Daily Anesthesiology Research Analysis
Three studies advance perioperative and critical care anesthesiology: (1) latent class analysis identified two post–hematopoietic cell transplant ARDS phenotypes with distinct trajectories and 90-day mortality; (2) a double-blind RCT showed thoracoscopic intercostal nerve block provides superior early analgesia to erector spinae plane block after uniportal VATS lobectomy; and (3) a multicenter machine learning model predicted red cell transfusion risk in mitral valve surgery using routine preope
Summary
Three studies advance perioperative and critical care anesthesiology: (1) latent class analysis identified two post–hematopoietic cell transplant ARDS phenotypes with distinct trajectories and 90-day mortality; (2) a double-blind RCT showed thoracoscopic intercostal nerve block provides superior early analgesia to erector spinae plane block after uniportal VATS lobectomy; and (3) a multicenter machine learning model predicted red cell transfusion risk in mitral valve surgery using routine preoperative variables.
Research Themes
- Phenotyping and precision stratification in critical care
- Optimization of regional anesthesia for thoracic surgery
- Perioperative machine learning for blood management
Selected Articles
1. Acute Respiratory Distress Syndrome Phenotypes After Stem Cell Transplantation: A Latent Class Analysis.
Using latent class analysis in 166 HCT recipients with ARDS, two phenotypes emerged with distinct physiology, timing, and outcomes. Class 1 showed worse gas exchange, higher bilirubin, later onset, more idiopathic pneumonia syndrome, and markedly higher 90-day mortality; a six-variable parsimonious model classified classes with 0.90 accuracy.
Impact: This work advances precision critical care by defining actionable ARDS phenotypes after HCT using routinely available variables, enabling stratification for trials and tailored management.
Clinical Implications: Phenotype-specific risk stratification may guide ventilatory strategies, immunomodulation, timing of escalations, and enrollment in targeted trials for post-HCT ARDS. The six-variable model could enable bedside classification.
Key Findings
- Two latent classes optimally described post-HCT ARDS, differing in gas exchange, bilirubin, timing, and outcomes.
- Class 1 had worse hypoxemia (P/F 157 vs 210), higher Pco2 (41 vs 36 mm Hg), higher bilirubin (1.4 vs 0.9 mg/dL), later onset, and higher 90-day mortality (72.8% vs 48.2%).
- A parsimonious six-variable model (WBC, platelets, bilirubin, Pco2, BMI, temperature) achieved 0.90 classification accuracy.
- Class 1 aligned with idiopathic pneumonia syndrome; class 2 aligned with peri-engraftment respiratory distress and neutropenia.
Methodological Strengths
- Multicenter cohort with rigorous latent class model selection (BIC, entropy, VLMR-LRT).
- Use of routinely available variables enabling real-world applicability and bedside translation.
Limitations
- Retrospective design within a single healthcare system; lack of external validation.
- No interventional testing; biological correlates were not directly measured.
Future Directions: Prospective external validation of the parsimonious classifier, integration with biomarkers to anchor biology, and phenotype-specific interventional trials in post-HCT ARDS.
2. Intercostal nerve block is superior than erector spinae plane block after uniportal video-assisted thoracoscopic surgery: randomized controlled trial.
In a double-blind RCT of 60 uniportal VATS lobectomy patients, thoracoscopic intercostal nerve block yielded lower resting and coughing pain scores at 4 and 8 hours versus ESPB, and delayed opioid initiation. Safety profiles and recovery metrics were similar.
Impact: Provides randomized, blinded evidence to guide regional analgesia choice after uniportal VATS, an expanding thoracic surgical approach.
Clinical Implications: For early postoperative analgesia after uniportal VATS lobectomy, ICNB may be favored over ESPB without compromising safety or recovery. Protocols may prioritize ICNB when feasible expertise and equipment are available.
Key Findings
- ICNB reduced resting and coughing VAS scores at 4 and 8 hours postoperatively compared with ESPB (p ≤ 0.017).
- ESPB patients initiated morphine earlier (1.5 vs 10 hours, p = 0.002) and had higher 24-hour cumulative consumption (11 vs 7 mg; trend, p = 0.103).
- No block-related adverse events; similar rates of PONV, complications, drainage duration, and hospital stay.
Methodological Strengths
- Double-blind randomized controlled design with standardized analgesic assessments.
- Direct head-to-head comparison of two contemporary regional techniques in a homogeneous surgical cohort.
Limitations
- Single-center, small sample size limits precision and generalizability.
- Primary benefits were confined to early postoperative hours; 24-hour opioid reduction did not reach statistical significance.
Future Directions: Larger multicenter RCTs with longer follow-up and patient-centered outcomes (e.g., recovery quality, persistent pain) to confirm and extend findings; cost-effectiveness and implementation studies.
3. Machine learning for the prediction of blood transfusion risk during or after mitral valve surgery: a multicenter retrospective cohort study.
Across 1,477 mitral valve surgery patients from eight centers, a LightGBM model selected via mRMR achieved AUC 0.734 in validation and 74.2% accuracy prospectively, driven by ten routine preoperative variables. SHAP improved interpretability, highlighting hematocrit, RBC count, and body habitus as top predictors.
Impact: Offers a pragmatic, interpretable tool to anticipate transfusion needs using widely available preoperative data, potentially improving perioperative blood management and resource allocation.
Clinical Implications: Preoperative risk stratification can guide blood product preparation, cell salvage planning, and patient optimization (e.g., anemia management), reducing delays and wastage.
Key Findings
- Among multiple ML models, LightGBM provided the best performance (training AUC 0.935; validation AUC 0.734; prospective accuracy 74.2%).
- Top predictors (via SHAP) included hematocrit, RBC count, weight, BMI, fibrinogen, hemoglobin, height, age, left ventricular dilation, and sex.
- Feature selection via mRMR identified ten key preoperative variables from an initial set of thirty.
Methodological Strengths
- Multicenter cohort with prospective dataset assessment and model interpretability via SHAP.
- Systematic feature selection (mRMR) and evaluation of multiple ML algorithms.
Limitations
- Retrospective development with performance drop from training to validation indicates potential overfitting.
- Generalizability beyond participating centers and healthcare systems remains to be established; calibration metrics not reported.
Future Directions: External validation across diverse populations, incorporation of intraoperative variables for dynamic updating, and impact analyses on clinical workflow and blood product utilization.