How can Personal Health Records help maximise value from Mental health (open) data


Written by Dr. Janak Gunatilleke

Since starting my role at Mindwave Ventures in June, I have been thinking about how patient-centered solutions such as Personal Health Records can help maximize the value from healthcare data, building upon the existing open data in the area.

The current state of open data in the NHS

The NHS publishes a number of datasets — for example, NHS Digital and Public Health England publish various operational and performance datasets including waiting times, GP prescribing and information on health-related behaviors such as smoking and alcohol intake.

Within Mental health, the Clinical Record Interactive Search (CRIS) system provides access to anonymized information from the South London and Maudsley NHS Foundation Trust patient record system[1].

Social media platforms present another interesting dimension. On one hand, information shared specifically — for example, a woman who shared a video sharing her experience of living with a particular mental health condition[2] — provides opportunities for peer support through sharing of similar experiences. On the other hand, it has been demonstrated data from publicly available Twitter feeds can be collected and analyzed to provide useful insights into conditions such as bipolar, PTSD and depression[3].

Image by Gerd Altmann from Pixabay

Underlying technology trends and opportunities

I believe that there are 3 key elements.

First, with the increasing use of smartphones — 72% accessed the internet on their mobiles in 2018 compared to 17% in 2008[4] — is the opportunity to collect and use additional data in real-time, and deliver interventions through smartphones. For example, as described by Huckvale (2019), the relationship between smartphone sensor and location data, and established outcome measure such as the PHQ-9 have been studied[5]. This leads to potential predictive value in linking activities such as sleep to severe depressive episodes and relapses.

Second, is the need for personalization. As well as genomics-based precision medicine, there are also opportunities to better understand and provide information relevant to what service users are seeking. For example, a study by Crangle (2015) explored the different types questions services users asked and demonstrated that even well-respected sources such as the US National Institute of Mental Health didn’t result in effective answers to the specific questions asked[6]. This highlights opportunities to better understand nuances in the meaning behind questions and respond with individualized content.

Third, are wider opportunities that can be delivered through developments in technologies including Artificial Intelligence (AI). Woebot is a chatbot and a randomised control trial with university students in 2017 demonstrated a 20% improvement in PHQ-9 scores within 2 weeks[7]. High levels of engagement — students were using the app almost every day — were thought to have contributed to the results, highlighting the need to consider user adoption and engagement when designing and implementing solutions.

Challenges to optimizing the use of open data

The wider healthcare ecosystem has 5 main stakeholder groups with varying priorities:

  1. Citizens — do they feel safe about their data being open and do they trust those getting access?
  2. Clinicians — do they feel threatened by the solutions (i.e. AI) or do they see them as helping them meet their challenging workloads?
  3. Government & NHS leadership — are the resulting solutions addressing the right challenges?
  4. Other commercial organizations (i.e. Pharmaceutical firms) — do open methodologies give enough confidence and incentives to invest in developing solutions?
  5. Technology companies — do open methodologies provide the right incentives (i.e. financially) and are the right tools and support readily available?

There are 3 key challenges that arise from the above complexities:

  1. Trust — citizens tend to consider data sharing initiatives involving their health data with a level of distrust, as exemplified by the high profile failure of the program which attempted to join up data cross primary and secondary care. Interestingly, a study by Ostherr (2017) found that the public expressed much less concern when sharing data with commercial organization that sold devices (i.e. fitness trackers) compared to sharing data for scientific research[8]. Context (i.e. consumers already deciding to use a solution in ‘exchange’ for data privacy) and increased awareness of risk during scientific research were highlighted as potential causes.
  2. Incentives — good quality data is required to improve algorithms and models used in Artificial Intelligence solutions. However, open data is often found to be of dubious quality and requires extra effort to use meaningfully[9]. The study highlights one of the potential reasons which is that, by the nature of open data, it can often be used for a different purpose to the original intention at the time of data collection. Furthermore, there is currently no clear incentives alignment between data producers and data consumers to improve quality. Having said that, due to access, open data may be facing extra scrutiny on its quality compared to closed data.
  3. Support and infrastructure — currently there are very few ecosystems, policies and standards that take into account all of the above stakeholders. This siloed approach makes adoption and meaningful scaling across the system challenging.

Potential solutions to addressing the challenges

  1. Trust — ‘Data Trusts’ have been defined as ‘a legal structure that provides independent stewardship of data’[10]. Pilots by the Open Data Institute found that there is significant interest in the concept and that independence helps manage conflicting interests. On the technology front, startups such as are using blockchain to create a secure personal data store to enable individuals to share their data with third parties, giving them control, providing extra visibility and enabling them to receive rewards.
  2. Incentives — there are opportunities to maximize existing data sources and better align incentives across various stakeholders. A recent spin-off from the University of Oxford is planning to do just this. Cristal Health will enhance value of the CRIS dataset mentioned above by bringing together 3 stakeholder groups; NHS Trusts who will generate revenue, Pharmaceutical companies that will improve early-stage drug discovery efficacy and costs, and tools for academia to extract better insights from data[11]. At an individual level, appropriate rewards can be provided through personal data storage solutions as described above.
  3. Support and infrastructure — MyData ( ) is a model which extends beyond health data to include financial, retail and other consumer data. Through a focus on consent management, it provides individuals the ability to control and benefit from their data and supports organizations to protect individual interests during big data analytics[12]. Additionally, it promotes usable data through standardized APIs and supports the creation of a collaborative business environment through interoperability.

Personal Health Records can help accelerate improvement

Personal Health Records (PHRs) provide individuals access to their clinical records (which are held within hospital or GP systems), opportunities to enrich the record by adding and collecting further information (i.e. manually or through fitness trackers) and to share the record with relatives and other healthcare professionals.

PHRs could support many elements of the potential solutions described above.


Dynamic consent is described as enabling users to give or change consent over time as projects and their own situation changes[13]. This approach gives users more time to think, provides visibility of what’s happening with their data and confidence that their data is being managed appropriately. It gives data processors flexibility and opportunities to get additional users engaged. With a PHR that covers many clinical pathways or more than one organization, it provides a good ‘digital front door’ to present and gain dynamic consent.

‘Federated learning’ is analyzing the data within the storage environment of each organization rather than transferring and processing it at a central and ‘third party’ location. This could provide both individual organizations and the citizens more confidence that their data is secure. For example, Owkin — — is a startup that is applying machine learning to medical research using Federated learning. The benefits of this approach fit it well where a single PHR platform is used across many clinical pathways and disparate organizations.


It has been demonstrated that providing tools to enable analysis and data visualization can motivate users to donate data[14]. PHRs are a great backdrop to provide visualisations of user data and insights based on trends over time. For example, this could include charting various metrics such as sleep and mood over time, and also super-imposing other collated data points including changes to medications.

Additionally, PHRs provides options, with appropriate consent and safeguards, to connect users with other similar users to share experiences and foster belonging.

PHRs also present an opportunity to improve the quality of data being shared. Data archetypes[15] — a specification for the various data elements that need to be captured — can provide consistency and improve the ability to share data across systems.

PHRs could be an interesting solution to resolving conflicts of interests mentioned earlier. For example, if developed by an independent party to the established hospital and other provider systems, it could be an objective aggregator across regions and disparate systems enabling the secure and efficient sharing of data. The opportunity to create a thriving ecosystem could be further strengthened by adopting open technology standards that enable maximum interoperability between systems.

Support and infrastructure

MedMij is a PHR technical framework and set of standards that is established in the Netherlands which aims to create an ecosystem covering apps across primary care, hospitals, pharmacies, etc[16].

In the NHS, there are opportunities to build on the groundwork including allocated funding through the Global Digital Exemplars program, publishing of standards and best practice including the PHR adoption toolkit, and national IT contracts based on interoperability such as the GP IT futures program [17].

Bigger picture / expanding the concept

Open source integration engines such as MIRTH Connect (NextGen Connect) — — provides a framework to enable to connect disparate healthcare information systems and to establish interoperability. A number of API ecosystems — from HumanAPI ( and Validic ( that focusses more on connections to healthcare institutions and consumers provide infrastructure to integrate more data sources into a PHR solution.

Over time and with adequate functionality, as more users consider a PHR as a single ‘digital front door’, it presents a great opportunity to be scaled further across an increased scope and more varied data sources.

If you want to chat or get involved in the work Mindwave Ventures are doing to develop data archetypes for mental health, please drop me an email via


[1] NIHR Biomedical Research Centre (2019). Clinical Record Interactive Search (CRIS) [Online]. Available at: [Accessed: 30th July 2019]

[2] Glyn Elwyn et al. (2014). Crowdsourcing health care — hope or hype? [Online]. BMJ. Available at: [Accessed: 30th July 2019

[3] Coppersmith, G., Dredze, M. and Harman, C., 2014, June. Quantifying mental health signals in Twitter. In Proceedings of the workshop on computational linguistics and clinical psychology: From linguistic signal to clinical reality (pp. 51–60)

[4] Ofcom (2018). A decade of digital dependency [Online]. Available at: [Accessed: 07th September 2019]

[5] Huckvale, K., Venkatesh, S., Christensen, H., 2019. Toward clinical digital phenotyping: a timely opportunity to consider purpose, quality, and safety. npj Digital Medicine, 2: 88

[6] Crangle, C.E. and Kart, J.B., 2015. A questions-based investigation of consumer mental-health information. PeerJ, 3, p.e867

[7] Fitzpatrick, K.K., Darcy, A. and Vierhile, M., 2017. Delivering cognitive behaviour therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial. JMIR mental health, 4(2), p.e19

[8] Ostherr, K., Borodina, S., Bracken, R.C., Lotterman, C., Storer, E. and Williams, B., 2017. Trust and privacy in the context of user-generated health data. Big Data & Society, 4(1)

[9] Sadiq, S. and Indulska, M., 2017. Open data: Quality over quantity. International Journal of Information Management, 37(3), pp.150–154

[10] Open Data Institute (2019). Data trusts: lessons from three pilots [Online]. Available at: [Accessed: 7th July 2019]

[11] Department of Psychiatry, University of Oxford. (2019). Cristal Health uses industrial scale data science to improve lives [Online]. Available at: [Accessed: 28th July 2019]

[13] Kaye, J., Whitley, E.A., Lund, D., Morrison, M., Teare, H. and Melham, K., 2015. Dynamic consent: a patient interface for twenty-first century research networks. European Journal of Human Genetics, 23(2), p.141.

[14] Bietz, M., Patrick, K. and Bloss, C., 2019. Data Donation as a Model for Citizen Science Health Research. Citizen Science: Theory and Practice, 4(1).

[15] Leslie, H (2012). Introduction to Archetypes and Archetype classes [Online]. Available at: [Accessed: 28th July 2019]

[16] von Grätz, PG (2018). Personal health records in Europe: National or beyond? [Online]. Available at: [Accessed: 28th July 2019]

[17] NHS Digital (2019). Future GP IT systems and services [Online]. Available at: [Accessed: 31st July 2019]