The cohort started from 100 deidentified inpatient users.
Evidence Brief
Acute Inpatient Evidence.
PI was applied to a deidentified MIMIC-based inpatient cohort. This page summarizes what the latest hybrid cohort artifacts support, what PI demonstrated, and what it does not claim.
This brief reflects the latest hybrid inpatient cohort report dated April 22, 2026. It is an evidence summary, not a clinical validation claim.
At a glance
What this evidence covers.
PI reviewed a multi-domain inpatient cohort instead of forcing one narrow disease lens first. It looked for the current organizing burden, the broader structure around it, and the points where action should still stay gated.
Only cases with enough structure moved into the full hybrid pass.
The usable cohort spanned 177 admissions.
The latest report also covered 70 ICU stays.
What PI demonstrated
What the run shows in plain English.
The main result is not a single score. It is that PI could orient around the live inpatient picture, keep explanation and action separate, and stay honest when the action lane was not ready.
The main story could emerge across systems
PI did not force the case into one domain up front. Infection, renal, respiratory, metabolic, and hemodynamic pressure could all compete to explain what mattered most.
Stage 3 used more than single edges
The run combined confirmed direct relationships, grounded interactions, recurrent pathways, and structural reinforcement before summarizing the case.
Explanation stayed stronger than intervention
PI often produced a clear explanation and an actionable-with-caution lane, while still withholding recommendation-grade action when governance did not support it.
Current posture
What the usable cohort looked like.
- 29 cases showed stronger mixed support.
- 8 cases remained exploratory only.
- 36 cases surfaced an actionable-with-caution lane.
- No usable case released recommendation-grade intervention.
Why this matters
Orientation can still be useful before intervention is ready.
In an inpatient setting, it is better for PI to say what seems to be organizing the case, what deserves attention first, and why it is stopping short of governed intervention than to force a stronger action claim than the evidence supports.
Representative cases
Two examples from the latest cohort report.
These two cases show the current strength of the inpatient run: PI can orient around the live picture, show supporting structure, and keep action separate from explanation.
Case A
Respiratory pressure shaped the main inpatient story.
179-day episode
108
Confirmed edges
9
Confirmed interactions
1,357
Confirmed chains
- The case cleared full Stage 3 sufficiency across 179 observed days.
- The current anchor target was Sustained Acute Deterioration Pattern.
- The main explanation centered on Respiratory Compromise Pattern driving that anchor target.
- PI separately surfaced Infection Signal Pattern to Creatinine as actionable with caution.
- Recommendation-grade action on the main focus target remained gated.
Case B
Renal stress sat underneath the deterioration picture.
65-day episode
183
Confirmed edges
42
Confirmed interactions
823
Confirmed chains
- The case cleared full Stage 3 sufficiency across 65 observed days.
- The main explanation centered on Blood Urea Nitrogen plus Sustained Renal Instability Pattern.
- Creatinine kept showing up as a hidden lever and watch signal in the same case.
- PI surfaced Sustained Renal Instability Pattern to Oxygen Saturation as actionable with caution.
- Recommendation-grade action on the main focus target remained gated here as well.
Cohort map
What the inpatient cohort showed overall.
Above the individual cases, PI also summarized what kept repeating across the cohort: the main burden families, the most common mechanisms, and the structural roles that kept appearing.
Recurring burden families
- Sepsis or infection appeared in 35 of 37 usable cases.
- Renal structure appeared in 34 of 37 usable cases.
- Metabolic or electrolyte signals appeared in 20 cases.
- Respiratory and hemodynamic structure still repeated across the cohort.
Recurring mechanisms
- Infection Signal Pattern to Sustained Acute Deterioration repeated in 24 patients.
- Infection Signal Pattern to Sustained Renal Instability repeated in 23 patients.
- Sustained Renal Instability to Sustained Acute Deterioration repeated in 22 patients.
- Blood Urea Nitrogen to Creatinine repeated in 21 patients.
The same pressure points kept showing up.
Sustained Acute Deterioration Pattern most often behaved like the cohort bridge node, Infection Signal Pattern most often behaved like a hidden lever, and Creatinine appeared repeatedly as a vulnerability point.
Structure was present even without full analytic support.
Across the usable cohort, PI surfaced 1,827 bootstrap-confirmed edges and 905 structurally reinforced confirmed edges, while SEM support was not available in this slice.
The run was stronger for explanation than intervention.
The latest report explicitly reads as explanation over intervention: most cases still supported orientation and caution more strongly than governed action release.
Governance
What kept the report honest.
- 63 users were sufficiency-gated before Stage 3.
- Relative-day freshness was used because dates were deidentified and shifted.
- Explanation, recommendation, and actionable-with-caution lanes stayed separate.
- Abstention reasons were preserved instead of hidden.
- Interaction and pathway summaries were only surfaced when supported.
Claim boundaries
What this page does not claim.
- This is not a diagnosis or treatment recommendation page.
- This is not a clinical validation claim.
- The current cohort supports explanation more strongly than governed intervention.
- SEM support was not available in this run.
- Many abstentions came from missing recommendation-grade target support, missing unit-safe effect expression, and incomplete clinical grounding.
Discuss the evidence
Talk through the inpatient findings.
The best discussion is usually about fit and next evidence: what PI already demonstrated here, what remains gated, and what should improve in the next clinical cohort.