Multi-Site Coordination

Selective Centralization and the Coordination Cost Problem in Distributed Health Systems

As health systems grow through acquisition, merger, and affiliation, coordination becomes the dominant management challenge. A single-site hospital has coordination problems — handoffs, department boundaries, communication breakdowns — but these problems exist within a shared physical space, a shared governance structure, and usually a shared culture. Multi-site systems face coordination problems that are qualitatively different: each site has its own history, its own community relationships, its own workforce culture, its own EHR customizations, its own informal workarounds, and often its own identity. The management question is not whether to coordinate — clearly you must — but what to coordinate centrally, what to leave local, and how to structure the boundary between them.

The default instinct is to pick one of two extremes: centralize everything for efficiency, or decentralize everything to preserve autonomy. Both are wrong. Full centralization creates standardization gains but destroys local responsiveness, alienates acquired workforces, and imposes uniform solutions on non-uniform problems. Full decentralization preserves local identity but produces duplication, inconsistency, inability to share scarce resources, and coordination failures at the boundaries between sites. The answer — well-established in organizational theory but persistently misapplied in healthcare — is selective centralization: a deliberate, analytically grounded decision about which functions to centralize, which to decentralize, and which to share.

The Theoretical Foundation: Differentiation and Integration

Lawrence and Lorsch (1967), in their landmark study of organizational design, established the principle that effective organizations must simultaneously differentiate (allowing subunits to adapt to their specific environments) and integrate (ensuring that subunit activities are coordinated toward organizational goals). The key finding was not that differentiation or integration was better — it was that the appropriate balance depends on the environment. Organizations facing heterogeneous environments require more differentiation; organizations facing homogeneous environments require more integration. The failure mode is mismatch: insufficient differentiation in a heterogeneous environment produces rigid solutions that do not fit local conditions; insufficient integration in a complex environment produces fragmentation and coordination failure.

Galbraith (1973, 1977) extended this work into a design framework: the amount of information processing an organization must perform is a function of its uncertainty and interdependence. When tasks are uncertain and interdependent, organizations need coordination mechanisms — lateral relationships, liaison roles, integrating managers, matrix structures — that increase information processing capacity. The design choice is not centralize vs. decentralize but rather: what coordination mechanisms match the information processing demands of this particular function?

Applied to multi-site health systems, the implication is precise. Each function has a coordination cost (the cost of standardizing and managing across sites) and a local adaptation value (the benefit of allowing each site to tailor the function to its specific context). Functions where coordination cost is low and local adaptation value is low are obvious centralization candidates. Functions where coordination cost is high and local adaptation value is high should remain local. The difficult decisions are the functions in the middle — and that is where Galbraith’s framework earns its keep.

Shortell et al. (1993, 1996), studying organized delivery systems, documented this dynamic empirically in healthcare. Systems that achieved “functional integration” — shared support services, common information systems, unified financial management — performed better than systems that attempted either full clinical integration (standardizing clinical practice across sites) or pure administrative consolidation. Enthoven’s (1993) concept of the integrated delivery system pointed toward the same conclusion: the value of integration is not uniformity but the ability to manage across the continuum of care, which requires standardized infrastructure supporting locally adapted clinical practice.

The Selective Centralization Framework

The decision criterion for each function is the ratio of coordination cost to local adaptation value. This is not a spreadsheet calculation — the variables are partly quantitative, partly qualitative — but it is a structured analysis that prevents the “centralize everything” and “decentralize everything” defaults.

Centralize when standardization creates value and local variation creates risk. Functions in this category share a common characteristic: the cost of inconsistency across sites exceeds the benefit of local customization. Supply chain procurement benefits from volume purchasing power and standardized contracts — local variation in vendor selection produces higher costs without corresponding quality gains. Credentialing and privileging must meet the same legal and accreditation standards regardless of site; duplication across five sites means five separate credentialing offices doing identical work with identical regulatory requirements. IT infrastructure — networks, security, EHR hosting, data backup — gains nothing from local variation and loses economies of scale. Compliance reporting is driven by external requirements that do not vary by site; centralizing it eliminates redundant effort and reduces the risk of inconsistent filings. Financial reporting and consolidation require standardized chart of accounts and reporting timelines regardless of site preferences.

Decentralize when local context determines effectiveness. Functions in this category share the opposite characteristic: the value of local fit exceeds the efficiency of standardization. Hiring and recruitment depend on local labor market conditions — the recruitment strategy that works in a regional hospital’s urban labor market is wrong for a Critical Access Hospital (CAH) sixty miles away competing for a different candidate pool with different compensation expectations and different community ties. Patient scheduling and access must accommodate community preferences — a rural FQHC whose patients are agricultural workers needs evening and weekend hours that an urban specialty clinic does not. Clinical protocols must account for local disease burden, acuity mix, available specialist backup, and transfer capabilities — a CAH managing acute MI with a 45-minute transport time to PCI needs different protocols than a regional hospital with an on-site cath lab. Community engagement depends on relationships with specific community leaders, faith organizations, school systems, and employer groups that are inherently local.

Share when the function benefits from both coordination and local input. Some functions occupy the middle ground — they require cross-site standardization for comparability but local flexibility for relevance. Quality metrics should be standardized in their definitions (so sites can be compared) but interpreted in the context of each site’s patient population and acuity mix. Workforce analytics benefit from centralized data infrastructure and standardized measures but must be acted on locally by managers who understand their unit’s specific dynamics. Grant administration in systems that pursue federal or state funding benefits from centralized grant-writing expertise and compliance infrastructure while requiring local input on community needs assessments, program design, and relationship management with funders.

Healthcare Example: A Five-Site Rural Health Network

Consider a five-site rural health network in the Pacific Northwest: one 85-bed regional hospital, two Critical Access Hospitals (25-bed each, 30 and 55 miles from the regional hub), and two Federally Qualified Health Centers. The network formed over six years through a combination of acquisition (one CAH), affiliation agreements (the second CAH and both FQHCs), and organic growth. Each site has its own history, its own community board or advisory council, and its own staff who identify primarily with the local site rather than the network.

The network CEO faces the standard multi-site dilemma: the system is spending significant administrative overhead on duplicated functions across five sites, but previous attempts at consolidation met fierce resistance from local staff and community stakeholders who saw centralization as the regional hospital “taking over.”

What they centralized. After a six-month analysis using the coordination cost framework, the network centralized four functions. (1) Credentialing: five sites were independently credentialing many of the same providers, with different processes, different timelines, and different documentation requirements. Centralizing to a single credentialing office with one standardized process eliminated 2.3 FTE of duplicated effort and reduced credentialing cycle time from an average of 97 days to 41. (2) IT infrastructure: each site had its own IT support arrangement — the regional hospital had a four-person IT department, the CAHs each had one IT contractor, and the FQHCs relied on their FQHC network’s shared IT. Centralizing IT under the regional hospital’s department with remote support for satellite sites improved security posture, standardized EHR configuration, and reduced IT cost per site by 30%. (3) Compliance reporting: CMS cost reports, state licensure filings, and accreditation documentation were produced independently by each site, often by clinical staff with compliance as a secondary responsibility. A central compliance office with two dedicated staff replaced fragmented effort across twelve people who each spent 10-20% of their time on compliance. (4) Financial consolidation: a single finance team producing standardized monthly reports for all five sites replaced five different reporting processes with five different chart-of-accounts structures.

What they decentralized. Three functions were explicitly designated as local decisions with no central override. (1) Hiring: each site retained full authority over recruitment, interviewing, and selection. The CAHs recruit from small communities where personal relationships and community reputation are decisive — the HR director at the regional hospital has no advantage in those labor markets. (2) Patient scheduling: the FQHCs serve predominantly Medicaid and uninsured populations with scheduling needs (walk-in availability, evening hours, bilingual scheduling staff) that differ fundamentally from the regional hospital’s specialist scheduling requirements. (3) Clinical protocols: the CAHs manage stabilize-and-transfer cases that the regional hospital manages definitively. Imposing the regional hospital’s clinical protocols on CAHs with different capabilities, different specialist availability, and different transfer times would be clinically inappropriate.

What they shared. Three functions were structured as shared services with central coordination and local execution. (1) Quality metrics: a standardized set of fifteen quality measures was defined centrally, with data collection automated through the shared EHR. But quality improvement initiatives were designed and executed locally, because the root causes of quality gaps differ by site — the regional hospital’s readmission problem is a care transition problem, while one CAH’s readmission problem is a medication reconciliation problem. (2) Workforce analytics: a central analyst produces monthly workforce dashboards for all five sites using standardized definitions (turnover rate, vacancy rate, time-to-fill, overtime ratio). Site managers use the dashboards to identify local problems and design local interventions, with cross-site comparison enabling identification of outliers and sharing of effective practices. (3) Grant administration: a central grants coordinator manages compliance, reporting timelines, and financial tracking for all federal and state grants across the network. But grant applications are developed collaboratively, with local sites providing community needs data and relationship management and the central coordinator providing writing expertise, budget development, and regulatory compliance.

Eighteen-month outcomes. Administrative overhead (measured as administrative FTE per clinical FTE) declined 20% across the network. Each site retained its community identity — local signage, local advisory boards, local leadership with authority over local decisions. Cross-site referral completion (the percentage of referrals from CAHs and FQHCs to the regional hospital that result in a completed appointment) improved from 62% to 84%, driven primarily by the shared EHR configuration and a centrally designed referral tracking workflow. Staff satisfaction on the “organizational support” dimension of the engagement survey improved at four of five sites. The one site where satisfaction declined was a CAH where the local administrator perceived the centralization of IT as a loss of control — a reminder that even well-designed centralization decisions produce political costs that must be managed.

The Traveling-Resource Model

Multi-site networks face a specific capacity problem: each site needs access to specialists, specialized equipment, or specialized administrative expertise, but no single site has enough volume to justify full-time presence. The traveling-resource model — rotating providers, equipment, or expertise across sites on a scheduled basis — creates capacity without duplication.

A psychiatrist who provides on-site consultations at each of two CAHs and one FQHC two days per week each creates behavioral health access at three sites that could not independently recruit a psychiatrist. A mobile ultrasound unit that rotates across sites on a weekly schedule provides diagnostic imaging capacity that no individual site could justify purchasing. A credentialing specialist, revenue cycle auditor, or infection preventionist who serves all five sites on a rotating basis provides expertise that small sites cannot afford independently.

The operational challenge is scheduling coordination. A traveling psychiatrist’s schedule must account for patient demand patterns at each site, travel time between sites, and the need for continuity of care that requires consistent day-of-week presence. This is a resource allocation problem with constraints — precisely the type of problem addressed in Operations Research Module 5 (scheduling and resource allocation). The OR framework applies directly: the psychiatrist is a shared resource with capacity constraints, the sites are demand nodes with different volume patterns, and the optimization objective is to maximize patient access across the network subject to provider availability and travel constraints.

The traveling-resource model also interacts with the network topology concepts from OR Module 4. The multi-site system is a network whose nodes are sites and whose edges are the transportation, communication, and referral linkages between them. Coordination cost is topology-dependent — a hub-and-spoke network (all sites connected through the regional hospital) has different coordination properties than a mesh network (sites connected to each other directly). The five-site network in the example above is a natural hub-and-spoke topology, with the regional hospital as the hub. This topology minimizes the number of coordination relationships (four, rather than the ten that a fully connected five-node network would require) but creates a single point of failure: if the hub’s coordination capacity is overwhelmed, the entire network’s coordination degrades.

Integration Points

Operations Research Module 4: Network Flow. The multi-site health system is a network with computable topology. The coordination cost between any two sites is a function of the network distance between them — not just geographic distance, but information distance (do they share an EHR? do they share governance?), cultural distance (do they trust each other?), and operational distance (are their workflows compatible?). OR M4’s network flow analysis provides the analytical machinery to identify bottlenecks in cross-site coordination, quantify the cost of adding or removing network edges (affiliations, data sharing agreements), and determine whether the current network topology matches the information processing demands that Galbraith’s framework identifies. A health system that models its multi-site coordination as a network problem rather than an organizational chart problem will make better structural decisions.

Operations Research Module 5: Scheduling and Resource Allocation. The traveling-resource model is a scheduling optimization problem. The provider, equipment, or expertise that rotates across sites must be scheduled to maximize utilization while respecting constraints: travel time, minimum session duration at each site, patient demand patterns by day of week, and provider preferences. This is a variant of the vehicle routing problem or the multi-facility scheduling problem, both well-studied in OR. The practical implication is that traveling-resource schedules should not be built by hand (which inevitably under-optimizes) but should be generated using scheduling algorithms that account for all constraints simultaneously and can be re-optimized when demand patterns shift.

Product Owner Lens

What is the workforce problem? Multi-site health systems face coordination overhead that grows with the number of sites, and the default responses — centralize everything or decentralize everything — both fail. The result is either rigid standardization that ignores local context or fragmented autonomy that wastes resources and produces coordination failures at site boundaries.

What system mechanism explains it? Lawrence and Lorsch’s differentiation-integration framework (1967) establishes that effective multi-unit organizations must simultaneously differentiate (adapt to local environments) and integrate (coordinate toward system goals). Galbraith (1973) links the required coordination mechanism to information processing demands. The coordination cost of each function varies — some functions (credentialing, IT, compliance) have low local adaptation value and high standardization benefit; others (hiring, scheduling, clinical protocols) have high local adaptation value. The selective centralization decision should be driven by this ratio, not by organizational politics or historical defaults. Shortell et al. (1993, 1996) confirmed empirically that functional integration outperforms either full centralization or full decentralization in organized delivery systems.

What intervention levers exist? The primary lever is the centralization decision itself — which functions to centralize, decentralize, or share. Secondary levers: designing shared-function governance that gives local sites voice without veto; implementing traveling-resource models for scarce expertise; building cross-site communication infrastructure that reduces coordination cost without requiring centralization; and establishing cross-site quality and workforce analytics that enable comparison without mandating uniformity.

What should software surface? (a) Cross-site coordination dashboard: referral completion rates between sites, shared-resource utilization, credentialing cycle times, compliance filing status — the metrics that reveal whether coordination is working or breaking down. (b) Traveling-resource scheduler: optimization-based scheduling for providers and equipment that rotate across sites, with demand forecasting and constraint management. (c) Function-level centralization analyzer: for each administrative function, display the current cost structure (FTE, vendor spend, cycle time) at each site to identify where centralization would reduce cost and where local variation reflects genuine local adaptation rather than mere historical accident. (d) Cross-site workforce comparison: standardized workforce metrics (turnover, vacancy, overtime, engagement) displayed by site with statistical controls for patient volume and acuity, enabling identification of site-level workforce problems that site managers may not recognize as outliers.

What metric reveals degradation earliest? Cross-site referral completion rate is the leading indicator of coordination failure in multi-site networks. When referrals from satellite sites to the hub (or between sites) begin falling through — patients not scheduled, results not returned, follow-up not completed — the coordination infrastructure is failing. This metric degrades before administrative overhead metrics rise, because coordination failure manifests first as dropped handoffs and only later as rework and duplication.

Warning Signs

These indicators suggest multi-site coordination is failing or the centralization balance is wrong:

• Cross-site referral completion drops below 75% — patients are falling through the gaps between sites, indicating coordination infrastructure failure
• Satellite sites create workarounds for centralized functions — local staff building shadow processes means the centralized service is not meeting local needs
• Centralized functions impose timelines that ignore site-level operational realities — a credentialing office demanding documentation by Thursday from a CAH that has one part-time administrative staff member
• Traveling providers report spending more time on logistics than patient care — the scheduling coordination overhead is consuming the capacity the model was designed to create
• Local site identity erodes after centralization — community stakeholders, patients, or staff report feeling that “our hospital was taken over,” indicating centralization extended into functions where local identity creates value
• Duplication persists in functions designated as centralized — sites maintaining their own compliance files, their own vendor relationships, or their own IT workarounds alongside the central function, paying the cost of both
• Quality metrics are standardized but quality improvement is not happening — measurement without local action means the shared-function model is producing data without producing decisions
• Network leadership cannot articulate which functions are centralized and why — the absence of a deliberate framework means decisions were made politically rather than analytically, and will be re-litigated continuously