Simple Scoring Algorithm

Description

The simple scoring algorithm is a specialization of a quantitative decision. It takes multiple inputs, assigns an integer value to each, and sums the results to arrive at a score.

Problem

In medicine there are a large number of simple algorithms used to assess risk, assign probability to various conditions, and predict outcomes. One common form of these algorithms is based on a "scorecard." (See also: http://dmg.org/pmml/v4-3/Scorecard.html predictive models). The models are formulated in terms of a (usually brief) list of relevant clinical findings. Each of these findings is associated with a value (typically a small integer) and the scoring algorithm works through the simple expedient of adding together the values of all of the positive findings and testing the result against a threshold. Because of their pure computational nature, such algorithms can be modeled as DMN decisions.

For example, The Revised Geneva Score is a score used in the diagnostic workup of possible pulmonary embolism (PE). A group of eight findings frequently associated with PE are each assigned an integer weight (see Figure 38 – Patient with Suspected PE). The clinician sums the weights of the positive findings and uses the resulting score for further decision making. If the score is greater than or equal to 11, an imaging exam is indicated. If less, a blood test (the D-dimer) is used to further assess the probability of PE. Figure 38, “Patient with Suspected PE,” shows an algorithm for the workup of suspected pulmonary embolism. The initial testing decision is based on a score produced by adding the values associated with the findings that are positive. Totals under 10 lead to a blood test (the D-dimer); totals of 11 or greater suggest the need for a definitive radiology exam (the computed tomography-pulmonary angiogram (CTPA)). The graphic goes on to describe further diagnostic decision making after the initial testing decision has been made.

Applicability

This pattern applies under the following conditions:

· Pure flow of information that takes multiple inputs and maps each one to a numeric value, based on a defined rule;

· The numeric values are aggregated into a numeric score;

· No "side effects" other than the evaluation of the score; and

· The collection of the inputs, as well as the use of the outputs, are outside the scope of the model.

Pattern

The scoring algorithm is encapsulated within a single Decision node. Model relevant findings are indicated as data inputs to the decision. The scorecard is modeled as a vertical decisions table with a Collect (sum) Hit Policy. Figure 39 shows the simplest form of the DRD. This pattern shows a generic Decision Requirements Diagram displaying the input of a series of simple variables into a scoring algorithm. A scoring algorithm compatible with risk scores (e.g., the RGS) is displayed in Figure 40. In this table, for each row whose rule evaluates to "true," the output value will be added to the sum representing the total score for the table. If available, any reference to the algorithm logic should be modeled as a Knowledge Source attached to the main Decision node.

Figure 2 – the simplest form of the DRD for this pattern

A decision table that uses the Collect hit policy with the sum operator is displayed in Figure 62. It shows a pattern for a table-based scoring algorithm that produces a sum of the numbers in all of the rows where the individual rules come true. This represents a common approach to risk scoring for medical conditions. In this table, for each row whose rule evaluates "true," the output value will be added to the sum representing the total score for the table.

Figure 3 – Table Representation of PE Decision

Specific Example

Geneva Score. Figure 42 (Revised Geneva Score Decision Table) depicts the decision table used to calculate the RGS.

v Note the use of the Collect hit policy with the sum operator to indicate that the points in the outcome column will be summed for all rules that evaluate positive.

Figure 4 – Revised Geneva Score DRD

The first six rules in Figure 42 are based on a single finding whose presence is tested for. The seventh rule requires the presence of two findings and rules eight and nine test different ranges of the heart rate.

Figure 5 – Revised Geneva Score Decision Table

Meta Data

<blank>

Discussion

<blank>

Related Patterns

<blank>