The Use of Postnatal Weight Gain Algorithms to Predict Severe or Type 1 Retinopathy of Prematurity: A Systematic Review and Meta-analysis

Importance: The currently recommended method for screening for retinopathy of prematurity (ROP) is binocular indirect ophthalmoscopy, which requires frequent eye examinations entailing a heavy clinical workload. Weight gain-based algorithms have the potential to minimize the need for binocular indirect ophthalmoscopy and have been evaluated in different setups with variable results to predict type 1 or severe ROP. Objective: To synthesize evidence regarding the ability of postnatal weight gain-based algorithms to predict type 1 or severe ROP. Data Sources: PubMed, MEDLINE, Embase, and the Cochrane Library databases were searched to identify studies published between January 2000 and August 2021. Study Selection: Prospective and retrospective studies evaluating the ability of these algorithms to predict type 1 or severe ROP were included. Data Extraction and Synthesis: Two reviewers independently extracted data. This meta-analysis was performed according to the Cochrane guidelines and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) guidelines. Main Outcomes and Measures: Ability of algorithms to predict type 1 or sever ROP was measured using statistical indices (pooled sensitivity, specificity, and summary area under the receiver operating characteristic curves, as well as pooled negative likelihood ratios and positive likelihood ratios and diagnostic odds ratios). Results: A total of 61 studies (>37000 infants) were included in the meta-analysis. The pooled estimates for sensitivity and specificity, respectively, were 0.89 (95% CI, 0.85-0.92) and 0.57 (95% CI, 0.51-0.63) for WINROP (Weight, IGF-1 [insulinlike growth factor 1], Neonatal, ROP), 1.00 (95% CI, 0.88-1.00) and 0.60 (95% CI, 0.15-0.93) for G-ROP (Postnatal Growth and ROP), 0.95 (95% CI, 0.71-0.99) and 0.52 (95% CI, 0.36-0.68) for CHOP ROP (Children's Hospital of Philadelphia ROP), 0.99 (95% CI, 0.73-1.00) and 0.49 (95% CI, 0.03-0.74) for ROPScore, 0.98 (95% CI, 0.94-0.99) and 0.35 (95% CI, 0.22-0.51) for CO-ROP (Colorado ROP). The original PINT (Premature Infants in Need of Transfusion) ROP study reported a sensitivity of 0.98 (95% CI, 0.91-0.99) and a specificity of 0.36 (95% CI, 0.30-0.42). The pooled negative likelihood ratios were 0.19 (95% CI, 0.13-0.27) for WINROP, 0.0 (95% CI, 0.00-0.32) for G-ROP, 0.10 (95% CI, 0.02-0.53) for CHOP ROP, 0.03 (95% CI, 0.00-0.77) for ROPScore, and 0.07 (95% CI, 0.03-0.16) for CO-ROP. The pooled positive likelihood ratios were 2.1 (95% CI, 1.8-2.4) for WINROP, 2.5 (95% CI, 0.7-9.1) for G-ROP, 2.0 (95% CI, 1.5-2.6) for CHOP ROP, 1.9 (95% CI, 1.1-3.3) for ROPScore, and 1.5 (95% CI, 1.2-1.9) for CO-ROP. Conclusions and Relevance: This study suggests that weight gain-based algorithms have adequate sensitivity and negative likelihood ratios to provide reasonable certainty in ruling out type 1 ROP or severe ROP. Given the implications of missing even a single case of severe ROP, algorithms with very high sensitivity (close to 100%) and low negative likelihood ratios (close to zero) need to be chosen to safely reduce the number of unnecessary examinations in infants at lower risk of severe ROP..