RiskBenefit Rule 2 Confidential Do Not Distribute 1

Risk/Benefit Rule #2 Confidential - Do Not Distribute 1

Risk/Benefit Rule #1 Confidential - Do Not Distribute 2

Risk/Benefit Rule #3 Confidential - Do Not Distribute 3

Risk/Benefit Rule #3 Confidential - Do Not Distribute 4

Risk/Benefit Rule #3 Confidential - Do Not Distribute 5

Implementing the Tools Confidential - Do Not Distribute 6

Implementing AI: Step 1 1. Choose • Demand outcomes data • Compare feature-tofeature • Ensure workflow alignment • Model ROI Confidential - Do Not Distribute 7

Implementing AI: Step 2 2. Use • Define Key Performance Indicators (KPIs) • Design & implement SOPs • Iterate, working closely with tech partner Confidential - Do Not Distribute 8

Implementing AI: Step 3 Optimize • Measure & analyze KPIs • Implement & control changes • Expand scope Confidential - Do Not Distribute 9

Implementing AI Choose • Demand outcomes data • Compare featureto-feature • Ensure workflow alignment • Model ROI Use • Define KPIs • Design & Implement SOPs • Iterate, working closely with tech partner Confidential - Do Not Distribute Optimize • Measure & analyze KPIs • Implement & control changes • Expand scope 10

Case Studies: Predictions & Engagement Confidential - Do Not Distribute 11

“Rare Data” in Diabetes: Predicting A 1 c Confidential - Do Not Distribute 12

What About Behavioral Health? Confidential - Do Not Distribute 13

“Rare Data” in Behavioral Health AI: Predicting Suicidality Confidential - Do Not Distribute 14

Using Perceptrons to Drive Patient Engagement Confidential - Do Not Distribute 15

Positive Externalities Include Patient Satisfaction 8. 3 (n = 3, 743) 6. 9 (n = 3, 643) “I feel there is daily “I feel cared for & contact with my cared about. Not just doctor. ” a condition or “The easy way to diagnosis” “I was very happy that report the when I stated my cough information without had deteriorated I was having to login in a contacted by Barb computer. I get so quickly, so far I am very busy at work I tend impressed. ” forget to do it. This “I think it’stoexcellent “I think this is an just as is. way Not is tooso easy. ” outstanding service! intrusive…gets to the Thank you for point and doesn’t keeping track of your take a lot of time. Patients where other Thanks a lot for “I like it because show's feel safe because asking!” I doctors“I don’t” feel that my doctor is [sic] that you all care about next to me even thou your patients, and it gives I am 2 hrs away from me sense of relief knowing that you all are there for him. Different city. ” us, love it. ” 16

Recap Confidential - Do Not Distribute 17

AI Is on the Rise Confidential - Do Not Distribute 18

AI Is Conceptually Straightforward & Clincally Complementary A / C o n ? 99. 5% accuracy Confidential - Do Not Distribute 19

Careful Selection Will Pay Dividends Confidential - Do Not Distribute 20

Procedural Implementation Is Crucial Choose Use Confidential - Do Not Distribute Optimize 21

Prediction, Engagement, and Satisfaction Are Within Reach 8. 3 (n = 3, 743) 6. 9 (n = 3, 643) Confidential - Do Not Distribute 22

Evidence-Based Digital Health • • • Investor Deck Blake Marggraff CEO blake@epharmix. com +1. 925. 899. 2669 Confidential - Do Not Distribute 23

Portfolio of Epharmix Interventions • Chronic Disease Management – Epx. Diabetes (w/ variations) – Epx. Hypertension – Epx. COPD – Epx. Heart. Failure – Epx. Asthma – Epx. Med. Tracking § Emergency Surgery ◦ ◦ Medicine & Epx. Discharge. Referral Epx. Pneumonia Epx. Post. Discharge Epx. Surgery § Nephrology ◦ Epx. Dialysis ◦ Epx. CVCto. Fistula § General Medicine ◦ Epx. Wellness ◦ Epx. Fall. Prevention • Psychiatry & Neurology – Epx. Depression – Epx. Substance. Use Confidential - Do Not Distribute Product Overview 24

Epharmix Clinical Workflow Provider Identifies Patients Subsidizes and incentivizes the implementation/usag e by those patients' primary care providers; Epharmix operationalizes with PCPs Epharmix Partner Enrolls patients on Epharmix platform Patients Answer messages on their phone, sending in clinicallyrelevant data Epharmix Categorizes at-risk individuals and triggers alerts for Care Managers Confidential - Do Not Distribute Team Members Respond with appropriate intervention Clinical 25

A Step Beyond • While machine learning toolsets are increasingly commoditized, data is the key differentiator to making meaningful models • Crafting a workflow that makes use of the model prediction to drive meaningful outcomes is also vital • ROI for Custom er Epharmix is uniquely positioned to achieve both points: – No dependence on external data sources like EHR: we collect our own data – A tight-knit research center working with us to trial these new models & workflows Insi ght Mo del Interventi ons Rare Epharmix Data Product Strategy Confidential - Do Not Distribute 26

Predicting Outcomes • The longitudinal data we collect from patients are prime for use in predicting their future outcomes – Direct data: reported biometrics, signs & symptoms – Meta-data indicative of behavior: response rate, response time, alert rate, and more • A variety of relevant models constructed so far – Using 1 month’s interaction with Epx. Depression to predict suicidality/self-harm in the next two months (see fig. 1) • 85% accuracy – Using 4 weeks’ interaction with Epx. Diabetes to predict Hb. A 1 c in 6 months • 90% accuracy fig 1. Predicting 2 -Month Suicidality using Epx. Depression Product Strategy Confidential - Do Not Distribute 27

Applying the Models • Case Study #2 • Case study #1 – Using the Epx. Depression suicidality prediction model – Identify patients likely to report self-harm, Engagin triageg, them to a high needs high risk category on the touch care Epharmix dashboard Engaging, Nonfine with –engaging, Preemptive counseling, low hard to prioritizedtouch outreach impact care – A client has a high-touch, hightech care program designated for high-risk patients – Program is expensive, so there is a huge need to identify types of patient to make the most impact (ROI) – A proposed 1 -year learning initiative, using Epharmix as the precursor program to predict the “types” of patients in terms of care impact – Funnel patients identified in Epharmix as most impactful to the high-touch program Product Strategy Confidential - Do Not Distribute 28

Provider Workflow Enroll � Provider can manually enroll a patient under 60 s, or utilize other enrollment methods such as email or selfenrollment portal. As patients respond, provider will be alerted about critical values (usually via email). Provider can also log in anytime to view past data reported. � Monitor Intervene At any given time, provider can focus on the patients who need help and work top-oflicense, while knowing that patients who are doing OK at the moment are still monitored. � Per SOP, provider can contact the patient for the alert, then resolve it on the dashboard, or further dispatch to upstream providers (like PCP). Confidential - Do Not Distribute � Manage Product Overview 29

Consultative Implementation Process 1. ) Identification • • • What is the patient inclusion criteria? Who is running the report for eligible patients? How often should the list be updated? 2. ) Enrollment • • • What current processes are already in place to add opt-out Epharmix enrollment? Who can telephonically reach out to patients and enroll into program? What automated enrollment techniques are relevant for the target population? 3. ) Standard Operating Procedures • • • What can the care manager do when an alert is triggered? When does a physician need to be notified? How often should a care manager reach out to a patient in red vs. yellow? 30

ED Referral: >2 x increase in post-discharge adherence 31

Hypertension: 10. 8 & 6. 6 -point decrease in SBP & DBP 32

Diabetes: 1. 15% drop in Hb. A 1 c in 4 months 33

COPD: 61% reduction in hospitalization risk N=168 Percentage of patients hospitalized in Epx. COPD vs. Days in Trial 34