CASP Module 15, Section 3: Predictive Adherence and Workflow AI Tools
MODULE 16: THE DIGITAL PHARMACIST: TECHNOLOGY & AI IN PRACTICE

Section 15.3: Predictive Adherence and Workflow AI Tools

Examining AI tools that analyze patient data to predict non-adherence risk, enabling proactive interventions, and optimizing pharmacy workflows through intelligent task routing and resource allocation.

SECTION 15.3

Predictive Adherence and Workflow AI Tools

From Reactive Dispensing to Proactive Care: Using AI to Manage Patients and Your Pharmacy.

15.3.1 The “Why”: The Two-Front War of Pharmacy Practice

As an experienced pharmacist, you are simultaneously fighting a war on two fronts every single day. The first is the Clinical Front: a public health crisis of medication non-adherence. You know that 50% of patients with chronic diseases do not take their medications as prescribed, leading to over $300 billion in avoidable healthcare costs and hundreds of thousands of deaths annually. You are acutely aware that a prescription dispensed is not a prescription taken, but you are largely blind to what happens after the patient leaves the counter.

The second is the Operational Front: a daily battle against workflow chaos. You are drowning in a sea of “dumb” tasks—data entry, counting, insurance rejections, and endless phone calls—all arriving in a chaotic, unpredictable stream. This operational burden is the single greatest barrier preventing you from fighting the clinical war. You know you should be calling that high-risk patient on Humira, but you literally do not have the time because you have 50 prescriptions in your queue and two technicians are on lunch.

This is the central paradox of modern pharmacy: the more clinically valuable our services become (MTM, immunizations, adherence counseling), the less time our operational model gives us to perform them. We are trapped in a system that values speed of dispensing over quality of care.

This section explores the AI tools designed to solve this exact paradox. It is a two-part solution:
1. Predictive Adherence AI: This is your new “clinical radar.” It answers the question: “Of the 2,000 patients I am responsible for, which 20 need my help today?” It finds the clinical “signal” in the data “noise.”
2. Workflow Optimization AI: This is your new “charge nurse.” It answers the question: “How can I find the 30 minutes I need to actually help those 20 patients?” It automates and intelligently routes the 90% of low-value tasks to create clinical capacity.

Mastering these tools is not optional. It is the only way to transform your practice from a reactive prescription factory into a proactive clinical care hub.

Pharmacist Analogy: The AI-Powered “Charge Nurse”

Imagine you are the only pharmacist in a busy, 30-bed hospital unit. You have no “charge nurse.” Patients are just randomly assigned to beds as they arrive. A patient with a simple dressing change is put in the critical care room, while a patient crashing from septic shock is put in a hallway bed. You are forced to run from room to room, treating every patient with the same level of urgency because you have no triage system. This is the “First-In-First-Out” (FIFO) model of a traditional pharmacy workflow. It’s insane, inefficient, and dangerous.

Now, imagine you are given an AI-Powered Charge Nurse. This AI “co-pilot” sits at the central station and manages the entire unit for you and with you.

1. It Manages the Patients (Predictive Adherence):

  • The AI is connected to every patient’s vitals. It doesn’t wait for a “code blue.”
  • It taps you on the shoulder and says: “Go to Room 201. That patient’s adherence score just dropped to 95%.” (You discover their first refill of an anticoagulant is late).
  • It then says: “Don’t worry about Room 202. That patient has a 10-year perfect adherence history. An automated text message is sufficient.”
  • It finds the high-risk patients before they crash, allowing you to intervene proactively.

2. It Manages the Staff (Workflow AI):

  • While you are in Room 201, the AI is managing the workflow.
  • It sees a new “admission” (an eRx) for a simple, stable refill. It routes this task directly to the “automated dispensing cabinet” (the robot).
  • It sees another admission for a complex C-II. It flags this and puts it in your “high-priority” queue for when you are finished with your clinical intervention.
  • It sees that your technician is overwhelmed with data entry, so it re-routes simple tasks to another, less-busy technician.

This AI “Charge Nurse” doesn’t replace you. It liberates you. It triages the patients and the tasks, allowing you to stop being a “task-rabbit” and start being the clinical expert, applying your skills only to the patients and problems that actually need your brain.

15.3.2 Deep Dive: Predictive Adherence Modeling (The “Clinical Radar”)

For decades, our only tool for measuring adherence has been Proportion of Days Covered (PDC) or Medication Possession Ratio (MPR). Your community pharmacy experience is defined by these metrics. You know them as crude, retrospective measures. You can only see that a patient has already been non-adherent for the last 3 months. It’s like looking at a car crash in the rearview mirror. It’s too late to prevent it.

Predictive Adherence Modeling flips the script. It is a forward-looking tool that uses machine learning to identify the patients who are most likely to become non-adherent in the future. This allows you to intervene before the crash.

The “Data Ingredients” for an Adherence Model

An adherence model is a complex algorithm fed by thousands of data points. Your skill in reviewing a patient profile is a micro-version of this. The AI just does it at a massive scale, finding patterns you would never have the time to see. The more data sources, the more accurate the prediction.

Masterclass Table: Data Sources for Predictive Adherence Models
Data Source Key Data Points (Variables) Predictive Power (What It Tells the AI)
1. Pharmacy Management System (PMS) – Refill Gap History (days late)
– First-Fill vs. Refill
– Drug Class (e.g., Antidepressant, Statin)
– Formulation (e.g., Injectable vs. Oral)
– Patient Pay (Copay Amount)
– Days Supply (30 vs. 90)		 This is the strongest predictor.
Key Pattern: The single highest predictor of long-term abandonment is a late first refill.
Key Pattern: A copay over $50 increases abandonment risk by >30%.
Key Pattern: Patients on 90-day supplies are significantly more adherent than those on 30-day.
2. Electronic Health Record (EHR) – Diagnosis Codes (ICD-10)
– Number of Comorbidities
– Number of Prescribers
– Lab Values (e.g., A1c, LDL)
– Clinical Notes (via NLP)		 This provides the “clinical context”.
Key Pattern: A diagnosis of Depression or Bipolar Disorder is a massive predictor of non-adherence for other, unrelated chronic meds (e.g., statins).
Key Pattern: A patient seeing >3 prescribers (e.g., PCP, Cardio, Endo) is at high risk due to fragmented care.
3. Patient Demographics – Age
– Gender
– Language Spoken
– Communication Preferences (Text vs. Phone)		 This helps tailor the intervention.
Key Pattern: Younger, tech-savvy patients may ignore a phone call but respond instantly to a text nudge.
Key Pattern: Patients >75 may be at risk due to cognitive decline or polypharmacy.
4. Social Determinants of Health (SDoH) – Zip Code (as a proxy)
– Transportation Access (e.g., mail order vs. pickup)
– Income / Payer Type (Medicaid vs. Commercial)
– Housing Stability		 This provides the “socio-economic context.”
Key Pattern: Patients in “food deserts” or “transportation deserts” are at high risk. The problem isn’t forgetfulness; it’s access.
Key Pattern: Patients on Medicaid may have high adherence (due to $0 copay) but face other barriers like health literacy.
Clinical Pearl: The “PDC Fallacy” vs. The Risk Score

The Old Way (The PDC Fallacy): You run a report and see two patients:
• Patient A: 85% PDC for their statin. (Looks good, no action taken).
• Patient B: 60% PDC for their statin. (Looks bad. You call them).
This is a reactive model, and you are missing the real problem.

The New Way (The AI Risk Score): The AI runs a report:
• Patient A: 85% PDC… BUT… they just added an antidepressant, their copay just tripled, and they are 3 days late on their first refill since the change. The AI gives them a 98% (HIGH) Risk Score. The AI flags them for an immediate high-touch pharmacist call.
• Patient B: 60% PDC… BUT… they are on 90-day fills, have been stable for 5 years, and their “late” fills are always exactly 2 weeks late when their Social Security check arrives. The AI gives them a 30% (LOW) Risk Score, recognizing this as a stable (though poor) pattern. The AI flags them for a low-priority tech call to discuss med sync.

The AI model allows you to ignore the “stable bad” patient and focus your limited time on the “newly unstable” patient who is about to be lost to follow-up. It prioritizes risk of change over historical performance.

The Output: The Proactive Intervention Queue

The AI model’s output is not a static report. It is a dynamic, actionable “to-do list” that should become the central dashboard for your clinical pharmacy team. It integrates directly into your pharmacy system and refreshes daily, just like your fill queue. Your new “workflow” is to manage this “Adherence Queue” with the same urgency as your “Fill Queue.”

15.3.3 Tutorial: The Proactive Intervention Playbook

You have the “Adherence Queue.” Now what? A score is useless without a corresponding action. This is the “playbook” that translates a risk score into a specific, tiered intervention. This is how you operationalize the data.

Masterclass Table: The Tiered Adherence Intervention Playbook
Risk Tier & Score Patient Profile (Example) Likely Root Cause(s) The Intervention (The “Play”) Assigned To:
TIER 1: CRITICAL
(Risk Score: 90-100%)		 – New Start: Patient “John D.” just prescribed Januvia.
– First Fill: Has not picked up the script, 2 days late.
– Data: High copay ($75), new diagnosis.		 First-Fill Abandonment:
1. Cost Shock
2. Fear/Anxiety (new med)
3. Lack of Understanding		 “First-Fill Welcome Call” (High-Touch)
1. Empathize: “Hi John, this is your pharmacist. I’m calling about the new medication Dr. Smith sent. I saw the high copay and wanted to help.”
2. Solve: “I’ve already run a search and found a manufacturer copay card that will make this $15. I’ve applied it for you.”
3. Counsel: “While I have you, this is an important medication. Can I take 2 minutes to talk about what to expect?”		 Pharmacist
(High-Skill, High-Empathy)
TIER 2: HIGH-RISK
(Risk Score: 75-89%)		 – Established Patient: “Mary S.” on Eliquis.
– Refill Gap: 5 days late on her refill (PDC just dropped to 80%).
– Data: First-time gap. No recent hospitalizations.		 Unstable Adherence:
1. Forgetfulness
2. Side Effect (e.g., new bruising)
3. Ran out of refills		 “Adherence Check-in Call” (Tech or RPh)
1. Identify: “Hi Mary, this is [Tech] from the pharmacy. We noticed your Eliquis refill is a few days late. We want to make sure you’re not out.”
2. Triage:
  If “I forgot!” $\rightarrow$ “Great, I can get that ready for you. Would you like to enroll in our Med Sync program?”
  If “I’m worried about bruising…” $\rightarrow$ “That’s a great question for the pharmacist. Let me transfer you.” (Warm handoff to RPh).		 Trained Technician
(Triage & Med Sync)
Pharmacist
(If clinical barrier)
TIER 3: AT-RISK
(Risk Score: 50-74%)		 – Established Patient: “Robert P.” on Losartan.
– Refill Gap: 0 days late.
– Data: Has a pattern of being 3-4 days late every 90 days. PDC is stable at ~85%.		 Stable (but imperfect) Adherence:
1. Not synchronized
2. Procrastination		 “Proactive Nudge” (Low-Touch / Automated)
1. Automated Text/App: “Hi Robert, your Losartan is due for refill in 3 days. Press ‘1’ to refill now for pickup.”
2. Batch Call: An automated voice call (“IVR”) with the same message.		 Automated System (AI/Software)
TIER 4: LOW-RISK
(Risk Score: < 50%)		 – Established Patient: “Susan B.” on Metformin.
– Refill Gap: 0 days late.
– Data: 5-year perfect adherence. 90-day fills. Auto-refill enrolled.		 Adherent No Intervention Needed
1. Patient remains in standard automated refill workflow.
2. Do not waste staff time.		 No One (Standard Workflow)

15.3.4 Deep Dive: AI-Powered Workflow Optimization (The “Charge Nurse”)

The adherence playbook is revolutionary, but it requires one precious resource: pharmacist and technician time. You cannot make those high-touch calls if you are drowning in the fill queue. This is the second, critical half of the AI solution: using AI to create that time by optimizing the pharmacy’s internal operations.

The traditional pharmacy workflow is a “dumb” system. It is governed by two primitive rules:

  1. First-In-First-Out (FIFO): A simple refill for atorvastatin that arrived at 10:00 AM is handled before a STAT antibiotic for a septic child that arrived at 10:01 AM.
  2. “Siloed” Staff: Your “Data Entry” tech only does data entry. Your “Fill” tech only fills. Your “Pharmacist” only verifies. If one silo gets clogged, the entire system stops, even if the other silos are empty.

AI-powered workflow optimization shatters this model. It replaces “FIFO” with “Intelligent Task Routing” and “Siloes” with “Dynamic Resource Allocation.”

15.3.5 Core Functions of a Workflow AI Platform

This AI acts as the “brain” of your pharmacy, sitting *on top* of your PMS and directing traffic. It sees every new prescription, every staff member’s status, and the current queue times, and makes millisecond-level decisions to optimize the entire system.

Function 1: Intelligent Task Triage & Routing

This is the most important function. The AI *reads and categorizes* every single task as it arrives (eRx, PA request, phone call) and routes it to the *most appropriate, lowest-cost, available resource*. This is the end of FIFO.

Workflow Visualization: The AI-Powered “Intelligent Task Router”
The “Old” FIFO Model vs. The “New” AI-Triage Model
“OLD” FIFO QUEUE

1. Atorvastatin Refill

2. Norco (C-II) New Script

3. Amoxicillin (STAT)

4. Xeljanz (PA Req’d)

ONE BIG QUEUE FOR EVERYONE

(STAT script is stuck behind a refill)

“NEW” AI-TRIAGE

All Tasks Arrive…

AI Routing Engine

Route 1: Atorvastatin Refill $\rightarrow$ Robot Queue
Route 2: Amoxicillin (STAT) $\rightarrow$ STAT Pharmacist Queue
Route 3: Xeljanz (PA) $\rightarrow$ PA Tech Queue
Route 4: Norco (C-II) $\rightarrow$ C-II Pharmacist Queue

Function 2: Dynamic Resource Allocation & Load Balancing

This function moves beyond *tasks* to manage your *people*. The AI maintains a “digital twin” of your pharmacy, knowing who is working, what their skills are, and what their current workload is.

  • Load Balancing: The AI sees that your “Data Entry” queue has 50 tasks, but your “Fill” queue is empty. It automatically re-routes 10 simple data entry tasks to the “Fill” technician’s workstation, cross-training them and balancing the load without a human manager intervening.
  • Skill-Based Routing: The AI knows that “Tech A” is a certified, experienced tech, while “Tech B” is a new trainee. It automatically routes complex tasks (e.g., compounding calculations, insulin data entry) to Tech A, while routing simple tasks (e.g., atorvastatin data entry) to Tech B. This optimizes for both safety and training.
  • Presence-Aware Routing: The AI is integrated with your phone or computer status. It sees that “Pharmacist A” has just answered a 10-minute phone call. It automatically re-routes all incoming STAT prescriptions to “Pharmacist B” to ensure no delay in verification.

Function 3: Predictive Workload & Staffing

This is the “manager-level” function. The AI uses its ML models, trained on your pharmacy’s specific history, to predict the future. Your skill in “anticipating the 5 PM rush” is now quantified and super-charged.

  • Predicting the Rush: The AI knows your pharmacy’s “rhythm.” It knows that on the first Friday of the month (payday) at 4:00 PM, your prescription volume will increase by 150% and your wait time will triple.
  • The AI Intervention (Staffing): On Thursday, it sends an alert to the pharmacy manager: “PREDICTIVE WORKLOAD ALERT: Tomorrow’s 4-7 PM volume is predicted to be 150% above average, exceeding available staff capacity. Recommendation: Add one technician from 3-7 PM to maintain target wait times.”
  • The AI Intervention (Workflow): At 2:00 PM on Friday (before the rush), the AI *proactively* routes all non-urgent tasks (e.g., refills for next week) to the back of the queue, clearing the decks for the incoming wave of new, “waiter” prescriptions.

15.3.6 Governance, Risk, and the “Human-in-the-Loop”

These tools are revolutionary, but they are not magic. They are powerful algorithms that can create new and subtle forms of risk. As the pharmacist-in-charge, you are not just responsible for your human staff; you are now responsible for the safe practice of your AI co-worker. Your role as an auditor is paramount.

The #1 Risk: Algorithmic Bias in Adherence Models

This is the most significant ethical and clinical risk you will face. An AI model is only as fair as the data it was trained on. If the “historical data” reflects historical biases, the AI will learn, codify, and amplify those biases at a massive scale.

Scenario: The “SDoH Bias”
• The Data: The AI is fed 5 years of data showing that patients from a specific low-income zip code (who may face transportation and food security barriers) have a high rate of non-adherence.
• The AI’s (Biased) Lesson: The AI learns a simple, dangerous pattern: “Patient from Zip Code 12345 = High-Risk.”
• The Catastrophic Result: A perfectly adherent patient moves into that zip code. The AI *automatically* flags them as “High-Risk,” wasting your team’s time. Even worse, this score could be shared with payers, who might use it to deny them access to high-cost drugs (“This patient is high-risk for non-adherence, so we won’t approve this $5,000 medication”).

The Mitigation (Your Role as AI Auditor):
1. Demand Transparency: You must *never* use a “black box” model. Your vendor *must* be able to tell you the “Top 5 Risk Factors” for each patient’s score.
2. Audit the “Why”: If you see “Zip Code” as the #1 risk factor, you must flag this as a biased-run. The intervention should be different (e.g., offer free delivery) and you should *not* treat the patient as “non-compliant.”
3. The Human-in-the-Loop: The AI score is a recommendation, not a command. Your job is to use your human judgment. “I see the high-risk score, but I’ve spoken to this patient, and they are not a risk. I will override the AI’s recommendation.”

The #2 Risk: Over-Optimization and Brittle Workflows

An AI optimized *only* for speed can be a nightmare to work with. It can burn out your best people and grind to a halt when something unexpected happens.

Scenario: The “Burnout Bot”
• The AI’s Logic: “Pharmacist A is my fastest verifier. Tech A is my fastest filler.”
• The Result: The AI routes 70% of the workload to Pharmacist A and Tech A, while Pharmacist B and Tech B get the “easy” scraps. Your “A-Team” burns out and quits in 3 months. The AI has optimized for speed, not for sustainability or fairness.

The Mitigation (The Human “Override Switch”):
1. You Define the Goals: As the manager, you must be able to configure the AI’s “goals.” You should be able to set the goal not just as “Lowest Wait Time,” but as “Balanced Workload.”
2. The “Charge Pharmacist” Override: The pharmacist on duty must have a “master dashboard” that shows the AI’s automated routing. They must have a “drag-and-drop” ability to manually re-assign tasks, overriding the AI’s logic if they see a problem (e.g., “I see the AI is swamping Tech A. I’m moving 20 of their tasks to Tech B to help out”).

Your Transformation: The Conductor of a “Smart” Pharmacy

For your entire career, you have been one of the “musicians” in the orchestra, playing your instrument (verification, counseling) as fast as you can, often while reading from messy sheet music (the workflow).

These AI tools transform you into the Conductor. The AI hands you a perfectly clean, triaged, and prioritized score. It manages the “logistics” of the orchestra—ensuring the right musician has the right music at the right time (workflow AI). It also uses “precognition” to warn you which musicians are likely to miss a note (adherence AI).

Your job is no longer to just “play your part.” Your new, elevated job is to listen to the AI’s signals, guide the human team’s interventions, and conduct the entire symphony of patient care. You are finally freed from the “tyranny of the queue” to practice at the absolute peak of your clinical license.