Accurate training of the Cox proportional hazards model on vertically-partitioned data while preserving privacy.

BACKGROUND: Analysing distributed medical data is challenging because of data sensitivity and various regulations to access and combine data. Some privacy-preserving methods are known for analyzing horizontally-partitioned data, where different organisations have similar data on disjoint sets of people. Technically more challenging is the case of vertically-partitioned data, dealing with data on overlapping sets of people. We use an emerging technology based on cryptographic techniques called secure multi-party computation (MPC), and apply it to perform privacy-preserving survival analysis on vertically-distributed data by means of the Cox proportional hazards (CPH) model. Both MPC and CPH are explained. METHODS: We use a Newton-Raphson solver to securely train the CPH model with MPC, jointly with all data holders, without revealing any sensitive data. In order to securely compute the log-partial likelihood in each iteration, we run into several technical challenges to preserve the efficiency and security of our solution. To tackle these technical challenges, we generalize a cryptographic protocol for securely computing the inverse of the Hessian matrix and develop a new method for securely computing exponentiations. A theoretical complexity estimate is given to get insight into the computational and communication effort that is needed. RESULTS: Our secure solution is implemented in a setting with three different machines, each presenting a different data holder, which can communicate through the internet. The MPyC platform is used for implementing this privacy-preserving solution to obtain the CPH model. We test the accuracy and computation time of our methods on three standard benchmark survival datasets. We identify future work to make our solution more efficient. CONCLUSIONS: Our secure solution is comparable with the standard, non-secure solver in terms of accuracy and convergence speed. The computation time is considerably larger, although the theoretical complexity is still cubic in the number of covariates and quadratic in the number of subjects. We conclude that this is a promising way of performing parametric survival analysis on vertically-distributed medical data, while realising high level of security and privacy.

Cross-population evaluation of cervical cancer risk prediction algorithms.

BACKGROUND: Cervical cancer is a preventable disease, despite being one of the most common types of female cancers worldwide. Integrating existing programs for cervical cancer screening with personalized risk prediction algorithms can improve population-level cancer prevention by enabling more targeted screening and contrive preventive healthcare innovations. While algorithms developed for cervical cancer risk prediction have shown promising performance in internal validation on more homogeneous populations, their ability to generalize to external populations remains to be assessed. METHODS: To address this gap, we perform a cross-population comparative study of personalized prediction algorithms for more personalized cervical cancer screening. Using data from the Norwegian and Estonian populations, the algorithms are validated on internal and external datasets to study their potential biases and limitations when applied to different populations. We evaluate the algorithms in predicting progression from low-grade precancerous cervical lesions, simulating a clinically relevant application of more personalized risk stratification. RESULTS: As expected, our numerical experiments show that algorithm performance varies depending on the population. However, some algorithms show strong generalization capacity across different data sources. Using Kaplan-Meier estimates, we demonstrate the strengths and limitations of the algorithms in detecting cancer progression over time by comparing to the trends observed from data. We assess their overall discrimination performance in personalized risk predictions by analyzing the accuracy and confidence in individual risk estimates. DISCUSSION AND CONCLUSION: This study examines the effectiveness of personalized prediction algorithms across different populations. Our results demonstrate the potential for generalizing risk prediction algorithms to external populations. These findings highlight the importance of considering population diversity when developing risk prediction algorithms.

Nordcan.R: a new tool for federated analysis and quality assurance of cancer registry data.

Aim of the article: We present our new GDPR-compliant federated analysis programme (nordcan.R), how it is used to compute statistics for the Nordic cancer statistics web platform NORDCAN, and demonstrate that it works also with non-Nordic data. Materials and methods: We chose R and Stata programming languages for writing nordcan.R. Additionally, the internationally used CRG Tools programme by International Agency for Research on Cancer (IARC/WHO) was employed. A formal assessment of (GDPR-compliant) anonymity of all nordcan.R outputs was performed. In order to demonstrate that nordcan.R also works with non-Nordic data, we used data from the Netherlands Cancer Registry. Results: nordcan.R, publicly available on Github, takes as input cancer and general population data and produces tables of statistics. Each NORDCAN participant runs nordcan.R locally and delivers its results to IARC for publication. According to our anonymity assessment the data can be shared with international organizations, including IARC. nordcan.R incidence results on Norwegian and Dutch data are highly similar to those produced by two other independent methods. Conclusion: nordcan.R produces accurate cancer statistics where all personal and sensitive data are kept within each cancer registry. In the age of strict data protection policies, we have shown that international collaboration in cancer registry research and statistics reporting is achievable with the federated analysis approach. Undertakings similar to NORDCAN should consider using nordcan.R.