Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Letter to the Editor
  • Published:

Predictive models for severe intraventricular hemorrhage in preterm infants

Journal of Perinatology volume 34, page 802 (2014)Cite this article

Subjects

It was with great interest that we read the recent article by Luque et al.1 on a prediction model for severe intraventricular hemorrhage (SIVH) and noticed that it is quite consistent with our similar publication on a predictive model for SIVH.2 Both models demonstrate the utility of routinely-collected clinical data in predicting SIVH among vulnerable preterm infants and show good discrimination between infants with SIVH and those without (0.79 for Luque et al.1 and 0.84 for Singh et al.2). Further, the Singh model showed good calibration as indicated by a comparison of the observed and expected risk. Calibration of the Luque model appears more marginal, as indicated by the discrepancy between the observed SIVH risk and the SIVH risk quintile in Figure 2. In both the models, the benefits of indomethacin for decreasing SIVH risk are apparent, more so as the risk decile increases. The differences between the models can be explained by differences in sample definition, as well as differences in variable selection. Singh et al.2 enrolled all infants born less than 32 weeks (and ⩾500 g). This sample definition yielded an SIVH incidence of about 5%. Luque et al.1 restricted their sample to more vulnerable infants, that is, those with birth weights between 500 and 1249 g. This yielded an SIVH incidence of about 15%.

We wish to address two interpretations by the authors. One concerns the investigators’ reference to the cut point of 17%, which they indicate maximizes sensitivity and specificity. There are several issues regarding this cut point and their interpretation that require clarification. First, maximizing sensitivity and specificity is a model-specific determination that minimizes classification errors (that is, false positives (FP) and false negatives (FN)) more or less equally. For clinical practice, it may not represent the optimal cut point. Maximizing sensitivity and specificity does not account for the relative cost of the classification errors, nor the fact that indomethacin use may have detrimental side effects, particularly on this vulnerable population. If the cost of FN errors (for example, clinically, financially or as perceived by parents) is considered greater than FP errors, then one may wish to select a cut point that enhances sensitivity (that is, reduce FN) and tolerate a greater FP fraction. If the reverse perception is true, then the appropriate cut point would favor specificity over sensitivity. Thus, the optimal cut point for clinical practice should be based on the relative costs of the errors, as well as the estimated prevalence of the condition.3

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

References

  1. Luque MJ, Tapia JL, Villarroel L, Marshall G, Musante G, Carlo W et al. Neocosur Neonatal Network. A risk prediction model for severe intraventricular hemorrhage in very low birth weight infants and the effect of prophylactic indomethacin. J Perinatol 2014; 34 (1): 43–48.

    Article  CAS  Google Scholar 

  2. Singh R, Gorstein SV, Bednarek F, Chou JH, McGowan EC, Visintainer PF . A predictive model for SIVH risk in preterm infants and targeted indomethacin therapy for prevention. Sci Rep 2013; 3: 2539.

    Article  Google Scholar 

  3. Zweig MH, Campbell G . Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin Chem 1993; 39 (4): 561–577.

    CAS  Google Scholar 

  4. Sinclair JC . Weighing risks and benefits in treating the individual patient. Clin Perinatol 2003; 30: 251–268.

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Division of Neonatology, Department of Pediatrics, Pediatric Residency Research, Tufts University School of Medicine, Baystate Children’s Hospital, Springfield, MA, USA

    R Singh

  2. Department of Epidemiology, Baystate Medical Center, Tufts University School of Medicine, Springfield, MA, USA

    PF Visintainer

Authors
  1. R Singh
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. PF Visintainer
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to PF Visintainer.

Ethics declarations

Competing interests

The authors declare no conflict of interest.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Singh, R., Visintainer, P. Predictive models for severe intraventricular hemorrhage in preterm infants. J Perinatol 34, 802 (2014). https://doi.org/10.1038/jp.2014.152

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/jp.2014.152

This article is cited by

Search

Advanced search

Quick links