In Cureus
BACKGROUND : Difficult and traumatic neuraxial blocks and procedures are not uncommon. Although score-based prediction has been attempted, the practical application of those has remained limited for various reasons. The aim of this study was to develop a clinical scoring system from the strong predictors of failed spinal-arachnoid puncture procedures assessed previously using artificial neural network (ANN) analysis and analyze the score's performance on the index cohort.
METHODS : The present study is based on the ANN model analyzing 300 spinal-arachnoid punctures (index cohort) performed in an academic institute in India. For the development of the score, i.e., Difficult Spinal-Arachnoid Puncture (DSP) Score, the coefficient estimates of the input variables, which showed a Pr(>|z|) value of <0.01, were considered. The resultant DSP Score was then applied to the index cohort for receiver operating characteristic (ROC) analysis, Youden's J point determination for best sensitivity and specificity, and diagnostic statistical analysis for the cut-off value for predicting the difficulty.
RESULTS : A DSP Score incorporating spine grades, performers' experience, and positioning difficulty was developed; the minimum and maximum scores were 0 and 7, respectively. The area under the ROC curve for the DSP Score was 0.858 (95% confidence interval 0.811-0.905), Youden's J point for cut-off was at 2, which showed a specificity and sensitivity of 98.15% and 56.5%, respectively.
CONCLUSION : The ANN model-based DSP Score developed for predicting the difficult spinal-arachnoid puncture procedure showed an excellent area under the ROC curve. At the cut-off value 2, the score had a sensitivity plus specificity of approximately 155%, indicating that the tool can be useful as a diagnostic (predictive) tool in clinical practice.
Karim Habib Md R
2023-Jan
interpretable machine learning, lumbar puncture (lp), orthopedic anesthesia, regional anesthesiology, scores, spinal deformiites, spinal puncture
