Hospital-at-home (HaH) delivers active hospital-level treatment in a patient’s residence as a substitute for conventional inpatient care. It is distinct from routine home health, remote monitoring after discharge, or a virtual ward that lacks hospital-level clinical responsibility. Early US demonstration work showed that selected acutely ill older adults could receive substitutive care at home with feasible clinical operations and favorable outcomes (1). A subsequent randomized trial found that home hospital care for selected acutely ill adults was associated with lower cost, fewer laboratory orders and consultations, and lower 30-day readmission, although the single-system sample and model design limit broad generalization (2).

Evidence accumulated across different countries and models. A 2012 meta-analysis of randomized trials reported lower mortality, readmission, and cost and higher satisfaction, but combined heterogeneous populations and forms of hospital substitution (3). An updated Cochrane review of admission-avoidance HaH concluded that it probably makes little or no difference to mortality, probably reduces subsequent residential care, and may improve satisfaction; HaH treatment duration was often longer even though initial facility stays were shorter, and cost estimates varied (4). A systematic review focused on chronic disease presentations found lower readmission and long-term-care admission without a mortality difference, again with heterogeneity across interventions (5).

The strategic appeal is clear. HaH can create an alternative to scarce physical beds, reduce exposure to hospital-associated harms, and support patient preference. Yet expansion converts a local clinical program into a distributed hospital. Every capability normally concentrated inside the facility—nursing, physician oversight, pharmacy, diagnostics, equipment, oxygen, transport, communication, security, infection control, escalation, and quality surveillance—must reach the home reliably.

The federal policy context has also changed. CMS launched the Acute Hospital Care at Home (AHCAH) initiative in November 2020. Its 2024 congressionally required evaluation found generally favorable mortality, readmission, and post-discharge spending patterns while identifying demographic and geographic differences in participation (6). MedPAC’s 2024 report examined program experience and policy choices, including patient selection, quality, payment, and data needs (7). In 2026, Congress extended AHCAH flexibilities through September 30, 2030, and CMS released additional program data covering April 2023 through September 2025 (8). This longer horizon supports investment but does not eliminate the need for evidence, standards, or local economic discipline.

This review examines HaH as an enterprise operating model. It asks what capabilities are required for safe scale, how leaders should interpret heterogeneous evidence, and how capacity, equity, workforce, caregiver burden, and payment should be governed together. We present this article in accordance with the narrative review reporting checklist.

A targeted narrative search was completed on August 12, 2026. PubMed/MEDLINE and Crossref were searched for peer-reviewed evidence. CMS, MedPAC, and Congress.gov were searched for current federal program, data, and statutory information. Reference lists of trials, systematic reviews, and policy analyses were screened for foundational studies.

Search concepts combined hospital at home, home hospital, hospital-level care at home, admission avoidance, early discharge, randomized trial, systematic review, implementation, patient selection, caregiver, equity, rural, cost, workforce, escalation, quality, and payment. English-language sources published between January 1999 and August 2026 were eligible. Controlled trials, quasi-experimental evaluations, systematic and qualitative reviews, economic evaluations, implementation frameworks, and official policy reports were prioritized. Older literature before 1999 was excluded because the modern US evidence base and cited pilot literature began in 1999.

Sources were excluded if they described post-discharge monitoring without substituting for hospital-level care, routine skilled home health, residential care, or inpatient-at-home programs lacking sufficient information to identify clinical accountability. Promotional accounts without methods were not used as evidence. The author screened records for relevance and synthesized findings into six operating domains: strategic capacity, selection, distributed clinical operations, quality and escalation, equity and caregivers, and payment. No meta-analysis or formal certainty rating was undertaken; conclusions reflect triangulation across evidence types.

The completed search approach is summarized in Table 1.

The completed evidence-synthesis exhibits are presented in Table 4 and Supplementary Table S1.

“Hospital at home” is not one intervention. Programs differ in whether patients enter from the emergency department, an inpatient unit, or the community; whether care fully substitutes for admission or accelerates discharge; the conditions treated; daily visit frequency; physician model; technology; diagnostics; geography; and whether postacute transitional care is bundled. A finding from one configuration should not be attributed to every program.

Economic results illustrate this point. In a Presbyterian Healthcare Services analysis, HaH costs were 19% lower than costs for comparable inpatients, with equal or better selected outcomes (9). At Mount Sinai, an HaH episode bundled with 30 days of transitional care was associated with shorter acute length of stay, lower readmission, fewer emergency visits and skilled-nursing admissions, and better patient ratings than matched inpatient care (10). The bundle makes it difficult to separate the acute-at-home effect from the transitional-care effect, but it demonstrates the value of designing the full episode.

Not every contemporary evaluation shows lower utilization or shorter treatment. A 2025 Kaiser Permanente pilot found longer acute and combined acute-restorative duration and no significant difference in 30-day emergency or hospital use measured from the acute phase, although escalation was uncommon and most controls expressed interest in the model (11). This does not invalidate HaH. It shows that program scope, restorative services, comparators, start-up stage, and endpoint definitions change the result.

Executives should therefore request an evidence map before approving a program. It should match target diagnoses, entry route, acuity, geography, age, caregiver assumptions, and service bundle to the studies being cited. Claims such as “HaH lowers cost” are incomplete without stating whose cost, over what period, and whether technology, transport, caregiver time, and postacute services were counted. Capacity claims should specify whether a home admission actually substitutes for an occupied staffed bed or adds an episode that would otherwise have been observed or treated as outpatient.

The appropriate objective is not to recreate every hospital process in the home. It is to provide the clinical functions required for the selected episode with equal or greater reliability. Some facility routines exist because resources are centralized, not because the patient needs them. HaH redesigns the production model; it should not merely move the room.

An HaH program needs a stated enterprise purpose. Common purposes include relieving seasonal bed constraint, supporting emergency flow, avoiding capital expansion, improving experience, reducing harm, extending specialty reach, or succeeding under population-based payment. Each implies a different design. A ten-census pilot may meaningfully relieve a small hospital but be immaterial to a large academic center. A program built for chronic obstructive pulmonary disease and heart failure may not address surgical congestion.

The capacity thesis should quantify the bottleneck. Leaders should model eligible arrivals by hour and day, consent, deployment success, average daily census, length of treatment, escalation, and substitution. The relevant denominator is not every inpatient; it is the population whose clinical and home circumstances fit the service. Start-up assumptions should be conservative because clinician referral and patient acceptance develop over time.

Value should be measured across the episode. Direct HaH cost includes clinical labor, command-center coverage, vehicles, courier, mobile imaging, laboratory, pharmacy, equipment, oxygen, technology, vendor fees, installation, retrieval, and waste. Shared hospital costs may not fall when one patient moves home. Physical capacity creates financial value only if a bed constraint is relieved, fixed expansion is deferred, or staffing can be adjusted safely. Under fee-for-service payment, the revenue and cost relationship differs from global budget or risk contracts.

Boards should require a sensitivity analysis. Variables include daily census, travel time, payer mix, nurse productivity, technology cost, escalation, length of episode, and whether freed beds are filled. A model that works only at perfect census or zero escalation is not resilient. The program should define minimum viable scale and a stop, redesign, or partnership threshold.

Eligibility is often presented as a checklist. In practice it is a clinical prediction, operational feasibility assessment, and consent process repeated over time. Diagnosis and physiologic stability matter, but so do trajectory, treatment intensity, cognition, mobility, medication access, home safety, utilities, geography, communication, and willingness. Selection must account for the program’s actual capabilities, not an abstract definition of home care.

Programs need explicit inclusions, exclusions, and gray-zone escalation. Criteria should be version controlled and reviewed against actual transfers, declined referrals, and near misses. An exclusion may indicate an essential safety boundary or an addressable service gap. If patients are repeatedly excluded because mobile imaging is unavailable after hours, the organization should decide whether to build that capability or narrow the program—not quietly vary practice.

Consent must be meaningful. Patients should understand that they are receiving inpatient-level care, how staff will enter the home, what monitoring occurs, who responds at night, possible out-of-pocket obligations, the role expected of family, and how return to hospital happens. Declining HaH should not disadvantage the patient. Qualitative evaluation of a randomized trial found that patients valued comfort, sleep, mobility, and relationship benefits while also identifying concerns that inform program communication (12).

Selection continues after arrival. A patient appropriate at noon may require escalation at midnight. Teams need trend thresholds, rapid reassessment, and a low-friction path back to the hospital. The goal is not a zero-transfer rate. An appropriate early transfer is evidence that surveillance and escalation worked. A suspiciously low rate may indicate conservative selection or delayed recognition.

6.1 Command capability and decision rights

The command function maintains situational awareness across homes. It should know patient status, scheduled and completed visits, medication and equipment delivery, pending tests, staff location, alerts, risks, and escalation readiness. Coverage must be continuous, with one accountable clinical leader per shift and unambiguous decision rights among the home team, remote physician, emergency medical services, vendors, and brick-and-mortar hospital.

Communication redundancy is essential. Consumer broadband or a single tablet cannot be the sole route for urgent care. Programs need tested alternatives for connectivity, power, severe weather, and vendor outage. The home address, access instructions, hazards, language, and emergency contacts must be available to responders while respecting privacy.

6.2 Mobile clinical supply chain

The HaH supply chain is a clinical function. Pharmacy, specimens, imaging, oxygen, meals when required, durable equipment, and waste must arrive within clinically meaningful windows. Inventory that is trivial inside a hospital can become a critical missing item miles away. Service-level agreements should define order cutoffs, urgent delivery, chain of custody, temperature control, failed access, retrieval, cleaning, and downtime.

Route design affects both economics and equity. Large geographic coverage increases access but reduces productive clinical time. Organizations can use geographic pods, mobile teams, community partners, or scheduled route density while preserving rapid response. Travel time should be measured as clinical capacity, not hidden in a vendor invoice.

6.3 Workforce model

HaH requires clinicians who can work independently in variable environments and collaborate through a command structure. Roles may include hospitalists, advanced practice clinicians, registered nurses, paramedics, pharmacists, therapists, social workers, care coordinators, and logistics personnel. The model should define which encounters must be in person, which may be virtual, who performs urgent assessment, and who can administer treatments under applicable law and policy.

Workload is not simply visits per shift. It includes travel, setup, documentation, communication, home complexity, specimen handling, equipment issues, and escalation. Safety provisions should address driving, lifting, infection exposure, animals, weapons, harassment, unsafe neighborhoods, weather, and lone work. Staff need check-in procedures and the authority to leave an unsafe environment without improvising patient abandonment.

Implementation evidence emphasizes early policies, stakeholder relationships, efficient admission, communication, skilled staffing, and attention to caregiver impact (13). These are core infrastructure, not change-management accessories. Referring clinicians must trust the program; otherwise eligible patients are never offered it.

6.4 Information and technology

Technology should support a clinical model rather than define it. Typical functions include vital-sign capture, video, messaging, scheduling, task management, escalation alerts, and documentation. Data should enter the legal health record with minimal reconciliation. Alert thresholds need named responders, expected times, and false-alert monitoring.

Technology failure must not make the home unsafe. Selection should not rigidly exclude people because they lack a device, broadband, or digital skill if the program can provide equipment, connectivity, training, or non-digital alternatives. A published HaH framework identified multiple dimensions—including people, population, delivery, technology, quality, and data—that help compare models more accurately (14). The lesson is to specify the operating configuration rather than market a device.

Table 3 summarizes the minimum operating capabilities required for reliable delivery.

Facility measures do not automatically fit the home. HaH quality should include mortality, escalation, emergency use, readmission, adverse drug events, falls, pressure injury, infection, delirium, function, experience, caregiver outcomes, and post-discharge use. Process reliability should include deployment delay, missed or late visits, unavailable medication, equipment failure, critical-result closure, and response time.

Denominators matter. Reporting only completed HaH episodes excludes people who were eligible but never offered care, accepted but could not be deployed, or transferred early. A full funnel—screened, eligible, offered, accepted, deployed, completed, transferred—reveals access and selection. Outcomes should be attributed from the initial decision to treat at home, with transfer counted as part of the episode rather than a separate failure that disappears from the data.

Risk adjustment and comparison are difficult because selection is intentional. HaH patients may be healthier than all inpatients but more complex than a narrow control group. Randomized evidence remains important, while operational programs should use carefully matched comparisons and transparent criteria. CMS’s 2024 evaluation and subsequent data releases are valuable because they create national surveillance beyond individual success stories (6,8).

Quality review should distinguish preventable and appropriate escalation. Categories may include expected diagnostic need, clinical deterioration despite appropriate care, delayed recognition, logistics failure, patient preference, home-safety change, or technology failure. A multidisciplinary review can then modify selection, capability, or protocol. The aim is learning, not suppression of transfers.

CMS reported that AHCAH participants were more likely to be White and urban and less likely to receive Medicaid or low-income subsidies than comparison beneficiaries, although program geography and state coverage contribute to these patterns (6). Equity cannot be assessed only after enrollment because selection and offer are major filters. Programs should stratify the full funnel by race, ethnicity, language, disability, age, payer, income proxy, rurality, broadband, housing, and caregiver availability where lawful and appropriate.

An inequity may arise from rigid criteria. Requiring broadband, a smartphone, an English-speaking caregiver, a private room, or family labor can systematically exclude people. Some requirements are necessary for a specific model, but the executive question is whether the hospital can provide an enabling support: hotspot, interpreter, paid aide, meal, transportation, home modification, alternative monitoring, or community partnership. A national policy analysis highlighted equity, caregiver, legal, operational, technology, quality, and payment considerations as interdependent scale issues (15).

Caregivers should not become an unpaid substitute for hospital staffing. A representative US survey found broad hypothetical acceptability of HaH but also underscored the tasks respondents might be expected to perform (16). A secondary analysis of a randomized trial found no evidence of greater caregiver burden in the home-hospital group, but the sample and program supports matter (17). Programs should explicitly list required, optional, and prohibited caregiver tasks; assess willingness and ability separately; train and support caregivers; provide 24/7 contact; and monitor burden.

Patients living alone should not be automatically excluded if a safe professional support plan is possible. Conversely, the mere presence of family does not establish capacity. Caregiver availability can change during an episode. Refusal or inability to provide unpaid work should trigger support or alternative care, not blame.

Rural design demands additional safeguards. A 2025 randomized clinical trial demonstrated that hospital-level care could be delivered to adults in rural settings with high experience ratings and greater activity, while cost and readmission were not significantly different (18). Rural scale may require longer travel, local workforce partnerships, redundant connectivity, and explicit transport thresholds. It may improve access for some patients while being unsafe for others at excessive distance; geography must be designed, not ignored.

As of August 12, 2026, federal law extends Medicare AHCAH flexibilities through September 30, 2030 (8). The extension provides a more credible planning window than repeated short increments. It also requires continued study and data collection, reinforcing that payment permanence and model design remain policy questions.

Payment should preserve hospital-level accountability for an inpatient-level episode while avoiding incentives for duplicative capacity or inappropriate selection. Paying the full inpatient amount can support readiness and start-up but may overpay lower-cost episodes if efficiency gains are not shared. A reduced rate may deter participation or encourage selection of the least complex patients. Bundled, episode-based, or population-based models can align incentives across acute and transitional care but require risk adjustment, stop-loss protection, quality safeguards, and clear attribution.

MedPAC has identified issues including program standards, data, payment, quality, and beneficiary protections (7). A durable approach could combine a base payment for the acute episode, explicit recognition of required readiness, and performance accountability across transfers and post-discharge outcomes. Payment should not depend on caregiver labor that is neither measured nor supported.

Payers should align eligibility and reporting to reduce fragmentation. A hospital cannot safely run fundamentally different clinical programs by payer. Contract differences may affect payment and prior authorization, but core safety, staffing, escalation, and quality standards should remain consistent. Medicaid participation is especially important for equity; state variation can otherwise make HaH disproportionately available to Medicare and commercially insured populations.

The program’s economics should be tested under alternative futures: fee-for-service, global budget, accountable-care risk, and withdrawal or modification of a waiver. Assets should be modular where possible. A hospital should know how staff, vehicles, technology, and vendor contracts would be redeployed if payment changes.

9.1 Standardization and research priorities

The contemporary US landscape includes diverse models and rapidly evolving technology, making clear definitions and comparable reporting essential (19). Research should identify which program components produce benefit for which patients, rather than treating the HaH label as the intervention. Broader reviews of alternatives to inpatient hospitalization can help separate evidence for substitutive HaH from observation units, rapid follow-up, and other models (20). Admission avoidance and early-discharge HaH should also remain analytically distinct because their starting points, fixed-cost effects, and caregiver demands differ (21).

Patient-selection research needs prospective validation across conditions and populations. Expert work has emphasized clinical suitability, home conditions, and social support while warning that digital readiness and literacy can become inequitable exclusion criteria (22). Modern criteria should build on early pilot experience (23) but be recalibrated to current diagnostics, therapeutics, monitoring, and response capacity. Programs should publish the number and characteristics of people screened out and why.

Preference research is equally important. Reasons patients accept or decline HaH include perceived safety, burden on family, comfort, prior home experience, and trust in the clinical team (24). Consent materials and enabling services should be tested with patients who decline, not only with successful participants.

Finally, regulation should establish a common floor for accountability, quality reporting, data, emergency response, and patient protection without freezing one operational model. Policy analysis has called for a deliberate regulatory course that preserves safe, high-quality care as the field expands (25). Multi-year national surveillance should link program configuration with outcomes, transfers, caregiver effects, equity, and total spending.

Table 4 lists board-level assurance questions and guardrails before scale.

The first phase is design and assurance. Leaders define the target population, capacity thesis, clinical model, geography, quality measures, equity supports, and payment scenarios. A multidisciplinary hazard analysis tests common and severe failures. Patient and caregiver representatives review consent, communication, and burden assumptions. The board approves bounded pilot authority and stop rules.

The second phase is a low-census reliability pilot. The objective is not immediate savings. It is to prove screening, consent, deployment, medication, diagnostics, visit completion, alert response, documentation, escalation, and home closure. Every delay and workaround is logged. Simulations test deterioration, power loss, connectivity failure, missed delivery, unsafe home, and transport delay.

The third phase is disciplined growth. Census increases only after reliability thresholds are met. Eligibility broadens one clinical cohort or geographic pod at a time. Staffing and logistics grow before demand, not after failures. Finance validates full episode cost and substitution. Contracting aligns payer rules where possible.

The fourth phase is integration. HaH becomes part of daily bed management, emergency flow, specialty planning, pharmacy, quality, and capital strategy. Referral is embedded in clinical work rather than dependent on a champion. Performance is compared with appropriate facility care and reported to the board. The program remains capable of pausing safely when weather, staffing, technology, or transport conditions make home care unreliable.

Benefits should be reviewed at defined census milestones rather than at an arbitrary anniversary. At each milestone, leaders should compare the original assumptions with observed eligibility, acceptance, travel, staffing, episode duration, transfer, payer contribution, caregiver demand, and net bed substitution. Expansion should pause when clinical reliability falls below threshold, even if demand is strong. Conversely, a clinically reliable program with weak economics should be redesigned transparently rather than preserved through hidden cross-subsidy.

The maturity test is whether HaH behaves as part of the hospital during pressure. During a respiratory surge, the service should accept appropriate patients without lowering selection standards. During a technology outage, teams should know the census and continue time-critical care. During severe weather, admission and visit thresholds should change before routes become unsafe. These operational responses should be rehearsed, measured, and incorporated into the hospital’s emergency operations plan.

This review integrates trial, review, implementation, economic, caregiver, equity, rural, and current federal-policy evidence. It treats HaH as a distributed hospital service and translates evidence heterogeneity into explicit executive design questions. The search strategy, full operating framework, and supplementary search details support reproducibility and practical use.

The review is narrative rather than systematic. It did not include duplicate screening, formal quality grading, or meta-analysis. HaH definitions, populations, and comparators vary substantially, limiting generalization. Several influential studies were conducted in integrated systems or mature programs and may not predict start-up performance. Observational comparisons remain vulnerable to selection bias, and early federal program results may reflect participating organizations with greater readiness. Cost studies differ in included resources and perspective. Policy and program requirements may change after August 12, 2026.

Hospital-at-home can be a credible substitute for conventional inpatient care for selected patients, but it is not a telehealth project. It is a distributed hospital operating model. Safe scale requires a quantified capacity thesis, dynamic selection, 24/7 command, mobile diagnostics and therapeutics, reliable logistics, integrated information, rapid escalation, workforce protection, caregiver safeguards, and an equity strategy that funds enabling support.

Boards should evaluate HaH on completed and failed episodes, net substitution, full cost, quality, workforce, and access—not census or technology adoption alone. Payment should support hospital-level responsibility while rewarding efficient substitution and avoiding selection or caregiver cost shifting. The 2030 federal extension creates time to build durable capability. Organizations should use that time to establish evidence, reliability, and accountability before treating HaH as routine capacity.