CCPP Module 11, Section 2: Using Risk Stratification Tools and Data Analytics
MODULE 11: PATIENT IDENTIFICATION AND RISK STRATIFICATION

Section 11.2: Using Risk Stratification Tools and Data Analytics

From Fishing with a Net to Fishing with a Spear: The Art and Science of Precision Targeting.

SECTION 11.2

Using Risk Stratification Tools and Data Analytics

A deep dive into using EHR-based registries, predictive analytics, and risk scoring models to pinpoint high-risk patients within a larger population.

11.2.1 The “Why”: Beyond Population Identification to Precision Intervention

In the previous section, we established the strategic imperative of selecting target populations. We learned to focus our efforts on high-impact disease states and patient groups where our interventions can deliver the greatest clinical and financial return. This is the equivalent of a fisher deciding to cast their nets in a part of the ocean known to be teeming with a specific, valuable type of fish. It’s a sound strategy that dramatically increases the odds of success compared to fishing in an empty sea.

This section, however, is about upgrading your equipment from a wide net to a precision-guided spear. It is not enough to simply identify a large population, such as “all patients with diabetes.” Within that group of thousands, there is a vast spectrum of risk. There is the 62-year-old with well-controlled type 2 diabetes on metformin alone, whose A1c has been stable at 6.8% for five years. Then, there is the 62-year-old with a 15-year history of diabetes, an A1c of 11.2%, complicated by nephropathy, retinopathy, two recent hospitalizations for DKA, and significant financial barriers to affording insulin. To treat these two patients with the same level of intensity is a profound misallocation of your most precious resource: your clinical time.

Risk stratification is the process of using data and analytical tools to parse a large, identified population into distinct tiers of risk. It is the science of predicting which patients are most likely to experience negative outcomes—hospitalization, disease progression, adverse drug events, or death—in the near future. By applying these tools, you transform your approach from generic population management to targeted, individualized care. You can now focus your most intensive, hands-on interventions (the spear) on the small percentage of patients who are driving the majority of poor outcomes and high costs, while deploying more efficient, technology-enabled strategies (the net) for the stable majority. Mastering risk stratification is the key to scaling your impact, proving your value in a data-driven healthcare landscape, and practicing at the absolute top of your license.

Pharmacist Analogy: The Pharmacy’s High-Value Coupon System

Imagine you are the manager of a large retail pharmacy, and your goal is to increase front-end sales.

Population Identification (The Old Way): Your first strategy is to print 10,000 generic flyers that say “10% Off Your Purchase!” and mail them to every household in your zip code. This is like targeting “all patients with hypertension.” You’ve identified a population, and you’ve sent out a generic intervention. You’ll get some response, but most flyers will be thrown away, the cost is high, and the return on investment is low. You are fishing with a very wide net.

Risk Stratification (The CCPP Way): Your new strategy is to use your pharmacy’s data analytics.

  • Tier 3 (Low Risk/Stable): Your system identifies thousands of customers who shop regularly but only buy prescriptions. They receive an automated email with a coupon for “10% off any vitamin purchase.” This is a low-cost, automated, but still targeted intervention.
  • Tier 2 (Rising Risk): The system identifies customers who regularly buy diabetic testing supplies. They receive a personalized, high-value coupon in the mail for “$5 off any brand of sugar-free candy or diabetic socks.” The offer is more specific and more valuable because you’ve identified a specific need.
  • Tier 1 (High Risk/High Potential): The system identifies your top 50 customers—those who not only have chronic conditions but also frequently buy high-margin cosmetics, personal care items, and greeting cards. These are your “whales.” You don’t just send them a coupon. You, the pharmacy manager, personally call them. “Hi Mrs. Jones, this is Michael, your pharmacist. I noticed you’re a regular shopper, and I wanted to personally invite you to a private, after-hours event next week where we’ll be offering 30% off the entire store just for our best customers.”

This is risk stratification. You used data to segment your population into tiers and then matched the intensity of the intervention to the value and potential of each tier. The phone call (your intensive CMM visit) was reserved for the highest-risk, highest-value customer. The generic email (your automated adherence reminder) was used for the stable majority. This is precisely how we must approach managing patient populations—by deploying our most valuable resource (our personal intervention) on the patients who need it most and stand to benefit the most.

11.2.2 Foundational Tools: Mastering the Electronic Health Record (EHR) Registry

Before we delve into complex predictive models, we must master the most powerful and accessible tool at your disposal: the Electronic Health Record (EHR). Modern EHR systems (like Epic, Cerner, Athenahealth, etc.) are not just digital filing cabinets; they are vast, searchable databases of clinical information. Your first and most important data analytics skill is learning how to “query” this database to build patient registries. A registry is simply a dynamic list of patients who meet a specific set of criteria. Building and refining these registries is the foundational activity of population health management.

You do not need to be a computer programmer to do this. Most EHRs have user-friendly reporting workbenches or registry tools that allow you to build these lists using logical rules (“AND,” “OR,” “NOT”). Your role is to act as the clinical architect, defining the precise criteria needed to identify the exact patient cohort you want to target. This is a direct translation of your pharmacy skills; you are essentially writing a very specific, detailed prescription for a list of patients.

Masterclass Table: From Clinical Question to EHR Registry Query
Your Clinical Question (The Need) The Logical Components of Your EHR Query Example Patient You Will Find
“I want to find my uncontrolled diabetic patients who are at the highest risk for complications.” Primary Population: Patients with a diagnosis of ‘Type 2 Diabetes’ on their problem list.
AND
Lab Value: Most recent A1c result is > 9.0%.
AND
Lab Value: Most recent eGFR is < 60 mL/min.
AND
Demographic: Patient is currently ‘Active’ with the clinic. 		A 68-year-old male with T2DM, a recent A1c of 9.8%, and an eGFR of 45, who is an active patient in your primary care clinic. This patient is a prime target for SGLT2 inhibitor initiation.
“I need to identify HFrEF patients who are not on guideline-directed therapy.“ Primary Population: Patients with ‘HFrEF’ or ‘Systolic Heart Failure’ on problem list (or an echocardiogram report with EF < 40%).
AND
Medication: NOT on an ACEi, ARB, or ARNI.
OR
Medication: NOT on a Beta-Blocker.
OR
Medication: NOT on an MRA.
OR
Medication: NOT on an SGLT2 Inhibitor. 		A 72-year-old female with HFrEF who is on a beta-blocker and an SGLT2 inhibitor, but not on an ARNI or MRA. She represents a clear gap in care and an opportunity for therapy optimization.
“I want to find older adult patients who are at high risk for falls due to their medications.” Primary Population: Patients with Age ≥ 65 years.
AND
(Medication Class: Taking any Benzodiazepine
OR
Medication Class: Taking any Z-drug (zolpidem, etc.)
OR
Medication Class: Taking any Anticholinergic (using a high-risk list like Beers Criteria)
OR
Medication Count: Taking > 4 antihypertensive agents). 		An 81-year-old female taking lisinopril, amlodipine, HCTZ, metoprolol, and a nightly dose of trazodone. Her medication regimen places her at extremely high risk for orthostatic hypotension and subsequent falls.
Your First Meeting: Befriending the IT/Data Analyst

One of the most valuable professional relationships you can build is with your organization’s clinical informatics or data analytics team. You are the clinical expert who knows what to look for, and they are the technical experts who know how to build the reports to find it. Schedule a 30-minute meeting with them. Bring them your clinical questions (like the ones in the table above). Explain the “why” behind your request. When they understand that your goal is to prevent readmissions or improve quality scores, they become your partners in population health. This collaboration is a critical force multiplier for your efforts.

11.2.3 Intermediate Tools: Applying Validated Risk Scoring Models

Building registries based on a few criteria is a powerful first step. The next level of sophistication is to apply validated, evidence-based risk scoring models to your patient populations. These scores are medical algorithms that have been scientifically proven to predict a specific outcome. They take multiple patient factors, assign a specific weight or point value to each, and sum them to produce a score that quantifies a patient’s risk. Instead of just identifying “a patient with A-Fib,” you can now identify “a patient with A-Fib and a CHA₂DS₂-VASc score of 6, indicating an 9.8% annual risk of stroke.” This level of precision is incredibly powerful for prioritizing your interventions and communicating risk to providers and patients.

Risk Score Deep Dive: CHA₂DS₂-VASc for Stroke Risk in Atrial Fibrillation

What it is: An acronym-based scoring system used to estimate the annual risk of ischemic stroke in patients with non-valvular atrial fibrillation, guiding the need for oral anticoagulation.

Why it’s a Pharmacist’s Power Tool: Atrial fibrillation is incredibly common, and the decision to anticoagulate is one of the most high-stakes choices in primary care. As a pharmacist, you are often the first to notice a new diagnosis of A-Fib or to see a patient with A-Fib who is not on anticoagulation. Being able to quickly calculate their CHA₂DS₂-VASc score allows you to transform a vague concern into a specific, data-driven recommendation. You can say, “Dr. Smith, I see Mr. Johnson was diagnosed with A-Fib. I calculated his CHA₂DS₂-VASc score to be 4, which places him at a 4% annual risk for stroke. Per the guidelines, anticoagulation is strongly recommended. Would you like me to help facilitate a prescription for apixaban?” This is top-of-license practice.

Masterclass Table: Deconstructing the CHA₂DS₂-VASc Score
Factor Acronym Points Clinical Pearl for Pharmacists
Congestive Heart Failure C 1 Look for diagnoses like HFrEF/HFpEF on the problem list or medications like loop diuretics, ARNIs, or SGLT2is prescribed for a HF indication.
Hypertension H 1 A documented diagnosis or currently taking antihypertensive medication. This will apply to the vast majority of your older patients.
Age ≥ 75 years A₂ 2 Note the double weight. Age is a potent, non-modifiable risk factor.
Diabetes Mellitus D 1 A documented diagnosis or taking any anti-diabetic medications.
Stroke / TIA / Thromboembolism S₂ 2 This is the most powerful predictor of a future stroke. Any history of these events, even many years ago, confers double points. This is a critical factor to verify.
Vascular Disease (prior MI, PAD, aortic plaque) V 1 Look for prior cardiac stents, a history of claudication, or a carotid artery stenosis diagnosis. Often missed.
Age 65-74 years A 1 Even this age range confers risk. Do not combine with the A₂ category; a patient is either 1 point or 2 points for age, not both.
Sex Category (Female) Sc 1 Female sex is considered a “risk potentiator” rather than an independent risk factor. A score of 1 in a female (due to sex alone) does not automatically warrant anticoagulation.
Putting it into Practice: Your A-Fib Registry Workflow
  1. Build the Registry: Create an EHR registry of all active patients with a diagnosis of ‘Atrial Fibrillation’.
  2. Filter for Gaps: From that list, exclude all patients who are currently on an oral anticoagulant (warfarin, apixaban, rivaroxaban, dabigatran, edoxaban). The remaining list contains your potential gaps in care.
  3. Stratify by Score: For each patient on the filtered list, calculate their CHA₂DS₂-VASc score. A score of ≥ 2 for men or ≥ 3 for women is a strong indication for anticoagulation.
  4. Prioritize and Intervene: Start with the patients with the highest scores. Send a message to the provider with your calculated score and a specific recommendation. This is a highly efficient, high-impact workflow.

Risk Score Deep Dive: LACE+ Index for Readmission Risk

What it is: A validated tool used to predict the risk of hospital readmission or death within 30 days of discharge. The acronym stands for Length of stay, Acuity of admission, Comorbidities, and Emergency department visits.

Why it’s a Pharmacist’s Power Tool: Preventing hospital readmissions is one of the highest priorities for any health system. It is a major driver of cost and a key quality metric. Pharmacist-led transitions of care services are proven to reduce readmissions, but it’s impossible to provide this intensive service to every single discharged patient. The LACE+ score allows you to stratify all discharged patients and focus your limited TOC resources on those at the highest risk of returning. This is a perfect example of data-driven resource allocation.

Masterclass Table: Deconstructing the LACE+ Index
Factor Acronym Points Assignment Clinical Pearl for Pharmacists
Length of Stay (days) L 1 pt for 1 day, 2 for 2, 3 for 3, 4 for 4-6, 5 for 7-13, 7 for ≥14 days A long length of stay is a proxy for clinical complexity, deconditioning, and a higher chance of post-discharge complications.
Acuity of Admission A 3 pts if admitted via the Emergency Department (non-elective). 0 pts if elective. An emergent admission implies the patient was acutely ill and unstable, increasing their post-discharge fragility.
Comorbidities (Charlson Comorbidity Index) C 1 pt for 1 comorbidity, 2 for 2, 3 for 3, 5 for ≥4 comorbidities This measures the patient’s underlying disease burden. Pharmacists should focus on comorbidities that are heavily influenced by medications (Diabetes, HF, CKD, COPD).
Emergency Department Visits (in past 6 months) E 1 pt per visit, up to 4 points Frequent ED visits are a massive red flag. It signals poor chronic disease management, social instability, or both. This is a key indicator of a patient “failing” in the outpatient setting.
“+” Factor (Added later) + 2 pts if discharged on >10 medications. The “+” was added specifically to account for medication complexity. This is your domain. A high LACE+ score driven by the “+” factor is a direct call to action for a pharmacist intervention.
Using the LACE+ Score to Build a Tiered TOC Service

The LACE+ score is designed to be actionable. Many hospitals automate the calculation within the EHR for every discharged patient. You can use the score to create a tiered TOC program:

  • High Risk (LACE+ Score > 10): These patients receive the “Cadillac” service. This includes a bedside medication reconciliation and counseling session by a pharmacist before discharge, a follow-up phone call from the pharmacist within 48-72 hours, and a scheduled in-person or telehealth CMM visit within 7 days.
  • Moderate Risk (LACE+ Score 5-9): These patients may receive a follow-up call from a pharmacy technician or nurse, using a standardized script to check on medication access and identify any red flags that need to be escalated to the pharmacist.
  • Low Risk (LACE+ Score < 5): These patients might receive an automated text message reminder to pick up their prescriptions and to call with any questions.

This tiered approach ensures that your most intensive resources are reserved for the patients who are statistically most likely to be readmitted, maximizing your program’s efficiency and impact.

11.2.4 Advanced Tools: The Frontier of Predictive Analytics

Risk scores like CHA₂DS₂-VASc and LACE+ are powerful, but they are based on a limited number of variables and represent a snapshot in time. The cutting edge of risk stratification lies in predictive analytics and machine learning. These advanced computational techniques can analyze hundreds or even thousands of variables from the EHR, claims data, and demographic information to create highly accurate, continuously updated predictions about a patient’s future risk.

As a pharmacist, you will not be building these models yourself. However, you are a critical end-user and clinical validator of them. Your organization may invest in a predictive analytics platform that generates a daily list of “High-Risk Patients for Pharmacist Intervention.” Your job is to understand conceptually how these models work, what their limitations are, and how to combine their powerful predictions with your irreplaceable clinical judgment.

Masterclass Table: Traditional Risk Scores vs. Predictive Analytics Models
Feature Traditional Risk Score (e.g., LACE+) Predictive Analytics Model
Number of Variables Typically 5-10 manually selected variables. Can analyze hundreds or thousands of variables (diagnoses, labs, medications, procedures, demographics, etc.).
Data Sources Primarily clinical data found in the EHR. Can integrate EHR data, insurance claims data, social determinants of health data, and even patient-generated data.
How it Works Simple, linear point system based on pre-defined weights. Easy to calculate by hand. A complex, non-linear algorithm (machine learning) “learns” patterns from historical data to make predictions. A “black box” to the end user.
Output A single score (e.g., “LACE+ score is 11”). A dynamic risk percentage (e.g., “This patient has a 42% probability of being admitted in the next 90 days”) and a list of the top contributing risk factors.
Pharmacist’s Role Calculate the score and use it to stratify patients. Review the high-risk patient list generated by the model, validate the clinical context of the top risk factors, and intervene.
A Practical Example: The “High Cost Patient” Predictive Model

An Accountable Care Organization (ACO) uses a predictive model to identify patients who are most likely to become “high-cost” (exceed $50,000 in healthcare spending) in the next 12 months. The model runs every night and produces a list of the top 100 highest-risk patients.

As the ACO’s pharmacist, you receive this list. For one patient, the model assigns a 78% probability of becoming high-cost and lists the top three contributing factors as:

  1. New prescription for oral chemotherapy (imatinib).
  2. Two recent ED visits for “nausea and vomiting.”
  3. Comorbid diagnosis of depression.

The algorithm has done the heavy lifting of identifying the at-risk patient. Now, your clinical brain takes over. You immediately recognize this pattern: The imatinib is causing severe nausea, which is leading to ED visits and likely poor adherence, all of which is exacerbated by the patient’s depression. The model found the “what,” but you understand the “why.” Your intervention is now crystal clear: a CMM visit to proactively manage the chemotherapy side effects, collaborate with the oncologist on an antiemetic regimen, assess adherence, and coordinate with the behavioral health team. This is the powerful symbiosis of advanced analytics and expert clinical practice.

11.2.5 The Indispensable Human Element: Clinical Judgment as the Final Filter

Data, registries, risk scores, and predictive models are incredibly powerful tools. They provide the map and the compass for navigating your patient population. But they are not, and will never be, a substitute for your clinical judgment. An algorithm can tell you that a patient has a high risk score, but it cannot tell you that the patient is a non-English speaker with low health literacy who is the primary caregiver for a spouse with dementia. An EHR registry can identify a patient who is not on a statin, but it cannot know that the patient had a documented case of rhabdomyolysis from a statin ten years ago. You are the final, essential filter that translates data into safe and effective patient care.

“Trust, but Verify”: The Guiding Principle of Data-Driven Pharmacy Practice

You must treat the output of any risk stratification tool with the same professional skepticism you apply to a new prescription. Always ask yourself these critical questions when reviewing a list of high-risk patients:

  • Is the data accurate? Does the problem list reflect the patient’s current state? Are the lab values recent? Is the medication list reconciled? A model is only as good as the data it’s fed.
  • What is the clinical context? Why is this patient high-risk? Does the model’s prediction make clinical sense? A high LACE+ score is less concerning in a patient being discharged to a skilled nursing facility with 24/7 care than in a frail elderly patient being discharged home alone.
  • What does the patient want? The data may suggest that adding a fourth antihypertensive is the “right” thing to do, but the patient may be experiencing significant pill burden and side effects. Your role is to balance evidence-based guidelines with the patient’s goals of care and personal preferences.
  • Is there a “non-data” factor I’m missing? Risk scores rarely account for social determinants of health, caregiver support, health literacy, or cultural factors. Your holistic assessment of the patient is the piece that the algorithm can’t see.