Daily Anesthesiology Research Analysis
Three impactful anesthesiology-related studies stood out today: a British Journal of Anaesthesia analysis shows that selection bias dramatically inflates the apparent performance of intraoperative hypotension prediction models; a pragmatic crossover trial in Anesthesiology finds that automated intraoperative insulin dosing reminders do not improve postoperative glycemia and may be associated with more surgical site infections; and an Anesthesia & Analgesia mechanistic study functionally validate
Summary
Three impactful anesthesiology-related studies stood out today: a British Journal of Anaesthesia analysis shows that selection bias dramatically inflates the apparent performance of intraoperative hypotension prediction models; a pragmatic crossover trial in Anesthesiology finds that automated intraoperative insulin dosing reminders do not improve postoperative glycemia and may be associated with more surgical site infections; and an Anesthesia & Analgesia mechanistic study functionally validates a novel non-hotspot RYR1 mutation (p.Asp2730Tyr) linked to malignant hyperthermia.
Research Themes
- Validating and deploying AI/ML decision support in perioperative care
- Pragmatic trials of intraoperative decision support and glycemic management
- Genotype-to-function translation for malignant hyperthermia risk
Selected Articles
1. The effect of selection bias on the performance of a deep learning-based intraoperative hypotension prediction model using real-world samples from a publicly available database.
Using VitalDB, the authors show that selection bias in constructing test sets dramatically inflates the apparent performance of intraoperative hypotension prediction models. When evaluated on an unbiased sample, alarms increased and the positive predictive value of a deep learning model plummeted from 0.937 to 0.068 for 5-minute-ahead prediction.
Impact: This paper provides a rare, quantitative demonstration of how dataset curation can mislead model performance estimates, directly informing translational AI in anesthesiology.
Clinical Implications: Do not rely on published PPV/AUC without understanding sampling frames; require unbiased, prospectively collected evaluations before deploying hypotension prediction tools, and anticipate more frequent false alarms in real-world use.
Key Findings
- Unbiased testing increased alarm burden (18.1 vs 10.8 alarms/hour, P<0.001) for 5-minute-ahead prediction.
- Positive predictive value dropped from 0.937 (biased) to 0.068 (unbiased) for the deep learning model (P<0.001).
- Deep learning outperformed MAP-only statistically, but the difference was not clinically meaningful under unbiased testing.
Methodological Strengths
- Direct comparison of biased vs unbiased sampling frames using a public waveform dataset (VitalDB).
- Clear, clinically interpretable metrics (alarms/hour, PPV) at a defined prediction horizon.
Limitations
- Retrospective design using a single open dataset; not a prospective clinical deployment.
- Model calibration and thresholding may not generalize across institutions and monitoring practices.
Future Directions: Prospective, multi-center evaluations with unbiased sampling and workflow-integrated thresholds; benchmarking frameworks that penalize selection bias; and impact studies on clinician behavior and patient outcomes.
BACKGROUND: There are models to predict intraoperative hypotension from arterial pressure waveforms. Selection bias in datasets used for model development and validation could impact model performance. We aimed to evaluate how selection bias affects the predictive performance of a deep learning (DL)-based model and a model using only mean arterial pressure (MAP) as input (MAP-only model). METHODS: We used the VitalDB open dataset. A hypotensive event was defined as a MAP <65 mm Hg for 1 min. For the 'biased dataset', 'non-hypotensive events' needed to be (a) at the centre of a 'non-hypotensive period' with a MAP of >75 mm Hg for more than 30 continuous minutes and (b) at least 20 min apart from any hypotensive event. For the 'unbiased dataset', all samples were included unless the hypotensive event was already in the input segment. The alarms per hour and positive predictive values were compared between the DL and MAP-only models. RESULTS: The DL model generally performed better than the MAP-only model. For the prediction of intraoperative hypotension 5 min before the event with the DL model, using the unbiased vs the biased testing dataset resulted in 18.1 vs 10.8 alarms per hour (P<0.001) and a positive predictive value of 0.068 vs 0.937 (P<0.001). CONCLUSIONS: Both the DL model and the MAP-only model demonstrated worse predictive performance when tested on the unbiased dataset compared with the biased dataset. Although the DL model statistically performed better than the MAP-only model, the difference between the two models was not clinically meaningful. Clinicians should consider the potential impact of selection bias on the validation and the clinical performance of hypotension prediction models. CLINICAL TRIAL REGISTRATION: NCT02914444.
2. Perioperative Glucose Pragmatic (PROGRAM) Trial: Standardized Insulin Management in Surgical Patients.
In a large pragmatic crossover trial of 4,558 high-risk surgical cases, automated intraoperative insulin dosing reminders did not reduce early postoperative hyperglycemia compared with a glucose-check reminder. There were no differences in intraoperative glucose monitoring or insulin administration, while unadjusted surgical site infections were higher in the intervention arm.
Impact: This negative, pragmatic result directly informs perioperative informatics and quality improvement strategies, suggesting that simple nudges may be insufficient to improve glycemic control.
Clinical Implications: Do not expect automated insulin prompts alone to improve postoperative glycemia; consider protocolized insulin pathways, tighter closed-loop or team-based strategies, and monitor for unintended consequences such as potential infection risks.
Key Findings
- Primary outcome: no reduction in early postoperative hyperglycemia (OR 0.90; 95% CI 0.78–1.03; P=0.165).
- No differences in intraoperative glucose monitoring (OR 0.99) or insulin use (OR 1.00).
- Unadjusted odds of surgical site infection were higher with the dosing reminder (OR 2.52; 95% CI 1.37–4.64).
Methodological Strengths
- Pragmatic, sequential repeated crossover design embedded in routine care with >4,500 cases.
- Adjusted analyses accounting for demographics, surgery characteristics, and temporal effects.
Limitations
- Single-center provider-level crossover may have contamination or period effects; not patient-randomized.
- Infection signal based on unadjusted analyses; causal mechanisms not elucidated.
Future Directions: Test multi-component, closed-loop glycemic strategies in multi-center randomized designs; evaluate clinician workflow integration and potential impacts on infections and other outcomes.
BACKGROUND: Perioperative hyperglycemia is associated with adverse patient outcomes including surgical site infections. This study examined whether an automated insulin dosing reminder is associated with a lower risk for postoperative hyperglycemia and other secondary and safety outcomes in patients at high risk for intraoperative hyperglycemia. METHODS: The authors conducted a pragmatic trial using a sequential and repeated crossover design between October 5, 2022, and October 26, 2023. They sequentially assigned anesthesia providers to receive either an automated insulin dosing reminder (intervention) or a glucose check reminder (routine care) periodically throughout surgery for a consecutive sample of adult patients at high risk for intraoperative hyperglycemia undergoing major surgery at their quaternary medical center. The primary outcome was hyperglycemia (glucose greater than 180 mg/dl) at the first postoperative measurement 3 h or less postoperatively. The primary analysis studied the association between automated insulin dosing reminder and postoperative hyperglycemia adjusted for demographics, surgery characteristics, preoperative glucose, time period, and the interaction of intervention and time period. RESULTS: A total of 4,558 cases qualified for primary analysis: 2,611 cases in the routine care group and 1,947 cases in the intervention group. A total of 970 (37%) and 675 (35%) cases, respectively, experienced the primary outcome. The authors found no evidence of an association between treatment and postoperative hyperglycemia in the overall study period (odds ratio [OR], 0.90; 95% CI, 0.78 to 1.03; P = 0.165). There was no evidence of difference in intraoperative glucose monitoring (OR, 0.99; 95% CI, 0.83 to 1.19; P = 0.369) and intraoperative insulin use (OR, 1.00; 95% CI, 0.83 to 1.20; P = 0.995). The odds of surgical site infections were higher in the intervention group (overall unadjusted OR, 2.52; 95% CI, 1.37 to 4.64; P = 0.006). No difference in safety endpoints was observed between groups. CONCLUSIONS: Among surgical patients at high risk of intraoperative hyperglycemia, an automated insulin dosing reminder did not improve glycemic control or other outcomes compared with a glucose check reminder.
3. Beyond Hotspots: Functional Characterization of the Novel p.Asp2730Tyr Mutation in RYR1 Associated With Malignant Hyperthermia.
A newly identified non-hotspot RYR1 variant (p.Asp2730Tyr) increased sensitivity to caffeine and 4CmC in heterologous expression assays, with EC50 values lower than wild-type and comparable to a known pathogenic control. This functional evidence supports pathogenicity and expands the spectrum of MH-relevant RYR1 regions.
Impact: The study offers functional validation of a non-hotspot RYR1 variant, directly informing variant interpretation and MH risk stratification in anesthetic practice.
Clinical Implications: Consider non-hotspot RYR1 variants as potentially pathogenic when supported by functional assays; incorporate such evidence into MH counseling, genetic testing reports, and perioperative trigger avoidance strategies.
Key Findings
- p.Asp2730Tyr and p.Arg2508His each caused a leftward shift in caffeine and 4CmC dose-response curves (P<.001).
- Caffeine EC50: WT 2.56±0.04 mM vs p.Asp2730Tyr 1.12±0.09 mM (P<.001).
- 4CmC EC50: WT 43.2±1.90 μM vs p.Asp2730Tyr 21.8±1.04 μM (P<.001), indicating enhanced Ca2+ release.
Methodological Strengths
- Full-length RYR1 expression with a known pathogenic control mutation for benchmarking.
- Quantitative calcium imaging with EC50 estimation for two independent agonists (caffeine, 4CmC).
Limitations
- Heterologous expression in 293T cells may not fully recapitulate skeletal muscle physiology.
- Small number of patients; no in vivo or contracture testing data presented.
Future Directions: Correlate variant function with clinical contracture testing and anesthesia phenotypes; structural modeling and in vivo models to define mechanisms and guide ACMG classification.
BACKGROUND: Malignant hyperthermia (MH) is a life-threatening pharmacogenetic disorder triggered by certain anesthetics, characterized by muscle rigidity, elevated body temperature, and hypermetabolic crisis. This condition is primarily associated with genetic mutations in ryanodine receptor 1 (RYR1), which encodes the pivotal calcium release channel in the sarcoplasmic reticulum of skeletal muscle. While numerous hotspot mutations in RYR1 have been identified, the functional impact of nonhotspot mutations on channel activity related to MH remains insufficiently investigated. In this study, we identified a known pathogenic mutation (p.Arg2508His) and a novel variant (p.Asp2730Tyr), both located outside the conventional MH hotspots, in 2 patients with clinical suspicion of MH. Our objective was to investigate the functional implications of the p.Asp2730Tyr mutation in RYR1 on calcium release dynamics related to MH. METHODS: We engineered a recombinant wild-type (WT) plasmid (pcDNA3.1-3Myc-His-RYR1-WT) to express the full-length mouse skeletal muscle RYR1 using seamless multi-fragment cloning techniques. Two RYR1 mutations, p.Arg2508His (used as positive control) and p.Asp2730Tyr, were separately introduced into the WT plasmid, generating 2 mutant constructs (pcDNA3.1-3Myc-His-RYR1-p.Arg2508His and pcDNA3.1-3Myc-His-RYR1-p.Asp2730Tyr). We utilized 293T cells expression system to express either the WT or mutant forms of mouse RYR1. Fluo-4 calcium imaging was conducted to evaluate the alterations in calcium release in response to RYR1 agonists, caffeine or 4-chloro-m-cresol (4CmC), for each mutation compared to WT. RESULTS: Cells transfected with the p.Arg2508His or p.Asp2730Tyr mutation demonstrated a leftward shift in the caffeine and 4CmC concentration-response curves compared to WT, suggesting an increased channel sensitivity to caffeine and 4CmC (P < .001). The mean ± standard error of the mean (SEM) of the EC50 values for caffeine-induced calcium release was 2.56 ± 0.04 mM in WT, which significantly decreased to 1.32 ± 0.13 mM for p.Arg2508His (P < .001) and 1.12 ± 0.09 mM for p.Asp2730Tyr (P < .001). For 4CmC, the EC50 values were 43.2 ± 1.90 μM in WT, 17.2 ± 0.76 μM for p.Arg2508His (P < .001), and 21.8 ± 1.04 μM for p.Asp2730Tyr (P < .001), indicating enhanced calcium release in both mutations. CONCLUSIONS: The p.Asp2730Tyr mutation, situated beyond the established RYR1 hotspot regions, significantly alters calcium release dynamics related to MH. A comprehensive investigation into the structural conformations, functional assays, and in vivo mechanisms associated with this mutation could yield a more profound understanding of the molecular underpinnings of MH pathogenesis.