Across 4,366 US hospitals, coordination R is a deconfounded predictor of risk-adjusted mortality in community/regional Acute Care (2026 cross-section, partial r = −0.20, n=2,777). The readmission finding replicates across a nine-year longitudinal panel 2014–2023 (n=34,898 hospital-years, pooled β = −0.57, 95% CI excludes zero). Enter your hospital's CCN below to see where you sit relative to peers.
Partial r by stratum, year fixed effects, deconfounded for SVI and network density
Bands are 95% bootstrap CIs. Pooled panel with year fixed effects, Community Acute Care: β(R) = −0.566, 95% CI [−0.846, −0.293]; CI cleanly excludes zero. Replicates the cross-sectional readmission finding.
The lookup tool above shows your hospital's Coordination R and where it sits relative to peers in your stratum. This chart shows why R matters: across nine years of CMS data, in community Acute Care, hospitals with higher R consistently have lower 30-day readmission rates. The green line trends below zero across every measured year, and the shaded confidence bands sit clearly below the zero line — meaning the relationship is statistically robust across a full business cycle, not a single-year fluke.
For a community Acute Care hospital, a higher R is associated with measurably lower readmission, after controlling for community disadvantage (SVI) and PAC network density. For specialized facilities (Level I trauma centers, COTH major teaching hospitals), the signal is weaker — the amber line hovers near zero with confidence bands that cross it. Different optimization regimes likely apply at those facility types, which is why the per-hospital report frames recommendations differently for each stratum.
How to use both views together: the per-hospital report above tells you your hospital's R and where you sit on the readmission distribution. This chart tells you how trustworthy the link between the two is in your stratum. Both are inputs to a conversation about whether improving coordination is a meaningful lever for your facility — strong evidence for community Acute Care, more nuanced for specialized centers.
Select your role to see what the research means for the conversations you're in.
In the national longitudinal panel of 34,898 hospital-year observations (2014–2023), community / regional Acute Care hospitals' coordination metric R is associated with CMS Hybrid Hospital-Wide Readmission at β = −0.566 after year fixed effects (95% CI [−0.846, −0.294]). The coefficient is stable in the −0.07 to −0.10 range in every year individually. This is the CMS measure that determines HRRP penalty exposure — and a hospital sized 350–450 beds typically has $0–3M of annual readmission penalty risk on the table.
The signal is local-network, not regional. A parameter sensitivity analysis showed that narrowing the network definition to a 30 km radius strengthens the association by 18 percent; broadening to 60 km weakens it by 26 percent. Translation: the operational ROI lives in the providers already in your immediate 30 km PAC neighborhood. You do not need to wait for state-level infrastructure changes for Addie to deliver value.
Boarding-cost benchmarks: peer health systems your size bleed $12M+ annually in boarding costs (42-hour mean × $93/hour blended). Early Addie adopters compress boarding to under 24 hours — a recovery in the $5–7M range for a typical mid-size facility.
The longitudinal coupling is the credibility layer. The boarding math is the close.
Sources: Longitudinal panel n=34,898 hospital-years · Parameter sensitivity analysis (radius perturbation) · Industry boarding-cost benchmarks
A common reviewer concern is whether the coordination metric is just a proxy for "high-R hospitals happen to have better-quality PAC providers nearby." We tested this directly by adding distance-weighted HHA and SNF Care Compare star ratings as additional deconfounding controls. The community Acute Care partial r shrinks by only 1.1 percent — from −0.188 to −0.186, with the 95% CI still cleanly excluding zero. The coordination structure carries information independent of provider quality.
Translation for clinical leadership: improving your coordination structure delivers outcome benefits even if your individual provider quality stays the same. Addie acts on the structural arrangement of your network — Day-1 PAC need prediction, real-time matching by clinical capability, automated acceptance loops — not on rating your existing providers. The two interventions are complementary but distinct.
The mortality association in the 2026 cross-section, partial r = −0.203 (95% CI [−0.236, −0.170]) after deconfounding for SVI and network density, survived every sensitivity test in our supplementary battery (eight tests including E-value sensitivity, permutation, FDR correction, and ±30% parameter perturbation). The dual-null finding for Level I trauma and AAMC COTH major teaching hospitals replicates under two different mortality measures — a real organizational property, not a measurement artifact.
Structure first. Provider quality second. Addie acts on structure.
Sources: 2026 cross-section (n=2,777, partial r = −0.203) · PAC quality independence test (±1.1% shrinkage with star-rating controls) · 8-test sensitivity battery
A common technical question is what the model actually optimizes against. The coordination metric R we use in our research is statistically independent of PAC provider quality (1.1 percent shrinkage when controlling for distance-weighted star ratings). The model captures the spatial and structural topology of your PAC network — the arrangement, density, and clinical fit of providers — with provider-quality information as a secondary filter rather than the primary signal. This solves a different problem than referral-quality tools like Olio or WellSky's analytics suite, which optimize against provider performance directly.
Technical profile: 92 percent predictive accuracy on PAC need at admission, Mayo Clinic Platform-powered, Epic-native, deploys in 30 days. Compare to WellSky and naviHealth deployments that routinely take 4–6 months. Your PAC neighborhood has dozens to hundreds of providers within 80 km. The fragmentation signal in your downstream data is what Addie's real-time matching resolves: filter by clinical capability, bed availability, payer alignment, and geographic fit; send referrals automatically; track acceptance status in one unified view.
All findings backed by an eight-test sensitivity battery: E-value, permutation, multiple-comparison correction, PAC quality independence, residual quality deconfounding, negative-control outcomes, missing-data sensitivity, and Kuramoto parameter sensitivity. Each test confirms the structural-coordination signal is independent of the obvious alternative explanations.
Structural prediction at admission, not quality recommendation at discharge.
Sources: 8-test sensitivity battery (E-value, permutation, FDR, quality independence, residual deconfounding, negative controls, missing-data, Kuramoto parameters) · Model card (Mayo Clinic Platform validation)
Industry benchmarks place case-manager time at 2–3 hours per discharge on referral matching, phone tag, and faxing. Addie automates the matching and acceptance loop. At 30–50 discharges per day for a mid-size facility, that is 60–150 case-manager hours per day reclaimed — roughly 4–6 FTE of high-touch transition capacity at no incremental headcount. The hours go back into the high-judgment work case managers do best.
The measure that determines your HRRP penalty is CMS Hybrid Hospital-Wide Readmission. Our nine-year national panel shows that in community Acute Care hospitals β(R) = −0.566 (95% CI [−0.846, −0.294]), cleanly excluding zero across 2014–2023. Hospitals with higher coordination R have meaningfully lower risk-adjusted 30-day readmission. The workflow time you reclaim with Addie produces that signal at the hospital level — the structural improvement and the operational benefit are the same intervention.
Reclaim the hours. Move the measure.
Sources: Industry case-manager workflow benchmarks (2–3 hrs/discharge) · Longitudinal panel n=34,898 hospital-years (β = −0.566 readmission)
How the nine-year panel corroborates the cross-section, and where the two analyses diverge.
The cross-sectional analysis above is corroborated by a nine-year national panel constructed from CMS Care Compare annual refreshes spanning 2014 to 2023 (Rush Quality Analytics curated archive). After harmonizing condition-specific 30-day mortality measures (MORT-30-AMI, MORT-30-HF, MORT-30-PN, MORT-30-COPD, MORT-30-STK) into a within-year z-averaged composite and joining to CMS Hospital-Wide Readmission, we ran pooled OLS with year fixed effects and cluster-robust standard errors at the hospital level. The community Acute Care readmission finding is robust across all nine years (per-year partial r between −0.07 and −0.10, significant in 5 of 6 years post-2018; pooled β = −0.566, CI [−0.846, −0.293]). The mortality finding is directionally consistent but smaller under the noisier composite measurement.
The CMS Hybrid Hospital-Wide Mortality measure (Hybrid_HWM) used in the 2026 cross-section was introduced in FY2022 and is not available across the full 2014–2023 panel window. The composite-z mortality outcome we use for the longitudinal analysis is a defensible but methodologically distinct proxy. As a result, the dual-null finding for specialized facilities (Level I trauma centers + AAMC COTH major teaching) seen in the 2026 cross-section does not replicate cleanly in the panel composite — those facilities actually show the largest panel β. The most parsimonious interpretation is that the cross-section's dual null is specific to all-cause hospital-wide mortality, and the panel composite captures condition-specific outcomes where trauma and academic centers' specialization shows up. We discuss this distinction in the eventual paper but visitors browsing the lookup tool should treat the per-hospital report as the cross-section's interpretation, which uses Hybrid_HWM.
A 45-minute conversation with the PEP Inc. team can walk through your hospital's coordination profile, peer comparisons, and what improvement levers are realistic given your trauma level, teaching status, and case mix.
Schedule a Conversation →