Sharing Privacy Protected and Statistically Sound Clinical Research Data Using Outsourced Data Storage

Box 2

SCDA Protocol.

Preparation. Each hospital ℋ_𝑖 (1≤𝑖≤𝗇_ℋ) runs the CSA.Key(1^𝑛,1^𝑚,𝑞) algorithm to get a public key pair

𝑝𝑘_𝑖=(𝐀_𝑖,𝐌_𝑖) and a secret key 𝑠𝑘_𝑖=𝐓_𝑖. Each researcher ℛ_𝑗 (1≤𝑗≤𝗇_ℛ) runs the GHV∗.Key(1^𝑛,1^𝑚,𝑞) algorithm to

get a public key 𝑝𝑘_𝑗=𝐀_𝑗 and a secret key 𝑠𝑘_𝑗=𝐓_𝑗.

Data Publication. For all 𝗄 (1≤𝗄≤𝗇_𝖢) and 𝑙 (1≤𝑙≤𝗇_𝖳𝑖), each hospital ℋ_𝑖 (1≤𝑖≤𝗇_ℋ) runs the

CSA.Enc(𝑝𝑘_𝑖,𝐛_{𝑖,𝗄,𝑙}) algorithm to get a ciphertext pair (𝐜_{𝑖,𝗄,𝑙},𝐜^′_{𝑖,𝗄,𝑙}), where 𝐛_{𝑖,𝗄,𝑙} is the 𝑙th cell of the 𝗄th numeric

clinical data attribute of the 𝑖th hospital ℋ_𝑖. Then each hospital ℋ_𝑖 (1≤𝑖≤𝗇_ℋ) makes its data anonymous using

anonymity techniques for de-identification. Finally, each hospital ℋ_𝑖 (1≤𝑖≤𝗇_ℋ) outsources its data in the cloud

servers.

Query. The 𝑗th researcher ℛ_𝑗 sends a request for an aggregated data to the aggregator 𝒜𝒢𝒢. We assume that ℛ_𝑗 is

interested in the 𝗄th attribute and |ℐ| hospitals, ℋ_𝑖 (𝑖∈ℐ), have the data in which ℛ_𝑗 is interested. Each hospital ℋ_𝑖

(𝑖∈ℐ) has 𝑠_𝑖 tuples that meet the request, respectively.

Aggregation.  𝒜𝒢𝒢 retrieves all ciphertext pairs satisfying ℛ_𝑗's request. For each 𝑖∈ℐ, 𝒜𝒢𝒢 runs the

GHV∗.Add(𝐜_{𝑖,𝗄,𝑙1},…,𝐜_{𝑖,𝗄,𝑙𝑠𝑖}) and GHV∗.Add(𝐜^′_{𝑖,𝗄,𝑙1},…,𝐜^′_{𝑖,𝗄,𝑙𝑠𝑖}) algorithm to get (𝐜_𝑖,𝐜^′_𝑖). Then 𝒜𝒢𝒢 runs the

CSA.Agg({(𝐜_𝑖,𝐜^′_𝑖)}_𝑖∈ℐ,ℐ) algorithm to get 𝐜 and {𝐜^′_𝑖}_𝑖∈ℐ.

Consent.  𝒜𝒢𝒢 determines the order in which hospitals consented to ℛ_𝑗's request, then sends 𝐜 to the first hospital

and (𝐜^′_𝑖,𝐀_𝑗) to each hospital ℋ_𝑖 (𝑖∈ℐ). Each hospital ℋ_𝑖 (𝑖∈ℐ) in turn performs the dec-Aggregation phase in

our CSA protocol. If any hospital ℋ_𝑖 (𝑖∈ℐ) does not want ℛ_𝑗 to have the aggregated clinical data, it can deny the

request by simply not performing the dec-Aggregation phase.

Acquisition. After the consent procedure, the last hospital ℋ_|ℐ| sends 𝐜_|ℐ| to ℛ_𝑗. ℛ_𝑗 runs the GHV∗.Add(𝑠𝑘_ℛ𝑗,𝐜_|ℐ|)

to get 𝐛 that is an aggregated clinical data.