Prediction of Unexpected Emergency Room Visit after Extracorporeal Shock Wave Lithotripsy for Urolithiasis - an Application of Artificial Neural Network in Hospital Information System

Abstract

Extracorporeal shock wave lithotripsy (ESWL) for urolithiasis was developed for more than 30 years. It benefited most patients suffering from acute renal colic. The ESWL may be performed at outpatient based in most hospital in Taiwan. But the post-ESWL emergency room (ER) visits will be a painful experience for patient and the urologist, especially those patients visited ER immediately on the same day of ESWL. Though most guidelines for ESWL suggest the larger stone burden, the higher risk for post-ESWL ER visits, there are about 10% patients will come back to ER due to renal colic post-operatively. We use artificial neural network (ANN) to predict the post-ESWL ER visit for patient with urolithiasis. The result disclosed high sensitivity and specificity of prediction. In conclusion, it will decrease the rate of post-ER visit rate and patients’ suffer by using ANN to predict the post-ESWL ER visits.

Background

Though there are many benefits of ESWL for patients with urolithiasis, including minimal invasive, less pain, effective for smaller urolithiasis, and could be performed at outpatient based service^1,2, a few patients still suffered from renal colic, fever or other side effects that will let them come back to ER for help. It will increase suffering of patients and cost of medical care. Traditional guidelines only suggest avoiding the procedure in patients with large stone burden. But there are around 10% patients will come back due to post-ESWL renal colic even though they were treated within the suggested stone burden. How to decrease the rate of post-treatment ER visit would be an important issue for quality of care in patients with urolithiasis.

Methodology

From November 8, 2000 to April 4, 2006, all 1026 patients received ESWL for urolithiasis at Buddhist Tzu Chi General Hospital, Dalin branch, were included in the study as training set. All the dates of patients’ pre-ESWL and post-ESWL who visited emergency room due to renal colic were recorded. The patients’ characteristics, stone location, stone burden, stone shape (long to short axis ratio), pre-ESWL associated procedure and cost of admission, ESWL & ER visit were analyzed. All 506 patients who received ESWL during 2005/8/1~2006/4/28 at Buddhist Tzu Chi General Hospital, Taipei branch, were included as external validation set. We us ANN multilayer perceptron (MLP) model in the study. The outcome was presented as area under receiver operator characteristic curve (AUROC).

Results

Those possible encountered conditions were depicted as figure 1. The unexpected post-ESWL visits rate within one week are 13.6% and 29.8% for training set and external validation set respectively. The pre-ESWL ER visits, stone shape and stone burden are the important predictor of post-ESWL unexpected visits. The AUROC for training set, and external validation set are 0.87 and 0.66.

Figure 1.

All possible pre-ESWL& post-ESWL treatment for patients with urolithiasis & renal colic.

Discussion

The results showed high sensitivity and specificity of prediction of post-ESWL ER visit for patients with urolithiasis. We also found out the predictors: pre-ESWL ER visits and stone shape (long to short axis of stone), which was proposed by previous literature, are the important variable for outcome prediction. The developed model could decrease the patients’ post-ESWL suffering, expenditure of medical care and the patient-doctor relationship.

Figure 2.

Feature selection variable screening for training set.

Figure 3.

Area under receiver operator characteristic curve for training set, testing set and external validation set.

Figure 4.

Post-ER ER visit prediction using ANN.