EXCLUSIVE - Using cloud computing and big data to address a poorly understood medical condition
How can digital technology help clinicians in tackling a poorly understood, neglected disease, with a lack of comprehensive clinical information and agreed-on guidelines for diagnosis? How can technology facilitate international collaboration between clinicians separated by long distances across multiple countries and produce results benefiting patients around the world?
Recently, OpenGov learnt about the Collaborative Ocular Tuberculosis Study (COTS), a global initiative to address ocular tuberculosis (TB) leveraging cloud computing and big data.
COTS is led by uveitis  experts from 25 multinational Ophthalmology centres including Tan Tock Seng Hospital (Singapore), Postgraduate Institute of Medical Education and Research (India), Moorfields Eye Hospital (UK), Byer's eye centre in Stanford (USA), and many more. Participating centres in this international work group originate from over 10 countries spanning Singapore, India, UK, US, Turkey, Tunisia, Switzerland, Saudi Arabia, Italy, Spain, and Brazil.
Most people would think of TB as a disease affecting only the lungs. But few know that it can also affect other parts of the body, including the eye.
Uveitis could be indicative of latent TB. But the difficulty in diagnosing ocular TB often leads to delayed or missed diagnoses, resulting in poor clinical outcomes and missed opportunities to address TB infection at an early stage. The COTS group was created to address these problems.
OpenGov spoke to Dr Rupesh Agrawal (below left), a consultant ophthalmologist at Tan Tock Seng Hospital (TTSH) and communicated via e-mail with Dr Dinesh Visva Gunasekeran (below right), two of the leads on the study to learn more about the technology platforms used.
Dr Dinesh said that COTS-1 was effectively a pilot trial for the use of cloud-based data aggregation platform to facilitate this multinational clinical research collaboration.
Dr Agrawal started research on ocular TB in London in 2012. His team reviewed case records of more than 300 patients and published reports in different scientific journals. But this was a single centre study. Then he conducted a similar study in Singapore, looking at 60 patients over 5 years.
He said, “The conclusion was that we do not know anything about the disease. Let’s form a group, a consortium.”
The initiative was taken from Singapore and soon centres from US, Australia, India, the Middle East and more were on-board. Clinicians everywhere were facing the same challenges so joining forces and collaborating was seen as the way ahead.
The idea was proposed in 2015 and it took nearly three years, with 10-12 hours of time committed every week, to set up the consortium and complete the first stage of the study.
Designing a Smart Form
One of the key challenges was to design a secure, multi-user data aggregation system with a centralised data repository accessible by an administrator user. The data collection had to be uniform and standardised, so that data clean-up is minimised. In addition, patient confidentiality cannot be breached.
Data privacy requirements in the different geographies, such as the Personal Data Protection Act (PDPA) in Singapore, have to be complied with. Then the data has to be encrypted, to prevent anyone entering into the database and manipulating the data.
Finally, the form had to be simple enough for people to be motivated to do the data entry. Training requirements would have to be minimal, as clinicians are busy with seeing patients and they have limited time on their hands. Taking into account that research assistants and junior clinicians also help with the data entry, the form has to be straightforward with no scope for ambiguity.
After exploring multiple options, a cloud-based platform called Cognito Forms was selected. The team designed their own template for protocolised data entry, based on retrospective records by trained interpreters. The protocol was developed by the international experts on the steering committee of this study group, and was used by the team to design the ICT platform accordingly.
The form incorporated dropdown menus where possible and minimised the use of free text fields, so as to avoid situations where different jargons are used (e.g. one person enters right eye, another RE).
The form also had embedded logic to minimise keystrokes by modifying the questions presented for data entry based on responses to earlier questions (so that only relevant data is collected i.e. depth of relevant data instead of breadth of all data). For instance, if a patient informs that only one eye is involved, then automatically all entry fields related to the other eye become inactive.
The form also has embedded prompts to reinforce relevant criteria during data entry in accordance with the study protocol as well as important instructions, such as the process of anonymising patient data that is entered to the form.
The instructions specified how each site would code their patients, but no patient identifiers had to be entered. Each clinician would thus have access to the identity of the patients they have entered from their centre, but that information is not accessible by other sites and is also not required for the research project.
Each participating centre had to get their own ethics clearance, before they could start doing data entry. Only Dr Agrawal and Dr Dinesh have access to the central data repository.
We asked Dr Agrawal if they considered extracting data from Electronic Medical Records (EMRs). He replied that the team considered using machine learning or big data analytics to extract the data. But each centre uses a different EMR system.
“There is no common language in which these platforms can talk to each other. If they do not speak the same language, you cannot extract the data,” explained Dr Agrawal.
“Particularly in uveitis or TB, we do not have the same language, everyone writes their own language. So there needs to be a person for entering the information.”
Data processing and analytics was embedded into the form itself. A program for automated data processing/analytics was developed by a statistician using R-program based on the Microsoft Excel data file output of the data aggregation form.
A trial run was conducted with 100 patients. On conducting the automated analysis, a few important fields were found to be missing. Then the form was modified and it was then necessary to re-enter those information from the trial. Several iterations were required to arrive at the final form.
Dr Agrawal said, “We finally managed to get the form the way we wanted it to be. That form has given us a lot of data from 25 participating centres in 10 countries. With that data, we published our first report in the journal, JAMA Opthalmology. There are other reports in the pipeline, two more papers. One more is published and another will be published very soon. Three more are under review.”
Impact of the technology
Dr Dinesh told us that the form facilitated the international collaboration and overcame many limitations of existing scientific descriptions of this disease which are limited to small cohorts of patients from singular localities.
Some of its key benefits were ease of coordination of a multinational clinical investigation and protocolised data aggregation to extract meaningful data on a poorly understood subject matter.
Before this form, such a collaboration would have been incredibly difficult to coordinate. Many difficulties would have been encountered including insufficient data entry, security concerns with the transfer of large amounts of patient data between centres, and fatigue for data entrants.
Over 200 variables were collected for each study subject and without the smart form logic, this would have been a highly taxing exercise for the data entrants, who were often clinicians themselves with limited time.
Insights gained and future plans
The results of the study included novel findings that improve understanding of the disease and question existing doctrines that are based on limited available evidence from existing literature.
For example, the results revealed geographic variations in the way affected patients may present, and also suggest unusual behaviour of the condition in certain defined populations that will direct future study.
Around 80% of the patients entered into the records did very well after they were given anti-tubercular therapy, while 20% still had recurrences.
Another finding was that there was a particular subtype of uveitis which involves the back of the eye, called choroiditis, which is a strong indicator of TB.
PCR (polymerase chain reaction), a test which the researchers thought could be a diagnostic gold standard, was found to be very useful in diagnosing ocular TB. Similarly, another test QuantiFERON-TB gold, produced false positives.
Dr Agrawal said that through these insights, they realised that the disease needs to be studied in greater detail. The team is exploring a genetic study to see whether there is any genetic pre-disposition. They are also looking into novel ways to diagnose this disease.
COTS-1 was a retrospective study. Now the group is planning to launch a COTS registry for COTS-2, which will be a prospective study.
An app is being created, so that the participating centres can enter the data of any patient they see easily into the registry. The number of entries will be minimised to reduce demands on clinic time. This is also being done using the Cognito Forms based smart form.
“We have also created COTS-1A which is a survey questionnaire kind of form which we have circulated to general practitioners and to specialist doctors for their opinions. We recently got an education grant from a pharma company, for 45000 dollars. This is for a survey study called COTSCon (COTS consensus group meeting),” Dr Agrawal said.
For COTSCon, again a smart form has been created for circulation among experts with more than 10 years of experience in managing ocular TB from around the world.
All this will help accumulate evidence. At a closed-door meeting in PGI Chandigarh, India on November 15 this year, 25 experts from across the world will gather to discuss and generate evidence statements for this disease, which will guide all ophthalmologists in managing patients with uveitis.
To further ease data entry and make it less tedious and more enjoyable, Dr Agrawal said that gamification ideas are being implemented.
We enquired if images have been uploaded and whether the team considered using artificial intelligence algorithms on the images for diagnosis. Dr Agrawal replied that images were uploaded as part of the data entry. However, as the prevalence of the disease is not very high, it is difficult to gather the large number of images required to the train the system. Moreover, pattern recognition on the images is insufficient, unlike diabetic retinopathy.
But eventually, a machine learning algorithm will have to be built. It will have to go beyond the images to take into consideration multiple risk factors, such as geography and endemicity, and figure how the 200 variables correlate.
At the moment, the COTS group is looking to continue expanding on its collaborations, with technology solution providers, more ophthalmology centres, and more clinicians. Only by pooling together resources, knowledge, experience and data, can the understanding of this medical condition be improved.
 Inflammation of the middle layer of the eye that consists of the iris, ciliary body and choroid.