1 of 14

�������������Opioid and Substance Abuse analytics� �

AI/ML Based –Drug Prescription Fraud, Waste, Abuse application Development for Inspector General of Police-

Michigan Department of Health

2 of 14

Analytics-based high impact solution for early opioid abuse surveillance, intervention and outreach for Government, This application will Target Patients who are receiving Healthcare Benefits from Department of Health and Opioid prescribing Doctors who are claiming rebates and government subsidiary

Solution Scope --

3 of 14

Use Cases under consideration for Opioid Analytics

    • 1: Influencing Factors
    • 2: Target Population
    • 3: Program Integrity
    • 4: Member Outreach

4 of 14

Influencing Factors

Use Case Description:

Identify opioid population based on the claim & enrollment attributes for risk stratification & prioritize engagement.

User:

Government- Inspector General of Police office Dashboards

Department:

  • State Drug Surveillance , Monitoring & Control,
  • Disease Management Units
    1. Case Management Sub Units
    2. Care management Sub Units
    3. Disease surveillance, monitoring & control sub units

Pain Statement:

Due to complex nature of the data & manual processing of insights to take a decision, it has become Challenging to identifying & control health influencing factors , These factors in real world have a direct influence on patient’s risk of an opioid incident/addiction

Objective:

To Identify & control factor the puts the member at risk that could help Government in early identification of high risk member and engage member in Preventive Care

Solution Outline:

A weighted average based model that considers all attributes in claim & enrollment based on the payer’s goal , Based on the nature of the goal the model adjusts the weightage to identify and qualifies the population , which are considered for risk stratification & population segregation in subsequent steps

5 of 14

UC1: Analytics Solution

Are there any common patterns exhibited by patients who are taking higher doses of prescription opioids?

    • Do they exhibit any potential risk factors like history of previous addiction, mental illness or chronic diseases?
    • Do they have similar socioeconomic characteristics?

Machine Learning Algorithms

  • Decision Tree
  • Logistic Regression
  • Association Rule Mining

Influencing factors that lead to -

  • Long-term opioid dependence
  • Abuse

6 of 14

UC2: Target Population

Use Case Description:

Risk Stratify & prioritize segments for Engagement

User:

Payer- Inspector General of Police Office Dashboard

Department:

  • State Drug Surveillance , Monitoring & Control,
  • Disease Management Units
    1. Case Management Sub Units
    2. Care management Sub Units
    3. Disease surveillance, monitoring & control sub units

Pain Statement:

Each DM Campaigns has its own agenda and specific requirements collating the “at risk “members and prioritizing the members based on the Claim attributes involves time, effort and resources

Objective:

To provide pro active insights to

    • Identify members who are at Risk of Event of Interest-( Identify targets based on Campaign goals)
    • Provider best course of action which are observed on a Positive outcome observed (Claims+ Health record)
    • Prioritize population engagement so that payer’s can focus on preventive care measure on the targeted population

Solution

Evidence based care is a CDC recommended preventive care approach for treating opioid patients

Translating the EBC and its associated care plan into analytics:

    • Apply Positive outcome Claim thresholds on targets,
    • Model predicts the risk of an Opioid Event of Interest
    • Model recommends ideal treatment options & cost that will lead the target into positive health track

7 of 14

UC2: Analytics Solution

Predictive models are developed -

  • To generate risk score for each patient
  • To stratify & flag patients at risk of potential abuse
  • To plan effective care strategies to forestall addiction

High Risk

Medium Risk

Low Risk

Predictive Models -

  • Random Forest
  • Decision Tree
  • Logistic Regression
  • Naïve Bayes
  • Support Vector Machine
  • K-NN

Patient risk score

8 of 14

UC3: Program Integrity

Use Case Description:

Strengthening Program integrity based on the Rx

User:

Payer- Program Monitoring Staff

Department:

  • FWA-IG Units,
  • State Drug Surveillance , Monitoring & Control,
  • Disease Management Units
    1. Case Management Sub Units
    2. Care management Sub Units
    3. Disease surveillance, monitoring & control sub units

Pain Statement:

  • Most PHM solutions available does not have Rx based Program Integrity analysis which will help in preventive care planning & cost containment planning.
  • Limited Insights provided by Claim + clinical analysis at it tend to provider only care and cost insights whereas FWA analysis will provide proactive insights on potential abuse, misuse of Opioid by patient, provider,& pharmacies

Objective:

To strengthen the Target identification & improve the risk stratification results based on the feedback from FWA analysis by subjecting a claim into variety of ways an Opioid handling entities would abuse the system, Insight would help reduce exposure to opioids and prevent abuse .

Solution Outline:

Claim is subjected to various scenarios in which possible FWA could take place, Predictive Model will identify if claim is clean or a potential abuse /fraud of services and dollar. Model check the integrity of service rendered from Member, Prescriber and Pharmacy perspective.

9 of 14

UC3: Analytics Solution

Predictive Analytics based FWA Scenarios

Social Network Analysis

Investigative Reports

To investigate –

  • Prescription pattern
  • Utilization pattern
  • Billing pattern

Scenarios to highlight –

  • Suspicious claims
  • Anomalous providers/members

Predictive Models –

Random Forest, Decision Tree,

Logistic Regression, Naïve Bayes,

or Ensemble Models

To explore –

  • Relationship between physicians, members & pharmacies
  • Patients with multiple prescribers or pharmacies, early refill of prescriptions
  • Associations with proven fraudulent claimants, pharmacies or physicians
  • Distance between the patient and the provider location

To assess –

  • Performance of providers vis-à-vis their peer groups
  • Utilization by various demographic features
  • Consumption of specific drugs prone to abuse
  • Sudden spikes in usage, indicating anomalous behavior

10 of 14

UC4: Member Outreach

Use Case Description:

This use case aims to screen Members and collect patterns of prescription drug use using a survey. The responses are analyzed to identify non-medical use / problem patterns ranging from mild to severe. Based on the problem patterns, information booklets are sent to the Member.

User:

Medicaid Beneficiary

Department:

State Health Department

Pain Statement:

Detect emerging problems and intervene early before the condition can progress.

Well-supported scientific evidence shows that less severe forms of these conditions often respond to brief physician advice and other types of brief interventions. 

Objective:

To reduce risk factors and bolster protective factors among Members by providing --

1. Relevant literature in the form of booklets

2. Support Program information available in their State

Solution Outline:

1. Include “Prescription Drug Use” Survey for Members who were prescribed opioids x times for pain management in the last <15/30/45/60> days.

2. Send information booklet(s) that can be downloaded/viewed by the Member.

11 of 14

1

2

3

12 of 14

Overview

Factors

Identify Targets

New Members

Old Members

Influencing

Factors

Inhibiting Factors to follow care plan

Risk Scoring Algorithm

Prioritize

“At Risk” Members score

Risk Stratify

( Based on CM –Campaign Goals)

CS FWA

Investigate

Program Integrity

Re-Prioritize

“At Risk” Members Score

Identify & Engage Different Targets Based on Payer’s CM-Campaigns (Goals)

13 of 14

Key functions of Opioid Analytics

    • 1. Influencing Factors
    • Various claim and clinical attributes ( Encounter data from Hosptial)
    • Pick members based on influencing factor -new members
    • Inhibiting factors to care plan ( non adherence to care) .-Old Members
    • 2. Target Population
    • Predictive Models identify suitable candidates for outreach based on Risk scores
    • Buckets both old and new members in severity level
    • 3. Fraud, Waste & Abuse
    • Investigate potential abuse of services, resources and dollars
    • Target the non targeted/ low profile –Members who are in Early stages of Addiction cycle or Experimenting or low profile abusers .
    • 4. Member outreach
    • CE Portal Functionalities to assists case worker in .
    • Disease Management intervention
    • Care & Case Management

14 of 14

Thank You