We live in an age of information, in which data on our individual activities, preferences and attitudes are constantly being collected, sometimes with and sometimes without our knowing consent. In this digital era, ready access to personalized information has the ability to make many aspects of life easier and more efficient. I love that my browser routinely loads my favorite websites, my music preferences are preset, and Google Maps will give me directions to my home at the touch of a finger. Of course, the not inconsiderable downside is the loss of privacy and potential for unscrupulous use of personal data. As a society and as individuals we need to be vigilant about protecting ourselves from these dangers.
Nonetheless, as we continue to amass large datasets across specific populations, the promise of big data is intriguing. One of the most exciting potential applications lies in the search for medical advances and improved understanding of human disease. Big datasets allow us to perform analyses that may be more generalizable to larger segments of the population and provide more power to detect associations in human health. The rapid spread of electronic health records over the last decades has given us tremendous amounts of digital information on patient outcomes across the globe. Clinical care billing and third party payer datasets have also allowed analyses of current practice patterns and associations between disease and specific risk factors such as environmental exposures.
One of the most prominent ophthalmology-specific big data initiatives has been the American Academy of Ophthalmology’s Intelligent Research in Sight (IRIS) registry. This database contains deidentified data extracted from the electronic health records of participating ophthalmology clinics. The registry, first launched in April 2014, has grown rapidly. It currently enrolls 13 000 clinicians and encompasses more than 148 million patient visits from more than 37 million unique patients. The initial primary role of the IRIS system lay in its ability to help ophthalmologists comply with Physician Quality Reporting System requirements and Meaningful Use Clinical Quality Measures. Physicians enrolled in the IRIS registry could also compare their performance on prespecified benchmarks to the performance of other practices or within their own practice. As the database has grown, however, the opportunities to effectively its information for research purposes have also broadened. Several papers from IRIS registry data have now been published, with topics ranging from the rate of endophthalmitis after cataract surgery to treatment patterns for myopic choroidal neovascularization. The American Academy of Ophthalmology is currently engaged in an active search for IRIS Registry Analytics Teams to generate research projects in areas including the nature history of ophthalmic disease, rare disease prevalence, practice patterns, technology use and comparative effectiveness.
In the field of retina, we shouldn’t ignore our imaging datasets as potential future resources for big data repositories. Advances in retinal imaging now allow us to quickly and noninvasively obtain remarkably high resolution views of the retinal structure. the relative ease of fundus photography and optical coherence tomography, many of our patients routine, often on a monthly basis. The Beetham Eye Institute of the Joslin Diabetes Center, where I work, has accumulated 3 million digital retinal images. Deep learning algorithms can extract features from large image datasets that may not be appreciated clinically, but which may nevertheless be highly correlated with disease features. It’s been less than a year now since a group led by Google reported the use of a deep convolutional neural network to develop and validate a computerized algorithm for diabetic retinopathy detection from color fundus photographs. Although this algorithm has not been validated for prime–time clinical use, it represents a highly promising advance as we look to develop automated methods for screening and detection of retinal disease.
I’m fond of the word “challenge” as it applies to big data not because it implies difficulties (although difficulties certainly abound in this area), but because it also suggests a valuable opportunity to be met head on. It will indeed be a challenge to develop and use our big datasets responsibly, effectively, and efficiently over the next few years. I look forward to learning from the IRIS registry experience and other initiatives as we head into the brave new world of big data.