Preoperative prediction of postoperative urinary retention in lumbar surgery: a comparison of regression to multilayer neural network

Abstract

Objective: Postoperative urinary retention (POUR) is a common complication after spine surgery and is associated with prolongation of hospital stay, increased hospital cost, increased rate of urinary tract infection, bladder overdistention, and autonomic dysregulation. POUR incidence following spine surgery ranges between 5.6% and 38%; no reliable prediction tool to identify those at higher risk is available, and that constitutes an important gap in the literature. The objective of this study was to develop and validate a preoperative risk model to predict the occurrence of POUR following routine elective spine surgery.

Methods: The authors conducted a retrospective chart review of consecutive adults who underwent lumbar spine surgery between June 1, 2017, and June 1, 2019. Patient characteristics, preexisting ICD-10 codes, preoperative pain and opioid use, preoperative alpha-1 blocker use, details of surgical planning, development of POUR, and management strategies were abstracted from electronic medical records. A binomial logistic model and a multilayer perceptron (MLP) were optimized using training and validation sets. The models' performance was then evaluated on model-naïve patients (not a part of either cohort). The models were then stacked to take advantage of each model's strengths and to avoid their weaknesses. Four additional models were developed from previously published models adjusted to include only relevant factors (i.e., factors known preoperatively and applied to the lumbar spine).

Results: Overall, 891 patients were included in the cohort, with a mean of 59.6 ± 15.5 years of age, 52.7% male, BMI 30.4 ± 6.4, American Society of Anesthesiologists class 2.8 ± 0.6, and a mean of 5.6 ± 5.7 comorbidities. The rate of POUR was found to be 25.9%. The two models were comparable, with an area under the curve (AUC) of 0.737 for the regression model and 0.735 for the neural network. By combining the two models, an AUC of 0.753 was achieved. With a regression model probability cutoff of 0.24 and a neural network cutoff of 0.23, maximal sensitivity and specificity were achieved, with specificity 68.2%, sensitivity 72.9%, negative predictive value 88.2%, and positive predictive value 43.4%. Both models individually outperformed previously published models (AUC 0.516-0.645) when applied to the current data set.

Conclusions: This predictive model can be a powerful preoperative tool in predicting patients who will be likely to develop POUR. By using a combination of regression and neural network modeling, good sensitivity, specificity, and NPV are achieved.

Keywords: lumbar surgery; postoperative complications; risk factors; urinary catheterization; urinary retention.

FIG. 1.

Bar graph of the differences in rates of POUR based on patient demographics and medication use of the overall cohort (n = 891). Frequencies (N) and p values comparing those who did and did not develop urinary retention are listed in the label. Red bars = included in binomial logistic regression and neural network models. Blue bars = only included in neural network model. Figure is available in color online only.

FIG. 2.

Bar graph of the differences in rates of POUR based on planned surgical characteristics of the overall cohort (n = 891). Frequencies of the surgery type (N) and p values comparing those who did and did not develop urinary retention are listed in the label. Red bars = included in binomial logistic regression and neural network models. Blue bars = only included in neural network model. Figure is available in color online only.

FIG. 3.

Receiver operating characteristic curves for patients comparing regression model (dashed blue line), neural network model (dashed-dotted green line), and the stacked model (solid yellow line) combining the two. Figure is available in color online only.

FIG. 4.

Scatterplot for the testing set (N = 235) of the predicted probabilities of the regression model (y-axis) compared to the neural network model (x-axis) demonstrating (A) optimal cutoff points of 0.43 and 0.54, respectively, and (B) optimal cutoff points of 0.23 and 0.24, respectively, for combining the two models. By combining the two models using an “and” conjunction, the upper right quadrant (red shaded area) represents a positive test result. Thus, a green diamond in this area represents a true positive, whereas a blue circle in this area represents a false positive. A green or blue marker in the other quadrants represents a false negative or a true positive, respectively. Figure is available in color online only.