Search

ImportanceDiagnostic information from administrative claims and electronic health record (EHR) data may serve as an important resource for surveillance of vision and eye health, but the accuracy and validity of these sources are unknown.ObjectiveTo estimate the accuracy of diagnosis codes in administrative claims and EHRs compared to retrospective medical record review.Design, Setting, and ParticipantsThis cross-sectional study compared the presence and prevalence of eye disorders based on diagnostic codes in EHR and claims records vs clinical medical record review at University of Washington–affiliated ophthalmology or optometry clinics from May 2018 to April 2020. Patients 16 years and older with an eye examination in the previous 2 years were included, oversampled for diagnosed major eye diseases and visual acuity loss.ExposuresPatients were assigned to vision and eye health condition categories based on diagnosis codes present in their billing claims history and EHR using the diagnostic case definitions of the US Centers for Disease Control and Prevention Vision and Eye Health Surveillance System (VEHSS) as well as clinical assessment based on retrospective medical record review.Main Outcome and MeasuresAccuracy was measured as area under the receiver operating characteristic curve (AUC) of claims and EHR-based diagnostic coding vs retrospective review of clinical assessments and treatment plans.ResultsAmong 669 participants (mean [range] age, 66.1 [16-99] years; 357 [53.4%] female), identification of diseases in billing claims and EHR data using VEHSS case definitions was accurate for diabetic retinopathy (claims AUC, 0.94; 95% CI, 0.91-0.98; EHR AUC, 0.97; 95% CI, 0.95-0.99), glaucoma (claims AUC, 0.90; 95% CI, 0.88-0.93; EHR AUC, 0.93; 95% CI, 0.90-0.95), age-related macular degeneration (claims AUC, 0.87; 95% CI, 0.83-0.92; EHR AUC, 0.96; 95% CI, 0.94-0.98), and cataracts (claims AUC, 0.82; 95% CI, 0.79-0.86; EHR AUC, 0.91; 95% CI, 0.89-0.93). However, several condition categories showed low validity with AUCs below 0.7, including diagnosed disorders of refraction and accommodation (claims AUC, 0.54; 95% CI, 0.49-0.60; EHR AUC, 0.61; 95% CI, 0.56-0.67), diagnosed blindness and low vision (claims AUC, 0.56; 95% CI, 0.53-0.58; EHR AUC, 0.57; 95% CI, 0.54-0.59), and orbital and external diseases (claims AUC, 0.63; 95% CI, 0.57-0.69; EHR AUC, 0.65; 95% CI, 0.59-0.70).Conclusion and RelevanceIn this cross-sectional study of current and recent ophthalmology patients with high rates of eye disorders and vision loss, identification of major vision-threatening eye disorders based on diagnosis codes in claims and EHR records was accurate. However, vision loss, refractive error, and other broadly defined or lower-risk disorder categories were less accurately identified by diagnosis codes in claims and EHR data.

Purpose To develop a deep learning semantic segmentation network to automate the assessment of 8 periorbital measurements Design Development and validation of an artificial intelligence (AI) segmentation algorithm Methods A total of 418 photographs of periorbital areas were used to train a deep learning semantic segmentation model to segment iris, aperture, and brow areas. These data were used to develop a post-processing algorithm that measured margin reflex distance (MRD) 1 and 2, medial canthal height (MCH), lateral canthal height (LCH), medial brow height (MBH), lateral brow height (LBH), medial intercanthal distance (MID), and lateral intercanthal distance (LID). The algorithm validity was evaluated on a prospective hold-out test set against 3 graders. The main outcome measures were dice coefficient, mean absolute difference, intraclass correlation coefficient, and Bland-Altman analysis. A smartphone video was also segmented and evaluated as proof of concept. Results The AI algorithm performed in close agreement with all human graders, with a mean absolute difference of 0.5 mm for MRD1, MRD2, LCH, and MCH. The mean absolute difference between graders is approximately 1.5-2 mm for LBH and MBH and approximately 2-4 mm for MID and LID. The 95% confidence intervals for all graders overlapped in most cases, demonstrating that the algorithm performs similarly to human graders. The segmentation of a smartphone video demonstrated that MRD1 can be dynamically measured. Conclusions We present, to our knowledge, the first open-sourced, artificial intelligence system capable of automating static and dynamic periorbital measurements. A fully automated tool stands to transform the delivery of clinical care and quantification of surgical outcomes.

BACKGROUND Self-reported questions on blindness and vision problems are collected in many national surveys. Recently released surveillance estimates on the prevalence of vision loss used self-reported data to predict variation in the prevalence of objectively measured acuity loss among population groups for whom examination data are not available. However, the validity of self-reported measures to predict prevalence and disparities in visual acuity has not been established. OBJECTIVE This study aimed to estimate the diagnostic accuracy of self-reported vision loss measures compared to best-corrected visual acuity (BCVA), inform the design and selection of questions for future data collection, and identify the concordance between self-reported vision and measured acuity at the population level to support ongoing surveillance efforts. METHODS We calculated accuracy and correlation between self-reported visual function versus BCVA at the individual and population level among patients from the University of Washington ophthalmology or optometry clinics with a prior eye examination, randomly oversampled for visual acuity loss or diagnosed eye diseases. Self-reported visual function was collected via telephone survey. BCVA was determined based on retrospective chart review. Diagnostic accuracy of questions at the person level was measured based on the area under the receiver operator curve (AUC), whereas population-level accuracy was determined based on correlation. RESULTS The survey question, “Are you blind or do you have serious difficulty seeing, even when wearing glasses?” had the highest accuracy for identifying patients with blindness (BCVA ≤20/200; AUC=0.797). The highest accuracy for detecting any vision loss (BCVA CONCLUSIONS Although survey questions are not considered to be sufficiently accurate to be used as a diagnostic test at the individual level, we did find relatively high levels of accuracy for some questions. At the population level, we found that the relative prevalence of the 2 most accurate survey questions were highly correlated with the prevalence of measured visual acuity loss among nearly all demographic groups. The results of this study suggest that self-reported vision questions fielded in national surveys are likely to yield an accurate and stable signal of vision loss across different population groups, although the actual measure of prevalence from these questions is not directly analogous to that of BCVA.

To investigate whether associations between diabetic retinopathy (DR) and dementia and Alzheimer's disease (AD) remain significant after controlling for several measures of diabetes severity.Retrospective cohort study.ACT is a prospective cohort study of adults ≥ 65 years, randomly selected and recruited from the membership rolls of Kaiser Permanente Washington, who are dementia free at enrollment and followed biennially until incident dementia. ACT participants were included in this study if they had at enrollment or developed type 2 diabetes mellitus during follow-up, and data were collected through September, 2018 (3,516 person-years of follow-up). Diabetes was defined by ≥2 diabetes medication fills in one year. Diagnosis of DR was based on International Classification of Diseases Ninth and Tenth Revision codes. Estimates of microalbuminuria, long-term glycemia, and renal function from longitudinal laboratory records were used as indicators of diabetes severity. AD and dementia were diagnosed using research criteria at expert consensus meetings.A total of 536 participants (median baseline age 75 [interquartile range 71-80], 54% women) met inclusion criteria. Significant associations between DR5 years duration with dementia (hazard ratio 1.81 [95% confidence interval 1.23-2.65]) and AD (1.80 [1.15-2.82]) were not altered by adjustment for estimates of microalbuminuria, long-term glycemia, and renal function (dementia: 1.69 [1.14-2.50]; AD: 1.73 [1.10-2.74]).Among people with type 2 diabetes, DR itself appears to be an important biomarker of dementia risk in addition to glycemia and renal complications.

To create an unsupervised cross-domain segmentation algorithm for segmenting intraretinal fluid and retinal layers on normal and pathologic macular OCT images from different manufacturers and camera devices.We sought to use generative adversarial networks (GANs) to generalize a segmentation model trained on one OCT device to segment B-scans obtained from a different OCT device manufacturer in a fully unsupervised approach without labeled data from the latter manufacturer.A total of 732 OCT B-scans from 4 different OCT devices (Heidelberg Spectralis, Topcon 1000, Maestro2, and Zeiss Plex Elite 9000).We developed an unsupervised GAN model, GANSeg, to segment 7 retinal layers and intraretinal fluid in Topcon 1000 OCT images (domain B) that had access only to labeled data on Heidelberg Spectralis images (domain A). GANSeg was unsupervised because it had access only to 110 Heidelberg labeled OCTs and 556 raw and unlabeled Topcon 1000 OCTs. To validate GANSeg segmentations, 3 masked graders manually segmented 60 OCTs from an external Topcon 1000 test dataset independently. To test the limits of GANSeg, graders also manually segmented 3 OCTs from Zeiss Plex Elite 9000 and Topcon Maestro2. A U-Net was trained on the same labeled Heidelberg images as baseline. The GANSeg repository with labeled annotations is at https://github.com/uw-biomedical-ml/ganseg.Dice scores comparing segmentation results from GANSeg and the U-Net model with the manual segmented images.Although GANSeg and U-Net achieved comparable Dice scores performance as human experts on the labeled Heidelberg test dataset, only GANSeg achieved comparable Dice scores with the best performance for the ganglion cell layer plus inner plexiform layer (90%; 95% confidence interval [CI], 68%-96%) and the worst performance for intraretinal fluid (58%; 95% CI, 18%-89%), which was statistically similar to human graders (79%; 95% CI, 43%-94%). GANSeg significantly outperformed the U-Net model. Moreover, GANSeg generalized to both Zeiss and Topcon Maestro2 swept-source OCT domains, which it had never encountered before.GANSeg enables the transfer of supervised deep learning algorithms across OCT devices without labeled data, thereby greatly expanding the applicability of deep learning algorithms.

The objective grading of anterior chamber inflammation (ACI) has remained a challenge in the field of uveitis. While the grading criteria produced by the Standardization of Uveitis Nomenclature (SUN) International Workshop have been widely adopted, limitations exist including interobserver variability and grading confined to discrete categories rather than a continuous measurement. Since the earliest iterations of optical coherence tomography (OCT), ACI has been assessed using anterior segment OCT and shown to correlate with slit-lamp findings. However, widespread use of this approach has not been adopted. Barriers to standardization include variability in OCT devices across clinical settings, lack of standardization of image acquisition protocols, varying quantification methods, and the difficulty of distinguishing inflammatory cells from other cell types. Modern OCT devices and techniques in artificial intelligence show promise in expanding the clinical applicability of anterior segment OCT for the grading of ACI.

COVID-19 mortality risk stratification tools could improve care, inform accurate and rapid triage decisions, and guide family discussions regarding goals of care. A minority of COVID-19 prognostic tools have been tested in external cohorts. Our objective was to compare machine learning algorithms and develop a tool for predicting subsequent clinical outcomes in COVID-19. We conducted a retrospective cohort study that included hospitalized patients with COVID-19 from March 2020 to March 2021. Seven Hundred Twelve consecutive patients from University of Washington and 345 patients from Tongji Hospital in China were included. We applied three different machine learning algorithms to clinical and laboratory data collected within the initial 24 h of hospital admission to determine the risk of in-hospital mortality, transfer to the intensive care unit, shock requiring vasopressors, and receipt of renal replacement therapy. Mortality risk models were derived, internally validated in UW and externally validated in Tongji Hospital. The risk models for ICU transfer, shock and RRT were derived and internally validated in the UW dataset but were unable to be externally validated due to a lack of data on these outcomes. Among the UW dataset, 122 patients died (17%) during hospitalization and the mean days to hospital mortality was 15.7 +/− 21.5 (mean +/− SD). Elastic net logistic regression resulted in a C-statistic for in-hospital mortality of 0.72 (95% CI, 0.64 to 0.81) in the internal validation and 0.85 (95% CI, 0.81 to 0.89) in the external validation set. Age, platelet count, and white blood cell count were the most important predictors of mortality. In the sub-group of patients > 50 years of age, the mortality prediction model continued to perform with a C-statistic of 0.82 (95% CI:0.76,0.87). Prediction models also performed well for shock and RRT in the UW dataset but functioned with lower accuracy for ICU transfer. We trained, internally and externally validated a prediction model using data collected within 24 h of hospital admission to predict in-hospital mortality on average two weeks prior to death. We also developed models to predict RRT and shock with high accuracy. These models could be used to improve triage decisions, resource allocation, and support clinical trial enrichment.

This work explores a student-teacher framework that leverages unlabeled images to train lightweight deep learning models with fewer parameters to perform fast automated detection of optical coherence tomography B-scans of interest. Twenty-seven lightweight models (LWMs) from four families of models were trained on expert-labeled B-scans (∼70 K) as either “abnormal” or “normal”, which established a baseline performance for the models. Then the LWMs were trained from random initialization using a student-teacher framework to incorporate a large number of unlabeled B-scans (∼500 K). A pre-trained ResNet50 model served as the teacher network. The ResNet50 teacher model achieved 96.0% validation accuracy and the validation accuracy achieved by the LWMs ranged from 89.6% to 95.1%. The best performing LWMs were 2.53 to 4.13 times faster than ResNet50 (0.109s to 0.178s vs. 0.452s). All LWMs benefitted from increasing the training set by including unlabeled B-scans in the student-teacher framework, with several models achieving validation accuracy of 96.0% or higher. The three best-performing models achieved comparable sensitivity and specificity in two hold-out test sets to the teacher network. We demonstrated the effectiveness of a student-teacher framework for training fast LWMs for automated B-scan of interest detection leveraging unlabeled, routinely-available data.

Purpose To compare the rate of postoperative endophthalmitis after immediately sequential bilateral cataract surgery (ISBCS) versus delayed sequential bilateral cataract surgery (DSBCS) using the American Academy of Ophthalmology Intelligent Research in Sight (IRIS) Registry database. Design Retrospective cohort study. Participants Patients in the IRIS Registry who underwent cataract surgery from 2013 through 2018. Methods Patients who underwent cataract surgery were divided into 2 groups: (1) ISBCS and (2) DSBCS (second-eye surgery ≥1 day after the first-eye surgery) or unilateral surgery. Postoperative endophthalmitis was defined as endophthalmitis occurring within 4 weeks of surgery by International Classification of Diseases (ICD) code and ICD code with additional clinical criteria. Main Outcome Measures Rate of postoperative endophthalmitis. Results Of 5 573 639 IRIS Registry patients who underwent cataract extraction, 165 609 underwent ISBCS, and 5 408 030 underwent DSBCS or unilateral surgery (3 695 440 DSBCS, 1 712 590 unilateral surgery only). A total of 3102 participants (0.056%) met study criteria of postoperative endophthalmitis with supporting clinical findings. The rates of endophthalmitis in either surgery eye between the 2 surgery groups were similar (0.059% in the ISBCS group vs. 0.056% in the DSBCS or unilateral group; P = 0.53). Although the incidence of endophthalmitis was slightly higher in the ISBCS group compared with the DSBCS or unilateral group, the odds ratio did not reach statistical significance (1.08; 95% confidence interval, 0.87–1.31; P = 0.47) after adjusting for age, sex, race, insurance status, and comorbid eye disease. Seven cases of bilateral endophthalmitis with supporting clinical data in the DSBCS group and no cases in the ISBCS group were identified. Conclusions Risk of postoperative endophthalmitis was not statistically significantly different between patients who underwent ISBCS and DSBCS or unilateral cataract surgery.

IMPORTANCE: Approximately 2 million cataract operations are performed annually in the US, and patterns of cataract surgery delivery are changing to meet the increasing demand. Therefore, a comparative analysis of visual acuity outcomes after immediate sequential bilateral cataract surgery (ISBCS) vs delayed sequential bilateral cataract surgery (DSBCS) is important for informing future best practices. OBJECTIVE: To compare refractive outcomes of patients who underwent ISBCS, short-interval (1-14 days between operations) DSBCS (DSBCS-14), and long-interval (15-90 days) DSBCS (DSBCS-90) procedures. DESIGN, SETTING, AND PARTICIPANTS: This retrospective cohort study used population-based data from the American Academy of Ophthalmology Intelligent Research in Sight (IRIS) Registry. A total of 1 824 196 IRIS Registry participants with bilateral visual acuity measurements who underwent bilateral cataract surgery were assessed. EXPOSURES: Participants were divided into 3 groups (DSBCS-90, DSBCS-14, and ISBCS groups) based on the timing of the second eye surgery. Univariable and multivariable linear regression models were used to analyze the refractive outcomes of the first and second surgery eye. MAIN OUTCOMES AND MEASURES: Mean postoperative uncorrected visual acuity (UCVA) and best-corrected visual acuity (BCVA) after cataract surgery. RESULTS: This study analyzed data from 1 824 196 patients undergoing bilateral cataract surgery (mean [SD] age for those

Author(s): Lin, Andrew C; Lee, Cecilia S; Blazes, Marian; Lee, Aaron Y; Gorin, Michael B | Abstract: PurposeOptical coherence tomography (OCT) is widely used in the management of retinal pathologies, including age-related macular degeneration (AMD), diabetic macular edema (DME), and primary open-angle glaucoma (POAG). We used machine learning techniques to understand diagnostic performance gains from expanding macular OCT B-scans compared with foveal-only OCT B-scans for these conditions.MethodsElectronic medical records were extracted to obtain 61 B-scans per eye from patients with AMD, diabetic retinopathy, or POAG. We constructed deep neural networks and random forest ensembles and generated area under the receiver operating characteristic (AUROC) and area under the precision recall (AUPR) curves.ResultsAfter extracting 630,000 OCT images, we achieved improved AUROC and AUPR curves when comparing the central image (one B-scan) to all images (61 B-scans). The AUROC and AUPR points of diminishing return for diagnostic accuracy for macular OCT coverage were found to be within 2.75 to 4.00 mm (14-19 B-scans), 4.25 to 4.50 mm (20-21 B-scans), and 4.50 to 6.25 mm (21-28 B-scans) for AMD, DME, and POAG, respectively. All models with g0.25 mm of coverage had statistically significantly improved AUROC/AUPR curves for all diseases (P l 0.05).ConclusionsSystematically expanded macular coverage models demonstrated significant differences in total macular coverage required for improved diagnostic accuracy, with the largest macular area being relevant in POAG followed by DME and then AMD. These findings support our hypothesis that the extent of macular coverage by OCT imaging in the clinical setting, for any of the three major disorders, has a measurable impact on the functionality of artificial intelligence decision support.Translational relevanceWe used machine learning techniques to improve OCT imaging standards for common retinal disease diagnoses.

People who have Alzheimer's disease neuropathologic change (ADNC) typically associated with dementia but not the associated cognitive decline can be considered to be “resilient” to the effects of ADNC. We have previously reported lower neocortical levels of hyperphosphorylated tau (pTau) and less limbic‐predominant age‐related TDP‐43 encephalopathy neuropathologic change (LATE‐NC) in the resilient participants compared to those with dementia and similar ADNC as determined by current NIA‐AA recommendations using traditional semi‐quantitative assessments of amyloid β and pathological tau burden. To better understand differences between AD‐dementia and resilient participants, we developed and applied a deep learning approach to analyze the neuropathology of 14 brain donors from the Adult Changes in Thought study, including seven stringently defined resilient participants and seven age‐matched AD‐dementia controls. We created two novel, fully automated deep learning algorithms to quantify the level of phosphorylated TDP‐43 (pTDP‐43) and pTau in whole slide imaging. The models performed better than traditional techniques for quantifying pTDP‐43 and pTau. The second model was able to segment lesions staining for pTau into neurofibrillary tangles (NFTs) and tau neurites (neuronal processes positive for pTau). Both groups had similar quantities of pTau localizing to neurites, but the pTau burden associated with NFTs in the resilient group was significantly lower compared to the group with dementia. These results validate use of deep learning approaches to quantify clinically relevant microscopic characteristics from neuropathology workups. These results also suggest that the burden of NFTs is more strongly associated with cognitive impairment than the more diffuse neuritic tau commonly seen with tangle pathology and suggest that additional factors may underlie resilience mechanisms defined by traditional means., To better understand the neuropathology of individuals with Alzheimer's disease (AD) dementia and "resilient" individuals who have AD pathology but are cognitively normal, we developed deep learning algorithms to quantify the level of phosphorylated TDP‐43 and pTau in whole slide imaging. The models performed better than traditional techniques for quantifying pTDP‐43 and pTau, and the pTau model was able to segment pTau lesions into neurofibrillary tangles (NFTs) and tau neurites (neuronal processes positive for pTau). While both groups had similar quantities of pTau localizing to neurites, the resilient group had significantly less pTau burden associated with NFTs compared to the group with dementia.

Background: Vascular disease is a risk factor for Alzheimer’s disease (AD) and related dementia in older adults. Retinal artery/vein occlusion (RAVO) is an ophthalmic complication of systemic vascular pathology. Whether there are associations between RAVO and dementia risk is unknown. Objective: To determine whether RAVOs are associated with an increased risk of developing vascular dementia or AD. Methods: Data from Adult Changes in Thought (ACT) study participants were analyzed. This prospective, population-based cohort study followed older adults (age ≥65 years) who were dementia-free at enrollment for development of vascular dementia or AD based on research criteria. RAVO diagnoses were extracted from electronic medical records. Cox-regression survival analyses were stratified by APOE ɛ4 genotype and adjusted for demographic and clinical factors. Results: On review of 41,216 person-years (4,743 participants), 266 (5.6%) experienced RAVO. APOE ɛ4 carriers who developed RAVO had greater than four-fold higher risk for developing vascular dementia (Hazard Ratio [HR] 4.54, 95% Confidence Interval [CI] 1.86, 11.10, p = 0.001). When including other cerebrovascular disease (history of carotid endarterectomy or transient ischemic attack) in the model, the risk was three-fold higher (HR 3.06, 95% CI 1.23, 7.62). No other conditions evaluated in the secondary analyses were found to confound this relationship. There was no effect in non-APOE ɛ4 carriers (HR 1.03, 95% CI 0.37, 2.80). There were no significant associations between RAVO and AD in either APOE group. Conclusion: Older dementia-free patients who present with RAVO and carry the APOE ɛ4 allele appear to be at higher risk for vascular dementia.

The eye and brain share common mechanisms of aging and disease, thus the retina is an essential source of accessible information about neurodegenerative processes occurring in the brain. Advances in retinal imaging have led to the discovery of many potential biomarkers of Alzheimer’s disease, although further research is needed to validate these associations. Understanding the mechanisms of retinal disease in the context of aging will extend our knowledge of AD and may enable advancements in diagnosis, monitoring, and treatment.

In response to the COVID-19 global pandemic, recent research has proposed creating deep learning based models that use chest radiographs (CXRs) in a variety of clinical tasks to help manage the crisis. However, the size of existing datasets of CXRs from COVID-19+ patients are relatively small, and researchers often pool CXR data from multiple sources, for example, using different x-ray machines in various patient populations under different clinical scenarios. Deep learning models trained on such datasets have been shown to overfit to erroneous features instead of learning pulmonary characteristics – a phenomenon known as shortcut learning. We propose adding feature disentanglement to the training process, forcing the models to identify pulmonary features from the images while penalizing them for learning features that can discriminate between the original datasets that the images come from. We find that models trained in this way indeed have better generalization performance on unseen data; in the best case we found that it improved AUC by 0.13 on held out data. We further find that this outperforms masking out non-lung parts of the CXRs and performing histogram equalization, both of which are recently proposed methods for removing biases in CXR datasets.

Purpose Delayed rod-mediated dark adaptation (RMDA) is a functional biomarker for incipient age-related macular degeneration (AMD). We used anatomically restricted spectral domain optical coherence tomography (SD-OCT) imaging data to localize de novo imaging features associated with and to test hypotheses about delayed RMDA. Methods Rod intercept time (RIT) was measured in participants with and without AMD at 5 degrees from the fovea, and macular SD-OCT images were obtained. A deep learning model was trained with anatomically restricted information using a single representative B-scan through the fovea of each eye. Mean-occlusion masking was utilized to isolate the relevant imaging features. Results The model identified hyporeflective outer retinal bands on macular SD-OCT associated with delayed RMDA. The validation mean standard error (MSE) registered to the foveal B-scan localized the lowest error to 0.5 mm temporal to the fovea center, within an overall low-error region across the rod-free zone and adjoining parafovea. Mean absolute error (MAE) on the test set was 4.71 minutes (8.8% of the dynamic range). Conclusions We report a novel framework for imaging biomarker discovery using deep learning and demonstrate its ability to identify and localize a previously undescribed biomarker in retinal imaging. The hyporeflective outer retinal bands in central macula on SD-OCT demonstrate a structural basis for dysfunctional rod vision that correlates to published histopathologic findings. Translational relevance This agnostic approach to anatomic biomarker discovery strengthens the rationale for RMDA as an outcome measure in early AMD clinical trials, and also expands the utility of deep learning beyond automated diagnosis to fundamental discovery.

We thank Dr. Soliz for his expedient comments (1) regarding our work evaluating seven artificial intelligence (AI) algorithms in a head-to-head fashion (2). We welcome the opportunity to clarify these points of concern. We agree that the two U.S. Food and Drug Administration–approved diabetic retinopathy (DR) systems, one of which was tested in our study, use a threshold of moderate nonproliferative DR (NPDR) or higher for referral. We set the threshold to any DR versus no DR because that represents the clinical practice at the Veterans Affairs (VA) teleretinal system, as stated in our methods. All participating companies were aware of the study threshold prior to submitting their algorithms and were allowed to adjust their algorithms accordingly. As the VA manages one of the largest teleretinal screening systems in the U.S., …

Importance Big data studies may allow for the aggregation of patients with rare diseases such as uveitis to answer important clinical questions. Standardization of uveitis-related variables will be necessary, including the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes used to identify patients of interest. There are currently limited data on the uniformity of diagnosis mapping to ICD-10 codes for uveitis diagnoses among different health systems. Objective To assess the degree of uniformity in mapping of uveitis clinical concepts to ICD-10 codes across health care systems using the same electronic health record (EHR) system. Design, setting, and participants This multicenter survey study was conducted between September 14 and October 9, 2020, at 5 academic health care systems that use the Epic EHR. Researchers from the University of Washington, Harvard University, Stanford University, Yale University, and the University of California, San Francisco queried 54 uveitis-related diagnostic terms and recorded the associated ICD-10 codes. Main outcomes and measures The degree of uniformity for uveitis clinical concepts and associated ICD-10 codes. Results Fifty-four uveitis-related diagnostic terms were queried within the Epic EHR at 5 different health care systems. There was perfect agreement among all 5 centers for 52 of the 54 diagnostic terms. Two diagnostic terms had differences in ICD-10 coding: juvenile idiopathic arthritis associated chronic uveitis and intermediate uveitis. Intermediate uveitis was associated with codes H20.1x (ICD-10 description: chronic iridocyclitis) or H20.9 (ICD-10 description: unspecified iridocyclitis) in 3 centers while being associated with code H30.2x (ICD-10 description: posterior cyclitis) at the 2 remaining centers. The discrepancies appear to be related to a recent update in diagnostic mapping in the Epic EHR. Conclusions and relevance This study suggests that ICD-10 code mapping to uveitis diagnostic terminology appears to be highly uniform at different centers with the Epic EHR. However, temporal changes in diagnosis mapping to ICD-10 codes and a lack of 1-to-1 mapping of diagnosis to ICD-10 code add additional sources of complexity to the interpretation of big data studies in uveitis.