Roswell Park Data Ecosystem RFI #27-28.docx
DOCX document 617 KB Posted
- Attached to
- Enterprise Data Ecosystem
- Solicitation number
- 2140335
- Issued by
- Erie County, New York
About this file
This Request for Information (RFI #27-28) is issued by Roswell Park Comprehensive Cancer Center, a standalone NCI-designated 157-bed cancer center in Buffalo, New York. Roswell Park seeks qualified respondents to provide information regarding the evolution and maturation of its Enterprise Data Ecosystem, a foundational component supporting the organization's planned transition to a unified enterprise electronic health record platform and modernization of clinical, operational, financial, and research analytics environments. The RFI was advertised on October 28, 2026, with responses due by Thursday, November 12, 2026, at 2:00 p.m. EST via email to Rob Henschel. Questions from respondents are accepted through Wednesday, November 4, 2026, at 2:00 p.m. EST, with all questions and answers distributed collectively to all respondents through written addenda. The RFI does not anticipate a single vendor will provide all ecosystem components; respondents should propose architectures combining their products and services with partner technologies while making ownership, accountability, and contractual relationships explicit.
Roswell Park serves 54,000 patients annually with 332,000 outpatient visits and 5,400 admissions and operates within an environment where clinical care and research are closely integrated. The organization currently maintains a distributed enterprise data environment with on-premise and vendor-hosted elements, including a MariaDB Enterprise Data Warehouse on Microsoft SQL Server databases, with Power BI, Microsoft Fabric, and StrataJazz serving as primary analytics and reporting mechanisms. Respondents must address detailed requirements across 19 major categories including data virtualization, medallion architecture, data products, enterprise data warehousing, Power BI integration, metadata catalogs, governance, legacy data archival, EHR conversion services, research environments, AI capabilities, interoperability standards, and performance monitoring. The RFI explicitly requires respondents to provide ten-year cost projections including implementation and recurring costs, staffing plans by role, three planning scenarios (minimum viable, recommended, and expanded implementations), and cost sensitivity analyses. No pricing is provided in the RFI as it is an information-gathering exercise to inform future procurement strategy and requirements development.
View the file
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
REQUEST FOR INFORMATION
RFI # 27-28
ENTERPRISE DATA ECOSYSTEM
RFI Advertised: October 28, 2026 Due Date and Time: Thursday, November 12, 2026 by 2:00pm EST
Contact: Rob Henschel Rob.Henschel@RoswellPark.org
1. Instructions Submissions must be submitted via email to Rob Henschel @ rob.henschel@roswellpark.org by Thursday, November 12, 2026 at 2:00pm EST.
Roswell Park will accept questions from Respondents regarding this RFI from the date of issuance through Wednesday, November 4, 2026 at 2:00 p.m. EST. Questions must be submitted via email to: Rob Henschel @ rob.henschel@roswellpark.org. Respondents must reference the RFI number and title in the subject line of their email. As part of Roswell Park’s response to questions, ALL questions, and the associated answers, will be shared collectively with all Respondents via issuance of a written Addendum to this RFI.
2. Purpose Roswell Park Comprehensive Cancer Center is seeking information from qualified respondents regarding its desire to evolve and materially mature its Enterprise Data Ecosystem. The Enterprise Data Ecosystem is a foundational component of our organization’s planned transition to a unified enterprise electronic health record platform (an active effort) and the modernization of its clinical, operational, financial, and research analytics environments, foundational support to our cancer care and research missions.
Roswell Park does not assume that a single vendor will provide all required components of our future ecosystem. Respondents should propose an architecture that combines their respective products and services with partner technologies where appropriate, making ownership and accountability explicit while avoiding unnecessary dependencies on a single proprietary layer.
Respondents must clearly identify:
· Each proposed product and technology component.
· The vendor responsible for each component.
· Whether the capability is native, partner-provided, third-party, or custom-developed.
· The contractual and support relationship among participating vendors.
· The party responsible for end-to-end solution architecture.
· The party responsible for implementation and integration.
· The party responsible for ongoing operations and support.
· How issues spanning multiple products or vendors will be governed and resolved.
The purpose of this RFI is to assist Roswell Park in:
· Refining its future-state enterprise data architecture.
· Understanding available technology platforms and implementation approaches.
· Identifying potential strategic partners.
· Understanding how proposed solutions would integrate with a future unified EHR platform.
· Understanding the roles of data virtualization, data ingestion, cloud platforms, enterprise data warehouses, lake houses, semantic layers, analytics tools, research platforms, and AI technologies, and how they technically and functionally integrate and cooperate to form an Enterprise Data Ecosystem.
· Determining how existing investments should be preserved, incorporated, or decommissioned.
· Developing requirements and evaluation criteria for a potential future competitive procurement.
· Understanding estimated cost, internal and external, to implement and maintain the described ecosystem.
3. Background Roswell Park Comprehensive Cancer Center is a stand-alone NCI-designated 157-bed comprehensive cancer center with overlapping missions of healthcare operations and cancer research. In 2025, Roswell served 54,000 patients with 332,000 outpatient visits and 5,400 admissions. Roswell Park operates an environment in which clinical care and research are closely connected, and the future Enterprise Data Ecosystem must support both traditional healthcare operational requirements and cutting-edge cancer research needs. The ecosystem must support data generated through and beyond the enterprise EHR as well as multi-modal data to include:
· Financial and revenue cycle functions
· Quality and regulatory reporting
· Basic and translational research
· Clinical trials
· Laboratory services
· Pathology
· Imaging
· Biospecimens
· Genomics and molecular testing
· Registries
· Population health
· Cohorted data commons, e.g., OMOP CMD data and otherwise, both broadly and for a specific research use cases
· Custom-developed applications (on premise and cloud hosted)
· External data sources It is not our intention to duplicate enterprise EHR reporting; instead consider where we may extend it to other purposes.
4. Strategic Context The data ecosystem must be planned as a coordinated workstream in parallel with the unified EHR platform transformation. It must support continuity during coexistence of legacy and future platforms, enable data migration and validation, and establish a sustainable post-transformation architecture. The proposed ecosystem must be aligned to complement the future unified EHR platform but not be dependent upon it. The architecture should ideally provide immediate enterprise value while avoiding investments that would need to be discarded following implementation of the future EHR platform.
Legacy data supporting systems will be utilized during the course of the proposed data ecosystem and concurrent unified EHR implementations. The Enterprise Data Ecosystem implementation as proposed must give care for:
· Legacy system and unified EHR platform coexistence
· Data migration from legacy systems, mapping and conversion into the unified EHR platform
· Archival of legacy data not converted into the unified EHR platform to support a legal patient record
· Data validation and reconciliation
· Longitudinal patient data
· Testing
· Operational readiness
· Source-to-target reconciliation
· Historical-data access Business continuity Following the EHR implementation, the ecosystem data, including the EHR and the proposed solutions, should provide a sustainable, long-term architecture incorporating:
· Data originating from the unified EHR
· Historical clinical information
· Retained enterprise systems (on premise or vendor hosted)
· Research platforms
· Financial and operational platforms
· External datasets including CMS and competitive analysis data
· Multimodal data
· Enterprise analytics
· AI and data science
5. Current State Environment Roswell Park currently maintains a distributed enterprise data environment consisting of on-premise, vendor hosted, and Microsoft Fabric data elements. Roswell maintains a MariaDB Enterprise Data Warehouse that supports enterprise reporting, integrates data from multiple (often siloed) source applications, supports source-specific and departmental data marts, and provides data to enterprise analytics and reporting processes that are developed in-house as well as vendor hosted applications such as StrataJazz.
Roswell Park’s current EDW resides upon Microsoft SQL Server databases. These databases and many associated legacy clinical systems will eventually be converted to the future unified EHR or its companion, cloud-based archive with contextual relevance. Care must be given to maintain access to legacy data for clinical care and research purposes.
Roswell Park currently uses Power BI, Microsoft Fabric, and StrataJazz as the primary mechanisms to service enterprise analytics and reporting needs. These technologies support dashboards, enterprise and operational reporting, data analytics, decision support, and data preparation workloads. Roswell Park expects Power BI and StrataJazz to remain important enterprise visualization and reporting platforms.
Enterprise information is distributed across multiple applications, databases, and technology platforms. This data fragmentation creates challenges related to data discovery, integration, access, governance, and research access. Lack of consistent metric definitions and user self-service analytics are current challenges.
6. Future State and Guiding Principles The desired future state ecosystem is a scalable, governed, and interoperable Enterprise Data Ecosystem that can integrate data from the future EHR platform when necessary, as well as retained enterprise on-premise or cloud clinical, research, and financial platforms. The architecture must support structured and unstructured data; enable enterprise, financial, revenue cycle, and clinical reporting; support research discovery and data delivery; enable advanced analytics and computing; enable artificial intelligence; and allow access by custom applications both cloud-based and on-premise.
Roswell Park envisions an architecture in which data may remain at its source, be virtualized, or be selectively ingested according to business, technical, performance, governance, research, and cost requirements. Respondents may recommend alternative architectural approaches where they believe Roswell Park's objectives can be achieved most effectively.
Respondents should consider the following principles when developing their recommendations:
· The architecture should complement the future EHR while maintaining the ability to incorporate information from EHR and non-EHR systems.
· Research requirements should be incorporated into the architecture rather than implemented as a separate downstream data environment.
· The architecture must support combinations of data virtualization, zero-copy access, data ingestion, caching, and materialized/persisted enterprise datasets
· Data must be governed, discoverable, classified, traceable, controlled/authorized, and auditable.
· The architecture must minimize unnecessary vendor lock-in and support open standards and commonly adopted healthcare and research integration patterns (e.g., OMOP CDM, FHIR, HL7, REST).
· Data models, metrics, semantic definitions, integration patterns, and data products should be reusable across multiple business and research use cases.
· Authorized enterprise data should be made available to AI and advanced analytics platforms through secure and governed mechanisms.
· Make infrastructure, platform, consumption, support, and implementation cost drivers measurable and forecastable.
The diagram below provides a conceptual schematic of the Roswell Park future state Enterprise Data Ecosystem. While this may be used as a reference or starting point, Roswell Park encourages respondents to propose their own architecture diagrams.
Vendor-agnostic Enterprise Data Ecosystem example.
7. Requirements
7.1 Data Virtualization and Logical Data Models
Roswell Park seeks a data virtualization capability that can provide a common access layer across distributed enterprise information regardless of where the underlying data physically resides. Respondents should describe their ability to:
· Virtualize databases (e.g., relational, document-oriented, vector), cloud platforms, lake houses, object stores, and other data sources.
· Query data without physically copying it.
· Combine virtualized and persisted information within a common logical model and make that information available to governed semantic models or semantic layers. Perform query federation across heterogeneous systems.
· Apply query pushdown and other performance optimization techniques.
· Cache data selectively where appropriate.
· Expose virtualized datasets through SQL, APIs, JDBC/ODBC, or other standard interfaces.
· Support on-premises and cloud-hosted applications consuming virtualized data.
· Provide resilient access to source platforms.
· Handle source-system outages.
· Monitor and tune query performance to identify poorly performing source systems or queries.
Roswell Park also seeks the ability to construct logical data representations that combine information from heterogeneous enterprise sources. For example, enterprise patient, longitudinal cancer patient record, provider or encounter representation, revenue cycle claim, and clinical trial participant. Respondents should explain how logical models can remain stable while underlying source systems change during the EHR transformation and describe.
7.2 Medallion Architecture
The proposed architecture should support a standard medallion-style architecture including bronze, silver, and gold layers. Respondents should explain how medallion layers are implemented, lineage is maintained, transformations are versioned, and incremental processing is supported.
7.3 Reusable Data Products
Roswell Park seeks an architecture supporting reusable governed data products (i.e., a self-contained, high-quality data asset). Respondents should describe how a data product can be created and reused across Power BI, research, AI, data science, and custom applications without requiring separate data preparation for each use case. Describe creation, versioning, governance/access controls, metadata, APIs, and data product endorsement.
7.4 Enterprise Data Warehouse and Analytics
Roswell Park seeks a platform capable of functioning as a virtualized Enterprise Data Warehouse (EDW) that provides a unified, governed access layer for enterprise reporting and analytics across heterogeneous data sources, regardless of where the underlying data physically resides. The platform should allow clinical, financial, operational, research, and other enterprise data to be logically integrated and presented through consistent business-oriented models without requiring all source data to be replicated into a single physical repository. The virtualized EDW should support a combination of live or zero-copy access, data federation, and persisted curated datasets. The EDW is not intended to duplicate the capabilities of the unified EHR.
Respondents should describe how their proposed platform can support enterprise-scale reporting and analytics workloads. This includes integration with Microsoft Power BI Fabric, creation of reusable logical data models and semantic layers for operational reporting and research, historical and cross-domain analysis, fine-grained security and data masking, metadata, lineage, and governance. Workload management, query optimization and pushdown, caching, performance monitoring, and resilience when underlying source systems are unavailable or experience performance degradation. Respondents should also explain how they determine when data should remain virtualized versus materialized, and provide examples of comparable healthcare or research organizations where their platform has been successfully deployed as a virtualized or logical EDW.
7.5 Power BI Integration
Power BI is expected to remain a significant enterprise reporting and visualization tool. Respondents should describe how Power BI business analysts and reporting experts would access both virtualized and persisted information. The response should also describe how the proposed architecture would minimize duplication of business logic between the data platform and Power BI. While Roswell Park currently utilizes Microsoft Fabric to host semantic models, we welcome recommendations for alternative platforms.
7.6 Enterprise Data, Metric, and Research Metadata Catalog
Roswell Park seeks a searchable, governed enterprise catalog that enables business, clinical, operational, financial, research, and technical users to discover, understand, evaluate, access, and appropriately reuse enterprise information. At minimum, this catalog should enable technical and non-technical users and should support discovery of data assets, dashboards, reports, metrics, and KPIs.
The catalog should encompass source-system data; virtualized and persisted datasets; reports, dashboards, metrics, and KPIs; semantic models; data products; APIs; and institutionally created or transformed research resources such as OMOP datasets, cohorts, phenotypes, variables, and analytical artifacts.
Respondents should describe how the proposed catalog captures, connects, governs, and presents the following information:
· Data assets: Source systems, databases, schemas, tables, fields, files, interfaces, reports, dashboards, metrics, semantic models, data products, research datasets, cohorts, phenotypes, derived variables, and analytical artifacts, including legacy, future-EHR, and institutionally created assets.
· Business, clinical, and research meaning: Plain-language definitions, clinical context, intended purpose, calculation or cohort logic, applicable populations, appropriate and prohibited uses, limitations, and relationships to enterprise glossary terms and standard terminologies.
· Technical metadata: Physical and logical structures, data types, interfaces, formats, models, vocabulary and terminology versions, refresh schedules, data latency, dependencies, and environment information.
· Source-to-target lineage: Field-level and dataset-level mappings, transformations, derivations, reconciliation rules, and traceability from legacy systems through staging, legacy data conversion, the future EHR, the EDW, OMOP, semantic models, reports, and downstream research or analytical products.
· Data quality: Profiling results, completeness, conformance, validity, plausibility, timeliness, reconciliation status, quality thresholds, validation history, known issues, remediation status, and fitness-for-purpose assessments for clinical, business, and research use.
· Governance: Owners, stewards, custodians, authoritative sources, certification or endorsement status, sensitivity and classification, access requirements, approval workflows, permitted uses, retention requirements, applicable IRB or protocol restrictions, and auditability.
· Provenance: Originating systems, responsible pipelines and processes, creation and modification history, transformation and code versions, vocabulary releases, refresh history, contributors, and evidence necessary to reproduce derived datasets, metrics, and research results.
The catalog must support the EHR transformation by maintaining a durable inventory and historical record of data assets as legacy applications coexist with, migrate to, or are replaced by the future unified EHR. Respondents should explain how catalog entries, identifiers, definitions, mappings, lineage, quality evidence, governance assignments, and access pathways will be versioned and preserved as source systems and technologies change. Users should be able to determine where an asset originated, whether it remains in a legacy system, has been converted or archived, is available in the future enterprise EHR or another governed platform, and which version is appropriate for a specific operational, clinical, or research purpose.
The proposed capability should make data assets findable, accessible, interoperable, and reusable throughout the transformation and in the long-term future state. Respondents should describe catalog search, filtering, tagging, business-glossary, and natural-language discovery capabilities; integration with identity, access-request, governance, and workflow tools; use of open standards and machine-readable metadata; and mechanisms to identify trusted, certified, deprecated, restricted, or ungoverned assets. The response should also explain how metadata can be exported, integrated through APIs, or transferred to avoid dependence on a single proprietary catalog and to support reuse across Power BI, clinical operations, research environments, OMOP, analytics, AI, and custom applications. Finally, respondents should explain capabilities that empower non-technical users to determine whether required data already exists, identify definitions, owners, sources of truth, and understand how access can be requested/granted.
7.7 Reporting Modernization and Self-Service
Roswell Park seeks to reduce manual reporting and improve user self-service access. Respondents should describe how the proposed ecosystem would help to:
· Reduce manual data extracts and spreadsheet-based reporting.
· Establish reusable metrics to reduce repeated data preparation and duplicative effort.
· Enable use of trusted self-service datasets.
· Allow business users to discover available data.
· Improve report development turnaround.
· Improve transparency regarding available reports, datasets, and metric/KPI definitions.
· Position current data reporting builders to become higher-order data scientists and analytics, or data provisioners for self-service workforce members.
7.8 Data Governance and Security Controls
The ecosystem must provide strong enterprise governance and access controls and address integration with enterprise identity and access-management platforms:
· Role-based access control including use of Active Directory groups
· Fine-grained row-level controls
· Data masking
· De-identification support
· Audit logs
· Access reviews Respondents should describe how governance policies can be enforced consistently across virtualized datasets, Power BI, research environments, AI, and custom applications. The proposed solution should support least privilege, identity-based access, encryption, auditability, data minimization, environment separation, retention controls, and policy enforcement across virtualized and persisted access.
7.9 Legacy Clinical Data Archive
Roswell Park requires a sustainable approach for maintaining historical clinical information following the retirement of our legacy clinical and ancillary applications. This legacy data archive will serve as the legal historical patient record, support the servicing of patient release of information (ROI) requests by health information management (HIM), and support historical research data needs including access to clinical data extending 10–20 years or longer. Data in the archive must be accessible from within the future unified EHR platform through API, FHIR, REST or other mechanism to support seamless clinician access.
Respondents should describe the proposed archive architecture, supported clinical data types, user presentation of historical patient information, search capabilities, integration with the future EHR platform, release-of-information workflows, data retention, and long-term vendor support.
7.10 Legacy EHR Data Conversion and Migration
Roswell Park requires capabilities and implementation services that can assist with extraction of legacy data from the existing EMR and ancillary systems, data mapping, conversion, validation, cleansing, exception handling, testing, data quality reporting, and reconciliation of information during the EHR transition. The proposed solution should provide a controlled and auditable framework for managing data conversion activities throughout the EHR implementation lifecycle and should support both automated and manual processes required to prepare legacy information for migration. Respondents should explain how their proposed platform, tools, and implementation services will support the end-to-end data conversion process, including:
· Extraction of data from legacy EHR platform and other ancillary clinical applications, including relational, file-based sources, or vendor hosted platforms.
· Profiling of legacy data to identify data quality issues, incomplete records, inconsistent values, duplicate information, and other conditions that may affect conversion.
· Mapping of legacy source data elements to the corresponding target structures, fields, concepts, and workflows within the future EHR.
· Management of source-to-target mapping specifications, transformation rules, value translations, crosswalks, terminology mappings, and business rules.
· Transformation and normalization of data prior to loading into the future unified EHR platform.
· Creation and management of conversion HL7 data streams or other extracts in the formats required by the future EHR vendor.
· Retention and versioning of mapping specifications and transformation logic used during successive conversion cycles.
· Automated and manual validation of converted information.
· Reconciliation of source-system records against information loaded into the target EHR.
7.11 Role-Based Workspaces
Roswell Park seeks support for provisioning of role- and domain-oriented workspaces to support clinical operations, quality, finance, revenue cycle, payer relations (collectively beyond the capability of the unified EHR), and research needs. Respondents should describe:
· Workspace provisioning
· Access controls
· Role management
· Dataset assignment/mounting within the workspace
· Compute and storage isolation
· Development tools
· Promotion between development, testing, and production
· Cost allocation and enforcement
· Monitoring
7.12 Research Data Environment
The Enterprise Data Ecosystem must support Roswell Park's research mission as a first-class citizen and address support for historical clinical information extending 10–20 years or longer as required for longitudinal cancer research.
Controlled research access and honest-broker workflows:
IRB- and protocol-aware access: Route requests through documented review, approval, amendment, expiration, renewal, and revocation workflows that enforce protocol scope, consent limitations, data-use agreements, minimum-necessary access, and role-based permissions.
Honest-broker workflows: Support authorized honest brokers in identifying eligible records, preparing approved datasets, applying inclusion and exclusion criteria, documenting disclosure decisions, and delivering data without exposing direct patient identifiers to research teams unless specifically authorized. This can also include privacy transformations like date-shifting and masking.
Study-specific identifiers: Transform the enterprise patient identifier into a unique identifier created for each approved study. The same patient should receive a consistent identifier within a study but a different, non-derivable identifier across studies to reduce unauthorized linkage. These enterprise ID-to-study-identifier crosswalks need to be stored separately within a restricted honest-broker or trusted linkage service; encrypt and tightly control linkage keys; prohibit researcher access to the crosswalk; and support re-identification only through an authorized, documented workflow.
Project-specific provisioning: Create governed datasets and workspaces limited to the approved cohort, variables, dates, data domains, users, and retention period, with controls for export, onward sharing, and project closeout.
Auditability and provenance: Record the request, approvals, source data, cohort logic, identifier transformation, de-identification rules, linkage events, users, queries, exports, disclosures, dataset versions, and disposition actions needed to reproduce and audit each release.
De-identification of Unstructured Data: Roswell Park seeks information regarding solutions and qualified third-party services that support de-identification of unstructured and multimodal data under the HIPAA Expert Determination method, including clinical notes and reports, pathology reports and whole-slide images, radiology reports and images, and associated metadata. Respondents should describe the qualifications and independence of the expert; the methods used to identify and mitigate direct and indirect re-identification risk across text, pixel data, embedded annotations, headers, metadata, and linked datasets; support for statistically defensible risk assessment and documented determination that the risk of re-identification is very small; validation and quality-assurance procedures; handling of residual risk and exceptions; audit trails, reproducibility, and ongoing reassessment as data sources or uses change; and the deliverables provided to support institutional review, governance, and permitted research use. Responses should clearly distinguish native platform capabilities from services performed by a third party and identify applicable implementation, certification, recurring review, and data-volume costs.
Combining data across modalities: Respondents should describe how researchers can combine clinical data with other datasets including genomics, molecular results, clinical trials data, registry data, and external research datasets.
OMOP common data model: Roswell Park also seeks support for mapping appropriate clinical and research data to the OMOP Common Data Model for a standardized representation, interoperability, reproducibility, cohort discovery, and participation in multi-institutional and collaborative research networks. Respondents should describe capabilities and services for mapping Roswell Park’s internal structured data entities, local codes, and source values to OMOP standard concepts and vocabularies, including automated and curator-assisted mapping, source-to-concept lineage, mapping governance and versioning, validation and quality assurance, exception handling, and ongoing maintenance as internal source systems and OMOP vocabularies change.
7.13 Computing and Data Science
The ecosystem should support advanced analytical and computational workloads either in the cloud, on premises, or through a hybrid model. Respondents should explain how enterprise data can be safely made available to advanced computing environments without unnecessary duplication and describe support for Python, R, Jupyter notebooks, machine/deep learning, GPU workloads, research computing, and large-scale data processing.
7.14 Artificial Intelligence
Roswell Park anticipates increased use of artificial intelligence across clinical, operational, research, financial, and administrative functions. The Enterprise Data Ecosystem should support governed access to trusted enterprise information for AI workloads. Respondents should describe capabilities related to:
· Machine/Deep learning
· Predictive analytics
· Generative AI and AI Agents
· Natural-language querying of enterprise data
· Model management and monitoring
· AI access to approved enterprise information
7.15 Interoperability and Integration
Respondents should describe supported integration standards and technologies. These may include but are not limited to HL7, FHIR, SMART on FHIR, Bulk FHIR, REST APIs, C-CDA, Database connectivity, Change data capture, SFTP, Flat files, etc. Roswell Park also seeks to understand capabilities for secure data-product delivery through SQL, APIs, files, events, notebooks, BI models, applications, agents, and controlled data-sharing mechanisms.
7.16 Performance Monitoring
Roswell Park seeks capabilities to monitor data access, troubleshoot performance bottlenecks, and optimize data movement. Respondents should describe support for monitoring query and pipeline performance, source-system performance, data freshness, failed jobs, capacity utilization, cost monitoring and controls, and user activity.
Explain how administrators can determine whether performance issues originate within the virtualization platform, source systems, networks, data pipelines, reporting platforms, or user-generated workloads.
7.17 Multi-Vendor Governance
When the proposed solution incorporates multiple vendors, the respondent should provide a responsibility matrix. At minimum, identify responsibility for: overall architecture, platform implementation, data integration, virtualization, security, data governance, research capabilities, OMOP, data archival, EHR conversion, and production support.
The respondent should identify the organization that will serve as the overall solution architect and integration authority and describe the escalation model when an issue spans multiple vendor products.
7.18 Licensing Model and Preliminary Costing Information
Respondents should describe the licensing model for each major component and identify whether pricing is based on users, data volume, compute, storage, queries, enterprise licensing, etc. Identify any additional costs associated with non-production environments, disaster recover, data transfer, cloud resource consumption, API calls, rate limits, AI integration, partner products, support, and training. Respondents should also identify cost drivers that may materially increase total cost over time.
Roswell Park understands that final pricing cannot be established through an RFI, but requests that respondents provide estimated planning estimates for initial implementation cost range, annual recurring cost range, major assumptions, and cost drivers over a ten-year planning horizon. Infrastructure, platform, consumption, support, and implementation of cost drivers are measurable and forecastable.
7.19 Staffing Plan
Respondents should estimate staffing needs for implementation as well as ongoing maintenance and support for the proposed Data Ecosystem solution. Staffing needs should specify role, number of FTE or partial FTE, and note any assumptions.
7.20 Additional Cost Submission Requirements
Respondents shall provide:
· A ten-year planning horizon showing implementation, licensing, infrastructure, consumption, support, maintenance, managed services, and third-party costs.
· Costs separated into one-time implementation costs and recurring annual costs.
· Costs presented by implementation year or fiscal year.
· A detailed breakdown of costs for:
· Platforms licensing
· Data virtualization capabilities
· Data ingestion and orchestration
· Data storage and compute
· Metadata, governance, catalog, and lineage capabilities
· Security and identity services
· EHR integration, migration, validation, and coexistence activities
· Legacy data archival capabilities
· Research capabilities including OMOP, nSight, and related integrations
· Training and knowledge transfer
· Disaster recovery and business continuity
· Managed services and ongoing support
· Third-party products and subcontractors
· Three planning scenarios:
· Minimum viable implementation
· Recommended implementation
· Expanded future-state implementation post-two years
· Cost sensitivity scenarios illustrating the impact of:
· Increased data volume
· Increased storage consumption
· Increased compute utilization
· Increased query volume
· Increased user counts
· Artificial intelligence usage
· Disaster recovery requirements
· Additional environments (development, testing, training)
· Identification of:
· Minimum contractual commitments
· Committed-spend requirements
· Price escalation assumptions
· Renewal assumptions
· True-up mechanisms
· Exit, migration, transition, and data-export costs
· Estimated Roswell Park staffing requirements including:
· Implementation staffing by role
· Ongoing operational staffing by role
· Required skills by role
· Assumptions regarding managed services versus internal support
· Assumptions regarding automation tools that can alter ongoing staffing needs
8. Representative Use Cases
1. Longitudinal Oncology Patient Record: Combine clinical encounters, diagnoses, procedures, medications, laboratory results, pathology, molecular results, imaging metadata, and historical records into a reusable longitudinal oncology representation.
2. Enterprise Revenue Cycle Analytics: Combine line-item billing information, claims, remittance, denials, payer contract details, and finance data to create reusable revenue cycle datasets and metrics consumable through Power BI.
3. Research Cohort Discovery: Allow an authorized researcher to identify patients meeting clinical, treatment, molecular, demographic, and outcome criteria while enforcing research and governance requirements. Roswell Park will use an internally developed data driven discovery and cohort definition platform for integrated clinical, research, regulatory, biospecimen, genomic, and related data. The selected approach should allow internally developed and future research applications to consume governed data through stable interfaces without requiring each application to independently recreate ingestion and transformation logic.
4. Historical Research: Support analysis involving patients treated 10+ years earlier without requiring continued operation of retired clinical applications.
5. OMOP: Map clinical information into OMOP and allow researchers to conduct standardized analyses and participate in multi-institutional studies.
6. Data Virtualization: Combine information from an on-premises database, a cloud platform, the future EHR, and an external CMS dataset without requiring that every dataset be copied into a single physical repository.
7. Enterprise Metrics: Create a governed metric once and reuse the definition in multiple Power BI dashboards, datasets, operational reports, and applications.
8. AI Data Access: Provide an authorized AI application governed access to approved clinical and operational information with appropriate logging, provenance, and security to support audits.
9. Legacy EHR Conversion Validation: Compare information extracted from a legacy clinical platform with converted information in the future EHR to identify missing, incorrectly mapped, or inconsistent records.
10. Clinical Data Archive: Allow an HIM user to retrieve historical patient records for release-of-information purposes after the originating clinical system has been decommissioned.
11. Multimodal Cancer Research: Allow an IRB approved research project to combine clinical phenotype, genomic information, specimen information, imaging, pathology, and outcomes of information.
12. Governed Self-Service: Allow an authorized business analyst to locate an existing trusted dataset, understand its definition and lineage, request access, and use it from Power BI without requiring a new custom extract.
9. Implementation and Transition Services Respondents proposing implementation or advisory services should address the following workstreams and clearly identify inclusions, assumptions, dependencies, Roswell Park responsibilities, subcontractors, and deliverables.
9.1 Discovery and Planning
· Current-state assessment and data/source inventory
· Stakeholder, use-case, and workload discovery
· Architecture options and decision framework
· Platform and implementation roadmap aligned to the EHR program
· Business case, benefits, risks, and total-cost model
9.2 Foundation
· Cloud/on-premises landing zones and connectivity
· Identity, security, privacy, and environment design
· DevSecOps, source control, CI/CD, observability, and service management
· Metadata, catalog, governance, and data-quality foundation
9.3 Build and Migration
· Priority source onboarding and EHR data integration
· Virtualized views and federated access
· Persisted pipelines and curated layers
· Semantic models, certified metrics, APIs, and data products
· Migration or refactoring of selected EDW, marts, reports, and workflows
· nSight, Imagine, and OMOP integration or modernization
9.4 Validation and Deployment
· Unit, integration, performance, security, resilience, data-quality, and user-acceptance testing
· EHR conversion and report reconciliation
· Parallel operations, cutover, rollback, hyper care, and stabilization
· Operational acceptance criteria and production-readiness review
9.5 Adoption and Knowledge Transfer
· Role-based training for administrators, engineers, analysts, data scientists, stewards, product owners, and consumers
· Architecture, configuration, mapping, code, runbook, support, and governance documentation
· Paired delivery, mentoring, and measurable knowledge transfer
· Transition to Roswell Park operations or managed services
9.6 Program Delivery
· Integrated schedule, milestones, dependencies, critical path, and decision points
· Resource plan and role matrix across all parties
· Risk, issue, change, scope, quality, and financial management
· Coordination with EHR and other strategic-program governance
10. Detailed Requirements List
10.1 Data Virtualization and Logical Data Models
10.1.1 Describe the capability to virtualize data stored in relational databases, cloud platforms, and lake houses, object stores and other supported data sources.
10.1.2 Explain how users and applications can query source data without physically copying or relocating the information.
10.1.3 Describe how virtualized and persisted data can be combined within a common logical data model.
10.1.4 Describe support for federated queries across heterogeneous source systems and technologies.
10.1.5 Describe support for query pushdown and identify the source platforms for which pushdown is supported.
10.1.6 Describe other performance optimization techniques used for virtualized or federated queries.
10.1.7 Describe the ability to selectively cache data, including the criteria and controls used to determine when caching is appropriate.
10.1.8 Describe how virtualized datasets can be exposed through SQL, APIs, JDBC, ODBC, and other standard data access interfaces
10.1.9 Describe how on-premises and cloud-hosted applications can consume virtualized data.
10.1.10 Describe mechanisms for providing resilient access to source platforms.
10.1.11 Explain how the solution detects, manages, and recovers from source-system outages.
10.1.12 Describe capabilities for monitoring virtualized and federated query performance.
10.1.13 Describe capabilities for tuning query performance and identifying poorly performing queries or source systems.
10.1.14 Describe how logical enterprise representations can be created by combining information from heterogeneous sources.
10.1.15 Describe support for logical representations such as enterprise patient, longitudinal cancer patient record, provider, encounter, revenue cycle claim, and clinical trial participant.
10.1.16 Explain how logical data models and access interfaces remain stable while underlying source systems change during the EHR transformation.
10.1.17 Describe how source mappings, transformations, and logical models are versioned and governed during source-system transitions.
10.2 Medallion Architecture
10.2.1 Describe how the proposed platform implements a medallion-style architecture.
10.2.2 Explain how data lineage is maintained across bronze, silver, and gold layers.
10.2.3 Explain how transformations, mappings, and processing logic are versioned.
10.2.4 Describe support for incremental data processing across the medallion layers.
10.2.5 Explain how the platform supports reprocessing, recovery, and reconciliation when incremental processing fails or source data changes.
10.3 Reusable Data Products
10.3.1 Describe how governed, self-contained, high-quality data products are created within the proposed ecosystem.
10.3.2 Describe how a single data product can be reused by Power BI solutions.
10.3.3 Describe how a single data product can be reused for research purposes.
10.3.4 Describe how a single data product can be reused by artificial intelligence and machine-learning workloads.
10.3.5 Describe how a single data product can be reused by data science and advanced analytics workloads.
10.3.6 Describe how a single data product can be reused by custom cloud-based and on-premises applications.
10.3.7 Explain how the architecture avoids separate or duplicative preparation of the same data for different use cases.
10.3.8 Describe data-product versioning and change-management capabilities.
10.3.9 Describe governance and access controls applicable to data products.
10.3.10 Describe the technical and business metadata maintained for each data product.
10.3.11 Describe how data products can be accessed through APIs and other standard interfaces.
10.3.12 Describe the process for reviewing, certifying, endorsing, publishing, and retiring data products.
10.3.13 Describe how data-product ownership, stewardship, service levels, quality expectations, and support responsibilities are assigned and communicated.
10.4 Enterprise Data Warehouse and Analytics
10.4.1 Describe how the proposed platform can serve as a virtualized Enterprise Data Warehouse for enterprise reporting and analytics.
10.4.2 Describe the proposed logical or virtual EDW architecture and its major components.
10.4.3 Describe integration with on-premises databases, applications, and other data sources.
10.4.4 Identify existing EDW workload types that can be migrated without major redesign and those that would require redevelopment.
10.4.5 Describe support for on-premises data caching to improve performance.
10.4.6 Describe support for dimensional data models and normalized enterprise data models.
10.4.7 Describe support for slowly changing dimensions, including supported dimension-handling approaches.
10.4.8 Describe query optimization capabilities for enterprise analytical workloads.
10.4.9 Describe workload isolation capabilities and how the platform prevents one workload from adversely affecting another.
10.4.10 Describe how the EDW architecture scales for increased data volume, users, query concurrency, and processing demand.
10.4.11 Describe disaster recovery capabilities, recovery options, and architectural dependencies.
10.4.12 Identify supported recovery-point and recovery-time objectives and any related licensing or infrastructure requirements.
10.5 Power BI Integration
10.5.1 Describe how Power BI business analysts and report developers access virtualized and persisted data.
10.5.2 Identify supported Power BI connectivity modes, interfaces, and authentication methods.
10.5.3 Describe how semantic models integrate with the proposed data ecosystem.
10.5.4 Explain how business rules, transformations, calculations, and metric definitions can be centralized to minimize duplication between the data platform and Power BI.
10.5.5 Describe support for reusable and governed semantic models that can serve multiple reports and dashboards.
10.5.6 If an alternative semantic-model platform is recommended, describe the platform, rationale, benefits, limitations, migration requirements, and effect on existing Power BI investments.
10.5.7 Describe security-policy propagation between the data platform, semantic models, and Power BI reports.
10.5.8 Describe expected performance, capacity-management, and monitoring considerations for Power BI consumption.
10.6 Enterprise Data, Metric, and Metadata Catalog
10.6.1 Describe the proposed searchable enterprise catalog architecture, its major components, and how it supports business, clinical, operational, financial, research, and technical users.
10.6.2 Describe how the catalog inventories and harvests metadata from available data assets (e.g., source systems, databases, schemas, tables, fields, files, interfaces, APIs, semantic models), analytical assets (e.g., reports, dashboards, metrics, KPIs), and research assets (study datasets, cohorts, code, models).
10.6.3 Identify the structured, unstructured, document, imaging, pathology, genomic, biospecimen, and other data types that can be represented in or referenced by the catalog.
10.6.4 Describe how business, clinical, operational, financial, and research definitions are documented, governed, and presented in the catalog.
10.6.5 Describe support for an enterprise business glossary and the association of glossary terms with physical data elements, logical models, metrics, reports, research variables, and data products.
10.6.6 Describe how data owners and data stewards are documented and associated with catalog assets as how this information is presented to users.
10.6.7 Describe how the catalog communicates an asset’s status: trusted, endorsed, certified, under review, restricted, deprecated, archived, superseded, or ungoverned.
10.6.8 Describe the technical metadata captured and how it is captured (e.g., if automated) for cataloged assets, including data types, formats, structures, keys, relationships, interfaces, dependencies, environments, and storage locations.
10.6.9 Identify supported metadata scanners, connectors, APIs, agents, and integration mechanisms, including any technical or licensing limitations.
10.6.10 Describe how metadata is maintained: how new, changed, moved, or deleted assets are identified and owners are notified
10.6.11 Describe support for user-contributed metadata, annotations, tags, ratings, usage guidance, documentation, and collaboration.
10.6.12 Explain how non-technical users can determine whether the data, metrics, calculation logic, reports, research variables, cohorts, or datasets they require already exists.
10.6.13 Describe search, filtering, faceted navigation, tagging, glossary-based discovery, and natural-language discovery capabilities.
10.6.14 Describe how search results can be tailored according to user role, organizational function, authorization, research protocol, and permitted use.
10.6.15 Explain how users can identify data asset and metric owners, stewards, and sources of truth.
10.6.16 Describe how technical and business lineage is captured and displayed at the system, dataset, table, field, report, metric, and data-product levels.
10.6.17 Describe automated lineage capture across ingestion, transformation, virtualization, orchestration, semantic modeling, reporting, analytics, and research workflows.
10.6.18 Describe how access requests are routed, reviewed, approved, granted, tracked, and audited.
10.6.19 Describe any data quality rules and support for documenting quality-based metrics (completeness, validity). Describe if data quality contributes to the assets status.
10.6.20 Describe how known data-quality issues, exceptions, remediation plans, responsible parties, and resolution status are communicated to users.
10.6.21 Describe how refresh history, data latency, update frequency, effective dates, and data currency are communicated.
10.6.22 Describe support for sensitivity classification, including PHI, PII, confidential business information, research-restricted information, and other institutional classifications.
10.6.23 Describe how IRB protocols, consent limitations, data-use agreements, contractual restrictions, and study-specific restrictions can be associated with research assets.
10.6.24 Describe audit logging for metadata changes, approvals, certifications, access requests, governance decisions, downloads, exports, and administrative actions.
10.6.25 Describe how the catalog records the originating source system, extraction process, pipeline, transformation, code version, responsible process, and creation date for each derived asset.
10.6.26 Describe how catalog entries will identify whether an asset remains in a legacy system, is included in conversion, has been migrated, has been archived, has been replaced, or has been retired.
10.6.27 Describe support for persistent identifiers, standardized metadata, controlled vocabularies, machine-readable descriptions, and reusable metadata templates.
10.6.28 Describe how catalog metadata…
This is the start of the file's text. The full file is on GovTribe.
File details come from the government source that posted it. Updated .