People acknowledge the value of data and the benefits of data collaborations
. Yet, concerns about data privacy and security are increasing .
Especially when dealing with sensitive data, these concerns are often justified.
To solve these challenges, the leaders in the field of confidential computing
[https://confidentialcomputing.io/] establish new technologies to make data
collaborations simple and safe .
This blog focuses on applications of the avato
platform in the health care sector. A particular focus lies on challenges
associated with clinical trials.
We start with a straightforward example of a use-case termed Confidential Random
Oracle. Then, we are looking at a more challenging but equally important
use-case called Model Verification. Confidential Random Oracle and Model
Verification are often concatenated. The last use-case is of statistical
analysis on multi-party private data sets of clinical trials using our platform
In summary, the use-cases herein are:
* Confidential Random Oracle
* Model Verification
* Statistical analysis of multi-party private data sets
True Randomization for Clinical Trials Assignment – Confidential Random Oracle
Clinical trials are a crucial part of discovering new treatments or test new
medications. In general, they are conducted in different phases. In Phase 0,
researchers are interested in what, for instance, the drug does to the human
body. It involves typically 10 to 15 participants. Phase I is seen as a phase to
screen safety including the identification of possible side effects. This phase
generally examines a group of 20 to 80 participants. Several hundreds of
patients (300 to 500) participate in Phase II for further evaluation of the
efficacy. Following Phase II is Phase III, in which typically thousands of
patients are monitored. Lastly, in Phase IV, the post-marketing studies are
Clinical trials in Phase II and III use a set-up in which one group receives the
new, presumably better, or even curing medication, whereas the other group
receives a standard therapy or a placebo. Recruitments often involve different
parties across different geographies.
Group assignments during clinical trials are not only essential to design a
valid trial but can be life-changing to the patients. Thus, posing many
challenges ranging from ethical to technical to the trial organizer.
Randomization, meaning by-chance assignment of patients to individual groups, is
often used during trial design. Unfortunately, the human brain is not well
equipped to perform randomization. Therefore, algorithms are taking over the
Avato`s Confidential Random Oracle removes bias and the potential to modify
group assignments for any involved party by leveraging confidential computing.
The picture below describes how the avato platform can be used as a Confidential
Random Oracle in a clinical trial.
1. Clinical trial organizer invites partners to participate in a clinical
2. Participating parties recruit patients for the clinical trial.
3. Avato sends out encryption key to each participating partner.
4. Local data encryption of the data on each partner's infrastructure.
5. The various partners are sending the encrypted data back to the avato
platform. The raw data remains on the local system.
6. Avato computes the Confidential Random Oracle assigning each entry to either
the drug or control group.
7. Each partner receives an encrypted result with the assignment of her
patients to the individual groups.
Avato Confidential Random Oracle calculations for patient assignment to
different groups in a clinical trial.Advantages of using the avato platform for
a Confidential Random Oracle are:
* Confidential Random Oracle computation on one or multiple datasets possible
* No need for revealing identity at any stage mitigating the need for
anonymization on the partner's side
* Flexible and Scalable
* Fast and easy to use via a simple browser-based interface
* Confidential Random Oracle Function provided by decentriq
Testing the Confidential Random Oracle for trial group assignment – Model
The Confidential Random Oracle described above designates an essential use-case
where Model Verification is crucial.
In general, Model Verification is used to prove that a given model or
calculation was applied to a given data set. Model Verification is useful in
cases where the owner of the model is not able or does not want to reveal
details of the model to the data owner.
Within the avato platform, there are two ways of Model Verification possible.
First, proving the identity of the model and second, confirming the ownership of
the model. Here we focus on the first verification.
Model Verification follows the following steps in avato:
1. Prior to using avato the parties decide which function they want to check
2. Avato sends the specific encryption keys to both parties.
3. The model-party and the data-party are using their specific encryption keys
for local encryption of the data and the model.
4. Both parties are sending the encrypted model and the encrypted data back to
the avato platform, respectively.
5. Avato performs Model Verification of the model on the data.
6. Avato sends the encrypted result of the Model Verification back to the
Avato confirms the function applied to the data using a Model Verification
processAdvantages of using the avato platform for Model Verification dealing with
clinical trial group assignments:
* Guaranteed, provable random group selection
* Guarantee of proof without any party, including decentriq, being able to
reveal the data or function
* Only the data-party receives the prove and guarantee of the model
Of course, Model Verification is not limited by the complexity of the model. To
learn more about general applications of Model Verification, get in touch with
us [https://decentriq.ch/]to receive a white paper about Model Verification.
Statistical analysis of a Clinical Trial – Multi-party Confidential Computing
Clinical trials designed and executed well are providing a great wealth of data.
Statistical analysis of this data may reveal differences between the efficacy of
a new drug versus a control treatment. Various statistical tests are performed
on clinical data sets. However, it would be beyond the scope of this blog to
explain all possible statistics, their differences, pros, and cons.
In this use-case, we are assuming that the different parties conducted a
well-designed randomized clinical trial. Randomize group assignment for the
trial was done using the Confidential Random Oracle function.
For the analysis of the data, the parties decided to run a given function
(statistical test) on the data. As described above, the avato platform enables
the data-parties to verify this model using the Model Verification function.
The following sequence in avato results in a complete statistical analysis of
the trial data:
1. At the agreed endpoint of the clinical trial, the organizer notifies the
partners and decentriq.
2. Avato sends out individual encryption keys to all parties involved.
3. Locally, each party, including the analysis-party, encrypts the data and the
model using the provided key.
4. All parties are sending back the encrypted data to the avato platform. The
raw data remains local on each partner's infrastructure.
5. Inside the avato platform, the statistical analysis is computed without
revealing the data or the function to any party, including decentriq.
6. The results are individually encrypted and sent to the eligible parties for
Avato performs confidential multi-party statistical analysis on multiple
datasets provided by the different parties.Advantages of using the avato
platform for multi-party confidential statistical analysis of clinical trials:
* Avato runs almost any test or model on the data.
* Avato does not reveal the data or the model to any of the parties including
* Avato uses encryption to protect the results from anyone not eligible to see
* Combined with Confidential Random Oracle and Model Verification, avato is
adding trust to the confidentiality in clinical trials.
As you can see, the possibilities of avato are very diverse and not limited by
complexity. In this blog, we focused on basic operations on highly sensitive
datasets in the health care space.
In our next blog, we are expanding the use-cases of avato for different