<p>Predictors of LISA.</p>

A dataset used to develop a machine learning algorithm for identifying Less Invasive Surfactant Administration (LISA) procedures in preterm infants using administrative healthcare data. The model was trained on 884 births from 2019-2023 and validated on a combined sample of 1,001 births, achieving an AUROC of 0.87. It was created by Xuezheng Sun and published on figshare in April 2026.

Use Cases

• Training a predictive model for LISA procedure identification based on administrative data variables.
• Validating the performance of a LASSO-based algorithm across different gestational age subgroups.
• Analyzing the relationship between 21 selected predictor variables and the method of surfactant delivery.
• Evaluating algorithm performance metrics like sensitivity, specificity, PPV, and NPV in a real-world clinical cohort.

Strengths

• Model performance is quantified with specific metrics: AUROC of 0.87, specificity of 96.8%, and PPV of 90.0%.
• Development and validation used a substantial sample size: 1,263 preterm infants who received surfactant.
• Algorithm performance was validated on a separate 2024 birth cohort (n=622) in addition to a testing set.
• Performance was evaluated consistently across gestational age subgroups.

Limitations

• The dataset is very small at 9.5 KB, indicating limited scope and likely a summary of model results rather than raw patient data.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for direct modeling.

Provenance

Source

figshare, author Xuezheng Sun.

Collection Method

Retrospective study using chart reviews as a gold standard, with data extracted from Kaiser Permanente Northern California (KPNC) facilities.

Time Range

Births between 2019 and 2024.

Freshness

Last updated 2026-04-15 17:44:27; freshness should be verified.

Geography

Kaiser Permanente Northern California facilities.