Premium
Predictive algorithms for early detection of retinopathy of prematurity
Author(s) -
Piermarocchi Stefano,
Bini Silvia,
Martini Ferdinando,
Berton Marianna,
Lavini Anna,
Gusson Elena,
Marchini Giorgio,
Padovani Ezio Maria,
Macor Sara,
Pignatto Silvia,
Lanzetta Paolo,
Cattarossi Luigi,
Baraldi Eugenio,
Lago Paola
Publication year - 2017
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/aos.13117
Subject(s) - retinopathy of prematurity , medicine , gestational age , birth weight , pediatrics , receiver operating characteristic , algorithm , pregnancy , genetics , computer science , biology
Purpose To evaluate sensitivity, specificity and the safest cut‐offs of three predictive algorithms ( WINROP , ROPS core and CHOP ROP ) for retinopathy of prematurity ( ROP ). Methods A retrospective study was conducted in three centres from 2012 to 2014; 445 preterms with gestational age ( GA ) ≤ 30 weeks and/or birthweight ( BW ) ≤ 1500 g, and additional unstable cases, were included. No‐ ROP , mild and type 1 ROP were categorized. The algorithms were analysed for infants with all parameters ( GA , BW , weight gain, oxygen therapy, blood transfusion) needed for calculation (399 babies). Results Retinopathy of prematurity (ROP) was identified in both eyes in 116 patients (26.1%), and 44 (9.9%) had type 1 ROP . Gestational age and BW were significantly lower in ROP group compared with no‐ ROP subjects ( GA : 26.7 ± 2.2 and 30.2 ± 1.9, respectively, p < 0.0001; BW : 839.8 ± 287.0 and 1288.1 ± 321.5 g, respectively, p = 0.0016). Customized alarms of ROPS core and CHOP ROP correctly identified all infants having any ROP or type 1 ROP . WINROP missed 19 cases of ROP , including three type 1 ROP . ROPS core and CHOP ROP provided the best performances with an area under the receiver operating characteristic curve for the detection of severe ROP of 0.93 (95% CI , 0.90–0.96, and 95% CI , 0.89–0.96, respectively), and WINROP obtained 0.83 (95% CI , 0.77–0.87). Median time from alarm to treatment was 11.1, 5.1 and 9.1 weeks, for WINROP , ROPS core and CHOP ROP , respectively. Conclusion ROPS core and CHOP ROP showed 100% sensitivity to identify sight‐threatening ROP . Predictive algorithms are a reliable tool for early identification of infants requiring referral to an ophthalmologist, for reorganizing resources and reducing stressful procedures to preterm babies.