Measuring Outcomes Using Digital Innovations

New digital tools have enormous potential to change the ways data collected in routine clinical care can be stored, analysed, and exploited to enhance research and improve patient outcomes, according to Dr Bruce Allan.

“By collecting data in a structured and protocol-driven manner and storing it electronically as part of routine clinical practice, the data begins to resemble that collected in clinical trials,” he said. “This presents an opportunity to use existing data for analysis - rather than spending time and money collecting new data for every research question.”

Highlighting the potential benefits of using data collected in routine clinical care, Dr Allan noted that under Article 9 of the General Data Protection Regulation (GDPR), anonymized healthcare data can be shared and used for analysis without the need for special measures for consent or Institutional Review Board (IRB) approval.

“This removes two of the biggest hurdles to accessing data - not just to track our outcomes but to perform analyses designed to improve them,” he said.

Ophthalmology, a field rich in image data, is particularly amenable to deep learning analysis. Labelling image data with diagnostic information can help train deep learning algorithms to perform classification and prediction tasks accurately. Dr Allan noted even untrained individuals, including medical students and high school students, can use code-free solutions to develop useful deep learning algorithms.

“The bottom line is you don’t need to be an expert in AI to engage with these tools and exploit the data you may have already collected in routine clinical care,” he said.

Anterior segment data lagging

Whether a novice or expert, users need access to well-ordered data sets, which Dr Allan observed is currently limited – particularly for the anterior segment.

“Nearly all of the available open-access data sets are for retina, so this is a ‘data deficit’ that needs to be addressed,” he said.

In recognition of the problem, he said the ESCRS recently announced a research call to collect open-access data sets. The goal is to create richly labelled data sets of imaging and clinical outcomes data for groups of 1,000 to 100,000 patients.

Machine learning has already made an impact in cataract surgery through the creation of more accurate biometry formulas for IOL power calculation, Dr Allan noted. The ESCRS recently launched a new tool that uses web scraping technology to allow users to see results from different IOL formula calculators in one place (see https://iolcalculator.escrs.org).

To further improve the accuracy of machine learning biometry algorithms, Dr Allan stressed the importance of collecting good quality data, including biometry, IOL details, and refraction measurement information.

“In machine learning, the size of the data sets needed diminishes if you have good-quality information available,” he said, adding data export in a common format remains problematic for contemporary biometry devices.

“We still have some work to do to get the data from the scanners into the cloud, where it can be properly exploited,” he said.

Streamline data collection

To collect data reliably, Dr Allan said, the process needs to be automated as far as possible. In the operating theatre, he suggested using barcodes on lens packs to streamline data collection and reduce the need for manual transcription. Machine learning algorithms can also, potentially, be trained to recognize steps in surgical procedures and assist with writing operating notes.

Solutions are also coming onstream to improve the quality of refraction data.

“Autorefraction is quick but not so accurate, whereas subjective refraction is the gold standard but is really slow to do and hard to integrate into high-volume practice,” he said. “However, there are new ways of [performing] refraction that [do not] rely on standard blur testing that are potentially faster, more accurate.”

Dr Allan also highlighted the importance of incorporating patient-reported outcome measures (PROMs) into the data collection process. For example, his primary facility, Moorfields Eye Hospital, developed a screenbased questionnaire with 11 questions to assess criteria such as emotional well-being, visual quality, comfort, and spectacle dependence.

By leveraging the wealth of data collected in routine clinical care and utilizing new analysis techniques, Dr Allan concluded, it is possible to gain insights and make improvements to benefit patients and the healthcare system as a whole.

Dr Allan presented at the 40th Congress of the ESCRS in Milan. Bruce Allan MD, FRCS is a consultant ophthalmic surgeon at Moorfields Eye Hospital, London, UK. bruce.allan@ucl.ac.uk