Outcome Calculator for Very Low Gestational Age Neonates born in Switzerland

This calculator was developed by Carlos Ochoa, Martina Steurer, Mark Adams, Thomas M. Berger and supported, in part, by a grant of the Swiss National Science Foundation.

Disclaimer

The predicted outcome probability and its 95% confidence interval provided here are by no means intended to be a conclusive or definitive source of information to decide on an individual infant’s outcome probabilities. Instead, this information should be used by clinicians as a supporting aid for better informed decisions.

The results displayed are based on historical values (i.e., the most recent 5-year-period) for Switzerland. The parameters used to predict these outcomes were calculated through alternative approaches, yielding always the same result. However, the limited and usually small sample sizes for most of the NICUs ask for a cautious interpretation of the displayed results. The effect of small sample sizes is reflected in the width of the confidence intervals. The internal validity of the prediction was also measured via two different methods, ROC/AUC and Brier’s score (see interpretation). An external validation was not possible because of the local characteristics of the data.

Please note that Cesarean section and outborn status were not considered in the model. There is no convincing evidence that Cesarean section is clearly beneficial for very low gestational age neonates and the prevalence of being outborn for this population is very low in Switzerland (5%) and would therefore produce reference groups with limited statistical power.

Interpretation

1. Predictive probability and 95% confidence interval: depending on the selected outcome, the results obtained represent the estimated probability of either survival or survival without severe morbidities and their respective 95% confidence intervals at three different time points (at birth, on admission to the NICU and on day of life 7). The correct interpretation of the probability, for example as a percentage, is: from a total of 100 patients with the given set of characteristics (as entered into the calculator), how many of them are expected to survive (or respectively, survive without severe morbidity). The 95% prediction interval (confidence interval for the predictive probability) can be interpreted as a 95% confidence that the probability of the outcome lies inside the interval.
2. Brier’s score: the lower the score, the better the predictions are calibrated. The values range between 0 (for a perfect calibration) and 0.25 (for a useless/random calibration).
3. ROC/AUC: a higher value means a better predictive validity of the method. The values range between 0.5 (for a useless/random prediction) and 1 (for a perfect/accurate prediction). Values between 0.7 and 0.8 are considered moderate, between 0.8 and 0.9 are good, and higher than 0.9 are very good.