A prediction model for thrombocytopenia after neurosurgery: a retrospective study
Chenlu Zhang1, Shoumei Jia2, Yuli Zang3
1Affiliated Dongyang Hospital of Wenzhou Medical University, Jinhua, Zhejiang Province, China.
Objective:
Thrombocytopenia frequently occurs after major surgery and is linked to negative clinical outcomes. This study aimed to predict thrombocytopenia occurrence in patients following neurosurgery using a logistic regression model.
Methods:
We retrospectively analyzed 1,109 postoperative patients who had neurosurgery between January 2010 and December 2020 at the Dong Yang People's Hospital. We obtained medical records, encompassing demographic details, clinical outcomes, and laboratory results, from the hospital's database. The variables included in the model were identified using least absolute shrinkage and selection operator (LASSO) regression. Data from the research participants were split into training and test sets. The training dataset was employed for constructing a logistic regression model, while the test dataset was utilized to test the model's performance. Additionally, we performed subgroup analysis by etiology.
Results:
Among 1,109 patients, 103 developed thrombocytopenia (9.3%). Patients with thrombocytopenia had a longer hospital stay and mechanical ventilation time than those without. The eight predictive variables selected through LASSO regression for modeling were: hypertension, Sepsis-related Organ Failure Assessment (SOFA) score, serum total bilirubin and albumin, International Normalized Ratio, thrombocytocrit, first measured systolic blood pressure and use of vasoactive drugs. The logistic regression model demonstrated satisfactory discriminative ability, showing an area under the curve of 0.916 (95% confidence interval [0.886-0.947]) for the training set and 0.883 [0.817-0.950] for the test set, coupled with high specificity (>0.9) but low recall (0.414 training, 0.364 test) at a cutoff value of 0.5. Calibration curves also indicated good predictive accuracy for both sets. Subgroup analysis revealed that the model for acute brain injury achieved a robust optimism-corrected AUC of 0.905, with precision improving from 0.545 to 0.664.
Conclusion:
The logistic regression models from this study have the potential to predict thrombocytopenia following neurosurgery and can serve as clinical decision-support tools for early intervention.
