1. Overview: Total Literature and Search Sources
By publication year (library items)Gray = before 2019; *2026 is an incomplete year
By theme (one paper may belong to several themes)
Search and Verification Process
- Literature search: 25 English search queries covering flow, layout, emergency, operating department, ICU, logistics, simulation, digital twins, generative design, normal–epidemic dual use, wayfinding, healing environments, green and smart hospitals, children's hospitals, Chinese hospitals, etc.; the top 100 results of each query, 2,447 raw records in total, were deduplicated and added to the library.
- Close-reading annotations: 4 Chinese and English thematic reviews with 173 annotated papers in total (method, key findings, design implications and evidence strength/verification grade), merged with the search results by DOI.
- Crossref check: each DOI was queried to retrieve title, journal, year, authors, citation count and reference list, from which the “citation links among library items” were derived. Records whose DOI is not in Crossref (ACM, Zenodo, arXiv, Chinese DOIs, etc.) are kept but marked “unverified”.
View the 25 search queries and result counts
2. Key FindingsSynthesized from four thematic reviews and the literature library
The starting point has moved from experience-based circulation to real care pathways and operational data, then tested by simulation. Representative work:
Layout optimization generates candidates, DES tests queues and resource occupancy, ABM handles behavior and route choice; they should be combined. Shorter distances do not necessarily shorten visit time.
Ophthalmology outpatient comparison: linear corridors reinforce order and sightlines but may create bigger bottlenecks under high load; clustered layouts are more flexible, but coordination drops when sightlines are cut.
Staff-to-staff visibility is an independent spatial performance; single rooms redistribute work and risk, with no outcome that is better across the board.
AGV fleet size, loading rules and replenishment frequency interact; under low demand net labor savings may be negative; BIM/IFC can generate multi-floor routing graphs.
Not just physical separation: care pathways, PPE transitions, links to medical technology/surgery/ICU, clean–dirty logistics, MEP systems and staffing must all work at once.
Elevator failures and functional reorganization change laboratory turnaround time; HDSS integrates traffic models, DES and network analysis into a design support system.
The validation ladder should be “generation → simulation → observation → causation”. Hospital-specific results center on diffusion-based generation of nursing unit functional blocks and GCN+LLM outpatient plan generation (120 plans).
EMR-driven renovation optimization, AnyLogic/queuing-theory indicator studies, normal–epidemic dual-use design strategies, whole-process infection control management. Chinese results are mostly grade-B bibliographic records and still need full-text checking.
Few post-occupancy validations; many efficiency metrics, few on safety and well-being; many static optima, few on failures and conversion; “separation” is often conflated with “reduced infection risk”; many single cases, few transferable models.
3. Mind MapClick a node to expand/collapse; click a leaf paper to open the article; “Filter in library” links to the library below
The center is the research overview; eleven theme branches plus “Chinese Research” and “Evidence Gaps”. Numbers on second-level nodes are matching library items; third-level leaves are representative papers (annotated papers first, then by citation count). Exportable as SVG; a full-screen version and the Markdown source are kept in the research team archive.
4. Medical Process Flow × Space: How Research Connects the Two
Medical planning (clinical workflow) answers “who does what when, and how materials and information flow”; spatial design answers “where these activities happen and how they connect”. The diagram below shows the key spatial issues of each stage along the patient and material main lines.
Waiting capacity, sightlines, online/offline diversion
Downstream admission, results-waiting area, infection separation
Equipment occupancy, dual flow of specimens and patients
Clean–dirty circulation, OR layout, turnover
Adjacent to surgery/emergency/imaging
Visibility, walking distance, collaboration
Elevators and access control, inventory location
Process stage – spatial object – method – representative literature matrix
Method comparison: what each type of tool is best for
| Method | Best suited for | Main limitations | Link to architectural design |
|---|---|---|---|
| Lean / value stream analysis | Finding repeated steps, waits, transport and handover waste | Limited in expressing future demand and randomness | Restructure the process before drawing the plan |
| Process mining | Identifying actual paths and frequencies from electronic records | System timestamps are not actual movement times | Producing data-based interdepartmental flow matrices |
| Discrete event simulation (DES) | Queues, resource occupancy, beds, examination and treatment stages | Unless space is explicitly modeled, it is only a fixed transport time | Comparing combined scenarios of layout + resources + process |
| Agent-based simulation (ABM) | Wayfinding, behavioral differences, spatial interaction, dynamic visibility | Demanding behavior rules and calibration | Testing corridors, waiting, work stations and crowd organization |
| Space syntax / visibility graphs | Accessibility, sightlines, spatial structure, wayfinding | Hard to predict capacity and waiting on its own | Cross-validated with observed behavior and DES/ABM |
| MIP / QAP / metaheuristics | Department location, floor allocation, multi-objective layout | Tends to oversimplify clinical and architectural feasibility | Generate candidates, then review by simulation and experts |
| Digital twin | Real-time state updates, prediction, operational intervention | High cost of data interoperability, maintenance and validation | Linking BIM/IFC, process data and asset status |
The method comparison in the table is drawn from Thematic Review 1 and cross-checked against the library.
5. Themes in DetailEach theme: research progress · evidence boundaries · design implications · annotated papers (click to view method and findings)
6. Chinese ResearchThematic Review 2: Chinese, items (incl. 11 English-language papers by Chinese scholars and 1 doctoral dissertation)
7. Frontiers and Research AgendaThematic Review 4 (2022–2026 frontiers) + overall assessment
Frontier routes: what has been shown and what has not
| Research route | Existing evidence shows | Not yet established |
|---|---|---|
| Hospital diffusion models | Preliminary generation of nursing unit functional-block plans | Code-compliant, buildable hospital-wide plans |
| GNN + LLM planning | Hospital-specific topology refinement and improved proxy scores (120 outpatient plans) | Better clinical outcomes or post-occupancy performance |
| Reinforcement learning layout | Sequential constrained spatial search; pretraining improves generalization | Hospital-scale reward design and external validation |
| BIM→simulation | Integration frameworks for space-aware process models | Widely replicated, continuously maintained operational twins |
| UWB / RTLS | Measuring real movement and contacts | Error-free activity inference; causal effects of architecture |
| Ambient computer vision | Automated measurement of perioperative events | Attributing process improvements to architectural design |
| Robots / AGV | Feasibility of hospital fleets and logistics use cases | Autonomy independent of the building; universal labor savings |
| Convertible design | Scenario frameworks and crisis experience | Standardized “conversion performance” benchmarks |
| Hospital at home | Service substitution for selected acute patients | Universal substitution ratio for physical beds |
| Net zero and climate resilience | Strategy frameworks, reviews and campus plan analyses | Consistently validated whole-life carbon and continuity outcomes |
Judged on the ladder “generated artifact → simulated performance → observed operation → causal intervention evidence”, the routes sit at different levels and must not be lumped together as “successful smart hospital demonstrations”.
Research Agenda (synthesis)
- Hospital-specific multi-floor benchmark datasets: geometry and multi-floor circulation, typed room functions and permitted adjacencies, clinical pathways, staff and resource calendars, logistics demand, epidemic configurations, measured movement and process outcomes, building services capacity and failures; train/test split by hospital or project.
- Typed, time-varying relationship graphs: distinguish physical adjacency, passage, transfer frequency, transfer urgency, restricted access, contamination compatibility and service dependency.
- Evaluate by distributions, not means: waiting time distributions, extreme congestion, interruptions and rework, urgent transfer delays, elevator and doorway contention, staff fatigue, failure recovery.
- Make hard constraints machine-checkable: automatically generated schemes must prove compliance through deterministic checks, not textual explanation.
- Close the “measure–simulate–design” loop: RTLS/UWB/CV observation → validation by direct observation → calibration of space-aware process models → generation of alternatives → robustness testing → implementation → post-implementation measurement.
- Distinguish prediction from causation: matched comparisons, phased implementation, controlled before–after designs, interrupted time series.
- Quantify conversion as a “service recovery” problem: time to safe operation, labor and materials, capacity unavailable during conversion, commissioning failures, post-conversion operating cost, time to return to normal.
- Couple distributed care with campus planning: which physical admissions are avoided, remaining inpatient complexity, coordination space, escalation receiving capacity, travel and care burden.
8. Literature LibraryTotal: 0 matching · click a row for key points · also available: CSV and RIS full download
| # | Paper | Year | Cited | In-library citations | Source |
|---|
“Cited” is the Crossref-wide citation count; “in-library citations” is the number of times other papers in this library cite it in their reference lists (only where Crossref provides DOI references, hence an underestimate).
9. Literature LinksTheme overlaps · in-library citation network · bridging papers
Theme co-occurrence networkThicker lines = more papers tagged with both (lines drawn only for ≥8 papers)
In-library citation network; larger nodes = more in-library citations, color = main theme, click a node to open the DOI
Top 20 most-cited within the library
Bridging papers (spanning ≥3 themes)
10. Notes and Sources
- Data files：literature.csv (all papers, UTF-8 with BOM, opens directly in Excel), literature.ris (importable into Zotero/EndNote), mindmap.html (full-screen mind map).
- Sources and reliability: “Key findings” and “key points” are compiled summaries, and verification grades/evidence strength are self-reported at compilation; all may contain errors, so check the original before citing. A DOI marked “unverified” has no Crossref record.
- Known limitations: (1) theme tags are matched automatically by keyword; (2) each query takes only the top 100 results, not the full database; (3) “China-related” is judged by author affiliation and title keywords; (4) 2026 is an incomplete year; (5) Chinese literature is less well covered in public searches than English, so the ratio of Chinese to international research cannot be calculated from this.
- Fangcang, Huoshenshan and other emergency construction results are mostly policy and practice analyses, not controlled trials, and cannot prove that any bed arrangement or corridor form is optimal in itself.