量身 AI← 案例库← Case library
案例 · AI 落地方案CASE · AI DEPLOYMENT PLAN · 学术健康科学中心 · 多伦多Academic health centre · Toronto

新宁布鲁克健康科学中心 SunnybrookSunnybrook Health Sciences Centre

从分散的 AI 项目,走向一个全院容量指挥中心From separate AI projects to one capacity command centre
基于公开资料的分析,与该企业没有合作关系。Based on public information; no affiliation with the company.
视频制作中,上线后会出现在这里。Video coming soon — it will appear here.
一句话判断The verdict 预测分散在不同项目里;指挥中心需要对「好结果」有一个统一的定义。Forecasts live in separate projects; a command centre needs one agreed definition of a good outcome.
起步顺序Order of work 先写清真实目标,再打通一条流程,最后扩展到床位、手术室与人力。Name the real target first, then join one flow, then widen to beds, theatres and staff.

公开事实Public facts 已核对Checked

  • 1,354
    病床数Hospital beds医院自述数据Sunnybrook's own figures
  • 59,486
    2025–2026 年度急诊就诊Emergency visits in 2025–2026官网绩效与报告页Performance and reporting page
  • 18,826
    2025–2026 年度手术台数Surgeries in 2025–2026官网绩效与报告页Performance and reporting page
  • 17,000+
    员工、医生、学习者与志愿者合计Staff, physicians, learners and volunteers官网口径Official site count

从预测,走到真正的目标From a forecast to the real target 判断Judgment

  • 预测需求Forecast demand:模型预测的内容:就诊量、床位需求、手术室时间。What the model predicts: arrivals, bed need, theatre time.(容易衡量Easy to measure)
  • 等到床位的时间Time to a bed:预测发挥作用之后,患者实际等待多久。How long a patient waits once the forecast acts.(有基线吗?Do we have a baseline?)
  • 每个人群都得到该有的照护Right care, every group:最需要的人优先,且各患者人群一致。The sickest get priority, across all patient groups.(谁来定义、谁来检查?Who defines and checks it?)

专家组裁决Expert panel 依据公开观点的推演Inference from public views

Ziad Obermeyer

算法在优化什么What it optimizes

先写清真实目标与指挥中心使用的代理指标;再按患者人群分别看结果。Write the real goal and the proxy; check results by patient group.

Ethan Mollick

人怎么用How people use it

在真实急诊夜班上测出 SIENA 与预测的失败点;由临床人员确认。Test forecasts on real emergency nights to find where they fail.

Erik Brynjolfsson

价值与度量Value & measurement

先设等待时间与占床日基线;预算要含流程再造与培训。Set baselines for waiting time first; budget for workflow redesign.

Andrew Ng

节奏与方法Pace & method

先选一条流程(急诊到入院床位),三到六个月内出数字。Start with one flow, emergency to bed; numbers in three to six months.

Thomas Davenport

整家企业怎么变Whole-enterprise change

这是床位、手术室与人力联合排布的改变;点名运营负责人,而非 IT 负责人。This changes how beds and theatres are planned. Name an operations owner.

Foteini Agrafioti

技术与业务战略对齐Strategy alignment

AI 实验室、IT 与临床项目的权责写成一页;每个项目挂到路线图目标。One page on who owns what between AI lab, IT and clinical teams.

保留的分歧Disagreement kept · 快试点 vs. 先审计Pilot fast vs. audit first

Ethan Mollick:尽早上真实班次,摸清边界。Put forecasts on real shifts early to find the edge.

Ziad Obermeyer:涉及患者优先级的模型先过目标与分组检查。Gate patient-priority models on target checks.

→ 尽早影子运行;凡排序患者的模型,检查前不上线。Shadow mode early; no patient ranking goes live before the check.

六位专家是谁、怎样思考、有什么铁律 →Who the six experts are, how they think, their rules →

三个目标Three priorities 建议Recommendation

1

写清真实目标Name the real target

为容量决策写明理想目标与代理指标;按患者人群分别看结果。Write the ideal goal and the proxy for capacity decisions; review results by patient group.

2

一次只打通一条流程Join one flow at a time

先在影子模式打通急诊到床位,再加入手术与人力预测。Start with emergency to bed in shadow mode, then add surgery and staffing forecasts.

3

一份全院模型清单One inventory of every model

列出所有临床与运营 AI 项目:负责人、数据、风险分级。List every clinical and operational AI project with an owner, data and risk tier.

场景组合Scenarios 建议Recommendation

场景Scenario做法Approach状态Status
容量指挥中心Capacity command centre
提效Efficiency体验Experience
把床位、急诊、手术与人力预测合在一个视图,辅助调度人员。Join bed, emergency, surgery and staffing forecasts in one view that advises planners.建议 · 公开路线图已列出Proposed · named in the public roadmap
SIENA 扩展为患者导航SIENA to patient navigation
体验Experience提效Efficiency
把急诊聊天机器人扩展为多语言引导,并对接 Novari 转诊软件。Extend the emergency chatbot to multilingual guidance, linked to Novari referral software.建议 · SIENA 试点之后Proposed · after the SIENA pilot
更公平的手术排程Fairer surgical scheduling
提效Efficiency降风险Risk
扩大 SMART-S 时长预测,并按病例复杂度与医生工作量做公平检查。Widen SMART-S duration forecasts, with fairness checks by case complexity and surgeon workload.建议 · 基于 SMART-SProposed · builds on SMART-S
数字病理复核Digital pathology review
提效Efficiency降风险Risk
利用全数字病理实验室,标出供病理医生复核的关注区域。Use the fully digital pathology lab to flag regions of interest for a pathologist to review.建议 · 医院已在探索Proposed · already being explored

12 个月路线12-month roadmap 建议Recommendation

奠基Foundation

第 1–3 月Months 1–3
  • 写清真实目标与代理指标Real target and proxy written
  • 带风险分级的清单Inventory with risk tiers
  • 一页权责One page of ownership
◆ 目标签字Target sign-off

影子运行Shadow

第 4–6 月Months 4–6
  • 急诊到床位预测影子运行Emergency-to-bed forecast in shadow
  • 等待时间基线Baselines for waiting time
  • 真实班次临床测试Clinician tests on real shifts
◆ 结果评审Results reviewed

上线Go live

第 7–9 月Months 7–9
  • 调度人员每班使用Planners use it every shift
  • 加入手术预测Surgery forecasts added
  • 按患者人群报告结果Results by patient group
◆ 临床安全签字Clinical safety sign-off

扩展Extend

第 10–12 月Months 10–12
  • SIENA 导航试点评审SIENA navigation pilot review
  • 加入人力预测Staffing forecast added
  • 重排路线图优先级Re-rank the roadmap
◆ 进入下一年Next year begins

治理与合规Governance 建议Recommendation

  • 所有临床与运营模型的清单与风险分级Inventory of every clinical and operational model, with risk tier
  • 结果按患者人群分别审视,而非只看总体Results reviewed by patient group, not only overall
  • 影响患者优先级的步骤由临床人员确认Clinicians confirm any step that affects patient priority
  • 健康隐私义务请与法务及隐私办公室确认Confirm health-privacy duties with counsel and the privacy office

三层指标Three-layer metrics 建议Recommendation

采用Adoption每班使用预测的调度人员Planners using the forecast each shift · 试点中临床人员对 SIENA 的使用Clinician use of SIENA in the pilot
过程Process预测与实际床位需求的差距Forecast versus actual bed need · 从决定到入床的时间Time from decision to bed · 临床人员推翻建议的比例Clinician override rate
结果Outcome各患者人群的等待时间Waiting time by patient group · 在目标等待时间内完成的比例Cases within target wait

目标值由企业自定。Targets are the company's to set.

风险Risks 判断Judgment

优化了错误的目标Optimising the wrong target

预测可能提高周转,而最长等待总落在同一些人群;先写清目标与代理指标。A forecast can speed throughput while the longest waits fall on the same groups; name target and proxy.

项目始终无法合流Projects that never join

三个独立试点各自成功,也未必能合成一个指挥中心。Three separate pilots can each succeed and still not add up to a command centre.

只增加提示,不减轻工作Alerts without relief

预测若只增加提示而不减少工作,临床人员会不再看;先在真实班次测试。If forecasts add alerts without removing work, clinicians stop looking; test on real shifts.

前 90 天First 90 days 建议Recommendation

第 1–2 周Weeks 1–2点名运营负责人;写目标与代理指标Name an operations owner; write target and proxy
第 3–4 周Weeks 3–4盘点所有 AI 项目与数据Inventory every A I project and its data
第 5–10 周Weeks 5–10选定急诊到床位;设基线;规划影子测试Pick emergency-to-bed; set baselines; plan shadow test
第 13 周Week 13与临床人员首次复盘First review with clinicians
今天只需要一个决定:One decision today: 为容量指挥中心指定一位运营负责人name one operations owner for the capacity command centre

我们可能错在:We could be wrong if: 临床人员发现预测只增加提示、不节省时间,或基线显示等待主要源于出院延迟而非预测不准。clinicians find the forecast adds alerts but no time, or baselines show waits come from discharge delays, not prediction.

本页是基于公开资料的分析,与所述企业没有合作关系,不代表其官方立场或内部计划;企业数字多来自其自述或媒体转述;「专家组」是依据六位专家公开观点的推演,不是他们的原话,也不代表他们或其雇主的立场。资料整理日期:2026-11-14。This page is an analysis based on public information. It has no affiliation with the company and does not represent its official position or internal plans. Company figures come mostly from its own statements or media reports. The expert panel is an inference from six experts' public views, not their words, and does not represent them or their employers. Compiled 2026-11-14.

想给你的公司也做一份?Want one for your company?

同样的方法:公开事实、专属判断、12 个月路线和前 90 天。公司大小都可以聊,中文沟通。Same method: public facts, a company-specific judgment, a 12-month roadmap and the first 90 days. Any size of business; we can talk in Chinese or English.

Telegram @johnsunlei邮件 Email [email protected]