Diagnostic Accuracy of Tele-ophthalmology Alternatives for Diabetic Retinopathy Screening: A Meta-analysis

Abstract

 

Purpose

 

Tele-ophthalmology has been proposed as a suitable alternative to leverage the delivery of specialized eye care in underserved populations. Thus, a meta-analysis was conducted to quantitatively synthesize the evidence pertaining to the diagnostic accuracy of tele-ophthalmology tools for diabetic retinopathy (DR) screening, and detect relevant factors that may influence its performance.

 

Methods

 

A structured search was conducted among six databases from January 1998 to July 2012. The searches were supplemented by handsearching the bibliographies of included studies, and grey literature was consulted. Two reviewers independently screened the abstracts and full-text of articles that met inclusion criteria. Data of 23 studies was extracted according to protocol and methodological quality was assessed using the QUADAS-2 tool. Based on detection threshold, the studies were stratified into two subgroups (1) Detection of any DR, and (2) detection of referable DR. A hierarchical bivariate random effects model was conducted for each category, and hierarchical summary receiver operator characteristic curves were also plotted separately. Likewise, subgroup analyses were conducted to explore the influence of pharmacologic mydriasis, number of fields taken per eye and methodological quality of studies on the reported diagnostic performance.

 

Results

 

For detection of any DR, the pooled sensitivity (Se) was 89% (95% CI: 81% to 93%) and specificity (Sp) 94% (95% CI: 89% to 96%). Alternatively, the pooled Se and Sp for the detection of referable DR was 91% (95% CI: 87% to 94%) and 92% (95% CI: 88% to 95%) respectively. Diagnostic performance was lower in studies that used a single field for the detection of any DR. Substantial heterogeneity was observed in both categories, partially explained by the threshold effect and study design differences. Methodological quality of studies was acceptable; however poor reporting of both patient demographics and index technology was noted in almost half of the studies.

 

Conclusions

 

Diagnostic accuracy of telemedicine strategies for DR screening in adults is satisfactory (Se >80%, Sp > 90%). Of note, summary estimates should be cautiously interpreted given the considerable heterogeneity observed in both categories. A comprehensive reporting of demographic information and index technology characteristics is highly recommended for future studies.

 

Keywords: 459 clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • 499 diabetic retinopathy