Monday, May 26, 2014

Modeling Clinical Rules with the Decision Modeling and Notation (DMN) Specification

In my previous post titled Modeling Care Pathways with BPMN, I explained the benefits of using BPMN constructs such as events, tasks, gateways, and data objects to model standardized care processes. An example of task in BPMN is the business rule task which can be executed by a business rule engine at run-time. For illustration purposes, I used a treatment algorithm titled Management of substance use disorder, Module C: General Health Care from the VA/DoD Clinical Practice Guideline (CPG) for the treatment of substance use disorders.

In this post, I discussed the Decision Model and Notation (DMN) specification recently published by the Object Management Group (OMG). While the BPMN is designed specifically for modeling business processes, the DMN provides a notation for modeling decision requirements and decision logic.

The DMN specifies the use of Decision Requirements Diagrams (DRG) for representing decision requirements and decision tables for expressing the decision logic. The latter can be expressed in a language called FEEL (Friendly Enough Expression Language) which is also specified in the DMN. The diagram  below from the DMN specification (click to enlarge) describes the relationship between the BPMN and the DMN and the Decision Requirements and Decision Logic levels of the DMN.

In the substance use disorder treatment algorithm in my previous post, there is a decision node with the question: Are Treatment goals achieved?. This decision node can be modeled in BPMN as a business rule task. We can then use the DMN to model the detailed requirements for the decision logic for determining whether treatment goals have been achieved or not. In mental health treatment, the Clinical Global Impression - Improvement scale (CGI-I) is a validated 7-point scale of health status improvement. According to Wikipedia:

The CGI-I requires the clinician to assess how much the patient's illness has improved or worsened relative to a baseline state at the beginning of the intervention. It is rated as: 1, very much improved; 2, much improved; 3, minimally improved; 4, no change; 5, minimally worse; 6, much worse; or 7, very much worse. 

For illustrative purposes, we consider that treatment goals are achieved if the CGI-I at 6 and 12 months from intake is 1, 2, or 3. Ideally, it would be nice to also include patient-reported outcome measures in determining the achievement of treatment goals. These rules can be executed by a production rule engine like Drools.

In addition to production rules, the decision logic can be modeled as a predictive analytics model in Predictive Model Markup Language (PMML) format. As an example, a logistic regression model represented in PMML can predict (based on the analysis of historical clinical data) which treatment options are more likely to lead to outcome improvement given the clinical profile of a specific patient.

The DMN fills an important gap in modeling complex business rules in a standardized way and will be an important tool in the Business Analyst's toolkit. The DMN has been added as a new technique in the public review release of version 3 of the Business Analysis Body of Knowledge (BABOK) of the International Institute of Business Analysis (IIBA).