The Future of Risk Adjustment Analytics: Using Dynamic Intervention Planning for Precision Targeting

Blog Posts  |  19 April 2021  |  By Amanda Cohen, MPH

Written By: Scott Stratton, MPH, Amanda Cohen, MPH, Cheryl Reifsnyder, PhD

Proper risk adjustment brings major benefits to health plans and their members. There are four core benefits of risk adjustment:

  1. Periodically confirming the presence of chronic illnesses and preexisting health conditions among members ensures that the plan is sufficiently funded for their care, be it through increased premiums (Medicare Advantage), risk transfer payments (Affordable Care Act), or capitations (Medicaid).
  2. Evidence permitting, risk adjustment can also ensure proper compensation by detecting unsubmitted diagnoses yet to be confirmed.
  3. Earlier detection of health conditions means earlier and more appropriate treatment for the affected members.
  4. Precise and efficient risk adjustment targeting focuses the attention of health plans and healthcare providers on the right patients at the right time for the right reasons, while minimizing interventions with low likelihood of succeeding.1, 2

In this article, we explore how risk adjustment analytics can be used to guide precision targeting, both when selecting patients, and when selecting interventions. We also share how Pulse8’s patent-pending Dynamic Intervention Planning empowers you to make the most effective use of your limited intervention budget—saving you money while helping to improve your enrollees’ health.3

Risk adjustment analytics promote appropriately funded, better care

Good risk adjustment analytics allow for precision coding. Precision coding is complete, timely, and accurate. On the financial side, conditions confirmed late or not at all will deprive the plan of the incremental funding for the needed care. For patients, missed or nonspecific diagnoses can deprive them of the care needed to treat their core condition, to prevent or lessen its progression, and to stave off complications.4, 5

Nevertheless, diagnostic errors are far too common. Studies have shown that ICD-10 coding in patient medical records can have error rates of around 20% to 50%.6 Studies have also shown that more than 40% of patients’ chronic conditions are not reported at all.7 Even among claims that do reflect accurate, specific patient diagnoses, at least one source states that more than 30% fail to pass the Centers for Medicare & Medicaid Services (CMS) validation process because they fail to supply sufficient supporting documentation.7

Sound risk adjustment analytics enable you to catch and correct these types of errors. When paired with technologies that share these gaps with providers and their staff, those gaps can be addressed and documented before submitting the next claim.

The risk adjustment analytics process can both decrease administrative waste and increase income for both the health plan and for providers that are part of increasingly common value-based contracts. Knowing when the patient’s last encounters occurred enables the provider’s staff to determine whether the current chart has what’s needed, or whether a new visit with the patient should be scheduled. Being able to close the loop by uploading charts with the supporting documentation ensure that the diagnoses will pass muster with CMS or state Medicaid authorities, and be worth their financial maximum.5

Comprehensive risk adjustment analytics leverage the breadth and depth of data that health plans can supply. Deep analytics can find otherwise undiscovered nuggets buried in the dense, voluminous raw data, that suggest the existence of an undiagnosed condition, along with the level of certainty of that inference. This certainty enables far more precise estimation of the outstanding opportunities (that Actuarial and Finance need to forecast reserves and realizable revenue), and prioritization of pursuit of patients who have the greatest probability of confirmation, hence incremental revenue.1, 2

Precision targeting of interventions

“Interventions,” in health plan risk adjustment, refers to actions that can be taken to realize the opportunities noted above. From light to heavy “touch,” from inexpensive to costly, they include reminders (voice, text, email, snail mail), encounter facilitation, retrospective medical record retrievals (MRRs), and in-home assessments typically by PAs or NPs (IHAs).2 With such a wide variety of techniques and costs, how do you choose the best intervention for each patient?

The answer is by using precision targeting.

At Pulse8, our patent-pending method of precision targeting is called Dynamic Intervention Planning.3, 8 Its purpose is to empower plans to target those combinations of intervention, provider, and patient that are the “best bets” to realize the identified opportunity.3

The process begins with Pulse8’s algorithms for finding and sizing risk adjustment opportunities. It “confirms” conditions using official software certified or supplied by the regulatory entity, CMS, or state Medicaid agency. Next it finds conditions, confirmed in previous payment periods, that should continue but have not yet been confirmed in the current one – “persisting” opportunities. The final “finding” step is the most sophisticated: identify any condition that we believe are likely risk-adjustable, but have not been labeled as a formal diagnosis, so it remains a “suspect”. This algorithm draws from as broad an array and number of years of data that the plan can make available, such as:3

  • Traditional medical, institutional, and pharmacy claims (mail and retail)
  • Lab test results
  • Medical records and health assessments
  • Sociodemographic information, including certain Social Determinants of Health (SDoH)
  • Various reports and return files from CMS servers (EDGE, RAPS, EDPS)

This applies thousands of codified clinical rules and heuristics based primarily on published, peer-reviewed literature.3 These are complemented and adjusted by findings from Pulse8 mining and modeling of millions of patient-years of data, looking for similarities among members with a shared condition. When it finds such similarities, it reviews the remaining data to see if any of the similarities might be predictive. “Data Mining” allows the algorithm to identify select factors shared by members with a target condition. It then uses those factors to identify other members who might also have that condition.3

For instance, suppose a member is taking a certain medication that is associated with a high likelihood of having a specific medical condition. The algorithm would tag that member as potentially having that medical condition. If, however, the medication the member is taking has off-label uses that are not associated with the specific medical condition, then, absent contrary data, the implied strength of inference will be relatively low. Other factors including age of the evidence further contribute to the scoring. More evidence for that condition would similarly raise the confidence of the inference.3

Confidence, RAF, and the final output

The amount of support – confidence – determines whether there is enough to warrant the formal suspicion of that condition and include it in consideration of intervention activities.

The confidence is also used to “size” and “prioritize” that opportunity: if the condition is worth, say 10 Risk Adjustment Factor (RAF) points and it has 35% confidence, then the expected value will be 35% of the RAF, or 3.5 RAF. These scores are combined with demographic and condition interaction scores (if applicable) to calculate various risk scores to quantify the performance of the overall program.

RAF points are converted to dollars according to the rules of the line of business.

  • Medicare Advantage: each point translates to added premium per month for that member
  • Affordable Care Act: we work with the plan’s actuaries to estimate the contribution to expected Risk Transfer Payment
  • Medicaid: the RAF contributes to expected future capitations based on the state’s formulae

The algorithm performs all of these analyses in the context of the most current risk adjustment methodology for that Line of Business and locale.8 The analytics algorithm produces a risk score for each patient,3 then calibrates the confidence level for each score. This process includes a review of both member and provider historical patterns, to see if they cast any doubt on the calculated scores. It also includes a big-picture review of the reliability of each data source, to eliminate any data that might push the analysis toward false positive or false negative results.8

The final output is a precisely calculated, confidence-adjusted clinical risk adjustment score associated with each “gap,” or problem area, that the algorithm identifies. This score can be used for precision targeting of interventions to the places where they will be most effective.

As precise are these estimates of value, they lack a major component that goes a long way in determining whether that opportunity will be realized, that gap will be closed: human behavior.

Pulse8’s Dynamic Intervention Planning

We noted that there are many types of interventions to close risk adjustment and other care gaps:

  • In-home medical assessments
  • Medical records review
  • Encounter facilitations such as wellness visits, telehealth consultations, visits for high-risk patients that have been especially important during the COVID-19 pandemic
  • Creative campaigns—for example, taking opportunities to call members and offer incentives for them to visit their primary care physicians

It is important to target interventions to the time and place where they will be most effective.8 Many vendors in risk adjustment make most of their margin, not from the effectiveness of their programs, but rather from the number of interventions performed – as they are the ones performing them. This means they are likely to recommend interventions even in situations where the gap is likely to close on its own.3 Doing so puts their financial interests at odds with the plan’s interest in doing just enough and only the most efficacious interventions.

At Pulse8, we analyzed three years’ data for both Affordable Care Act and Medicaid Advantage groups of enrollees—and found that up to 60% of gaps close naturally during the payment year, without any interventions3 – but vendors recommended strongly that the plans do them, which they did, wasting money, time, and significantly inconveniencing their providers.

Why do so many gaps close on their own? One example: patients with multiple chronic conditions are likely to see multiple providers each year, each a few times. With so many claims, the odds are good that the core diagnoses will be submitted by at least one of the providers. But many other factors come to play:

  • How reliable is the member at making/keeping appointments?
  • How well does the PCP or specialist document the visits?
  • Does their staff abstract diagnoses completely and code them accurately?

These and many other behavioral factors have a big say in odds of closure. And that’s why we built Dynamic Intervention Planning to mine and model them, to size and prioritize the opportunities so those most likely to close are atop the ranked list, and those least likely dwell at the bottom. Want to ensure you chase only gaps where the expected return is twice the cost of intervening? Easily done.

Armed with this and a bunch of supporting analytics of expected yields and lift, the plan can determine how much it wants to spend and can see the likely lift that will yield. Will more dollars for this program yield enough additional lift? Easy to calculate and demonstrate to suspicious CFOs. Pulse8’s Calcul8 and Collabor8 dashboards follow the programs’ progress so you can adjust tactics as needed, early enough to make a difference.

How does it work in the real world?

Let’s look at an example health plan member to see how Dynamic Intervention Planning targets the right intervention for that member.

The risk adjustment algorithm analyzes all available data for this enrollee: prior claims information, pharmacy transactions, labs, lab results, doctor visits, and types of doctors seen.

Based on this information, the algorithm predicts that this patient has chronic type 2 diabetes. It also supplies the estimated likelihood of this condition. Next, it reviews the member’s claims record and finds that his claims submissions do not include record of type 2 diabetes.

Dynamic Intervention Planning evaluates possible interventions to activate with this patient. For instance:

  • Option 1: Encounter Facilitation. The plan might call the member to schedule an appointment with an endocrinologist or primary care physician (PCP) who could perform an examination and conduct lab tests to determine whether the patient does, in fact, have type 2 diabetes.
  • Option 2: In Home Assessment (IHA). Alternatively, the plan might schedule an in-home health assessment with the member by a Physician Assistant or Nurse Practitioner where he could also be evaluated for type 2 diabetes.
  • Option 3: Medical Record Retrieval (MRR). Rather than spend several hundred dollars on an IHA, the plan might instead opt to wait towards the end of the plan year and request the medical records from PCP, specialist(s), and/or facilities to code independently to find the diabetes.

The plan may choose one or more of these interventions for that member. Different health plans have different strategies, and Pulse8 supports all of them. It’s also dynamic because everything is recalculated upon each data refresh to leverage the latest data to move the patient up/down in the intervention listings, affecting the plan’s decision of what to do next.8

When this type of precision targeting is performed on a macro scale, Dynamic Intervention Planning has the potential to generate sizable financial gains and eliminate an extraordinary amount of waste.

The bottom line

We have consistently seen the following occur with its clients after shifting to Pulse8:

  • Clients find and confirm more conditions and more RAF, hence increase revenue
    • Risk-adjustable conditions per member increase: we find more conditions than before
    • Disease-related risk scores increase: due to more conditions and higher RAF
    • Risk Adjustment-related revenues PMPY increase
    • RAF Yield increases: single best metric, as it reflects the combined results of intervening and deliberately not intervening
  • Clients cut intervention expenses – IHAs and MRRs – by >20% from pre-Pulse8 by
    • 10%: Let top decile (10%) of prospective opportunities close on their own
    • 10%: Omit bottom decile (or 2) as likely wastes of intervention resources
    • 3-10% or more: Ensure each intervention expects to return a threshold multiple over its cost, say 1.5x (50% over break-even)

A recent 2-year analysis across our then-book of business yielded these amazing results for just Medicare Advantage:

  • Increased clients’ RAF-based revenue by over $1 billion
  • Decreased clients’ intervention costs by over $250 million
  • Total benefit over 2 years was over $1.25 billion, again just for Medicare Advantage

Learn more about the power of precision targeting

Health plans today have limited budgets with which to fund interventions, making it more important than ever to target interventions to those times and places where they will have a significant financial impact.

The key to interventions that reduce waste and help improve health outcomes is the use of powerful precision targeting.

If you would like to learn more about how Pulse8’s Dynamic Intervention Planning can help you use precision targeting to select financially advantageous interventions for your enrollees, register for our webinar


  1. Mitigating waste in healthcare with analytics and interventions. Veradigm. Updated January 4, 2019. Accessed March 18, 2021,
  2. Wrathall J, Belnap T. Reducing Health Care Costs Through Patient Targeting: Risk Adjustment Modeling to Predict Patients Remaining High Cost. EGEMS (Washington, DC). April 20, 2017. 5(2):4. doi:10.13063/2327-9214.1279
  3. Pulse8. Analytic Logic: Risk Adjustment Analytics, Financial Reporting, and Risk Mitigation V18.11.0. Accessed March 25, 2021.
  4. Medicare risk adjustment. foresee Medical. Accessed March 18, 2021,
  5. Advancing Academy of Professional Coders. Risk Adjustment Factor. AAPC. Accessed March 20, 2021,
  6. Weed K. What is a RAF Score? RCxRules. Updated 2021. Accessed March 18, 2021,,to%20care%20for%20a%20patient.
  7. Cassano HJC. Factor HCC with a Two-pronged Approach to Risk Adjustment. AAPC Updated August 1, 2012. Accessed March 20, 2021,
  8. Calcul8 Risk Adjustment Analytics & Reporting. Pulse8. Accessed March 15, 2021,
Spread the word

Pulse8   risk adjustment   risk adjustment analytics   dynamic intervention planning   precision targeting   Calcul8  

Related resources