Decision Principles in Diagnosis
Chapter Seventy-Eight
Syllabus topic Module 2, "Decision principles in diagnosis"
Pages 276 to 279 of 378
In one line
Four decisions have to be made before a classification is reached: what to ask first, how much each answer counts, what to do when the answers conflict, and what follows from the answer.
In the wording you can write in an examination: the decision principles of a diagnostic scheme are the rules governing the order in which evidence is sought, the weight given to each item, the resolution of conflicting evidence, and the action that follows a classification. Charaka supplies an order of evidence in Vimānasthāna IV, an action rule in Sūtrasthāna I.61, and, in the structure of the attribute lists, an implicit weighting.
The four decisions
Every diagnostic procedure, classical or modern, has to settle these four. Naming them is most of the chapter.
| Decision | The question | Where the classical scheme answers it |
|---|---|---|
| order | what do I look at first? | Vimānasthāna IV: instruction, then observation, then inference |
| weight | how much does each item count? | nowhere explicitly; the lists are unweighted |
| conflict | what if the evidence points two ways? | nowhere explicitly; the scheme permits combinations |
| action | what follows from the answer? | Sūtrasthāna I.61: objects of adverse attributes, with place, measure and time |
Two of the four are answered and two are not, and saying which is which is the difference between an answer that has read the text and one that has read about it.
One: the order of questions
The classical answer is the order of means, not an order of individual questions: know the possibilities first, then observe, then infer. [Parīkṣā: How an Examination Is Structured] sets it out.
What it does not give is which attribute to ask about first. Nothing in the lists says to check dryness before heaviness.
The modern answer to that question is information gain, from [Decision Trees]: ask first whatever most reduces the uncertainty. That is a real addition and it is ours, not the text's, and [A Decision Tree Built From the Tridoṣa Attributes] applies it and reports what it yields.
And there is a second modern consideration the text does not have. Some questions are cheap and some are expensive. A procedure that minimises expected cost rather than expected questions is a different optimisation, and it is the right one when the tests differ in cost.
Two: the weighing of evidence
The lists are unweighted. Seven attributes for wind, and nothing says that dryness counts for more than keenness in classifying.
So the default is one each, which is what [Multi-Attribute Classification] uses and what the program in [Ayurvedic Classification as a Rule-Based Expert System] implements.
But the lists carry an implicit weighting anyway, and it is worth seeing. An attribute unique to one doṣa is decisive when present; an attribute shared by two is not. Uniqueness is a weighting, and it comes from the structure of the lists rather than from any statement.
Decision Principles in Diagnosis
| Attribute | How many doṣa have it | What its presence settles |
|---|---|---|
| dry, light, subtile, unstable, clear | wind only | wind, on one attribute |
| hot, soft, sour, liquid, bitter | bile only | bile, on one attribute |
| heavy, mild, watery, sweet, stable, slimy | phlegm only | phlegm, on one attribute |
| keen | wind and bile | narrows to two |
| cold | wind and phlegm, on this book's reading | narrows to two |
Sixteen of the eighteen attributes are unique to one doṣa. That is a remarkably clean design and it is what makes the scheme usable at all: most single observations settle the question.
Three: conflicting evidence
The scheme's own answer is that combinations are real. A case showing attributes of two doṣa is a case in which two predominate, which is one of the seven labels of [Prakṛti: Constitution as a Class Label]. So conflict is not an error state; it is an answer.
That is unusual and it is worth pausing on. Most classifiers are built to produce one label and treat a near-tie as a difficulty. This scheme treats the combination as a category in its own right, which is the multi-label reading.
What the scheme does not say is what to do when the evidence is thin rather than mixed: two attributes of wind and two of bile is a combination, and one attribute of each is a case where nothing has been established. The distinction is between a mixed answer and no answer, and the program in this book reports a tie rather than choosing, which is a decision the text does not make for us.
Four: the action
Charaka's rule, from Sūtrasthāna I.61:
Those changes (in wind, bile, and phlegm) that are curable may be set right by drugs possessing adverse attributes, administered according to (considerations of) place, measure, and time.
Read as a decision principle it has three parts.
A function from the classification to the response. The response is determined by the attributes matched: their opposites.
Three parameters that modify it. Place, measure and time. So the response is not a constant; it is a function of the classification and of three further variables.
And a precondition. "Those changes that are curable." The rule is stated for the cases it applies to, and the text elsewhere says the cure of incurable cases is not laid down here. A rule that declares its own scope is better drafted than most, and it is the same discipline as a specification stating its preconditions in [Formal Specification: Saying Exactly What a System Must Do].
Decision Principles in Diagnosis
Worked example: the four decisions, on one case
The case. A description reports dry and unstable.
Order. The possibilities are the three doṣa and their combinations, known in advance. The two attributes are observations. No inference has been needed yet.
Weight. Both attributes are unique to wind, so each is decisive on its own. Two of them agreeing is stronger still.
Conflict. None. Nothing points elsewhere.
Action. The opposites: moist, and stable. Applied with regard to place, measure and time.
And now a harder case. The description reports dry and sour. Dry is wind only; sour is bile only.
Conflict. Two decisive attributes pointing to different doṣa. Under the scheme, that is wind and bile both predominating, which is a label. Under a classifier built to emit one answer, it is a tie. The two treatments are different and the scheme's is the better model.
What these principles do NOT amount to
They are not an algorithm. Two of the four decisions are unanswered by the text, so a program implementing the scheme must supply them and must say that it has.
They are not a diagnosis. Classifying a description into the scheme's own categories is not identifying a condition, as [Āyurveda as a Śāstra, and What This Chapter Does Not Claim] says.
And they are not evaluated. No accuracy figure attaches to any of this, for the reason given in that same chapter.
Quick revision
- Four decisions: the order of questions, the weight of each item, the resolution of conflict, and the action that follows.
- Charaka answers two: the order of means, in Vimānasthāna IV, and the action rule, in Sūtrasthāna I.61.
- He does not answer weighting or conflict explicitly, so a program must supply both and declare that it has.
- The lists carry an implicit weighting: sixteen of the eighteen attributes are unique to one doṣa, so most single observations settle the question.
- Conflict is not an error state: a combination is one of the seven labels.
- The action rule has a function, three parameters, place, measure and time, and a declared precondition, curability.
Test yourself
1. Name the four decisions a diagnostic procedure must settle, and say which two Charaka answers.
The order in which evidence is sought, the weight of each item, the resolution of conflicting evidence, and the action following a classification. He answers the order, in Vimānasthāna IV, and the action, in Sūtrasthāna I.61.
2. What implicit weighting do the attribute lists carry?
Sixteen of the eighteen attributes belong to only one doṣa, so their presence settles the classification on its own, while the two shared attributes only narrow it to two. Uniqueness is a weighting even though no weight is stated.
Decision Principles in Diagnosis
3. How does the scheme treat conflicting evidence, and why is that unusual?
As a combination in which two doṣa predominate, which is one of its own seven labels. It is unusual because most classifiers are built to produce a single label and treat a near-tie as a difficulty rather than as an answer.
4. Set out the three parts of Charaka's action rule.
A function from the classification to a response, namely the opposites of the matched attributes; three parameters that modify it, place, measure and time; and a precondition, that the rule is stated for changes that are curable.
The rest of this subject
These notes are cut from the University's printed syllabus. Open the syllabus itself for the same subject.