# From Network Governance to Real-World-Time Learning: A High-Reliability Operating Model for Rare Cancers

## TL;DR

This paper introduces a model for improving rare cancer care by creating a learning system that tracks and improves treatment outcomes across hospitals.

## Contribution

The paper presents a novel operating model for rare-cancer care that ensures continuous learning and benchmarking without relying on a single metric.

## Key findings

- The model includes a minimal but high-quality data backbone for benchmarking and targeted improvements.
- It specifies data primitives and validity guardrails to prevent unsafe inferences in rare cancer benchmarking.
- The approach supports scalable, learnable excellence in rare-cancer care while protecting against gaming and inference drift.

## Abstract

Treating rare cancers is challenging because they are infrequent, biologically diverse, and care pathways are often fragmented as patients move between hospitals for diagnosis and therapy. As a result, avoidable delays, poorly coordinated steps, and inconsistent decision-making can harm patients and also make it hard to learn what works best from routine care. In this article, we propose a practical operating model for running rare-cancer care as a “learning system” that continuously measures its own performance and improves over time. Using sarcoma care in Switzerland as an example, we describe how coordinated multidisciplinary decision-making, reliable patient routing, and a minimal but high-quality data backbone can support fair benchmarking and targeted improvements without relying on a single metric. This blueprint may help networks improve outcomes, reduce avoidable patient burden and inefficiencies, and create better real-world evidence to guide future research, in order to continuously improve patient care.

Background: Rare cancers combine low incidence with high biological heterogeneity and multi-institutional care trajectories. These features make single-center learning structurally incomplete and render pathway fragmentation a dominant driver of preventable harm, variability, and waste. In this context, care quality is best understood as a property of pathway integrity across routing, diagnostics (imaging/biopsy planning), multidisciplinary intent-setting, definitive treatment, and surveillance—rather than as a department-level attribute. Objective: To define a pragmatic, transferable operating blueprint for a rare-cancer Learning Health System (LHS) that turns routine care into continuous, auditable learning under explicit governance, while maintaining claims discipline and protecting measurement validity. Approach: We synthesize an implementation-oriented operating model using the Swiss Sarcoma Network (SSN) as an exemplar. The blueprint couples clinical governance (Integrated Practice Unit logic, hub-and-spoke routing, auditable multidisciplinary team decision systems) with an interoperable real-world-time data backbone designed for benchmarking, pathway mapping, and feedback. The operating logic is expressed as a closed-loop control cycle: capture → harmonize → benchmark → learn → implement → re-measure, with explicit owners, minimum requirements, and failure modes. Results/Blueprint: (i) The model specifies a minimal set of data primitives—time-stamped and traceable decision points covering baseline and tumor characteristics, pathway timing, treatment exposure, outcomes and complications, and feasible longitudinal PROMs and PREMs; (ii) a VBHC-ready, multi-domain measurement backbone spanning outcomes, harms, timeliness, function, process fidelity, and resource stewardship; and (iii) two non-negotiable validity guardrails: explicit applicability (“N/A”) rules and mandatory case-mix/complexity stratification. Implementation is treated as a governed step with defined workflow levers, fidelity criteria, balancing measures, and escalation thresholds to prevent “dashboard medicine” and surrogate-driven optimization. Conclusions: This perspective contributes an operating model—not a platform or single intervention—that enables credible improvement science and establishes prerequisites for downstream causal learning and minimum viable digital twins. By distinguishing enabling infrastructure from the governed clinical system as the primary intervention, the blueprint supports scalable, learnable excellence in rare-cancer care while protecting against gaming, inequity, and inference drift. Distinct from generic LHS or VBHC frameworks, this blueprint specifies validity gates required for rare-cancer benchmarking—explicit applicability (“N/A”) rules, denominator integrity/capture completeness disclosure, anti-gaming safeguards, and escalation governance. These elements are critical in rare cancers because small denominators, high heterogeneity, and multi-institutional pathways otherwise make benchmarking prone to artifacts and unsafe inferences.