Population health analytics built on impactability rather than risk score, with comparison groups designed in, so a programme result is evidence of what the intervention did rather than a description of what would have happened anyway.
Risk stratification identifies who will cost the most. It does not identify who your intervention can help, and those are different populations. The members at the very top are frequently already in active treatment, already engaged with a provider, or in a trajectory no care management programme will alter.
How likely is the outcome, and how likely is an intervention to change it. Those are two different questions and they are not the same model. Most plans have built only the first.
Predicting cost is straightforward. Changing it is not, and a model optimized for the first tells you very little about the second.
Select the highest-cost members in a year and follow them into the next. Their cost falls substantially, on average, whether or not anything was done for them. That is a statistical property of selecting on an extreme value, not evidence of an intervention working.
Programmes measured on pre and post cost for an enrolled high-cost cohort will report savings reliably, and those savings will be largely unrelated to the programme.
Risk score is not actionability
High risk and high opportunity are different populations.
Pre and post measurement proves nothing here
Without a comparison group the result is arithmetic rather than evidence.
A risk score without its drivers is a ranking, not a decision
Care teams need to know why this member is high risk, what changed and what intervention could plausibly help.
Models optimize prediction, not response
A model can forecast cost accurately and still be poor at deciding whom to enrol.
Care management capacity is the constraint
The binding question is who receives the limited slots, not who qualifies.
Enrollment churn caps the horizon
An intervention with a payback beyond expected enrollment cannot return anything to the plan.
Build the comparison group before the programme launches.
A matched comparison group identified at enrolment, from members who met the same criteria and were not served, converts a programme from an assertion into evidence.
Identification is the easy step. The work that determines whether anything changes is qualifying identified members for impactability and reachability, and then matching each segment to an intervention that actually fits.
Step 1
Define the outcome the programme is trying to change, specifically enough to measure.
Step 2
Stratify by risk to establish the candidate population, using models appropriate to the outcome.
Step 3
Qualify for impactability: whose trajectory can an available intervention plausibly alter.
Step 4
Assess reachability and enrollment horizon.
Step 5
Detect change rather than only rank level.
Step 6
Segment by intervention fit, including the segment already receiving appropriate management.
Step 7
Size against capacity. Rank the list to the number of slots that exist.
Step 8
Define the comparison group at enrolment.
Step 9
Deliver member-level work into the systems care teams already use, with reason for identification attached.
Step 10
Measure against the comparison group and retire what does not work.
Rank to the number of slots you have.
A list of eight thousand candidates for four hundred slots is not a targeting output, it is a sorting problem presented as an opportunity.
Risk models are widely available and increasingly commoditized. What determines whether a population health programme returns anything is impactability, intervention matching and a measurement design that can distinguish effect from regression.
Identify and Qualify
Risk Stratification
Models selected for the outcome in question rather than a single general score.
Impactability Assessment
Which identified members have a trajectory an available intervention can plausibly change.
Reachability and Horizon
Contactability, engagement history and expected enrollment duration.
Rising Risk Identification
Members trending toward high cost rather than already there.
Match and Deliver
Intervention Segmentation
Members matched to the intervention that fits their need, including the segment where the correct answer is no intervention.
Condition and Cohort Analytics
Reusable cohort definitions governed across programmes.
Social and Community Need
Need identified where a corresponding resource exists to address it.
Workflow Delivery
Member-level work delivered into systems care teams already use, with supporting evidence attached.
Measure Honestly
Comparison Group Design
Matched comparison populations defined at enrolment.
Intervention Effectiveness
Outcomes measured against comparison by segment and intervention.
Engagement Analytics
Who was reached, who engaged and who completed.
Programme Portfolio Review
Which programmes are producing measurable effect, which are not, and which should be retired or redesigned.
What CaliberFocus does, and does not do?
We will tell you when a programme cannot be shown to work. We also do not run care management, and we will say when the constraint is capacity rather than targeting, because in that situation better analytics changes nothing except which members go unserved.
The honest assessment is what each programme can realistically achieve and how hard it is to prove. Several are worth doing for clinical reasons while being difficult to evidence financially.
| Programme | Who it targets | What to expect |
|---|---|---|
| Complex care management | Highest-risk, multi-condition members | Clinically valuable, financially hard to evidence. Regression dominates unless a comparison group exists. |
| Rising risk intervention | Members trending upward but not yet high cost | More responsive population and more room to change the trajectory. Frequently the better investment. |
| Condition-specific programmes | Defined clinical cohorts | Measurable clinical outcomes and a cleaner comparison group. |
| Transitions of care | Members leaving an inpatient setting | Short horizon, high responsiveness, and readmission is a measurable outcome. |
| Medication adherence | Members with fill gaps | Highly actionable, measurable and fast. |
| Avoidable utilization | Members with repeated emergency use | Requires understanding why. Access, behavioural health and social need drive much of it. |
| Behavioural health integration | Members with co-occurring needs | Substantial impact potential, confidentiality constraints, and specialist capacity is the limit. |
| Social need programmes | Members with identified unmet need | Only screen where a resource exists. Effect is real and slow, and financial evidence is difficult. |
Rising Risk Is Usually the Better Investment
Members trending upward, with conditions that are progressing rather than established and with utilization patterns that are changing, have more room to move and respond more often.
A risk score identifies candidates. Four filters turn that into a list worth working.
| Filter | The question | Who it removes |
|---|---|---|
| Risk | Who is likely to experience the outcome we are trying to prevent | Members with low likelihood |
| Impactability | Whose trajectory can an available intervention plausibly change | Members already in active treatment, already well managed, or in irreversible decline |
| Reachability | Can we actually contact and engage this member | Members with no working contact, repeated failed outreach, or no engagement history |
| Horizon | Will they remain enrolled long enough for the intervention to return anything | Members whose expected enrollment is shorter than the intervention payback |
The Signal Should Explain Itself
A flag arrives with the evidence that produced it and the route it implies. A care team given a score has to investigate before it can act, and that investigation is the work the analytics was supposed to remove.
Claims are complete, reliable and late. For a population health programme the interesting signal is usually forward-looking, which means pharmacy, clinical, engagement and utilization-pattern data carry more of the weight than the volume of claims data suggests.
Claims and encounters
Condition history, utilization patterns and cost, complete but lagged.
Pharmacy
Fill gaps, new starts, therapy changes and non-fills can precede a clinical event.
Clinical data
Results, values and assessments that carry severity and control information claims cannot express.
Member engagement data
Contact history, channel response, prior programme participation and preference.
Provider relationship
Attribution and engagement, so the intervention fits the member’s actual care context.
Social and community data
Collected where a corresponding resource exists, with source and consent recorded.
Integration principles
Capture the calendar as data. Match the grain to the decision. Preserve what was known and when. Keep training data reproducible. Do not reproduce actuarial projections.
Population health analytics decides who receives limited clinical resource. That is a consequential allocation, it can produce or reduce disparity depending on how it is designed, and it should be as governed as any other decision affecting members.
Model and cohort governance
- Purpose, population, inputs, outputs, validation, owner and review date
- Cohort definitions governed and versioned
- Subgroup performance measured
- Clinical ownership for models directing clinical intervention
- Intended use recorded and enforced
Equity
- Identification, reach, engagement and outcome by population
- Reachability filters reviewed for disparate effect
- Social need data used to offer support, never for coverage, pricing or cost decisions
- Programmes assessed for whether they widen or narrow gaps
Member protection
- Contact frequency capped at member level across programmes
- Consent and preference honoured across channels and programmes
- Sensitive conditions handled under stricter rules
Evidence and control
- Comparison group definition retained with the programme
- Results reported with the method
- A named owner per programme
- A retirement path for programmes that cannot be shown to work
Population health programmes are usually reported on members enrolled, touches delivered and pre-post cost change. The first two are activity and the third is regression. These measure whether the programme did anything.
| Category | What we measure | Why it matters |
|---|---|---|
| Effect against comparison | Outcome difference versus a matched group that met the same criteria and was not served | The only measure that distinguishes the programme from regression |
| Targeting quality | Share of enrolled members who were impactable, reachable and remained enrolled | Determines whether capacity was spent on members who could benefit |
| Engagement | Reached, engaged and completed rates by segment and population | Usually the binding constraint on effect |
| Capacity use | Slots filled, time per member, and members served against members identified | The productivity measure and the one that surfaces whether targeting is working |
| Equity | Identification, enrollment, engagement and outcome by population | Whether the programme narrows or widens existing gaps |
| Portfolio | Programmes with demonstrated effect, and programmes retired or redesigned | A portfolio where nothing is ever stopped has not been evaluated |
Honest expectation setting
Introducing comparison groups will reduce the savings your programmes report, in some cases substantially, because the previously reported figure included regression. That is not a decline in performance, it is the first accurate measurement.
We will take one programme, assess how its population was selected, test how much of its reported effect is explained by regression, and rebuild the targeting on impactability, reachability and horizon. We will also design the comparison group you will need to evidence the next result.
Bring us a specialty or clinical setting where clinicians are spending too much time creating notes. We will assess where ambient documentation fits, what must remain clinician controlled, how it should integrate with your EHR, and how to measure whether it is actually reducing burden.
One conversation with people who have run these deployments, and a written readiness view you can use with or without us.
- AI Agents and Workflow Automation
- Voice and Conversational AI
- Document AI and Intelligent Processing
- Generative AI and Enterprise Copilots
- AI Strategy and Governance
- HCC and Risk Adjustment Analytics
Security & Compliance
