, and the last 20th item requires responding to one among the three alternatives. Eleven of the twenty items assess structural and physiological characteristics, five assess psychological characteristics, and three assess the individual's behavioural characteristics. The minimum value, 1, on each scale, describes Vata Prakriti qualities, while the greatest value, 5, describes Kapha features. The values 2 and 4 indicate moderate Vata/Pitta and Pitta/Kapha traits, respectively. In addition, Item 20 has three brief explanations of psychological traits, each of which is supposed to be distinctive of a particular Prakriti type. Participants were instructed to select the item value (or item 20's description) most accurately represented them. For scoring, five new variables were created for each participant based on the frequency of each value across the 19 five-point items: Vata (value = 1 frequency across the 19 items), intermediate Vata/Pitta (value = 2 frequency), Pitta (value = 3 frequency), intermediate Pitta/Kapha (value = 4 frequency), and Kapha (value = 5 frequency). Based on the frequency pattern of the single dosa, the variable "Prakriti" was created (or intermediate value): If one answer outnumbered the others (frequency≥9), the person was categorized as a particular Prakriti type. If many values exhibited comparable frequency across items, making it difficult to develop a consistent profile, the subject was eliminated from further analysis. No attempt was made to standardize the questionnaire by the authors. Additionally, no information about the scoring and validation of the fifteen-item questionnaire, a shorter version of the twenty-item one, was available in the paper. The authors provided only the twenty-item questionnaire in the appendix of the paper.

Edwards and Streiner [60] devised a thirty-nine-item, self-reported questionnaire to evaluate the Prakriti of healthy individuals. The authors divided the questionnaires into three subscales – body-related items, mind-related items, and digestion-related items – each scale having thirteen items belonging to three doshas - Vata, Pitta, and Kapha. They evaluated the test-retest reliability using intra-class coefficients and internal consistency using Cronbach's alpha among the group of seventy-six and two hundred sixty-seven individuals, respectively. The values of Cronbach's alpha were calculated for Vata, Pitta, and Kapha items as 0.70, 0.58, and 0.50, respectively. Likewise, intra-class coefficients (ICC) for the absolute agreement were calculated for Vata, Pitta, and Kapha items in each of the three subscales as 0.78, 0.78, and 0.79 for the Mind subscale; 0.62, 0.92, and 0.82 for the Body subscale; and 0.82, 0.63, 0.77 for digestion subscale. The questionnaire is available through open source.

To sum up, the analysis encompassed several studies that collectively employed nineteen different questionnaires for the assessment of Prakriti. Among these questionnaires, six followed interview-style formats, eleven were based on self-reporting, and two utilized a combination of both approaches. Regarding the types of Prakriti being evaluated, three questionnaires focused on extreme, dual, and balanced Prakriti types, another three targeted extreme and dual types, eleven exclusively addressed extreme types, one solely assessed the dual type, and one questionnaire did not specify the type of Prakriti under consideration.

In terms of psychometric properties, six questionnaires exclusively established reliability indices, three focused solely on establishing validity indices, three addressed both reliability and validity indices, and seven questionnaires did not establish any reliability or validity indices.

Regarding the instructions provided to respondents for identifying Prakriti features based on lifespan, only eight questionnaires explicitly requested such information, while the remaining eleven questionnaires did not include any instructions related to identifying Prakriti features based on lifespan.

Furthermore, the availability of these questionnaires varied, with sixteen being open source, two being proprietary, and one being unavailable. The reference to these questionnaires across research studies ranged from a minimum of two to a maximum of two hundred ten instances.

In the realm of established and rigorously validated assessment instruments, the Mysore Psychological Tridosha Scale [56], as well as the Ayurveda Child Prakriti Inventory [58], has emerged as the foremost reliable and validated questionnaires for assessing Prakriti, followed closely by the Prototype Prakriti Analysis Tool (PPAT) [57]. The remaining questionnaires, however, have primarily demonstrated either reliability or validity but not both.

To ensure Prakriti evaluation questionnaires are of high quality, it is advisable to adhere to the following guidelines: (a) explicitly reference well-established theoretical frameworks such as Charaka Samhita, Suhruta Samhita, Ashtanga Hirdaya, etc., upon which the questionnaires are grounded, (b) outline detailed validation procedures, providing justification for the selection of specific psychometric properties for validation, as well as for any omission of validation prior to usage, (c) clearly specify whether the questionnaire is interview-based or self-reported, (d) provide explicit instructions for identifying Prakriti features based on lifespan, (e) describe the scoring methodology and the weighting assigned to different Prakriti features, (f) elucidate the criteria employed to classify individuals into extreme, dual, or balanced Prakriti types, (g) make the anonymized data accessible to facilitate replication of study findings.

3.1.2. Algorithm-based Prakriti assessment instruments

The five studies reviewed by us are listed and tabulated in Table 3, Table 4 The studies employed machine learning techniques to evaluate Prakriti classification, demonstrating initial results in terms of reliability, accuracy, and classification performance.

Farooque et al. [61] utilized seven features using facial images to identify extreme Prakriti types, showing good reliability and accuracy with the decision tree algorithm. Tiwari et al. [62] used machine learning to predict extreme Prakriti types (Vata, Pitta, and Kapha) based on questionnaire data, revealing distinct clusters for each type and achieving high accuracy with supervised algorithms. Gadre et al. [63] used visual facial features and achieved moderate to good accuracy. Madaan and Goyal [64] developed a machine-learning model based on a questionnaire, while Manisha, Umadevi, and Akshitha [65] analysed tongue images. These pioneering studies demonstrate the possibility of Prakriti assessment using machine learning techniques.

In a study by Farooque et al. [61], the Prakriti (constitution) of sixty-seven healthy individuals was evaluated using a questionnaire and physical observation based on 37 parameters conducted by an experienced Ayurveda physician using the same questionnaire. The authors aimed to classify the subjects into three extreme Prakriti types using various machine learning algorithms: the naive Bayes classifier, logistic regression, decision tree, and artificial neural network (ANN). However, validation was not performed, and the reported performance was solely based on the training data, limiting its usefulness as the code data was unavailable. Notably, the decision tree classifier relied on only seven features from the original 37 for the Prakriti classification. This study highlights the potential application of data mining and artificial intelligence in Ayurveda, demonstrating its capability to reduce the questionnaire's size.

Using unsupervised and supervised machine learning algorithms, Tiwari et al. [62] conducted a research study to predict extreme Prakriti types (Vata, Pitta, and Kapha). The study consisted of the VADU cohort with 528 individuals and a North India cohort with 850 volunteers. Qualified Ayurveda physicians performed data labelling using a questionnaire [48] comprising seventy-three items. For the unsupervised learning approach, clustering analysis was employed, generating three distinct clusters corresponding to the three extreme Prakriti types. Additionally, the authors applied three supervised machine learning algorithms, namely LASSO, Elastic Net, and Random Forest, to identify Prakriti types using questionnaire data. These models demonstrated high sensitivity and specificity for each Prakriti type during tenfold cross-validation, and when tested on a different cohort, they achieved very high accuracy. For the Kapha Prakriti, LASSO demonstrated a sensitivity of 93.1 % and a specificity of 100 %. The Elastic Net model exhibited a sensitivity of 96.55 % and a specificity of 100 %, while the Random Forests model had a sensitivity of 100 % and a specificity of 98.51 %. Moving to the Pitta Prakriti, LASSO achieved a sensitivity of 82.75 % and a specificity of 94.02 %, Elastic Net had a sensitivity of 86.2 % and a specificity of 97.01 %, and Random Forests showed a sensitivity of 79.31 % and a specificity of 98.51 %. Similarly, for Vata Prakriti, LASSO showed a sensitivity of 94.73 % and specificity of 91.37 %, Elastic net exhibited a sensitivity of 97.36 % and specificity of 93.1 %, and Random Forests showed a sensitivity of 97.37 % and specificity of 91.38 %. These results provide insights into the performance of the models in distinguishing different Prakriti types in North India data. It is worth noting that the high accuracy may be influenced by the fact that Ayurveda physicians used the same questionnaire to assign ground truth labels, potentially leading to target leakage and an overestimation of model performance. Similarly, if coded, a simple scoring method employed by Ayurveda physicians may have also yielded comparable results. Nonetheless, the emergence of three natural clusters corresponding to the three Prakriti types is a fascinating finding that warrants further exploration. Upon request, the dataset is available for further investigation,³ which is an important step in ensuring study replication. However, the study did not specify the availability of code or licensing for its use.

Gadre [63] conducted a study analysing the facial photographs of 330 individuals to determine their Prakriti using statistical models and machine learning algorithms. The data labelling method involved an Ayurveda physician assigning Prakriti types to the photographs. Various methods such as vision and deep learning, support vector machines, random forests, and neural networks were utilized to extract and analyse skin, hair, eyes, lips, nose, and facial shape features. The authors reported the confusion matrix of the classifiers used in the study and detection accuracies. The results showed detection accuracies of 0.69 in the Kapha-pitta class, 0.43 in the Vata Kapha class, 0.30 in the Vata-pitta class, 0.09 in the single dosha dominant Kapha prakriti category, 0.12 in the Pitta Prakriti category, and 0.00 each in the Vata Prakriti and Vata-Pitta-Kapha category. However, the study did not provide public access to the data and code, limiting replication and further exploration of the findings. Additionally, no validation method has been described by the authors, thus resulting in a non-reproducible methodology. Access to data and code is crucial for transparency, reproducibility, and scientific progress, emphasizing the importance of promoting open access in research.

Madaan and Goyal [64] developed a machine-learning model for the Prakriti assessment based on 28 characteristic features derived from a questionnaire developed by them. The study involved a sample size of 405 participants, and a questionnaire (validated by two Ayurveda experts) did the Prakriti assessment. Although the method to label the data is not described at all, it is likely that a scoring of the questionnaire was used to arrive at seven prakriti classes. The authors employed several machine learning algorithms to predict the Prakriti using the Questionnaire's responses to arrive at the ground truth labelling leading to potential target leakage. However, it is crucial to note that both the questionnaire used in the study and the code and data employed for the models were unavailable, limiting the ability to replicate and further investigate the findings. Transparency and reproducibility are essential elements of scientific research, and the availability of data and code allows for validation, refinement, and extension of the study's results. Moreover, since the Prakriti type can be determined by a simple scoring of the Questionnaire's responses, the rationale for using Machine Learning Models in the Prakriti assessment remains unclear.

Joshi, Umadevi, and Akshitha [65] conducted a study using tongue images from 107 healthy adult subjects to identify Prakriti types. The Ayurveda physician assigned labels to these images captured using mobile phones based exclusively on the tongue image. The study focused on extracting characteristic features of the tongue, including texture, cracks, pimples, and shape, which Ayurveda physicians use to identify Prakriti. The tongue images were segmented using the grab cut algorithm, an image segmentation algorithm based on computer vision. Classification algorithms such as the k-nearest neighbours (KNN) algorithm, neural network, and decision tree classifier were employed to classify the images. The dataset was divided into training and testing sets with an 80:20 proportion. Various features of the tongue images, such as colour, shape, coating, and pimples/buds/cracks, were extracted using different algorithms. The decision tree classifier outperformed other classifiers, achieving a sensitivity of 71.43 %, specificity of 95.00 %, precision of 83.33 %, and negative predictive value of 90.47 %. Overall, the study demonstrated the potential of tongue images and classification algorithms in assessing Prakriti types. However, the study did not specify the data and code availability, limiting the findings’ replicability. Further research and validation are necessary to establish the robustness and generalizability of these results.

In summary, the studies examined in this section employed machine learning algorithms and statistical models to evaluate Prakriti classification. These methods showed promise in determining Prakriti types based on (a) questionnaire responses, (b) facial characteristics, and (c) tongue images. Future research could consider including (d) body shape, (e) height, and weight data and possibly combine all these methods to assess Prakriti more comprehensively. However, certain studies encountered challenges related to target leakage, replicability, suitable ground truth labelling, weak methodologies, and unavailability of data and questionnaires, underscoring the need for transparency and accessibility in future research. The issue of data and code availability was also identified. Overall, these investigations contribute to the growing body of knowledge regarding the use of machine learning techniques for Prakriti assessment, offering valuable insights for further exploration in this field. Additionally, the field would greatly benefit from the creation of a comprehensive data repository with corresponding labels.

3.1.3. Devices and equipment-based Prakriti assessment instruments

One of the conventional methods used by Ayurveda Physicians to evaluate Prakriti is pulse examination (other being visual inspection, interrogation, etc.). The index, middle, and ring fingers representing Vata, Pitta, and Kapha are placed on the radial tubercle proximal to the wrist crease to assess pulses. The physician exerts suitable pressure on the fingers and gradually releases them to perceive pulsations. This method serves as a means of determining Prakriti. Lad [1] identified the characteristic features of Vata, Pitta, and Kapha pulses in healthy individuals, including pulse rate (vega), rhythm (tala), force (bala), tension and volume (aakruti), temperature (tapmana), and arterial wall hardness (kathinya). Notably, the Vata pulse exhibits cobra-like movements (sarpa gati), the Pitta pulse resembles frog-like movements (manduka gati), and the Kapha pulse resembles swan-like movements (Hansa gati). Different devices have been developed for Prakriti diagnosis based on this elucidated principle. However, our review includes only those devices cited in scientific studies.

Chaudhari and Mudhalwadkar [66] present a system designed for Nadi Pariksha, which involves the utilization of three photoplethysmography (PPG) sensors. The system enables the continuous measurement of the radial pulse at three specific points using these sensors. The collected signals are then filtered and subjected to analysis to extract important features like pulse rate, frequency, and amplitude. While the authors did collect preliminary data, they did not explain the procedure for Prakriti identification, nor did they make the system available for use.

Goyal and Agarwal [67] presented a study on designing and developing a pulse diagnosis system per Ayurveda principles. The system utilized a single piezoresistive sensor (MPXM2053D) to capture wrist pulse signals from 42 subjects. Components of the system included the sensor itself, an instrumentation amplifier, and a data acquisition card (DAQ), which were placed on the radial artery at the wrist. The study calculated the energy distribution within different frequency bands (0–30 Hz) and observed that the energy in the 3–12 Hz band played a significant role in distinguishing between healthy and unhealthy subjects. Several classification algorithms, such as K-nearest neighbour (KNN), support vector machine (SVM), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and decision tree (DT), were employed to classify subjects into healthy or unhealthy categories. Notably, the wrist pulse acquisition system accurately classified healthy and unhealthy subjects, with the KNN classifier achieving the highest accuracy of 96.29 %. Although the study did not specifically aim to measure Prakriti, it demonstrated the potential of using radial pulse for diagnostic purposes, following the traditional practices of Ayurveda physicians. However, it is important to note that the device used in the study is currently unavailable for further analysis and replication.

Khandai and Jain [68] attempted to compare the piezoresistive and piezoelectric sensors to acquire radial pulse data. However, the authors gathered initial data without evaluating the study participants' Prakriti. Furthermore, they did not conduct substantial analysis, and the device is unavailable.

Kadarmandalgi and Asaithambi [69] employed three Photoplethysmography (PPG) sensors interfaced with Arduino to capture wrist pulse signals. Three parameters were derived from the pulse sensor data: pulse wave velocity, augmentation index, and reflectivity index. The authors collected data sets from 100 healthy and 100 unhealthy subjects. They claimed that their system successfully diagnosed patients exhibiting symptoms related to dosha imbalances, comparing the results with Ayurveda-based theoretical outcomes. However, they didn't try to determine the participants' Prakriti. The study introduced the application of pulse wave velocity and reflectivity index in pulse acquisition, which presents intriguing parameters warranting further investigation. However, it should be noted that both the device and the data used in the study are not available for access, limiting the opportunity for additional analysis and replication.

Joshi and colleagues [70] introduced Nadi Tarangini, which utilized three piezoresistive sensors to capture pulse waveforms. The authors collected data from a limited number of subjects and conducted preliminary analysis; however, they did not provide a detailed explanation of the scoring or validation procedure [71]. Although the device is commercially available, it is proprietary, meaning specific details about its accessibility were not disclosed.

Sareen and colleagues [72] introduced the Nadi Yantra system, which utilizes piezoelectric pressure sensors to capture signals from the radial artery for pulse diagnosis. The system's mechanical design involves three finger-like projections that can be adjusted in the tip region to identify optimal signal-capturing locations. Springs integrated into the design contribute to damping and mimic the natural damping caused by muscles in the practitioner's fingers. Once the three best positions are determined, they are securely locked using an additional robust spring mechanism that minimizes motion artifacts. Incremental adjustments in pressure can be achieved by repositioning the lock towards the slant side. The study collected pre- and post-lunch data from five subjects and presented some heart rate variability (HRV) analysis; however, no specific algorithm for Prakriti was described, and the device itself is not currently available. This development represents a positive initial step in capturing the Nadi pulse, but further work is needed to advance this field of research.

Selvam and Begum [73] developed the Nadi Aridhal system, which uses automated methods to diagnose diseases based on Ayurvedic principles. By analyzing pulse waveforms, the system identifies dominant doshas (Vata, Pitta, and Kapha) and provides potential disease diagnoses non-invasively. However, specific numerical results from tests on three individuals were not mentioned. Thus, it poses a challenge to repeat the study for further research.

Kalange et al. [74] developed the Nadi Parikshan Yantra, which utilized three pressure sensors to collect radial pulse data on one hundred and twenty healthy subjects. Although the precise methodology for subject labelling was not extensively explained, the study reported a 62 % correlation between the electrical pulse data and diagnoses made by Nadi Vaidya. Frequency analysis revealed the presence of three distinct and prominent harmonic frequencies at 1.36 Hz, 2.750 Hz, and 4.078 Hz. Time domain analysis identified specific parameters like P2/PI as differentiators between Vata, Pitta, and Kapha doshas. Statistical analysis employing t-tests demonstrated significant differences (p < 0.05) among Vata-Pitta, Pitta-Kapha, and Vata-Kapha in terms of the measured parameters. The study showcased the potential of utilizing pulse waveform characteristics to identify the dominant dosha objectively. However, the study lacked extensive details on subject labeling and methodologies employed, although it successfully highlighted the feasibility of utilizing pulse waveform analysis for dosha identification.

Bhawankar, Dharmik, and Telrandhe [75] developed an IoT-based pulse examination system utilizing Nadi Pariksha for disease detection and prediction. This system involves registering patients, attaching wrist pulse sensors, securely transferring data to the cloud for pre-processing, and applying a classification algorithm, aiming to assist doctors in real-time disease prediction and diagnosis while enhancing the potential for improved healthcare outcomes.

Joshi and Bajaj [76] developed a Prakriti diagnosis system based on wrist pulse analysis, utilizing hardware and an Artificial Neural Network (ANN). The system achieved satisfactory sensitivity values of 85 % for Vata and Kapha dosha and 90 % for Pitta dosha in classifying individuals into Prakriti types using data from two hundred subjects.

Rao and colleagues [77] conducted an analysis of the Prakriti of fifty healthy individuals using a wrist pulse acquisition system equipped with three piezoelectric sensors. They compared the Prakriti classification based on pulse patterns with classifications performed by an expert Nadi Pariksha Ayurveda physician and a questionnaire separately. The focus of the study was to develop an automated instrument for non-invasive diagnosis based on Ayurvedic wrist pulse analysis. The aim was to digitize the three pulse patterns representing Vata, Pitta, and Kapha and classify the individual's body nature (Prakriti) into different categories. The study revealed substantial to moderate agreement between the pulse-based classifications conducted by the expert Ayurvedic physician and the developed instrument. Cohen's kappa values were reported as 0.719 for basic-level classifications and 0.454 for sub-level classifications. Furthermore, the instrument-based classification demonstrated moderate agreement with questionnaire-based Prakriti identification, yielding a kappa value 0.476. However, there was a poor correlation between the classifications based on the questionnaire and pulse readings. Overall, the developed instrument showed promise in assessing Prakriti and had the potential to assist in the early detection of dosha imbalances. This study stands out as the only one validating pulse-based Prakriti assessment against other measures. The authors expressed their willingness to support the replication of the study, which further highlights their commitment to advancing research in this area.

In summary, the cited studies present diverse approaches to Prakriti assessment, incorporating technologies such as photoplethysmography and pulse diagnosis aligned with Ayurveda principles. However, methodological transparency, device accessibility, and replicability pose limitations across these studies. While some demonstrate potential for dosha identification through pulse waveform characteristics, such as Kadarmandalgi and Asaithambi [69] and Kalange et al. [74], others like Goyal and Agarwal [67] and Khandai and Jain [68] lack comprehensive analysis and data availability. Joshi and colleagues [70] introduce Nadi Tarangini, albeit with limited data and proprietary devices. Sareen and colleagues [72] present Nadi Yantra's mechanical advancements but fall short in specific Prakriti algorithms and device availability. Rao and colleagues [77] offer promising automated pulse-based Prakriti assessment but underscore the need for further correlation. Overall, while these studies provide insights, addressing methodological, replicability, and operationalization challenges is crucial for advancing standardized and validated Prakriti assessment methodologies.

4. Discussion

As described in the Ayurveda textbooks, the assessment methods for determining Prakriti have exhibited a predominantly subjective nature. While modern approaches offer the potential for objectivity, they necessitate further research and the establishment of robust validation processes. A significant challenge in this pursuit lies in the absence of a reliable means to establish the ground truth, i.e., an individual's Prakriti. Some questionnaires have attempted to address this by incorporating Ayurveda Physician diagnoses and comparing them with questionnaire results to establish validity. However, it is noteworthy that studies have reported low Inter-rater reliability among Ayurveda physicians, typically within the range of 0.20–0.40 [6,7,78]. Prakriti, based upon three humours – Vata, Pitta, and Kapha – corresponds to ten types of personality - six dual dosha prakriti types, three single dosha, and a single balanced Prakriti type.

The initial step towards achieving credible and objective Prakriti assessment entails the meticulous investigation, standardization, and evaluation of subjective methods for Prakriti assessment. Once a suitable protocol for ground truth determination has been established, objective metrics for Prakriti assessment can be validated using this ground truth as a reference point. Subsequent assessments, including construct validity and other forms of validity, such as test-retest reliability and inter-rater reliability, can be conducted. Furthermore, this ground truth Protocol may be employed to curate a comprehensive database of various features (e.g., tongue images, PPG, ECG, pulse pressure), facilitating Prakriti assessment through Machine Learning methods based on diverse features. The ground truth serves as a critical validation tool for devices employed in the Prakriti assessment.

In the realm of dosha-based Prakriti assessment via questionnaires, a range of methodologies exists, accompanied by shared limitations and opportunities for enhancement. A notable concern is the insufficient construct validation, particularly in Prakriti measurement. The absence of a clear and universally accepted definition for Prakriti hampers standardization efforts [16]. Many questionnaires lack adequate validation, and addressing discrepancies in inter-rater reliability among Ayurveda experts is of paramount importance. To illustrate, several studies [[52], [53], [54],57] engaged multiple Ayurveda physicians for Prakriti assessment, but the mechanisms for resolving disagreements and the precise procedure for establishing the ground truth in Prakriti identification were not evidently delineated. In other words, there is a lack of clarity regarding the comprehensive methodology employed to establish Prakriti identification definitively. Additionally, the availability of datasets will help future studies to operationalize the Prakriti evaluation robustly. It is advisable to incorporate clear references to Ayurveda textbooks, establish criteria for Prakriti categorization, and undergo psychometric validation for self-administered and interview questionnaires [5,17]. Addressing these gaps mandates fundamental research aimed at devising a universally accepted and reliable approach to Prakriti assessment.

Similarly, the potential of machine learning and data mining in Prakriti assessment is promising, albeit accompanied by specific limitations. The absence of rigorous validation processes and an excessive reliance on training data may impede the generalizability of such models [79]. It is imperative to institute robust validation methodologies, facilitate data sharing, and exercise caution in the utilization of questionnaires [80]. Clear justification for complex models is essential, and innovative approaches, such as the utilization of facial and tongue images, should address data access constraints for validation purposes. Open access resources such as large, labelled datasets for Prakriti assessments will play a pivotal role in establishing credible Prakriti assessment methodologies. Furthermore, the accessibility of code and datasets via electronic repositories such as Kaggle [81], Github [82], and ensuing coding-oriented competitions aimed at establishing the ground truth will serve to enhance the robustness of the Prakriti identification methodology. Likewise, establishing correlation measures between the Prakriti measure and Western personality measures, such as the Big Five [59] and Sheldon's body types [83], may provide cross-disciplinary insights into the measurement of Prakriti in an objective, reproducible, and rigorous manner [84].

Exploration into pulse-based Prakriti assessment devices has yielded valuable insights, yet substantial limitations persist. Methodological inadequacies raise concerns regarding reliability, while limited data access poses challenges to reproducibility. To advance Prakriti assessment in this domain, transparency, accessibility of data and hardware design, and adherence to standardized procedures are imperative. Further research is warranted to establish robust methodologies for pulse waveform analysis and to validate them against established measures. Non-invasive pulse acquisition systems hold promise but necessitate rigorous validation and resource sharing to enhance Prakriti assessment methodologies.

4.1. Future work

Building on these observations, future research should prioritize replicable, rigorously validated methods of Prakriti assessment. Efforts to standardize questionnaires and algorithms—alongside meticulous psychometric and clinical validation—are essential for widespread acceptance. Data sharing from well-designed validation studies would further accelerate progress, while establishing standards for assessment devices will lay the groundwork for their incorporation in clinical practice. Investigations into potential alignments between the Tridosha-based Prakriti concept and contemporary psychological traits (e.g., the Big Five, Hippocratic temperament types) could foster a more nuanced and empirically grounded understanding of Prakriti. Such an integrative approach may strengthen the theoretical underpinnings of Prakriti and promote greater cross-cultural relevance.

5. Conclusion

Prakriti assessment remains a complex task requiring thorough validation and standardization. While questionnaires currently offer the most accessible option with Ayusoft being the most rigorously validated, followed by the open-source Prototype Prakriti Assessment Tool (PPAT), their subjective nature highlights the need for rigorous psychometric studies. Machine learning-based tools offer a promising yet complex alternative, facing challenges in accurately identifying authentic Prakriti types and overcoming validation and replicability challenges. Devices, despite offering objective assessments, are constrained by design limitations and a lack of established validation protocols. Properly established protocols, reliable devices, open-source resources, and cross-disciplinary collaborations will be key to refining how Prakriti is measured and understood.

Author contributions

AG: Methodology, Validation, Formal analysis, Investigation, Writing-Original Draft; VS: Supervision; SC: Supervision; RG: Conceptualization, Writing – Review & Editing, Supervision, Project administration.

Declaration of generative AI in scientific writing

During the preparation of this work, the first author used ChatGPT to improve the readability of the manuscript using grammatically correct words and sentences. The manuscript has been originally written by the author himself. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Funding sources

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article: