Visual loss from ROP may be prevented by early diagnosis and timely treatment, which emphasizes the importance of ROP screening in routine clinical practice [1]. Recently, Pivodic et al. developed an individual risk prediction model, DIGIROP-Birth, using only birth characteristics to describe a continuous hazard function for identifying TR-ROP [23]. This easy-to-use prediction model was built using Swedish National Patient Registry data, and validated in US and European cohorts, yielding satisfactory results. The present study validated the DIGIROP-Birth model in Chinese preterm infants, and found that the model had less satisfactory performance than previously reported (AUC = 0.634 in this study vs. AUC = 0.85 in the study by Pivodic et al.) [23].
Several reasons could account for the discrepancy. First, there were only 83 (5.4%) Asians among the 1535 infants that comprised the US validation group in Pivodic et al.’s study [23], while our study cohort consisted of 442 Chinese infants. Asian infants appear to be at higher risk of developing TR-ROP than white infants due to differences in ethnic ancestry and underlying genetic predisposition [20, 21]. Second, compared with the DIGIROP-Birth model’s training cohort, our cohort of Chinese infants had an older mean gestational age (28.8 weeks vs. 28.1 weeks) and lower mean birth weight (1119 g vs. 1237 g). This might be explained by the fact that in less-developed countries, severe ROP occurs in more mature and larger infants [3, 23]. Third, the quality of neonatal care is one of the most critical factors for ROP development and progression [28]. Besides a lack of neonatologists and nurses with neonatal care expertise, neonatal units are in short supply of enough equipment for continuous monitoring of preterm infants on supplemental oxygen [29–31]. The timing and duration of supplemental oxygen, oxygen concentration, and prolonged mechanical ventilation are among the most crucial risk factors for TR-ROP [32]. Consequently, preterm infants born in less-developed countries or regions are more likely to be exposed to postnatal risk factors for TR-ROP that are better controlled in industrialized countries [29, 31]. Similar characteristics of ROP and the corresponding clinical settings have been reported in other developing countries in Asia and Latin America [29–31]. Finally, the DIGIROP-Birth model did not consider postnatal risk factors for ROP, which could also account for its decreased predictive performance in our Chinese cohort. Modification of the DIGIROP-Birth model through the incorporation of postnatal risk factors might improve its applicability in less-developed countries.
Several studies have shown that complications of prematurity, such as bronchopulmonary dysplasia, apnea, intraventricular hemorrhage, and sepsis, are associated with the development of ROP [32–35]. Gestational age, birth weight, male sex, apnea, and intraventricular hemorrhage were found to be independent risk factors for TR-ROP in our cohort of Chinese infants with gestational ages of 24 to 30 weeks. Therefore, apnea and intraventricular hemorrhage were included as additional risk factors in our modified DIGIROP-Birth model. Our principal goal was to determine the sensitivity of the DIGIROP-Birth model. That is, its ability to rule out TR-ROP and determine the number of ROP screening examinations that could have been safely spared by using this model. With the cutoff value of 0.0084, the sensitivity of the DIGIROP-Birth model improved from 51.6–95.7%, with an NPV of 98.5%. Previous studies also revealed that apnea of prematurity and intraventricular hemorrhage were independently associated with a higher risk of ROP [34–36]. Infants with apnea are more likely to require oxygen therapy, which could induce ROP development due to immature antioxidant systems. Oxygen-related factors play a crucial role in TR-ROP, including the duration of supplemental oxygen, oxygen concentration, and prolonged mechanical ventilation [32]. Although several large randomized-controlled studies have compared different oxygen saturation target ranges, the ideal range that could reduce ROP occurrence without increasing preterm infants' mortality remains controversial [37–40]. Intraventricular hemorrhage occurs in 25–30% of preterm infants with birth weights < 1500g, often causing neurodevelopmental impairment [41]. Early control of intracranial pressure secondary to intraventricular hemorrhage may prevent TR-ROP development in infants with a combined diagnosis of ROP and intraventricular hemorrhage. This is because the progression of ROP may associate with reduced ocular circulation secondary to high intracranial pressure [42]. Thus, apnea and intraventricular hemorrhage, two important premature birth complications, could greatly improve the predictive ability of the DIGIROP-Birth model for TR-ROP in Chinese preterm infants.
The sensitivity of the DIGROP-Birth model in infants with a gestational age < 28 weeks or a birth weight < 1000 g was satisfactory. This suggests that the DIGROP-Birth model may also be valuable as an auxiliary tool for ROP screening in extremely preterm infants and infants with extremely low birth weight. In infants with a gestational age ≥ 28 weeks or birth weight ≥ 1000 g, the DIGIROP-Birth model was less effective, but could be modified with postnatal risk factors. Thus, the DIGIROP-Birth model still has the potential to decrease the frequency of ROP examinations in less-developed countries.
This study has several limitations. First, this validation study was conducted retrospectively. Despite the retrospective nature, however, the clinical data included in our analyses are routinely recorded in the neonatal intensive care units and could be collected reliably. Second, the single-center study had a relatively small sample size compared with other validation studies of ROP prediction models. Future multicenter prospective studies with large cohorts will enhance our findings.