Vestibule: A Post-Discharge Transition Agent with a 24h/48h/72h/7d Call Cadence

Voice-first structured follow-up calls in the window that the readmission literature identifies as load-bearing — pinned against Jencks et al.'s 19.6% Medicare 30-day readmission baseline.

Abstract Jencks et al. (NEJM, 2009)[1] established the modern post-discharge readmission baseline: 19.6% of Medicare beneficiaries are rehospitalised within 30 days, with 50.2% of medical readmits having no interim outpatient visit. Forster et al. (Annals, 2003)[9] reported that 19% of discharged patients experience a post-discharge adverse event, 66% of those being adverse drug events, with 6% preventable and another 6% ameliorable — placing the highest-leverage intervention window in the first three days. Jack et al.'s Project RED (Annals, 2009)[5] showed that a pharmacist call 2–4 days post-discharge reduces hospital utilisation by 30%; Coleman's Care Transitions Intervention[6] cut 30-day readmissions from 11.9% to 8.3% (OR 0.59); Naylor's APN-led transition[7] reduced rehospitalisations by ~36% and cost by $4,845/patient at 52 weeks; Schnipper's pharmacist-counseling-plus-followup[10] reduced preventable ADEs from 11% to 1% at 30 days. Vestibule is the AI-augmented operationalisation of this evidence: structured voice calls at 24h, 48h, 72h, and 7d post-discharge, condition-specific red flags, and escalation to a transitional-care nurse on uncertainty. Pass criterion: ≥ 50% absolute call-completion rate at 7d (vs Forster IVR's 64%[11]) and a documented trend toward Hansen meta-analysis[8] readmission reductions (3.6%–28% absolute) on a 200-patient pilot.

§ 1 Introduction

The post-discharge transition is one of the highest-leverage intervention points in modern care. Jencks et al.[1] showed that nearly one in five Medicare admissions are rehospitalised within 30 days; Krumholz[2] named the underlying phenomenon "post-hospital syndrome" and demonstrated that the cause of readmission matches the index diagnosis in only 37% of HF, 29% of pneumonia, and 36% of COPD cases — supporting condition-agnostic monitoring rather than narrow disease-specific protocols. The Wasfy et al. pre-post analysis[3] showed CMS's Hospital Readmissions Reduction Program shifted readmissions downward by 76.6 per 10,000 discharges per year after HRRP passage versus +0.5 before. The financial-and-clinical incentive is well-aligned.

The intervention literature is unusually consistent for healthcare research. Project RED[5], Coleman's Care Transitions[6], Naylor's Transitional Care[7], and Schnipper's pharmacist-counseling-plus-followup[10] all show meaningful effects with a common architecture: structured contact within the first week, medication reconciliation, follow-up appointment confirmation, and a clear escalation path. Hansen et al.'s systematic review[8] documents that multi-component bundles produce absolute risk reductions of 3.6%–28%, with post-discharge phone calls a common feature in all four positive RCTs.

What is missing is an AI-augmented operationalisation of this established intervention pattern at the scale and consistency human-staffed programs cannot reach. Vestibule fills that gap.

1.1 Contributions

1. A voice-first post-discharge agent with a four-call cadence (24h, 48h, 72h, 7d) derived from the Forster[9] and Schnipper[10] ADE-timing evidence.
2. An open implementation of the Project RED[5] and Coleman[6] intervention checklist as a deterministic agent skill suite.
3. A safety-pinned evaluation harness that measures both call completion (an AI-augmentation lever) and 30-day readmission (the binding clinical outcome).

§ 2 Background and Related Work

2.1 The Readmission Baseline

Jencks et al.[1] remains the canonical citation for 30-day readmission epidemiology — 19.6% Medicare, with 50.2% of medical readmits lacking interim outpatient contact. Krumholz[2] reframed the phenomenon as post-hospital syndrome, an acquired transient state of generalised risk that argues for condition-agnostic monitoring. Wasfy et al.[3] documents HRRP's downward effect at population scale. Wadhera et al.[4] sounds the counterweight: HRRP was associated with an absolute increase in 30-day post-discharge mortality of approximately +0.52 percentage points for heart failure and +0.44 percentage points for pneumonia (interrupted-time-series, comparing the 2012–2015 implementation period to the 2010–2012 baseline) — implying that pure readmission-avoidance can shift harm into the community. Vestibule's safety-monitoring design responds to this directly: the call cadence is for safety detection, not for readmission gaming.

2.2 Effective Transition Interventions

Project RED (Re-Engineered Discharge)[5] achieved a 30% reduction in hospital utilisation through a multi-component bundle including a pharmacist call 2–4 days post-discharge. The Coleman Care Transitions Intervention[6] cut 30-day readmission from 11.9% to 8.3% via a home visit plus three follow-up calls (OR 0.59). Naylor's transitional-care RCT[7] using APN-led transitions reduced rehospitalisations by ~36% and cost by $4,845/patient at 52 weeks. Hansen et al.'s systematic review[8] describes the meta-analytic effect-size range of 3.6%–28% absolute risk reduction across multi-component bundles.

2.3 The Medication Window

Forster et al.[9] remains the definitive source for post-discharge adverse-event timing: 19% of discharged patients have a post-discharge AE; 66% of those are adverse drug events; 6% preventable and another 6% ameliorable. The most-onset-dense window is within 72 hours. Schnipper et al.[10] showed that pharmacist counseling plus 3–5 day follow-up call reduced preventable ADEs from 11% to 1% at 30 days. Vestibule's 24h/48h/72h cadence is directly motivated by this evidence — the first three days are where the AE onset distribution peaks.

2.4 AI / IVR Follow-Up Systems

Forster et al.'s 2009 IVR feasibility study[11] achieved 64% completion of automated calls with problems detected in approximately 30% of respondents — establishing the baseline call-completion bar for any automated follow-up system. The Mistiaen Cochrane review[12] covered 33 RCTs and found heterogeneous effect: call design, scripting, and timing drive outcomes more than the existence of the call itself — directly motivating Vestibule's structured-skill architecture rather than a generic conversational agent.

2.5 Regulatory Context

CMS's HRRP[13] tracks six conditions — AMI, HF, pneumonia, COPD, CABG, elective hip/knee — with penalties up to 3% of base DRG payments. This is the ROI lever Vestibule operates against.

§ 3 Proposed Approach

3.1 Four-Call Cadence

3.2 Call Structure

Each call follows a fixed skill suite derived from the Project RED[5] and Coleman[6] intervention checklists: (1) condition-specific red-flag screen (HF weight, COPD dyspnea, AMI chest pain, post-op infection signs); (2) medication adherence and side-effect check, with explicit reconciliation against the discharge medication list; (3) follow-up appointment confirmation, scheduling assistance if not yet booked; (4) caregiver support check; (5) escalation pathway clearly stated. Call 3 (72h) adds a structured ADE inventory aligned with Forster's published distribution[9].

3.3 Escalation

Three escalation tiers, identical to Calline's pattern. Routine findings flow to a daily report the transitional-care nurse reviews. Escalate findings (deterministic red flag, missed PCP appointment, or model uncertainty above threshold) trigger an RN callback within four business hours or a clinic visit. 911 findings (life-threat red flags — crushing chest pain, severe dyspnea, focal neuro symptoms) trigger immediate hang-up-and-dial-911 with offer to conference.

§ 4 Evaluation Protocol

§ 5 Expected Contributions

1. System. An open AI-augmented post-discharge transition agent operationalising the Project RED / Coleman / Naylor / Schnipper intervention pattern.
2. Methodology. A safety-pinned evaluation harness with the Wadhera mortality gate as a hard non-negotiable.
3. Operational evidence. The first published cross-condition completion-rate and red-flag-recall numbers for a voice-AI transition agent at pilot scale.

§ 6 Limitations and Risks

The Wadhera et al.[4] finding that HRRP shifted harm into post-discharge mortality is the deepest design risk for Vestibule. A successful readmission-reduction tool that increases community mortality is not a successful tool. The mortality-monitoring gate is therefore architecturally non-optional, not a v0.2 addition. The Mistiaen Cochrane finding[12] that call effects are heterogeneous across designs means Vestibule's specific skill suite is the load-bearing claim; a future RCT would test the exact composition.

A separate concern is digital exclusion: voice-only delivery may miss patients without reliable phone access — exactly the high-readmission-risk cohort the intervention most needs. A SMS-and-portal-fallback path is required and is a v0.2 task.

§ 7 Conclusion

The post-discharge transition is the rare clinical problem where the intervention literature is consistent, the financial incentive is aligned, and the operational scale of human-staffed programs is the limiting factor. Vestibule is the AI-augmented operationalisation of that consensus pattern, with explicit safety gates derived from the Wadhera mortality finding. The result that the field can act on is not whether structured follow-up calls work — Project RED, Coleman, Naylor, and Schnipper already settled that — but whether AI augmentation can deliver them at the scale and consistency human nurse-led programs cannot.