Case-Mix Risk Stratification Table¶
Metric or dashboard — instantiates Risk-Adjustment and Benchmark Selection
A table or dashboard that groups cases by baseline severity or exposure before comparing outcomes.
When units differ in how hard their cases are, an overall comparison punishes whoever draws the tougher draw. Case-Mix Risk Stratification Table adjusts by partitioning rather than regressing. It names the baseline-risk dimensions that make cases harder or easier — severity, acuity, exposure profile — bins every case into strata along those dimensions, and compares outcomes within each stratum, where like really does face like. The comparison becomes a table: rows for strata, columns for the units being compared, cells holding stratum-specific outcome rates, and the stratified baseline serving as the fair standard. Its defining move is discrete grouping into risk bins and within-bin comparison, so that a unit handling sicker or riskier cases is judged against others handling equally sick cases, not against the easy average. It is a standing dashboard, not a one-shot regression: it shows where performance differs, stratum by stratum, rather than collapsing everything to a single residual.
Example¶
A hospital system compares surgical mortality across its five hospitals and finds one flagship center has the worst raw rate — an alarming headline. But that center is the regional referral hospital: it takes the sickest, highest-acuity patients the others transfer away. A Case-Mix Risk Stratification Table reframes the comparison. Each surgical case is binned by baseline risk — using severity markers like emergency versus elective admission, comorbidity burden, and age band — and mortality is tabulated within each risk stratum for all five hospitals.
Stratified, the picture inverts. Among the highest-risk cases, the flagship center actually posts the lowest mortality; its bad overall number came entirely from treating a caseload skewed toward the hardest strata. A community hospital that looked excellent overall turns out to be middling once you compare only its low-risk cases against everyone else's low-risk cases. The table's value is that it localizes performance: leadership can now see that the flagship excels on hard cases while a specific hospital lags on routine ones — a far more useful and fair signal than the raw league table it replaced.
How it works¶
- Name the stratifiers. Choose the baseline-risk dimensions that legitimately change expected outcomes — severity, acuity, exposure — before looking at results.
- Bin every case. Assign each case to a risk stratum, defining cutpoints or categories so each bin holds genuinely comparable cases.
- Build the stratified baseline. Compute each stratum's outcome rate as the within-bin standard, and lay units side by side within every stratum.
- Read across the table. Compare units stratum by stratum, and optionally recombine into a case-mix-adjusted summary that reweights each unit to a common case distribution.
Tuning parameters¶
- Number of strata — how finely cases are binned. More strata improve within-bin comparability but shrink cell counts toward small-sample noise; fewer strata are stable but leave residual case-mix inside each bin.
- Stratifying variables — which risk dimensions define the bins. Each added dimension sharpens fairness but multiplies cells and thins the table.
- Cutpoint placement — where category boundaries fall; poorly placed cuts leave heterogeneity inside a stratum.
- Standardization weights — the common case distribution used to recombine strata into one adjusted figure (e.g., direct vs indirect standardization), which shifts the summary.
- Minimum cell size — the floor below which a stratum's comparison is suppressed as too noisy to report.
When it helps, and when it misleads¶
Its strength is transparency: stratification is intuitive and auditable — anyone can see which bin a case fell in and how each unit did within it — and it localizes performance to specific strata instead of hiding it in one averaged number.[n1] It compares like with like without asking the reader to trust a model's coefficients.
Its failure mode is residual confounding within strata: bins are only as fair as the risk dimensions chosen, and an unmeasured severity factor still contaminates the within-stratum comparison — coarse bins can leave enough heterogeneity inside them to reproduce the very bias they were meant to remove. Fine stratification trades this for tiny, noisy cells and multiple-comparison temptations. The guarding discipline is to justify the stratifiers before seeing results, keep cells large enough to be stable, and treat a clean stratified table as fairer comparison, not proof of causation — the table shows adjusted differences, not why they occur.
How it implements the components¶
Case-Mix Risk Stratification Table fills the stratified-comparison side of the archetype — the machinery that makes a fair baseline by grouping rather than modeling:
risk_factor_specification— it names the baseline-risk dimensions (severity, acuity, exposure) that define the strata, making the assumed sources of difficulty explicit.benchmark_construction_rule— it builds each stratum's within-bin outcome rate as the stratified baseline, the standard each unit is compared against inside its bin.
It does not estimate a continuous residual from factor loadings (risk_adjustment_mapping, abnormal_residual_interpretation_rule) — that is Multi-Factor Performance Model — nor test whether the ranking survives alternative benchmark choices (alternative_benchmark_robustness_check), which is Alternative-Benchmark Sensitivity Grid.
Related¶
- Instantiates: Risk-Adjustment and Benchmark Selection — it constructs a stratified baseline that compares units within matched risk bins.
- Sibling mechanisms: Multi-Factor Performance Model · Style-, Sector-, or Case-Matched Benchmark · Benchmark Attribution Report · Alternative-Benchmark Sensitivity Grid · Pre-Registered Benchmark Policy · Out-of-Sample Benchmark Validation
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: The mechanism bins cases by baseline risk, computes within-stratum outcome rates, and compares units against like-for-like baselines, so its operative form is stratified analysis.
Nearest alternative: Representation, Specification & Plan — A table displays the results, but deriving risk-adjusted comparisons is the defining contribution.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Medicine & Healthcare
Origin pattern: Convergent development
Present-day reach: Specialized
Rationale: Healthcare outcomes practice established case-mix stratification so providers are compared within baseline-severity groups.
Related originating lineages:
- Statistics & Experimental Design — Stratified analysis supplies within-bin comparison, uncertainty, and residual-confounding cautions.
Review resolution: Medicine is primary because case-mix and acuity stratification were developed to classify patient populations for resource and outcome comparison. Statistics supplies the risk-grouping methods, so a convergent but specialized classification is warranted.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The stratification table and Style-, Sector-, or Case-Matched Benchmark both fight case-mix unfairness by grouping, and are easy to conflate. The difference is what they produce: the matched benchmark builds one comparator peer set that a single evaluated unit is measured against, whereas this table partitions the whole population into strata and compares every unit within each bin — a many-cell dashboard, not a single yardstick.
[n1] Case-mix adjustment — comparing outcomes only after grouping or standardizing for patient severity and baseline risk — is the standard method behind fair hospital and provider performance reporting, precisely because raw outcome rates confound harder caseloads with worse care. ↩