Quantify-Interpret-Recommend (QIR)

Description

The Quantify – Interpret – Recommend (QIR) pattern is a grouping of separate, but closely chained, pure cognitive processes. Because of their correlation, they can be modeled as individual Decisions or composed into one complex Decision.

In general, a quantitative decision takes multiple ("sensorial") inputs and aggregates them to generate a quantitative output. This quantitative output is then interpreted into a qualitative observation ("belief") that, in turn, drives (the "intent" of) an agent's action, or a recommendation thereof.

This is a generalization of the "Simple Scoring" pattern.

Problem

Some questions do not yield a finite number of qualitative, categorical answers but, rather, a quantitative, continuous value based on the complex aggregation of multiple diverse inputs that contribute in different roles and capacity, possibly with different weights. To be represented, these complex functions require possibly complex mathematical and/or statistical models, such as regression models or neural networks.

These quantitative predictions or estimates are used as assessments, to evaluate something about the patient that is not directly observable, and/or to better inform decisions regarding actions or interventions.

For example, the process of estimating the body mass index (BMI) of a patient can be modeled as quantitative decision corresponding to the simple question, "What is the patient's body mass index?" The resulting piece of information – a number – is not particularly useful on its own but is used by the following step, a classification cognitive task, whose corresponding decision might be, "What kind of body does the patient have?" The possible answers could be "underweight," "normal," "overweight," or "obese." This classification can be further used (e.g., to inform the choice of the best dietary and/or exercise program). Similarly, The CHA2DS2-VASc score is a quantitative decision that yields a score that is interpreted as the rough estimate of the likelihood of a patient's having a stroke. That, in turn, is used as the basis for deciding whether to prescribe an anticoagulant.

Applicability

This pattern applies when:

·       Observe-Assess-Deliberate cycle, or subsets thereof;

·       Multiple inputs are mapped to quantitative values, which are in turn aggregated by a predictive model;

·       A (qualitative) interpretation allows to assess and/or predict some aspect that is not otherwise directly observable;

·       The interpretation is used to determine the next best course of intervention; and

·       The actual execution – by intervention or communication – is out of scope.

Pattern

Figure 1 – Generic Decision Requirements Diagram illustrating the quantify/interpret pattern

Model the aggregation as a Decision with multiple data Inputs, and a Business Knowledge Model element that defines how the inputs should be aggregated. Notice that a decision table may not be expressive enough, in general, to define this model so other languages (e.g., PMML) could be considered.

The output of this first decision becomes a requirement for (and thus informs) a second-tier assessment/prediction Decision, which has an explicit Business Knowledge Model with mappings from (the value returned by) the quantitative decision to the qualitative classification or categorization. Reference material can be modeled as a Knowledge Source.

Figure 2 – Interpretation of Score Decision Table

Figure 2 illustrates a test score produced by a scoring algorithm using a simple threshold. Its output is the interpretation of the quantitative decision.

Figure 3 – Recommendation decision based on a quantitative sub-decision

Figure 4 – Generic Decision Table

Model the quantitative decision with multiple inputs, and a knowledge model that defines the mathematical formula used to arrive at the outcome based on those inputs. Use the outcome of the quantitative decision as an input into a recommendation decision. Create a knowledge model with mappings from the value returned by the quantitative decision to a list of alternative recommendations.

Figure 5— Quantify Interpret Recommend

The Figure 5 above (Quantify Interpret Recommend) illustrates a recommendation decision based on an interpretation sub-decision, which in turn is based on a quantitative sub-decision.

Figure 6 – Interpretation / Recommendation decision table

Model quantify-interpret pattern as described above. Model a third decision for the recommendation, with the outcome of the interpretation decision as an input.

Specific Example

Figure 7 – Quantify – Interpret: BMI

 The Figure 7 above (Quantify – Interpret: BMI) illustrates that the body mass index is calculated based on height and weight. The resulting numeric output is then used to classify the patient as underweight, normal weight, overweight, or obese.

 Figure 8 – Quantify – Recommend – CHADS2VASc

 The Figure 8 above (Quantify - Recommend – CHADS2VASc) is an example of a scorecard that takes a number of inputs, assigns a number to each (e.g., +1 for hypertension), and sums the numbers to arrive at a score. The score is used as the basis for anticoagulation-therapy recommendations.

Note that there can be an implicit interpretation; that is, each value of the score corresponds to a range of stroke risk. However, the actual numeric score is the input to the recommendation decision.

Figure 9 – Quantify – Interpret – Recommend: Pulmonary Embolism / Diagnostic Testing

 The Figure 9 above (Quantify - Interpret – Recommend: Pulmonary Embolism / Diagnostic Testing) shows the Geneva Score (Revised). The purpose of the Geneva Score (Revised) is to determine the pre-test probability of pulmonary embolism. Its value is used to assign the patient to a low-, moderate-, or high-risk category. If the RGS is < 11, the patient is considered low to moderate risk; if the RGS is 11 or greater, the patient is considered high risk. The patient's risk category is then used to determine the next action. For low- to moderate-risk patients, a blood test (the D-Dimer) is used to further assess the probability of PE; for high-risk patients, an imaging exam is recommended as the next stage in the workup.

Metadata

<blank>

Discussion

 <blank>

Related Patterns

<blank>