Transparency in Artificial Intelligence Reporting in Ophthalmology-A Scoping Review

Topic

This scoping review summarizes artificial intelligence (AI) reporting in ophthalmology literature in respect to model development and validation. We characterize the state of transparency in reporting of studies prospectively validating models for disease classification.

Clinical Relevance

Understanding what elements authors currently describe regarding their AI models may aid in the future standardization of reporting. This review highlights the need for transparency to facilitate the critical appraisal of models prior to clinical implementation, to minimize bias and inappropriate use. Transparent reporting can improve effective and equitable use in clinical settings.

Methods

Eligible articles (as of January 2022) from PubMed, Embase, Web of Science, and CINAHL were independently screened by 2 reviewers. All observational and clinical trial studies evaluating the performance of an AI model for disease classification of ophthalmic conditions were included. Studies were evaluated for reporting of parameters derived from reporting guidelines (CONSORT-AI, MI-CLAIM) and our previously published editorial on model cards. The reporting of these factors, which included basic model and dataset details (source, demographics), and prospective validation outcomes, were summarized.

Results

Thirty-seven prospective validation studies were included in the scoping review. Eleven additional associated training and/or retrospective validation studies were included if this information could not be determined from the primary articles. These 37 studies validated 27 unique AI models; multiple studies evaluated the same algorithms (EyeArt, IDx-DR, and Medios AI). Details of model development were variably reported; 18 of 27 models described training dataset annotation and 10 of 27 studies reported training data distribution. Demographic information of training data was rarely reported; 7 of the 27 unique models reported age and gender and only 2 reported race and/or ethnicity. At the level of prospective clinical validation, age and gender of populations was more consistently reported (29 and 28 of 37 studies, respectively), but only 9 studies reported race and/or ethnicity data. Scope of use was difficult to discern for the majority of models. Fifteen studies did not state or imply primary users.

Conclusion

Our scoping review demonstrates variable reporting of information related to both model development and validation. The intention of our study was not to assess the quality of the factors we examined, but to characterize what information is, and is not, regularly reported. Our results suggest the need for greater transparency in the reporting of information necessary to determine the appropriateness and fairness of these tools prior to clinical use.

Financial Disclosures

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Novel severe oculocutaneous manifestations of human monkeypox virus infection and their historical analogues

WHO has declared human mpox (formerly known as monkeypox) a global public health emergency since July, 2022. When case numbers were increasing, so did clinicians' exposures to new elements of the disease. Additionally, the burden of mpox is particularly apparent in immunocompromised patients, who can have more variable and severe manifestations of disease across organ systems. In this Grand Round, we report novel and severe oculocutaneous manifestations of mpox in this population, which are both sight and life threatening. Specifically, we highlight two patients with mpox and AIDS who had refractory skin necrosis that progressed to either ocular compromise or panfacial gangrene, or both. Both patients ultimately died due to systemic complications of their infections. Through clinical analogies, we show how past experiences with related orthopoxviruses, such as variola virus (smallpox) and vaccinia virus, can add useful context for understanding and treating these new disease states. We suspect that in patients who are immunocompromised, monkeypox virus can clinically evolve not only via viraemia but also through direct intradermal spread. We propose that intradermal spread occurs by a process clinically and immunologically analogous to progressive vaccinia, a complication previously seen after conventional smallpox vaccination. We share evidence in support of this theory and implications regarding early management and post-exposure prophylaxis for at-risk populations. Content note: this Grand Round contains graphic images of mpox lesions of the eyes and face.

Optical Coherence Tomography and Glaucoma

Early detection and monitoring are critical to the diagnosis and management of glaucoma, a progressive optic neuropathy that causes irreversible blindness. Optical coherence tomography (OCT) has become a commonly utilized imaging modality that aids in the detection and monitoring of structural glaucomatous damage. Since its inception in 1991, OCT has progressed through multiple iterations, from time-domain OCT, to spectral-domain OCT, to swept-source OCT, all of which have progressively improved the resolution and speed of scans. Even newer technological advancements and OCT applications, such as adaptive optics, visible-light OCT, and OCT-angiography, have enriched the use of OCT in the evaluation of glaucoma. This article reviews current commercial and state-of-the-art OCT technologies and analytic techniques in the context of their utility for glaucoma diagnosis and management, as well as promising future directions. Expected final online publication date for the Annual Review of Vision Science, Volume 7 is September 2021. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Common Microbes and Antibiotic Resistance in Ocular Infections at an Urban Public Tertiary Care Hospital

PURPOSE

We aim to describe ocular infection epidemiology for a public tertiary care hospital in New York City (NYC).

METHODS

We retrospectively reviewed 558 patients with ocular isolates from conjunctival, corneal, and intraocular culture from 2009 to 2017 for microbial growth and antimicrobial sensitivities.

RESULTS

In total, 185 ocular cultures (33%) had growth and the most commonly isolated microbes overall were Staphylococcus aureus (S. aureus) (23%), coagulase-negative Staphylococcus (CoNS) (23%), Pseudomonas aeruginosa (P. aeruginosa) (16%), and Streptococcus viridans (S. viridans) group (11%). The most common microbes within corneal (n = 61), conjunctival (n = 34), and intraocular isolates (n = 9) were P. aeruginosa (37.7%), S. aureus (35.3%), and S. viridans group (33.3%), respectively. Proportion of isolates exhibiting multi-drug resistance decreased over time (P = .006).

CONCLUSIONS

The microbial epidemiology of ocular infection of a public NYC hospital was distinct from other geographic locations, underscoring the importance of examining local profiles to more precisely inform empiric therapy.

Relationship between Caregivers' Oral Health Literacy and their Child's Caries Experience

OBJECTIVE

This study aimed to assess the association between caregivers' oral health literacy (OHL) and the dental caries experience of their child.

PARTICIPANTS

This cross-sectional study was conducted among 300 caregiver/child dyads at a paediatric dental centre in Saudi Arabia. The OHL was assessed using an Arabic translated version of Comprehensive Measure of Oral Health Knowledge (CMOHK-A) questionnaire. Test-retest reliability and internal consistency was assessed using the appropriate statistical methods.

MAIN OUTCOME MEASURE

Childrens' dental caries experience was assessed using the DMFT and deft index (decayed, missing/extracted, filled teeth) for permanent and primary teeth respectively.

RESULTS

The mean age of the caregivers and children was 37.9±7.9 years and 8.3±3.1 years respectively. Most children (87.7%) had dental caries experience (dmft ⟩ 0) and the mean DMFT/deft was 5.2±4.0. The means±standard deviations (SD) for untreated caries (DT/dt), missing/extracted teeth (MT/et) and filled teeth (FT/ft) were 2.6±2.5, 0.5±0.9 and 2.1±2.2 respectively. Multivariate linear regression models showed that caregivers educational levels and OHL scores were associated with the child's untreated caries levels. Caregivers' gender, educational levels and their perception of the child's oral health were significant predictors for child's lifetime caries experience. Children of caregivers with low OHL had more untreated caries than children of caregivers with adequate OHL.

CONCLUSION

This study found better caregiver OHL levels to be associated with lower caries experience for their child.

Right Ventricular Dysfunction Impairs Effort Tolerance Independent of Left Ventricular Function Among Patients Undergoing Exercise Stress Myocardial Perfusion Imaging

BACKGROUND

Right ventricular (RV) and left ventricular (LV) function are closely linked due to a variety of factors, including common coronary blood supply. Altered LV perfusion holds the potential to affect the RV, but links between LV ischemia and RV performance, and independent impact of RV dysfunction on effort tolerance, are unknown.

METHODS AND RESULTS

The population comprised 2051 patients who underwent exercise stress myocardial perfusion imaging and echo (5.5±7.9 days), among whom 6% had echo-evidenced RV dysfunction. Global summed stress scores were ≈3-fold higher among patients with RV dysfunction, attributable to increments in inducible and fixed LV perfusion defects (all P≤0.001). Regional inferior and lateral wall ischemia was greater among patients with RV dysfunction (both P<0.01), without difference in corresponding anterior defects (P=0.13). In multivariable analysis, inducible inferior and lateral wall perfusion defects increased the likelihood of RV dysfunction (both P<0.05) independent of LV function, fixed perfusion defects, and pulmonary artery pressure. Patients with RV dysfunction demonstrated lesser effort tolerance whether measured by exercise duration (6.7±2.8 versus 7.9±2.9 minutes; P<0.001) or peak treadmill stage (2.6±0.9 versus 3.1±1.0; P<0.001), paralleling results among patients with LV dysfunction (7.0±2.9 versus 8.0±2.9; P<0.001|2.7±1.0 versus 3.1±1.0; P<0.001 respectively). Exercise time decreased stepwise in relation to both RV and LV dysfunction (P<0.001) and was associated with each parameter independent of age or medication regimen.

CONCLUSIONS

Among patients with known or suspected coronary artery disease, regional LV ischemia involving the inferior and lateral walls confers increased likelihood of RV dysfunction. RV dysfunction impairs exercise tolerance independent of LV dysfunction.

Echocardiographic linear fractional shortening for quantification of right ventricular systolic function-A cardiac magnetic resonance validation study

BACKGROUND

Echocardiography (echo)-based linear fractional shortening (FS) is widely used to assess left ventricular dysfunction (LV_dys ), but has not been systematically tested for right ventricular dysfunction (RV_dys ).

METHODS

The population comprised LV_dys patients with and without RV_dys (EF<50%) on cardiac MRI (CMR): Echo included standard RV indices (fractional area change [FAC], TAPSE, S', and FS in parasternal long-axis (RV outflow tract [RV_OT ]) and apical four-chamber views (width [RV_WD ], length [RV_LG ]).

RESULTS

A total of 168 patients underwent echo and CMR (3±3 days); FAC (46±9 vs 28±11), TAPSE (1.9±0.4 vs 1.5±0.3), and S' (11.4±2.3 vs 10.0±2.6, all P≤.001) were lower among RV_dys patients, as were FS indices (RV_OT 32±8 vs 17±10 | RV_WD 40±11 vs 22±12 | RV_LG 16±5 vs 9±4%; all P<.001). FS indices yielded similar magnitude of correlation with CMR RVEF (r=.73-.56) as did FAC (r=.70), which was slightly higher than TAPSE (r=.47) and S' (r=.31; all P<.001). FS indices decreased stepwise vs CMR RVEF tertiles, as did FAC (all P<.001). In multivariate analysis, FS in RV_OT (regression coefficient .51 [CI 0.37-0.65]), RV_WD (0.30 [0.19-0.41]), and RV_LG (0.45 [0.20-0.71]; all P≤.001) was independently associated with CMR RVEF. FS indices yielded good overall diagnostic performance (AUC: RV_OT 0.89 [CI 0.82-0.97] | RV_WD 0.87 [0.78-0.96] | RV_LG 0.80 [0.70-0.90]; all P<.001) for CMR-defined RV_dy (RVEF<50%).

CONCLUSIONS

RV linear FS provides RV functional indices that parallel CMR RVEF. Parasternal long-axis RV_OT width, four-chamber RV width, and length are independently associated with RVEF, supporting use of multiple FS indices for RV functional assessment.

Echocardiographic Linear Dimensions for Assessment of Right Ventricular Chamber Volume as Demonstrated by Cardiac Magnetic Resonance

BACKGROUND

Echocardiography-derived linear dimensions offer straightforward indices of right ventricular (RV) structure but have not been systematically compared with RV volumes on cardiac magnetic resonance (CMR).

METHODS

Echocardiography and CMR were interpreted among patients with coronary artery disease imaged via prospective (90%) and retrospective (10%) registries. For echocardiography, American Society of Echocardiography-recommended RV dimensions were measured in apical four-chamber (basal RV width, mid RV width, and RV length), parasternal long-axis (proximal RV outflow tract [RVOT]), and short-axis (distal RVOT) views. For CMR, RV end-diastolic volume and RV end-systolic volume were quantified using border planimetry.

RESULTS

Two hundred seventy-two patients underwent echocardiography and CMR within a narrow interval (0.4 ± 1.0 days); complete acquisition of all American Society of Echocardiography-recommended dimensions was feasible in 98%. All echocardiographic dimensions differed between patients with and those without RV dilation on CMR (P < .05). Basal RV width (r = 0.70), proximal RVOT width (r = 0.68), and RV length (r = 0.61) yielded the highest correlations with RV end-diastolic volume on CMR; end-systolic dimensions yielded similar correlations (r = 0.68, r = 0.66, and r = 0.65, respectively). In multivariate regression, basal RV width (regression coefficient = 1.96 per mm; 95% CI, 1.22-2.70; P < .001), RV length (regression coefficient = 0.97; 95% CI, 0.56-1.37; P < .001), and proximal RVOT width (regression coefficient = 2.62; 95% CI, 1.79-3.44; P < .001) were independently associated with CMR RV end-diastolic volume (r = 0.80). RV end-systolic volume was similarly associated with echocardiographic dimensions (basal RV width: 1.59 per mm [95% CI, 1.06-2.13], P < .001; RV length: 1.00 [95% CI, 0.66-1.34], P < .001; proximal RVOT width: 1.80 [95% CI, 1.22-2.39], P < .001) (r = 0.79).

CONCLUSIONS

RV linear dimensions provide readily obtainable markers of RV chamber size. Proximal RVOT and basal width are independently associated with CMR volumes, supporting the use of multiple linear dimensions when assessing RV size on echocardiography.

The association between asthma-related emergency department visits and pollen and mold spore concentrations in the Bronx, 2001-2008

BACKGROUND

The incidence of asthma morbidity and mortality is highest among minority inner-city populations. Among New York City's five boroughs, the Bronx has the highest rate of asthma-related hospitalizations and mortality. Outdoor air pollutants have been associated with increased asthma-related ED visits (AREDV) in this borough.

OBJECTIVE

To better understand the contribution of pollen and mold to asthma severity in the Bronx.

METHODS

The numbers of daily adult and pediatric AREDV and asthma-related hospitalizations (ARH) from 2001 to 2008 were obtained from two Bronx hospitals. AREDV and ARH data were acquired retrospectively through the Clinical Looking Glass data analysis software. Daily counts for tree, grass and weed pollen and mold spore counts from March 2001 to October 2008 were obtained from the Armonk counting station. All data were statistically analyzed and graphed as daily values.

RESULTS

There were a total of 42 065 AREDV and 10 132 ARH at both Bronx hospitals. There were spring and winter peaks of increased AREDV. Tree pollen counts significantly correlated with total AREDV (rho = 0.3639, p < 0.001), and pediatric (rho = 0.33, p < 0.001) and adult AREDV (rho = 0.28, p < 0.001). ARH positively correlated with tree pollen counts (Spearman rho = 0.2389, p < 0.001).

CONCLUSIONS

There exists a significant association between spring AREDV and ARH and tree pollen concentrations in a highly urbanized area such as the Bronx. Early anticipation of spring pollen peaks based on ongoing surveillance could potentially guide clinical practice and minimize asthma-related ED visits in the Bronx.

Maternal periodontal status and preterm delivery: a hospital based case-control study

BACKGROUND AND OBJECTIVE

Recent studies have presented evidence that periodontal disease in pregnant women may be a determining factor for preterm delivery. However, this finding has not been consistently observed. The present investigation was carried out to explore the association between maternal periodontal disease and preterm delivery in the state of Kerala, India.

MATERIAL AND METHODS

The case-control study had a sample of 300 (100 cases and 200 controls) postpartum women over 18 years of age. Cases were women who had undergone spontaneous preterm delivery (< 37 wk of gestation) and controls were women who delivered at term (≥ 37 wk of gestation). Standard, clinical and periodontal examinations were performed at the maternity wards, and the existence of an association between periodontal disease and preterm delivery was evaluated by means of a multivariate logistic regression model that also considered other risk factors for preterm delivery.

RESULTS

Periodontitis was diagnosed in 25% of the mothers in the case group and in 14.5% of the mothers in the control group. Logistic regression analysis indicated a risk of nearly threefold for preterm delivery in mothers with periodontitis [adjusted odds ratio (OR(a) ) = 2.72; 95% confidence interval (CI): 1.68-6.84]. The other factors significantly associated with preterm birth were physical exertion (OR(a) = 2.80; 95% CI: 1.18-6.65), a previous history of preterm birth (OR(a) = 2.65; 95% CI: 1.20-5.83) and previous abortion/death of infant (OR(a) = 4.08; 95% CI: 1.56-10.65).

CONCLUSION

Periodontal disease is a possible risk factor for preterm delivery in this population.