Jan Franz Nygård

The proliferation of real-world healthcare data has substantially expanded opportunities for collaborative research, yet stringent privacy regulations hinder the pooling of sensitive patient records in a single location. To address this dilemma, we propose a multiparty homomorphic encryption-based framework for privacypreserving federated Kaplan–Meier survival analysis, surpassing existing methods by offering native floating-point support, a detailed theoretical model, and explicit mitigation of reconstruction attacks. Compared to prior work, our framework provides a more comprehensive analysis of noise growth and convergence, guaranteeing that the encrypted federated survival estimates closely match centralized (unencrypted) outcomes. Formal utility-loss bounds demonstrate that as aggregation and decryption noise diminish, the encrypted estimator converges to its unencrypted counterpart. Extensive experiments on the NCCTG Lung Cancer and the IKNL synthetic Breast Cancer dataset confirm that the mean absolute error (MAE) and root mean squared error (RMSE) remain low, indicating only negligible deviations between encrypted and non-encrypted federated survival curves. Log-rank tests further reveal no significant difference between federated encrypted and non-encrypted analyses, thereby preserving statistical validity. Additionally, an in-depth reconstruction-attack evaluation shows that smaller federations (2–3 providers) with overlapping data are acutely vulnerable, a challenge our multiparty encryption effectively neutralizes. Larger federations (5–50 sites) inherently degrade reconstruction accuracy, yet encryption remains prudent for maximum confidentiality. Despite an overhead factor of 8–19× compared to non-encrypted computation, our results show that threshold-based homomorphic encryption is feasible for moderate-scale deployments, balancing security needs with acceptable runtime. By furnishing robust privacy guarantees alongside high-fidelity survival estimates, this framework significantly advances the state of the art in secure, multi-institutional survival analysis.

Synthetic data is emerging as a cost-effective solution necessary to meet the increasing data demands of AI development, created either from existing knowledge or derived from real data. The traditional classification of synthetic data types into hybrid, partial or fully synthetic datasets has limited value and does not reflect the ever-increasing methods to generate synthetic data. The generation method and their source jointly shape the characteristics of synthetic data, which in turn determines its practical applications. We make a case for an alternative approach to grouping synthetic data types that better reflect privacy perspectives in order to facilitate regulatory guidance in the generation and processing of synthetic data. This approach to classification provides flexibility to new advancements like deep generative methods and offers a more practical framework for future applications.

An external control arm based on health registry data can serve as an alternative comparator in single-arm drug development studies that lack a benchmark for comparison to the experimental treatment. However, accessing such observational healthcare data involves a lengthy and intricate application process, delaying drug approval studies and access to novel treatments. Clinical trials typically comprise only a few hundred patients usually with high-cardinality features, which makes individual data instances more exposed to re-identification attacks. We examine whether synthetic data can serve as a proxy for the empirical control arm data by providing the same research outcomes while reducing the risk of information disclosure. We propose a reversible data generalization procedure to address these particular data characteristics that can be used in conjunction with any generator algorithm. It reduces the input data cardinality pre-synthesis and reverses it post-synthesis to regain the original data structure. Finally, we test a selection of state-of-the-art generators against a suite of utility and privacy metrics. The external control arm benchmark was generated using data from Norwegian health registries. In this retrospective study, we compare various synthetic data generation algorithms in numerical experiments, focusing on the utility of the synthetic data to support the conclusions drawn from the empirical data, and analysing the risk of sensitive information disclosure. Our results indicate that data generalization is advantageous to enhance both data utility and privacy in smaller datasets with high cardinality. Moreover, the generator algorithms demonstrate the ability to generate synthetic data of high utility without compromising the confidentiality of the empirical data. Our finding suggests that synthetic external control arms could serve as a viable alternative to observational data in drug development studies, while reducing the risk of revealing sensitive patient information.

PURPOSE: Cancer recurrence is an important long-term outcome of cancer survivors that is often not routinely collected and recorded by population-based registries. In this study, we review population-based studies to determine the current availability, landscape, and infrastructure of long-term outcomes, particularly metastatic recurrence, in women initially diagnosed with nonmetastatic breast cancer (MBC). METHODS: We reviewed the literature to identify studies that used population-based registry data to examine the distribution of metastatic recurrence in women diagnosed with non-MBC. Data on outcomes and methods of ascertainment were extracted. Registry infrastructure including sources and funding was also reviewed. RESULTS: A total of 23 studies from 11 registries in eight countries spanning Europe, North America, and Oceania were identified and included in the review. Most studies were retrospective in nature and collected recurrence data only for ad hoc studies rather than as part of their routine registration. Definition of recurrence and data sources varied considerably across studies: the cancer-free time interval between the start of follow-up and risk window ranged from the diagnosis of primary tumor (n = 7) to 6 months from diagnosis (n = 1); the start of follow-up differed between initial diagnosis (n = 16) and treatment (n = 7). CONCLUSION: Cancer surveillance should encompass outcomes among survivors for research and monitoring. Studies are underway, but more are needed. Cancer registries should be supported to routinely collect recurrence data to allow complete evaluation of MBC as an outcome to be conducted and inform health care providers and researchers of the prognosis of both nonmetastatic and metastatic patients with breast cancer.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To explore the standalone breast cancer detection performance at different risk score thresholds of a commercially available artificial intelligence (AI) system. Materials and Methods This retrospective study included information from 661,695 digital mammographic examinations performed among 242,629 female individuals screened as a part of x, 2004-2018. The study sample included 3807 screen-detected cancers (SDC) and 1110 interval breast cancers (IC). A continuous examination level risk score by the AI system was used to measure performance as the area under the receiver operating characteristic curve (AUC) with 95% CIs and cancer detection at different AI risk score thresholds. Results The AUC of the AI system was 0.93 (95% CI: 0.92-0.93) for SDC and IC combined and 0.97 (95% CI: 0.97-0.97) for SDC. In a setting where 10% of the examinations with the highest AI risk scores were defined as positive and 90% with the lowest scores as negative, 92.0% (3502/3807) of the SDC and 44.6% (495/1100) of the IC were identified by AI. In this scenario, 68.5% (10 987/16 029) of false positive screening results (negative recall assessment) were considered negative by AI. When 50% was used as the cut-off, 99.3% (3781/3807) of the SDC and 85.2% (946/1100) of the IC were identified as positive by AI, while 17.0% (2725/16 029) of the false positives were considered as negative. Conclusion The AI system showed high performance in detecting breast cancers within 2 years of screening mammography and a potential for triaging low-risk mammograms to reduce radiologist workload. ©RSNA, 2024.

The Cancer Registry of Norway (CRN) collects, curates, and manages data related to cancer patients in Norway, supported by an interactive, human-in-the-loop, socio-technical decision support software system. Automated software testing of this software system is inevitable; however, currently, it is limited in CRN’s practice. To this end, we present an industrial case study to evaluate an AI-based system-level testing tool, i.e., EvoMaster, in terms of its effectiveness in testing CRN’s software system. In particular, we focus on GURI, CRN’s medical rule engine, which is a key component at the CRN. We test GURI with EvoMaster’s black-box and white-box tools and study their test effectiveness regarding code coverage, errors found, and domain-specific rule coverage. The results show that all EvoMaster tools achieve a similar code coverage; i.e., around 19% line, 13% branch, and 20% method; and find a similar number of errors; i.e., 1 in GURI’s code. Concerning domain-specific coverage, EvoMaster’s black-box tool is the most effective in generating tests that lead to applied rules; i.e., 100% of the aggregation rules and between 12.86% and 25.81% of the validation rules; and to diverse rule execution results; i.e., 86.84% to 89.95% of the aggregation rules and 0.93% to 1.72% of the validation rules pass, and 1.70% to 3.12% of the aggregation rules and 1.58% to 3.74% of the validation rules fail. We further observe that the results are consistent across 10 versions of the rules. Based on these results, we recommend using EvoMaster’s black-box tool to test GURI since it provides good results and advances the current state of practice at the CRN. Nonetheless, EvoMaster needs to be extended to employ domain-specific optimization objectives to improve test effectiveness further. Finally, we conclude with lessons learned and potential research directions, which we believe are applicable in a general context.

The Cancer Registry of Norway (CRN) collects information on cancer patients by receiving cancer messages from different medical entities (e.g., medical labs, hospitals) in Norway. Such messages are validated by an automated cancer registry system: GURI. Its correct operation is crucial since it lays the foundation for cancer research and provides critical cancer-related statistics to its stakeholders. Constructing a cyber-cyber digital twin (CCDT) for GURI can facilitate various experiments and advanced analyses of the operational state of GURI without requiring intensive interactions with the real system. However, GURI constantly evolves due to novel medical diagnostics and treatment, technological advances, etc. Accordingly, CCDT should evolve as well to synchronize with GURI. A key challenge of achieving such synchronization is that evolving CCDT needs abundant data labelled by the new GURI. To tackle this challenge, we propose EvoCLINICAL, which considers the CCDT developed for the previous version of GURI as the pretrained model and fine-tunes it with the dataset labelled by querying a new GURI version. EvoCLINICAL employs a genetic algorithm to select an optimal subset of cancer messages from a candidate dataset and query GURI with it. We evaluate EvoCLINICAL on three evolution processes. The precision, recall, and F1 score are all greater than 91%, demonstrating the effectiveness of EvoCLINICAL. Furthermore, we replace the active learning part of EvoCLINICAL with random selection to study the contribution of transfer learning to the overall performance of EvoCLINICAL. Results show that employing active learning in EvoCLINICAL increases its performances consistently.