‘Secret Shares’ of Patient Health Data Enable Secure Multiparty Research

By Deborah Borfitz

November 20, 2024 | Scientists in Europe collaborated on the first international-level clinical study using secure multiparty computation (MPC), which enabled cross-border cooperation without sharing any personalized health data. The cryptographic method traces back to the late 1970s but has been “severely underused” up until now because it’s computationally complex and few data security experts are familiar with the technology, according to Hendrik Ballhausen, Ph.D., head of research administration at LMU University Hospital in Munich, Germany, and one of the lead players on the initiative.  

The goal of the so-called Federated Secure Computing initiative was to render the technology more user friendly for academic and public sector use. To this end, cloud computing startup bytes for life (the company chooses to leave its name uncapitalized) developed a modern client-server architecture for LMU Munich. The initiative was five years in the making, funded by Germany’s donors’ association Stifterverband. The architecture is now available to the public as free and open-source software. 

The recent pilot experiment demonstrated the value of MR-guided radiotherapy in the treatment of adrenal gland tumors, as reported in npj Digital Medicine (DOI: 10.1038/s41746-024-01293-4). Here, LMU Munich collaborated with LMU University Hospital and Gemelli University Hospital in Rome, together with cryptography experts at Estonia-based Cybernetica, whose industry-grade Sharemind MPC software was used to perform the cryptographic calculations, Ballhausen says. Estonia is one of the most digitally advanced countries in the world and Cybernetica is the leading supplier of commercial MPC solutions. 

This latest use case involved a modern cancer treatment available at few centers, meaning each institution on its own has a limited number of patients, he continues. Pooling that data via “secret shares” for encrypted calculations achieved better statistics regarding the efficacy and safety of the therapeutic approach.  

MPC was the best-case scenario, given that the alternative is “differential privacy” whereby noise gets added to a dataset to obscure entries about any individual patient, says Ballhausen. That statistical framework “works very well [only] if you have a database of many entries.” Secure multiparty computation, by contrast, guarantees that none of the underlying private or confidential data is revealed in the process of collaborative data analysis even if there are few data points to start with. 

Unlike blockchain, where every transaction is verified by all the participants, with MPC “we want to obscure the inputs and make sure by mathematical proofs that only the output of the joint computation will be revealed,” explains Ballhausen. Colleagues in Italy and Germany contributed patient data and were running their own secure compute node. 

“In our configuration, you need at least three independent nodes to guarantee security,” he says, so that neither one mischievous actor nor two colluding ones can get the third party’s information. “Before any information enters the system, it [effectively] gets distributed into little encrypted pieces and shuffled about,” preventing any one party from being able to reconstruct the individual data used in the calculation.  

‘Best of Both Worlds’

Collaborative, data-pooling research is faced with “two fundamental tradeoffs between cooperation and control and between privacy and the value of data,” says Ballhausen. “The question is can we have the best of both worlds; can we have cooperation but still maintain complete control over our own data? The classical answer is no”—not without secure multiparty computation, anyway. 

The cryptographic protocols can be secured even against “compromised actors” trying to game the system, he notes. But in a typical clinical scenario, “we usually assume that the actors are adhering to the protocol and are curious but honest.” 

Ballhausen and his team, within the German Consortium for Translational Cancer Research, performed the first secure MPC with patient data contributed by the radiation oncology departments of LMU Munich and Charité Berlin in 2019.  Scaling that to the European level came with technical challenges as well as many administrative data protection concerns.  

In the latest published paper, researchers “went the extra mile” to break down all the steps in the project that other groups could use as a blueprint to save some time on the administrative details, Ballhausen says. One of his personal interests currently is alleviating patients’ fear that once they share their data, they can never get it back. 

“One of the main features of this kind of calculation is you have to agree every time someone wants to do a calculation with your data; you yourself have to become active,” he says. “It’s not like you give someone your data and then they can do whatever they want [with it].” 

Since laws, companies, and data ownership can all change, says Ballhausen, patients need concrete assurance that “only by their active consent can other people use their data. This is a guarantee that no other system [e.g., IT security and contracts] can provide.” As with bitcoin, it can’t be moved unless they personally do the transferring, he offers as an analogy. 

Improving Survival

In the pilot study, the participants were 48 patients being treated for adrenal gland metastasis with MR-guided radiotherapy—half each from LMU University Hospital and Gemelli University Hospital. High local control of the cancer and low toxicity were observed, with a median overall survival of 19 months—in line with about 40% survival after two years in meta studies. Less than one-quarter of patients in the pilot study experienced any side effects and there were no reports of severe and life-threatening adverse events.  

CT scans have been the standard for radiation therapy planning for many years. “The state of the art is to take a cone beam CT right before each daily treatment fraction for patient positioning,” Ballhausen says. The problem is that soft tissue, including the tumors themselves, are constantly moving. Consequently, treatment planning and actual tumor position may diverge during the [roughly] 10 minutes it takes to move the patient into the treatment position and perform the radiation.”  

The advantage when using the online MR tomography technique is that it visualizes the tumor and surrounding organs at risk even during the treatment session, he adds. It also has great contrast and therefore provides a better picture of the soft tissue. In principle, oncologists can adjust the beam to accommodate any observed movement of the tumor, improving efficacy of treatment and minimizing the odds of adverse effects on nearby healthy tissue.  

Speed to Answers

One of the main reasons MPC has been underutilized in institutional healthcare settings is the difficulty in introducing the method into an operational environment guided by regular business logic, says Ballhausen. For the first trial run in Germany five years ago, it took months to complete a single calculation that left no one with the desire to do a second one. 

This led to the development of Federated Secure Computing, middleware architecture for non-expert business users—researchers, clinicians, study nurses, and IT students among them—which leaves all the difficult cryptography work to the experts and offloads its complexity to the server side, he says. It is a web-based application programing interface, or API, so collaborators all use their own server forming a secure peer-to-peer network and compute cluster that provides privacy and returns only the intended results.  

In the pilot study, it took non-expert colleagues in Italy less than 20 minutes to run the software to upload their data, Ballhausen reports. The “value-add” of the project is that it succeeded in making secure MPC usable for the public sector as well as small- and medium-size enterprises. The calculations themselves took only seconds—negligible overhead in terms of a clinical study that had been running for years.  

Critical to the success of the Federated Secure Computing project was the inclusion of data use and access committee members from the very beginning, says Ballhausen, above and beyond the usual ethics and study protocols and cooperation agreements between the partnering institutions. Every step of the way, he took the time to explain to everyone, including government data protection officials, what he and his team were going to do. A specialty law firm was even hired to evaluate the technology, which was particularly informative from the standpoint of patient consent, he adds.  

Researchers obtained written consent from patients after openly stating that their data would be shared between the cooperating parties in a “encrypted and pseudonymized” fashion, Ballhausen says. “We could have said anonymized, but we didn’t want to... compromise honest patient consent by saying that the technology would circumvent any problems that the patient might have.”   

Their extensive efforts were worth it in the end because they succeeded in creating a blueprint that could be used in the demanding context of clinical research on cancer patients—no easy feat, given strict European regulations on the protection of patient privacy and data protection, says Ballhausen. Probably only genetics data or anything related to children would have been more difficult, he adds. 

Replication Opportunities

The research team is now contemplating their first transatlantic use case, which may take more years to orchestrate, Ballhausen says. As the founder of bytes for life, he’d like to see Federated Secure Computing applied in the public sector as well as in the finance, manufacturing, and pharmaceutical industries.    

From the standpoint of LMU University Hospital, e-health applications are particularly appealing, says Ballhausen. Patients might be discharged sooner with electronic devices for collecting patient-reported outcomes and surveillance via telemedicine, “and of course we’d love to use and access that data.”   

The bigger vision here is to enable citizen science where people can opt to share healthcare data on their smartphone for research purposes. “Imagine a world where you .... can decide for yourself what institutions you enable to use your data without giving [it] away for free or forever,” he says. 

“In my mind [the open-source software project] is easy to replicate, and we have done everything we could to enable this,” says Ballhausen. A “propaedeutic” backend designed for many simple MPC calculations is included with the codebase for free, and users may choose any of the more powerful academic or commercial MPC backends such as Cybernetica’s Sharemind MPC. 

“Contemplate on the thought that privacy and collaboration are not at odds with each other anymore,” he advises. A world of cooperative opportunities has opened now that “doubts about the security and privacy of data sharing” have been alleviated.