25.2.6.4.2 Predicting Patient Onboarding Outcome

Invoke a machine learning model to predict patient onboarding outcomes and route the workflow.

The Patient Onboarding workflow uses the machine learning model in two different Invoke API activities. First, the Evaluate Insurance Risk calls the EVALUATE_RISK function to detect if historically a patient with a similar request to the current one was automatically rejected. Office staff tells you that this automatic rejection was due to the patient's insurance not reimbursing the procedure.

Figure 25-88 Two Combinations of Invoke API and Switch Using Machine Learning

Description of "Figure 25-88 Two Combinations of Invoke API and Switch Using Machine Learning"

The code for EVALUATE_RISK appears below. Predicting the outcome of the onboarding patient approval using the Neural Network machine learning model that won the AutoML experiment is a single SQL statement using the PREDICTION function. Its USING clause feeds the model the values of the current patient's insurance provider and initial procedure. Recall that these were the two columns that most strongly predicted the outcome. If the model predicts the STATUS to be "AutoRejected", then the function returns the outcome "Risky", otherwise it returns "Low Risk". The Evaluate Insurance Risk activity returns the Function Result into the V_RESULT workflow variable, and the subsequent Risk Level? Switch activity uses that variable containing the outcome value to decide the path to follow.

-- Result = 'Low Risk' or 'Risky'   
function evaluate_risk(p_patient_id number)  
return varchar2 is  
    l_result varchar2(30);          
begin  
    /*  
     | Machine Learning Model NN_CE379F672F was the winning model  
     | automatically identified by using AutoML.  
     */      
    select    
         prediction (NN_CE379F672F   
         using   
            pat.insurance_provider  as insurance_provider,  
            pat.initial_procedure   as initial_procedure  
         )  
      into l_result   
      from frc_patients pat  
     where pat.id = p_patient_id;
    if l_result = 'AutoRejected' then  
        return 'Risky';  
    else  
        return 'Low Risk';  
    end if; 
end;  

The Predict Approval activity uses Invoke API again to call a similar PREDICT_APPROVAL function. It accepts the previous prediction as a parameter. If that prediction was "AutoApproved" then it returns the outcome string "Auto-Approved", otherwise it returns the string "Manual. The Predict Approval activity returns the Function Result into the V_RESULT workflow variable, and the subsequent Auto‑Approve? Switch activity uses that variable containing the outcome value to decide the path to follow.