Main findings. The results of the study showed that in the population studied, the Peruvian customized charts 1) estimated the prevalence of SGA in foetuses more closely, and similarly to the Hadlock curve; 2) had a large effect size for the relationship between SGA and APO; 3) showed a low PPV, although it was higher than that of the Hadlock and FMF standards, and, like the others standards, had a high NPV which was higher than of the FMF standard. In addition, they revealed an LR + value with strong evidence to confirm the diagnosis, although without a supporting LR-. 4) Similar to the other curves, the customized charts showed poor performance predicting APO (AUC = 0.688), although their performance was superior to that of the FMF standard. However, they had a good performance in detecting an Apgar score < 7 at 5 minutes (AUC = 0.852), surpassing that of the WHO and FMF standards. Based on partial AUC, the Peruvian charts and the IG21 standard showed superiority over the other standards predicting APO, ICU admission and an Apgar score < 7 at 5 minutes.
Prevalence comparison. The prevalence of SGA is approximately 10% in different populations [15, 22]. This is compatible with the prevalence of 10.6% reported in a study carried out with the Peruvian population (n = 29,239) [23]. Therefore, it can be affirmed that the SGA prevalence of 10.5%, indicated by the Peruvian customized charts, best approximates the reality of the studied population. Taking these observations into account, we can affirm that the IG21 standard underestimates the proportion of SGA when applied to the Peruvian population (6.7%), causing unidentified foetuses to have a lack of prenatal surveillance. Likewise, the WHO and FMF standards overestimate this proportion (14.8% and 19.3%), which can lead to the unnecessary use of resources and overacting. [24]
Risk of perinatal adverse outcome based on small for gestational age diagnosis. It was found that, using the Peruvian charts, the ratio of APO vs. its absence was 9.9 times higher when a foetus was classified as SGA vs. normal weight (odds ratio).The calculated effect size corresponded to a Cohen's d > 0.8 (large effect) [21]. In this regard, several studies reported an increased risk of APO in SGA foetuses including that of Kabiri [15], who found a relative risk (RR) = 2.46 (95% CI, 1.9–3.1) for composite neonatal morbidity using a National Institute of Child Health and Human Development (NICHD) growth chart adjusted for ethnicity and a significantly lower RR when the FMF standard was used (RR = 1.47; 95% CI, 1.2–1.8). Additionally, Sovio [25] found that an estimated foetal weight less than the 10th percentile was associated with an increased risk of neonatal morbidity (RR = 1.60, 95% CI 1·22–2·09, p = 0.0012). It should be noted that we observed an absence of association between the detection of SGA by the IG21 standard and perinatal death, so prospective studies with a larger population should be carried out to corroborate this finding. However, these data should be taken into account when using the IG21 standard in the Peruvian population.
Accuracy of customized charts and growth standards. A diagnostic test with a high rate of true positives and a low rate of false-negatives is ideal for screening [26]. This was not the case for the curves studied since they showed sensitivities that did not exceed 50%. However, with the exception of the FMF standard, the curves showed high specificities. These findings are consistent with those of Kabiri [15], who also found low sensitivities in various growth standards for APO, ranging between 15% (NICHD) and 32% (FMF), while their specificities were also relatively high (77% for FMF and 94% for NICHD).
However, since sensitivity and specificity are intrinsic properties of a test [27], in actual clinical practice we do not have a priori data on which patients have APO. Therefore, it is more useful to know the probability that a foetus will have APO given that it has been classified as having SGA (PPV) [27]. In this regard, the Peruvian charts had a PPV = 52.5%, which was higher than the reference curve (Hadlock curve) and the FMF standard, while the probability that a foetus will not have APO given that it has been not classified as having SGA (NPV) was high for the Peruvian charts (90%), which was similar to the other standards. Kabiri [15], found similar PPVs of 21%, 16%, 18%, and 21%, for the WHO, FMF, Hadlock, and IG21 standards, respectively, with an NPV of 89% for all the standards studied.
However, predictive values are known to vary with changing disease prevalence [27]. To overcome this inconvenience, while also reflecting the sensitivity and specificity, the likelihood ratios were used [26]. On this point, the Peruvian charts indicated that the odds of APO increase 5.8 times when a foetus is classified as having SGA (LR+) [26], showing strong diagnostic evidence to confirm the diagnosis (LR + > 10: convincing diagnostic evidence; LR + > 5: strong diagnostic evidence)[26]. However, the LR- did not reach, in any growth curve, the ideal value to indicate how much the odds of APO decrease when the test is negative for SGA (ideal LR-: < 0.2). [28]
Performance of customized charts and growth standards for the prediction of adverse perinatal outcome in SGA foetuses. ROC curves are used to assess the overall performance of a diagnostic test, and it is considered that the test with the highest AUC has the best diagnostic yield [17]. The Peruvian charts, like the standards studied, showed poor performance in discriminating foetuses who had APO from those who did not (AUC = 0.688), although the performance was superior to that of the FMF standard. However, the Peruvian charts had a good performance in detecting an Apgar score < 7 at 5 minutes (AUC = 0.852), surpassing that of the WHO and FMF standards, therefore, if a foetus is classified as having SGA by the Peruvian charts, the paediatric team should be alerted for immediate care of the newborn. The AUCs in this study were higher than those reported by Kabiri[15], who found AUCs of 0.547, 0.550, 0.547 and 0.544 for the Hadlock, NICHD, IG21, and FMF standards, respectively. On the other hand, in medical practice it is interesting to know the performance of a test at the lowest rate of false-positives. Therefore, when comparing the AUCs in a range of less than 10% false-positives (partial AUC) [29], the Peruvian charts and IG21 standard were significantly superior to the other curves in the prediction of APO, admission to the ICU and the detection of an Apgar score < 7 at 5 minutes.
Based on these results, we can affirm that the Peruvian customized charts can be useful in the assessment of foetal weights when incorporated into the clinical field. If the weight is normal, a high NPV (90%) indicates that the foetus is unlikely to have an APO. If a foetus is classified as having SGA, a PPV = 52.5%, an LR + = 5.8 without a supporting LR-, and a poor AUC (although better than other growth curves, especially when the false-positive is < 10%) indicate that we must modify the estimated risk through more tests, such as Doppler ultrasonography [30, 31]. However, given the high performance of the Peruvian charts in predicting an Apgar score < 7 at 5 minutes (AUC = 0.852), we must prepare the necessary equipment for the immediate care of SGA newborns.
Strengths and weaknesses. The strength of this study is that the performance of the most widely used foetal growth standards in Peru were compared. Likewise, the diagnostic performance of the first Peruvian customized foetal growth charts was studied. However, this study also has weaknesses, such as classifying foetuses as having SGA considering only foetal weight, without including abdominal circumference, a parameter that is also accepted and recommended for this purpose [32]. Likewise, we used birth weight rather than weighted ultrasound estimates to compare the diagnostic accuracy of ultrasound-based foetal weight charts and standards. However, since the main clinical utility of charts is evidenced in the postnatal period, we consider birth weight to be an acceptable indicator.
In conclusion, the Peruvian customized charts more accurately estimated the prevalence of SGA in foetuses and had a large effect size by identifying a significant association between an SGA diagnosis and adverse perinatal outcome. The diagnostic metrics support the use of the Peruvian charts as a diagnostic and prognostic test in which the probability of APO is reduced when the diagnosis of SGA is excluded, and also as a test with a good ability to predict an Apgar score < 7 at 5 minutes. Finally, the FMF standard showed the lowest performance when applied in the studied population. These findings will need to be confirmed with prospective work and larger sample sizes.