Home / Resources / Articles / Part 2: Innovative Artificial Intelligence (AI) Tool Optimizes Care for T2D Patients

Part 2: Innovative Artificial Intelligence (AI) Tool Optimizes Care for T2D Patients

Oct 10, 2020

Author: Steve Freed, R.PH., CDE, et. al.

Part 2: Collaborative Analysis Concludes that the AI Tool Improves Outcomes for Type 2 Diabetes Patients  

A new clinical software using artificial intelligence is being used for those with Type 2 diabetes in Kentucky. It shows promising results in improving the Quality of Life for its residents and lowering treatment costs. This article on how AI improves outcomes for T2D is the second in a 3-part series. Click here to go to Part 1.


The project was designed to address through clinical intervention the issues experienced by patients with diabetes, such as high A1c levels. Through team-based care, clinical software support, and patient engagement, the intervention was projected to help control HbA1c and potentially lower treatment costs.  

The project design consisted of using an electronic tool (www.GlucosePath.com) to facilitate and recommend treatments to patients based on cost and social factors. The project was designed to have pharmacists use that decision tool to make recommendations for alternative regimens.  The use of pharmacists rather than the patients’ current physician was designed to address two potential issues: 

  • Cost – Pharmacists typically bill at a lower rate than physicians; 
  • Clinical inertia – The pharmacists didn’t recommend the patient’s current regimen, so they would be more open to changing that regimen if it wasn’t working. 

Another goal of the project was to determine which regimen would be the most cost-effective for both the patient and the insurer.  

Approximately 43 providers and 126 patients from St. Elizabeth Healthcare participated in the test group. The patients included were adults (age 18+) from Kentucky Medicaid, had type 2 diabetes, an A1C higher than 8%, and had previously completed an appointment with a Saint Elizabeth’s primary care physician (PCP). Other patients that did not meet these criteria were not able to participate. 

Pharmacists and patients used the electronic tool to identify treatment regimens (including diet and exercise, which were modeled as if they were medicines) consisting of 1 to 5 drugs.  After agreeing on a course of action, the pharmacist sent the regimen to the patient’s provider for final review.   

The study allowed for the possibility that the patient or provider would accept all, some, or none of the medicines in the selected regimen.  Patients were grouped into four main categories based on how much they followed the selected regimen.  The following table shows those categories and the number of patients in them.  





Full – the patient took exactly the medicines in the selected regimen identified by the tool 



Alternate – the patient followed a regimen that was different from their pre-intervention regimen and that did not include medicines recommended by the tool 



Partial – the patient took at least one (but not all) of the medicines in the selected regimen created by the tool 



No changes – The patient’s regimen remained the same after intervention as before. 



Alternate initially, full after nine months 



No changes, full after six months 



Table 1: Clinical Decision by the Provider 

Www.GlucosePath.com software used an algorithm that would focus on A1C control, body weight, adherence, cost, and side effects. The main goal and outcome consisted of clinical data and team-based decision support. For the clinical data, a longitudinal analysis was used to measure the patient’s changes over time, comparing the test group with the control group. 

As shown in the above table, providers used the www.GlucosePath.com software to obtain recommendations that would suit each patient. Based on the A1C measure’s changes, most of the providers used the full recommendation for 72 patients. For 22 patients, the alternate recommendation was used. Then, for 18 patients, they used a partial recommendation, and for 12 patients, they did not use the recommendation. The last two patients did not follow the recommendation immediately, but they did after a few months. 

Graph 1: Recommendations Used Among Groups 


The providers that used these recommendations for their patients noticed a statistically significant improvement in the A1C. Surprisingly, the partial recommendation showed the highest decrease in A1C, about 20.96%. Patients who had used the full recommendation from their providers had an A1C decrease of 17.80%, and the patients who used the alternate recommendation only showed a 13.60% decrease. The control group, 11.38%, only showed an improvement compared to the no-change group but not compared to the other groups.  

A T-test was also used to compare the full and partially test group’s differences versus the control group. The collaborative test group showed a statistically significant improvement decline in the A1C with a P-value of 0.04. This improvement could be the effect of the increase in sample size by combining the two groups. 

The project also asked pharmacists and providers a series of questions about the use of the clinical AI tool and whether it improves patient outcomes. 


Question: How significant is the CDS tool for improving patient outcomes? 




Strongly Disagree 









Strongly Agree 



Table 2: Responses from providers regarding the CDS tool 

Most providers mentioned that they were not aware of the CDS tool that would recommend these regimens for their patients. Most of their knowledge about the tool was introduced during the experiment. Approximately 82% of the providers mentioned that they agreed that the CDS tool would improve patient outcomes. Only 18% disagreed, saying that this tool would not improve patient outcomes.  

They were also asked if they felt comfortable with the regimen that their patients’ decision support tool chose. More than 85% agreed that they felt comfortable with those regimens. About 53% of the providers agree that this tool can be handy and will use it in their current practice. 

Providers had also mentioned that medication costs were one of their main concerns because every patient wants to save money and, at the same time, get the best diabetes regimen. The CDS tool would display up to ten regimen options. It is the provider’s and pharmacist’s duty to notify of any drug interactions that would interfere with the patient’s current medications. In the test, they also compared costs for diabetes medications between the test and control group. It was concluded that with the CDS tool, the mean cost for the test group was $712, while the mean cost for the control group was $2,419. An exciting discovery about the most commonly used diabetes medications in each regimen was Biguanide, followed by insulin (basal) and GLP1s, respectively. Pharmacists mentioned that the best part of www.GlucosePath.com.com is finding these types of regimen recommendations with the best costs available and lowering the A1C by 20%.  

Next-Part 3: Optimize Clinical Diabetes Management with AI Technology  

User testimonials about the experience and exciting findings using the software. A more in-depth examination of the benefits of applying decision-support software in clinical practice.  

Part 3 will also provide you a chance to use the software at no cost. 

Part 1:   Adherence-related issues are hemorrhaging the costs of diabetes-related treatment in America

Part 2: Collaborative Analysis Concludes that the AI Tool Improves Outcomes for Type 2 Diabetes Patients

Part 3: Optimize Clinical Diabetes Management with AI Technology



Steve Freed, R.PH., CDE, with:

Peter Jay Won Pharm.D. Candidate, University of South Florida, Taneja College of Pharmacy 

Aleksandra Kusic, PharmD Candidate, Florida A&M University, College of Pharmacy and Pharmaceutical Sciences  

Joan Prifti, PharmD Candidate, Lake Erie College of Osteopathic Medicine, LECOM School of Pharmacy