Predicting the Failure Risk of Internal Fixation Devices in Chinese Patients Undergoing Spinal Internal Fixation Surgery: Development and Assessment of a New Predictive Nomogram

Abstract

The current study is aimed at developing and validating a nomogram of the risk of failure of internal fixation devices in Chinese patients undergoing spinal internal fixation. We collected data from a total of 1139 patients admitted for spinal internal fixation surgery at the First Affiliated Hospital of Guangxi Medical University from May 2012 to February 2019. Of these, 1050 patients were included in the spinal internal fixation group and 89 patients in the spinal internal fixation device failure group. Patients were divided into training and validation tests. The risk assessment of the failure of the spinal internal fixation device used 14 characteristics. In the training test, the feature selection of the failure model of the spinal internal fixation device was optimized using the least absolute shrinkage and selection operator (LASSO) regression model. Based on the characteristics selected in the LASSO regression model, multivariate logistic regression analysis was used for constructing the model. Identification, calibration, and clinical usefulness of predictive models were assessed using C-index, calibration curve, and decision curve analysis. A validation test was used to validate the constructed model. In the training test, the risk prediction nomogram included gender, age, presence or absence of scoliosis, and unilateral or bilateral fixation. The model demonstrated moderate predictive power with a C-index of 0.722 (95% confidence interval: 0.644-0.800) and the area under the curve (AUC) of 0.722. Decision curve analysis depicted that the failure risk nomogram was clinically useful when the probability threshold for internal fixation device failure was 3%. The C-index of the validation test was 0.761. This novel nomogram of failure risk for spinal instrumentation includes gender, age, presence or absence of scoliosis, and unilateral or bilateral fixation. It can be used for evaluating the risk of instrumentation failure in patients undergoing spinal instrumentation surgery.

Figure 1

Typical cases of internal fixation failure: (a) screw fracture, (b) broken rod, (c) screw cap loosening, (d) screw rod connection loosening, (e) cross rod loosening, (f) pulling nail, (g) titanium plate nail loosening, (h) titanium plate loosening, (i) titanium cage loosening displacement, (j) neck pillow fusion nail loosening, (k) Harrington rod fracture, and (l) Harrington rod wire fracture.

Figure 2

Feature selection using the LASSO binary logistic regression model. (a) Feature selection by the LASSO binary logistic regression model. By verifying the optimal parameter (lambda) in the LASSO model, the partial likelihood deviance (binomial deviance) curve was plotted versus log (lambda). Dotted vertical lines were drawn based on 1 SE of the minimum criteria (the 1-SE criteria). (b) Feature selection by the LASSO binary logistic regression model. A coefficient profile plot was produced against the log (lambda) sequence in Figure 2(a). Four features with nonzero coefficients were selected by optimal lambda. LASSO: least absolute shrinkage and selection operator; SE: standard error.

Figure 3

Construction of a nomogram for the failure of spinal internal fixation devices.

Figure 4

Calibration curves for predicting the risk profile of spinal instrumentation failure in the cohort. The x-axis represents the predicted risk of spinal instrumentation failure. The y-axis represents the actual diagnosis of spinal instrumentation failure. The diagonal dashed line represents an ideal perfect prediction model. Solid lines represent the performance of the nomogram, where closer proximity to diagonally dashed lines represents better prediction.

Figure 5

The AUC for training and validation tests. (a) Training test. (b) Validation test. AUC: area under curve.

Figure 6

Decision curve analysis of failure risk nomogram of spinal internal fixation device. The decision curve showed that if the threshold probability of a patient and a doctor is >3% and<72%, respectively, using this failure risk nomogram in the current study to predict failure risk adds more benefit than the intervention-all-patients scheme or the intervention-none scheme.