Aaron Shepard
Design Director / Product Designer
cover_v1.jpg

anthem.ai

My work at Anthem.ai was mainly focused on the front portion of the design process, everything from Research to Prototypes. There was a multitude of different tools we were creating, some could be integrated into the EMR, for example Epic, others would be integrated in to anthem.com, or used as external web applications. My role was more about the discovery of those products, building out different concept models, wireframes & flows, not necessarily the delivery of.

 At the beginning of the project, while working with the Dir. of Product, and a team of data scientists who were building the data model, I started work on our Proof-of-Concept, which drove the budget for research and design. Then partnered with a new Dir. of Research to create the artifacts you will see shortly. Essentially, I played a major role in the entire process.

 

my role: Research Planning, Synthesis & Analysis, Key Insights, User Profiles, Product Strategy, Concept Models & Diagrams

 
 

Background - Bullet points

  • Building an AI driven Insight Optimization Tool

  • Helps steer decision making for clinicians

  • Provides guidance for patients

  • Based on 5 billion insurance claims

  • Compares 100 distinct combinations of treatment options 

 
 

Background - Descriptive

I was working on an AI driven Insight Optimization Tool, and then a Comparative Treatment Efficacy System to help steer decision making for clinicians and provide guidance for patients. This tool draws from as many as 86 million medical records & 5 billion insurance claims to identify patients who have similar demographics, medical histories, and diagnoses. Then compares over a hundred distinct combinations of treatment options to identify and predict outcomes from those treatments. This is extremely helpful for PCPs (Primary Care Physicians), Care Managers, Endocrinologists, and Nurse Practitioners who have patients with complex health conditions. 

 
 

The Problem

Providers face several challenges when making treatment decisions for their patients.

Clinical trials often have limitations that make them less applicable to real-world scenarios. These trials may not accurately represent the provider's patient population, and they rarely offer head-to-head comparisons of treatment choices. Additionally, healthcare providers must contend with an overwhelming volume of literature and clinical practice guidelines to stay current in their field.

 
 

The Process

To start our research journey, we created a Research Plan & Discussion Guide which included our test goals and what we were curious about.

A few examples are: How to use the tool, what insights are valuable to clinicians, what are their expectations in integration of a tool like this into they workflow, the barriers to using the tool, or any unmet needs when clinicians are recommending treatment options to patients.

 Also, we needed to determine how many participants should we seek to test, what type of test we should conduct, and any metrics we wanted to track. From here we created the discussion guide and a concept test, and through that test we validated how this tool would integrate into their workflow and how this tool compares to what they are doing today. Also, any way we could improve this tool.

 
 
 

User Profiles

Our user profiles are based on real data that has been captured, analyzed, and serves as a representation of the user. The data collected is descriptive of their demographics as well as their psychographics. User profiles help you understand how and why users interact with the product.

 
 

Key Insights

During research what we were discovering fell into the following categories:

 
 

Notable Key insight - “Total Care”

"Total care" extends far beyond prescription writing. It encompasses identifying optimal medication combinations, ensuring affordability, managing diet, exercise, weight management, promoting adherence (including timely prescription refills), encouraging compliance with medication schedules, working with dietitians, trainers, or lifestyle coaches. The path to achieving optimal care involves recognizing that personal behaviors often holds more significance than medical interventions.

 
 
 

Strategy

Most of our research lead us to creating a tool that needed to integrate with the EMR and work nicely with the clinician’s workflow. Also, a standalone product and a product based on populations of patients. In addition to three different tools for patients including a tool integrated into anthem.com, a standalone tool and a product that could be integrated into a corporate intranet. My focus was on the standalone product for the Clinicians

 With our strategy in mind, we created some design goals to provide some guardrails: understand both the needs of the patient & the clinicians, make optimal treatment insights discoverable & understandable, and remain true to the AI model.

 
 
 
 
 

Information Architecture

The Insight Optimization Tool is an integral component of the broader Health OS platform, which encompasses multiple channels. Specifically, this tool is housed within the Precision Insights channel.

 
 
 

User Flows & Wireframes

The provided flows are a partial representation of the Precision Insights channel within the Health OS platform. The selected flows include the welcome screen, search function, results display, and a low-fidelity version featuring the core concepts of the tool.

 
 
 

Design

Primary flow for clinicians:

To find insights and an optimal treatment option for a patient a clinician will start with a patient search. This will result in a patient record and an insights panel.

The patient record describes who the patient is and what insight type the clinician is looking for, in this case T2D. Also, any current treatments, A1C levels, types of medications, reasons not to treat, and medical history, basically an overview of the patient. Patient costs are essentially, what can the patient afford? For most clinicians, this was their number one responsibility. Treatment options shows current treatment, optimal & secondary treatment options, the expected change in A1C, and a treatment timeline and suggesting where this patient can expect to be in 3-6 months. In the insight details panel, we have our evidence, adverse outcomes, and the patient’s formulary, following with the patient’s treatment timeline

 

Feedback from Usability Testing

 
 

Areas to focus the design on:

Affordability & Cost – essentially, based on a lot of factors, what could the patient afford? For most clinicians, this was their number one responsibility, finding drugs their patients could afford. Also, is there a co-pay? Are there any assistance programs?

Adverse outcomes – What side effects will my patients be dealing with if a clinician adds another drug or 2? Will blood pressure increase? What will happen to kidney function? Will the drug help with weight loss?

Optimize trade-offs – How do we optimize for efficacy vs. quality of life / fewest side effects vs. lowest costs vs. lowest A1C vs. effects on Renal / Kidneys vs. taking the Fewest meds

Formulary – Was the prescribed drug in or out of the patient’s formulary?  What are the additional costs? How time consuming is this going to be for the clinician?

Evidence – Is the data real that we were showing or suggesting? - Is this evidence based?

Compliance - How can clinicians interact with the patient? Medication reminder?

 

 
 

Outcomes

This tool helped clinicians improve health outcomes for their patients and reduced costs of care by $70 million annually for Anthem. In addition, there was a 55% reduction in care gaps and improved patient compliance in multiple areas including wellness, prescriptions, and preventive care.