Jerome Carter, MD April 18, 2016
Now it is time to look at clinical work. Clinical work comes in many forms. Some happen through direct interaction with patients (e.g., taking a history, doing a physical exam, ordering a medication). Others are often done when patients are not around (e.g., results management, discussing patients with care team members, reviewing guidelines, deciding on diagnostic strategies). Since the goal for this installment is learning BPMN, we will use a simple example -- ordering a medication.
Giving a patient a medication involves a few standard steps. Once an indication for a drug arises, a familiar cascade follows:
1. Determine if a medication is warranted
2. Review co-morbidities
3. Decide drug type/class required
4. Review medication history
a. Review patient allergies
5. If no contraindications
a. Select drug from formulary (if required)
b. Check for interactions (drug-drug, drug-food)
c. Decide dose, route, and frequency
6. If inpatient
a. Finalize order
7. If outpatient
a. Send to pharmacy
For the current workflow model, I am assuming an outpatient setting with an EHR. Here is the scenario for the current model:
A 45 year-old man with a history of mild hypertension (12.5 HTCZ daily) presents with an earache. After determining the patient has bacterial otitis media, you decide that a course of antibiotics is required.
Since this model will incorporate user interactions with an EHR, I want to demarcate clearly those tasks that are user actions and those that are computer responses. With this in mind, I will use a modeling convention in which user actions are modeled as User tasks and EHR responses are modeled as Service tasks. In addition, task names use verbs that indicate the type of User action, Select (Menu) Formulary, or EHR action, Display Formulary, done directly for a User action. EHR actions that occur without direct user input are labeled using “Automatic” (Automatic Check – Interaction).
Only the clinician is represented in the main pool. If I desired to show the EHR more explicitly, it could be placed in its own lane. However, the current model is focused on clinician actions more than EHR features.
Figure 1. Earache Medication
The model begins with a message event signaling the patient’s request for a visit with the clinician. Interactions with the patient are shown using dashed message lines. Once the exam is completed, an XOR gateway branches the workflow based on whether an infection is present. Since the ear is infected, the clinician accesses the EHR to look at the problem list. After reviewing co-morbidities, the medication history and patient allergies are viewed in the EHR. Having completed these preliminary tasks, the clinician selects the formulary option from the navigation menu, selects a drug and completes the prescription with the EHR performing safety checks in the background. Finally, the patient is given a choice between having the Rx sent electronically vs. being given a paper Rx. (Since the gateway shows the choice explicitly, I chose not to model the choice with a patient message.) After being given instructions regarding side effects and potential adverse reactions, the patient leaves the practice, and the next appointment confirmation is sent by email.
Often it helps to classify tasks further than I have done here. For example, some user tasks – those with the verbs "Review" or "Decide" – are mental and may not always be useful in a model. However, if the goal were modeling for CDS, then explicit representation of mental tasks is critical. EHR interactions (not all) with tasks are shown with dotted lines. Like patient messages, they need not be explicit unless it serves some purpose. In models where EHR interactions are important, data store interactions make EHR use more explicit (along with verbs reserved for that purpose( e.g., “Display”).
Well, we have used all Level I BPMN elements, and more complex models require more nuance. In the next post, I will look at Level II BPMN elements.