Machine learning and SHAP-based risk assessment of PICC-related bloodstream infections in premature infants at the time of clinical suspicion
A machine-learning bedside tool using routine blood counts predicts dangerous bloodstream infections in premature infants with high accuracy.
A prospective NICU study (n=490 preterm infants) demonstrates that a random forest model using CRP, WBC count (CBC-derived), and respiratory rate can predict PICC-related bloodstream infections with AUC 0.973 at clinical suspicion time, offering an interpretable bedside risk stratification tool. WBC count as a key SHAP feature links this directly to CBC-based ML diagnostics watchlist.
What the study was
- Study design
- Prospective cohort with 4 ML algorithms, SHAP interpretability
- Population
- Preterm infants with PICC insertion, NICU, n=490
- Sample size
- 490
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Pediatric Research
Why it surfaced
ML + SHAP with CBC/WBC input for neonatal infection prediction; prospective design; directly matches CBC/ML watchlist topic.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.