“Measures by population” only work if they remain comparable across sites, regions, and acuity profiles. Otherwise, the library becomes a set of disconnected scorecards that can’t support contracting, improvement, or oversight. The goal is to add population nuance without breaking auditability. We’ll connect segmentation choices to indicator design and oversight expectations using Outcomes Frameworks & Indicators and the commissioner view in Using Data for Commissioning & Oversight.
Why segmentation is the core problem (not dashboards)
Community services populations are rarely homogeneous. Two counties can have the same program model but different housing instability, rural travel time, caregiver availability, and clinical acuity. If you publish a single “follow-up rate” or “engagement rate” without segmentation, the measure becomes a negotiation artifact rather than a management tool.
A good measures library includes segmentation rules that are: (1) consistent, (2) based on data you can reliably collect, and (3) stable enough that trends remain interpretable. The library should explicitly state which segmentations are required for the measure to be used in contracting, and which are optional “analysis layers.”
Two oversight expectations you must design for
Expectation 1: Measures must be reproducible under audit
Whether oversight comes from a state agency, a managed care organization, or an independent reviewer, a common test is reproducibility: can another analyst recalculate the same results from the same source data? Segmentation rules that rely on undocumented judgment calls (for example, “high need” defined informally by a supervisor) tend to fail this test. The library must specify objective segment definitions (fields, codes, time windows) and how missing data is handled.
Expectation 2: Contract measures must reflect controllable performance, not pure case mix
Funders increasingly want outcomes-led contracts, but they also want defensible fairness. If a measure is driven mainly by case mix (for example, higher ED use in the most medically complex segment), the library should either include adjustment logic or pair the outcome measure with process measures that reflect controllable practice (post-discharge follow-up timeliness, medication reconciliation completion, safety plan updates).
Segmentation approaches that work in real services
1) Population and eligibility segments
Start with segments that already exist in operating reality: eligibility group (IDD waiver, behavioral health, older adult supports), age band, and program model (mobile team, peer support, supportive housing). These are usually collectible and stable. The library should define which measures are universal and which are population-specific.
2) Risk and acuity segments
Risk segmentation is where libraries often break. The safest path is to use a small number of risk tiers derived from consistent fields (recent crisis events, comorbidity flags, housing instability markers, prior utilization) rather than complex proprietary scores that cannot be explained to funders. If you do use a formal score, document the calculation and version, and include a “plain-language interpretation.”
3) Access and geography segments
Operational performance is influenced by travel time, broadband access, and workforce availability. A measures library can include geography stratifications (urban/rural, distance bands to service hubs) so commissioners can interpret timeliness measures fairly and providers can justify resource allocation.
Operational Example 1: Segmenting follow-up measures after hospital discharge
What happens in day-to-day delivery: A provider serving multiple counties builds a library measure for “follow-up contact within X days after discharge.” The workflow begins with discharge notifications (ADT feeds or manual alerts) entering a centralized queue. A care coordination lead assigns each discharge to a responsible staff member within the same business day. The library specifies segments: discharge from psychiatric inpatient vs. medical inpatient, and “high-risk transition” flags (no stable housing, recent overdose, or multiple ED visits). Staff document follow-up attempts using standardized codes; the quality team reconciles notifications weekly to ensure no discharge event is missing.
Why the practice exists (failure mode it addresses): Discharge follow-up measures often look poor not because staff don’t follow up, but because discharge events are missed or documented inconsistently. Segmentation is needed because follow-up feasibility and risk differ dramatically between a planned medical discharge with a caregiver and an unplanned psychiatric discharge with housing instability.
What goes wrong if it is absent: A single blended follow-up rate becomes contentious: rural counties look “worse,” high-risk transitions drive misses, and commissioners assume performance failure. Meanwhile, the real failure mode (unreconciled discharge alerts, unclear ownership of follow-up) remains hidden. Teams waste time arguing about fairness rather than fixing workflow gaps.
What observable outcome it produces: With segmentation and reconciliation rules, the organization can show separate rates by discharge type and risk tier, plus the size of each segment. Commissioners can see whether improvements are occurring where they matter most (high-risk transitions), and audit trails demonstrate that every discharge event was captured and assigned.
Operational Example 2: Risk-tiered engagement measures for SMI community teams
What happens in day-to-day delivery: A community mental health team builds an “active engagement” measure but stratifies it into risk tiers defined by objective criteria (recent crisis contact, recent hospitalization, or medication nonadherence flag). The library defines what counts as engagement (clinical contact, structured outreach, verified collateral contact) and assigns different expected contact cadence by tier. Supervisors run weekly caseload reviews using a tiered roster; when a high-risk client has no qualifying contact within the expected window, the case is escalated for outreach planning and, if needed, supervisor-led intervention.
Why the practice exists (failure mode it addresses): In SMI services, “average engagement” can look acceptable while the highest-risk clients are disengaged and cycling through crises. Risk-tiering prevents the measure from being diluted by stable clients who require less intensive contact.
What goes wrong if it is absent: Teams hit a single engagement target by focusing on easier-to-reach clients. High-risk disengagement is noticed only after crisis events. Commissioners see rising utilization but the provider cannot demonstrate proactive management because the engagement measure is not sensitive to risk concentration.
What observable outcome it produces: Tiered measures create a clear operational signal: high-risk missed-contact exceptions. Improvement becomes visible through reduced exceptions, fewer crisis escalations, and better alignment between staffing effort and risk. Evidence includes tier rosters, outreach logs, and documented escalation actions.
Operational Example 3: Functional stability measures in older adult services with fair comparisons
What happens in day-to-day delivery: A home-based services organization tracks functional stability using a standardized assessment item set captured at intake and at scheduled intervals. The library stratifies outcomes by baseline risk (falls history, cognitive impairment flag, caregiver availability, and home safety risk indicators). Field staff enter structured findings; supervisors validate assessment completeness; and a quality analyst calculates change scores using library-defined rules (time window, minimum interval, handling of missed reassessments). Monthly reviews focus on whether higher-risk segments are stabilizing relative to baseline, not whether they match low-risk segments.
Why the practice exists (failure mode it addresses): Without baseline stratification, programs serving more complex older adults appear “worse” even when delivering high-quality stabilization. Baseline-based segments prevent misleading comparisons and help commissioners understand which investments reduce deterioration and crises.
What goes wrong if it is absent: Providers are penalized for serving higher-risk clients or are incentivized to avoid them. Measures become politically charged and the contract discussion drifts toward “your clients are different” rather than “what did we do and did it work?” Internally, teams cannot tell whether deterioration reflects baseline risk, missed follow-up, or gaps in home safety interventions.
What observable outcome it produces: Stratified reporting shows credible improvement within risk segments, supported by assessment timestamps, completeness audits, and documented intervention pathways (referrals completed, safety modifications verified). Commissioners can make funding decisions based on stabilized high-risk cohorts rather than comparing apples to oranges.
How to keep segmentation from turning into complexity overload
Limit segmentation to what drives decisions
A library can define many possible slices, but only a few should be “standard reporting strata.” A good rule is: if no one will take a different operational action based on the segment, it doesn’t belong in the standard pack.
Use “paired measures” to balance fairness and accountability
For outcomes that are heavily case-mix-driven, pair them with process measures that reflect controllable practice. For example, pair ED utilization with care plan review timeliness, post-discharge follow-up, or medication reconciliation completion. This satisfies funder expectations for accountability while acknowledging risk.
Document your adjustment logic and keep it stable
If you apply risk adjustment or weighting, document it in the library with version control. Even simple tiering is a form of adjustment; it must be stable across time periods to preserve trend meaning. Any changes should be logged with an effective date and a bridge analysis explaining what would have happened under the prior definition.
Funder-ready output: make segmentation part of the “evidence pack”
When funders ask for performance evidence, include segmentation definitions and population mix alongside results. A commissioner can only interpret outcomes responsibly if they can see the distribution of risk tiers, the completeness of the fields used for segmentation, and the operational actions triggered by exceptions. This turns segmentation from a defensive argument into a transparent governance practice.