The proposed system was applied and deployed into the mHealth app, which treats chronic insomnia based on cognitive behavioral therapy for insomnia (CBTi) [31]. After a favorable formal review by the Japanese Pharmaceuticals and Medical Devices Agency, research on the mHealth app was conducted by the digital therapeutics company, SUSMED, Inc. (Tokyo, Japan). Informed consent was obtained from the patients for publication of this study. The study has received ethical approval from the Ethics Committee and registered to clinical trial registry (UMIN000032951). All the methods were performed in accordance with the relevant guidelines and regulations.

The mHealth records collected from patients were divided into subjective and objective data. The subjective data, which include clinical indicators, sleep status, and the review of daytime activities, were collected through a self-administered questionnaire. The objective data, which include the results of the Psychomotor Vigilance Test [32], were evaluated by measuring the touch response using the function of mobile phones. For clinical indicators, Athens Insomnia Scale [33], Epworth Sleepiness Scale [34], and Quick Inventory of Depressive Symptomatology [35] were collected. For sleep status, time to go to bed, time to fall asleep, time to wake up, and time to get out of bed were recorded. Along with the medical information, timestamps of the app operation were collected. These clinical indicators were collected using a mobile phone app. All data were stored in the JavaScript Object Notation (JSON) format in the database.

The collected data from the patients’ devices were sent to the blockchain network via relay servers. We used three relay servers, and the app randomly selected two relay servers to send the data after the authentication of the client device. By deploying the relay proxy and setting the blockchain software development kits (SDKs) to write-only mode, the relay server sent the received data to the blockchain network. The authentication of the relay server was conducted with a single common authentication server. By configuring the Internet Protocol (IP) address restriction to the listed relay servers, the blockchain network, which contains the medical data, was protected against external attack. The blockchain network was made up of three organizations that contain two validating peers. Each account for nodes of the blockchain network and the relay servers were managed by independent departments in SUSMED, Inc. The strictness of the governance of the data management can be adjusted by individually managing the accounts of different stakeholders, such as pharmaceutical companies, contract research organizations, and regulatory agencies.

We used Hyperledger Fabric v1.0 to operate the blockchain network because Hyperledger is an open-source blockchain platform and has become widely used [36,37].

The blockchain network was administered by a collection of organizations. Each organization had multiple nodes. In this study, the network had three organizations and each organization had two nodes. As the state database, CouchDB was used to store the JSON document [38].

The nodes executed an installed chaincode and returned hash values generated from the execution result. The secure hash algorithm SHA-256 was used to compute the hash values encoded into the blocks of the blockchain [39]. To execute the transaction, each node followed the consensus algorithm, which was called endorsement policy [11], although the previous version of Hyperledger Fabric used Practical Byzantine Fault Tolerance as a consensus algorithm [40,41].

The endorsement policy was set in units of organizations, and the flexible set was available according to the needs of the app. Each organization issued one signature. The node that validated the transactions in each organization was called the endorser. In this study, the validation of each transaction required more than two signatures from three organizations.

Under the endorsement policy, transactions were validated and accepted in the following processes:

All data, including the client hashchain, were registered in the blockchain network via relay servers to secure the tamper-resistance of the data. In contrast, the secure string for the calculation of the hash value was preserved in the client device. At the end of the study, the secure string was sent to the blockchain network to verify the hashchain. The client hash values were calculated on the mobile phone based on medical data, the secure string, and the previous hash value using the SHA-256 algorithm [39].

There are issues regarding cybersecurity (ie, concerning external actors) and governance/authenticity (eg, internal actors like researchers) in medical data management. The tamper-resistance of the data registered in the blockchain network against external attack has been proven in previous studies. The reliability of the data against internal actors in the blockchain network can also be guaranteed by managing the accounts for each node by different stakeholders, such as pharmaceutical companies, contract research organizations, and regulatory agencies. Here, we evaluated how the data manipulation before registration to the blockchain network can be detected and distinguished by simulating the following malicious access. The artificial data were created for each scenario and tested if the fraudulent access were detected and the original data can be distinguished from the illegal data. Since the results of the manipulation of the data are deterministic due to collision-resistant hash functions of SHA-256 [42], we verified the result of a single manipulation in each scenario. Since the accounts for nodes of the blockchain network and the relay servers were managed by independent departments, both internal and external actors can be simulated by hacking each server.

In our previous study, the mHealth data obtained by a mobile phone were uploaded to the blockchain network. We then evaluated the robustness of the network and the tamper-resistance of the data in the blockchain network [24]. To advance further into the practical usage of blockchain in mHealth, it is necessary to design how the client devices, such as mobile phones, send medical data to the blockchain network. Client devices can send data to the blockchain network directly, but with this architecture, the mHealth app that patients installed to their mobile phone needs to have SDKs for blockchain. The blockchain network will also receive access from unspecified clients due to a lack of IP address restriction. In that situation, there are risks that unspecified clients can store medical information in the blockchain network using SDKs to read the data. The system is resistant against data tampering, but the operational costs increase since they must manage private keys for each client device (Figure 1). In order to overcome these obstacles, the usage of relay servers is one possible option. With the architecture using relay servers, the blockchain network can restrict the access by IP address selection and it is not necessary for the mHealth app to include SDKs for blockchain. With this architecture, blockchain networks were protected against unspecified access from the internet and the functions of SDKs in the relay server can be predetermined as write-only, resulting in the protection of the medical data stored in the blockchain network. On the other hand, there are risks that the hacking of the relay server will result in impersonation (Figure 1). To balance these trade-offs, we propose the following architecture of the client for the blockchain network in mHealth. The client devices, such as mobile phones, send their data to multiple relay servers and the data are compared between relay servers and verified. After verification, the relay server, which has permission to access by IP address restriction, sends the data to the blockchain network using write-only functions of SDKs. With this system, medical data stored in the blockchain network are protected against access from the internet and are resistant against the risks of server hacking (Figure 1).

For the usage of relay servers described above, it is indispensable to carefully design the authentication of client devices to send data to relay servers. It is possible that a single common server gives an authentication to client devices, but with this architecture, the system is vulnerable to server hacking. Malicious access to the single authentication server can result in impersonation (Figure 2). The alternative is to set an authentication server for each relay server. Although the risks of impersonation by server hacking can be reduced, the operational costs for authentication increase. In addition, it is not possible to verify the reliability of the original data if multiple authentication servers were maliciously accessed (Figure 2). In order to solve these problems, we implemented a single common authentication server with another method, using a hash value calculated on the client devices. As an initial setting, the client device generates and preserves a secure string. The client device calculates a hash value based on the medical data and the secure string, as well as the previous hash value using the SHA-256 hash algorithm. Thus, the hash value comprises the chain structure. The hash value was also registered in the blockchain network along with the medical data in order to guarantee tamper-resistance of the value, although the secure string was preserved in the client device. It is possible to verify the reliability of data using the secure string preserved in the client devices and to retrospectively reject impersonation after finishing the clinical trials. Even when a relay server was hacked by malicious access, we can verify the correct data based on the client hashchain and the secure string preserved in the client device. Since the hash value calculated in the client device makes up the chain structure, we called the technique “client hashchain” in contrast to blockchain (Figure 2). In the case of the device having been destroyed or disabled prior to the conclusion of the study, the secure string could be sent beforehand to the user’s personal storage, such as their email box.

In order to validate the system proposed above, we have implemented the architecture into the mHealth app. The app was designed to treat insomnia patients based on CBTi and collects medical data using mobile phones. The app generates a secure string at login and stores it on the client device. The app also calculates a hash value based on the medical data, the previous hash value, and the secure string so that the hash values make up the chain structure. The medical data collected with the app, as well as the hash value, were sent to the blockchain network via relay servers. We used three relay servers and the app selected two relay servers at random to send the data to the blockchain network. The blockchain network comprised three organizations, which contain two validating peers.

With these systems, we conducted the clinical trial on the mHealth app for insomnia patients. Informed consent was obtained from the patients and the app account was provided by the medical doctor. mHealth data were collected with the mobile app and sent to the blockchain network via relay servers along with the client hash value (Figure 3). In the client devices, client hash values were calculated based on mHealth data, the previous client hash value, and the secure string stored on the client device. The client hash value constitutes the chain structure and proves the origin of the sequential data.

Both mHealth data and the client hash value were sent to the blockchain network to be registered in the ledger. The ledger is made up of the blockchain, sequenced records in blocks, and a state database. Each node of the blockchain maintains a copy of the ledger. If the transaction was validated under the endorsement policy, the chaincode was executed and the block of the transaction was appended in each node. The mHealth data and the client hash value were stored in CouchDB and the blockchain. The block includes a hash of the block’s transactions, as well as a hash of the prior block. In this way, it is not possible to tamper with the ledger data without breaking the hash links.

Although the block size is limited in blockchain, it is enough for our medical data since the clinical indicators are stored as JSON data. In addition, the transaction throughput in the Hyperledger Fabric platform that we used is 2250 transactions per second [27]. In contrast, in a permissionless network or public network, such as Bitcoin and Ethereum, it takes 600 seconds and 10 seconds respectively to write a transaction on the ledger [28]. Since the number of transactions in our mHealth system occur several times per day for every patient, the transaction performance of the blockchain will not be a bottleneck.

To investigate the resistance of our proposed system, we first simulated an artificial attack on the relay server (ie, outside actors) or misconduct by the owner of the relay server (ie, inside actors) and evaluated if the malicious access was detected and if the original data are distinguishable from the illegal data. As described above, the client device sends the data to the blockchain network via multiple relay servers. We used three relay servers and the app selected two relay servers at random to send the data to the blockchain network. If one of the relay servers is hacked, the attacker can modify the data sent from the client device prior to blockchain submission and steal the authorization token. In this case, the secure string used in the calculation of the client hashchain was not stolen by the attacker.

As shown in Figure 4, an attacker hacked relay server #2 and modified some medical data as well as the client hash value. In this case, the client hash value was calculated with the previous hash value, modified medical data, and the incorrect string. The access fraud can be detected by the mismatch between the data sent from other relay servers. It is also possible to distinguish and reject illegal data from original data automatically using the client hashchain. At the end of the clinical trial, the device sent the secure string to the blockchain to make it possible to validate the uploaded data retrospectively. Since all the medical data and the client hash values were stored in the blockchain network, it is not possible to rewrite the hash value based on the secure string, which was sent via the relay servers. Legal data can be guaranteed by combining the client hashchain, which rejects impersonation, with the blockchain, which provides tamper-proof history.

We next simulated an artificial attack on the authentication server (ie, outside actors) or misconduct by the owner of the authentication server (ie, inside actors) during the clinical trial and evaluated whether the malicious access was detected if the original data are distinguishable from the illegal data. The authentication server possesses the authentication keys of every account. If the authentication key was stolen, the attacker can send illegal data from different devices. In this case, the secure string used in the calculation of the client hashchain was not stolen by the attacker.

As shown in Figure 5, using the stolen authentication key, multiple data with the same log ID were generated and sent to the blockchain network via relay servers by the attacker. Fraud detection can be achieved by identifying branched data. Furthermore, it is also possible to distinguish and reject illegal data from original data automatically using the client hashchain. At the end of the clinical trial, the device sent the secure string to the blockchain to enable the validation of the uploaded data retrospectively. Since all medical data and the client hash values were stored in the blockchain network, it is not possible to rewrite the hash value based on the secure string that was sent via relay servers.

To further investigate the resistance of our proposed system, we next simulated an artificial attack on the client device and evaluated if the malicious access was detected and if the original data are distinguishable from the illegal data. One of the most dangerous attacks is the malware root exploit, which enables the attacker to obtain the victim’s private key. In this case, the authentication key as well as the secure string were stolen by the attacker, resulting in a more serious situation.

As shown in Figure 6, using the stolen authentication key, multiple data with the same log ID were generated and sent to the blockchain network. Fraud detection can be achieved by identifying the branched data. However, it is not possible to distinguish illegal data from original data automatically using the client hashchain because the attacker has stolen the secure string to calculate the client hash value. In this case, however, it is possible to judge which are the original data by checking the data in the patient’s device offline, based on the fraud detection.

In this study, we have developed a secure and scalable mHealth system using relay servers and blockchain combined with a client hashchain. Although blockchain technology provides tamper-resistance to medical data [24], the security was limited to the registered data and it cannot distinguish between original data and impersonated data. In addition, scalability will be compromised if the client devices were dealt with as a node of blockchain network. With our proposed system, we have shown that these problems can be resolved.

Bitcoin was the first implementation of blockchain as a digital asset in widespread use. Although bitcoin can be used as a platform for preventing data tampering, it is not appropriate since it is an open network and massive computing power is necessary for proof of work (PoW) to obtain consensus [43]. Private blockchain networks, such as Hyperledger Fabric, are more appropriate for the management of medical data since the node of stakeholders can be controlled. In addition, it is possible to process more transactions in a private blockchain without PoW. Using a private blockchain network, we used relay servers to send the data from authorized clients. With this architecture, it is not necessary to incorporate SDKs on client devices. To make the system robust against hacking of the relay server, the app sent data to the blockchain network via multiple relay servers. Even when one of the relay servers was hacked, we could detect access fraud and distinguish the original data from modified data. In our study, we used three relay servers and two were randomly selected to send the data. The robustness of the system against server attack can be increased if we use more relay servers, for instance, if three out of five relay servers are randomly selected to send the data.

To further clarify the origin of the data, we combined the client hashchain with the private blockchain. Hash values combined with the blockchain have been used as the metadata in a previous study for the management of rights for digital contents [44]. In contrast, we used hash values calculated on the client device to protect against impersonation and verify the origin of the data by chaining them. Hash values with chain structure (client hashchain) enable the identification of the original data sent from a specific client, which stores the secure string. In addition, in combination with the blockchain, the system also ensured tamper-resistance and the reliability of the hash values to prevent impersonation. We have shown that fraudulent data by compromised relay servers can be detected and distinguished from original data using the client hashchain. Even when the secure string used in calculating the client hash value was stolen by the attacker with root exploit [45], it is possible to detect the malformation in the branching of the chaincode. Based on the detection of the malformation, the researchers can ask the patients and check which are the original data. Therefore, the system is highly resistant against impersonation and tampering.

In this study, we designed the architecture for mHealth and verified the performance in a clinical study. Although mHealth is suitable for collecting medical data, such as patient reported outcomes [46], by changing the client device from patients’ mobile phone to computers in medical institutions, the system can also be applied to clinical trials that use electronic data capture [47]. In addition, we could deploy smart contract, which is called chaincode in Hyperledger Fabric, to each node of the blockchain network to execute transactions. Since the smart contract may have the function to transform medical data into the determined format, it is possible to automatically complete the case report form of each patient if the app has access to additional medical data by deploying the smart contract for the clinical trial [25,48].

The system enables the verification of the accuracy of the medical data without confirmation by the third party, such as a contracted research organization, so that it is possible to reduce the cost of clinical trials as well as the possibility of human error. Thus, our system based on the blockchain technology combined with a client hashchain may enhance the development of drugs and medical devices.

Further studies are needed to verify the scalability of the system for conducting multiple clinical trials simultaneously. In Hyperledger Fabric v1.0, it is possible to partition the network and define a communication channel using an ordering service, which enables multiple clinical trials to be conducted in the same system [49]. Although the transaction throughput in the Hyperledger Fabric platform that we used is much higher than public blockchain, one drawback of the private network described here is preventing 51% of attacks in networks that are composed of a limited number of nodes without public validation.

Although our system is resistant against impersonation and tampering, hacking of the client device is a great threat. Root exploit is a type of malware attackers use to modify the Android operating system kernel such that attackers are able to gain super-user privileges. When attackers gain root of the operating system kernel, they also gain access to full administrator privileges. Through this, attackers are able to install other malware types, such as botnets, worms, or Trojans into the system. Further studies like root exploit detection [45] may be beneficial for the improved security of the system.

In this study, we designed a secure and scalable mHealth system using blockchain. A client hashchain was combined with the blockchain network to protect against impersonation, enabling the usage of relay servers and reducing the complexity of authentication of client devices for mHealth. The system was validated in the clinical trial, and the resistance against various fraud attacks was evaluated.