Early warning model construction and validation for urinary tract infection in patients with neurogenic lower urinary tract dysfunction (NLUTD): a retrospective study

Participants

Eligible patients with NLUTD admitted to Shenzhen Longcheng Hospital from December 2017 to December 2020 were retrospectively selected as construction and internal validation research subjects, and patients from January 2021 to June 2021 were selected as external validation research objects. The inclusion criterion was patients first diagnosed with NLUTD (ICD-10 code N31.901), while the exclusion criteria were (1) UTI co-existing with pulmonary infection, wound infection, etc., namely secondary UTI; (2) cannot obtain the required medical records, such as lack of blood routine, renal function, urological imaging data or urodynamic; (3) patients did not complete the 6-month follow-up.

Data collection

We collected the following predictors half-month after patients diagnosed with NLUTD: (1) General clinical data: including gender, age, body mass index (BMI), diabetes, hypertension, urinary calculi, and so on (2) Disease characteristics and bladder treatment-related data: including NLUTD type based on clinical manifestation (detrusor overactivity type, detrusor inactivity type and mixed type), catheterization method (voiding/intermittent catheterization and indwelling catheterization), preventive use of antibacterial drugs, time to start the bladder behavior training, serum albumin, serum creatinine, total serum bilirubin and blood urea nitrogen. Time to start the bladder behavior training <2 weeks means that the training started within two weeks of the first diagnosis of NLUTD. The bladder behaviour training included scheduled urination, delayed urination, reflex urination, anal stretch techniques, and pelvic floor muscle training. Patients with indwelling catheterization only performed anal stretch techniques and pelvic floor muscle training.

The end point outcome observation time is at six-month follow-up. All patients underwent routine urinalysis and urine culture tests. The clinical diagnosis of UTI met the relevant standards in the “Guidelines on urological infections” formulated by the European Association of Urology (EAU Guidelines, 2022). Clinical symptoms of UTI include urgent urinary, frequent urination, urinary retention, dysuria, lumbago, and urination pain. Physical signs included fever, costovertebral angle tenderness/pain and suprapubic tenderness. Patients with an axillary temperature of >37.5 were considered febrile. According to the above clinical manifestations, this study assesses UTI with “yes” if one of the following conditions is met and with one of above clinical manifestations: (1) Diagnosis of UTI in medical records; (2) Positive urine culture; (3) Fresh urine specimens were centrifuged and examined by phase contrast microscopy (400 ×). Bacteria were observed in half of every 30 visual fields.

Ethical approval

This study was approved by the Medical Ethics Committee of Shenzhen Longcheng Hospital (LCLL(3)–2021). Since this study is a retrospective study and there is no medical risk to the patients, the patient’s informed consent was exempted with the approval of the Medical Ethics Committee.

Statistical analysis

The data was analyzed by SPSS 22.0 software and R statistics version 3.6.3, utilizing the ROCR package for all analyses. The count data of patients’ age, BMI and diabetes were described by frequency and percentage. The comparison between UTI and non-UTI groups was performed by univariate regression analysis. P < 0.05 indicated that the difference was statistically significant. The logistic regression model was used to construct the risk prediction model. The prediction ability of the model was evaluated by the AUROC (Are Under Receiver Operating Characteristic) curve, C-index and the calibration curve. Bootstrapping with 1,000 resamples was done to calculate a relatively accurate C-index for internal cross-validation (Steyerberg, 2019; Van Calster et al., 2016; Van Calster & Vickers, 2015). The external validation was conducted by using sensitivity, specificity and accuracy (Steyerberg, 2019).

Univariate regression analysis of NLUTD patients between groups

The clinical data of patients with UTI and those without UTI was analyzed by univariate analysis. The results showed that the proportion of ≥65 years old, complicated with diabetes, urinary calculi, hypoproteinemia and female patients in the UTI group was significantly higher than that of the non-UTI group. The proportion of patients with intermittent catheterization and time to start bladder training <2 weeks was significantly lower in the UTI group than in the non-UTI group (P < 0.05). See Table 1.

Multivariate regression analysis among NLUTD patients with UTI

The statistically significant factors in the univariate analysis were used as independent variables, and whether the patient had a UTI during the follow-up period was used as the dependent variable, and binary logistic regression analysis was performed (see Table 2). The results showed that under the premise of controlling for other confounding factors, patients age ≥65 (OR = 2.478, 95%CI [1.215–5.050]), female patients (OR = 2.552, 95%CI [1.286–5.065]), diabetes (OR = 2.364, 95% CI [1.182–4.731]), combined urinary calculi (OR = 2.948, 95% CI [1.387–6.265]), indwelling catheterization embodiment (OR = 1.988, 95% CI [1.003–3.940]) and bladder behavior training intervention time ≥ 2 weeks (OR = 2.489, 95%CI [1.233–5.022]) were independent risk factors for UTI in patients with NLUTD (P < 0.05). The final formula was Y = 0. 907 × age + 0. 937 × female gender + 0. 860 × diabetic + 1. 081 × urinary calculi + 0. 687 × indwelling catheterization + 0. 912 ×Time to start bladder behavior training ≥ 2 weeks + 0. 547 × hypoproteinemia −2.570. See Table 2 for details. Y means the calculated UTI risk prediction score.

Internal evaluation of the early warning model

The AUROC of the model was evaluated, and the initial C-index was 0.832 (see Fig. 2). The Youden index was 0.82 + 0.59 − 1 = 0.41. The C-index of 1,000 Bootstrap internal validations was 0.828 (see Fig. 3). The calibration curve showed that the predicted value of the model and the observed value had a certain correlation (see Fig. 4), indicating that the model had a higher probability of predicting the occurrence of UTI and the actual probability of occurrence.

External validation of the early warning model

Thirty eligible patients with NLUTD treated in Shenzhen Longcheng hospital from January to June 2021 were included for external validation. There were 19 males and 11 females. The age ranged from 20 to 86 (51.72 ± 13.51). According to the formula of this prediction model, when calculated UTI prediction score ≥ 0.41, it is considered that the patient will occur UTI. This model predicts that 5 patients will occur UTI and 25 patients will not occur UTI within 6-month follow-up. The results were that UTI occurred in 8 patients and non-UTI occurred in 22 patients. Compared with the actual results, the sensitivity of the prediction model is 62.5%, the specificity is 100%, and the accuracy is 90%.

The research significance of the early warning model

UTI is one of the common complications of NLUTD. Because of a variety of complex factors, it can often manifest as recurrent episodes, which not only aggravates the clinical symptoms of NLUTD, but also affects the renal parenchyma damage, and ultimately lead to a decline in the quality of life of patients and shortened life expectancy (Van Calster & Vickers, 2015). Therefore, it must be actively prevented and controlled.

Identification of clinical risk factors for UTI is previously based on experts’ opinions without using contemporary modelling techniques, which is not helpful to accurately classify patients, leading to insufficient prevention and treatment of UTI (Chow, Gallo & Busse, 2018). The establishment of the predictive model shifts the focus from interventions when UTI occurs to the purposeful and focused prevention of UTI in people with high-risk NLUTD. This early warning model intuitively displays the relevant risk factors and their correlation with the occurrence of UTI, which could arouse the attention of medical staff, improve prevention, and finally reduce the workload of medical staff.

The prediction effect of the early warning model

In this study, the ROC curve was used to test the model’s fit, with specificity as the abscissa and sensitivity as the ordinate. The area under the ROC curve obtained in this study is 0.832, and the area under the ROC curve verified internally is 0.828. The optimal value of the area under the ROC curve is 1. Moreover, the results of the calibration chart in this study are better, and its slope is basically the same as the standard curve. The external validation results also showed a specificity of 100% and an accuracy of 90%.

Elderly female patients are prone to UTI

Elderly patients are prone to infections due to decreased estrogen levels, broad and short urinary tract, poor nutritional status, and reduced immunity. Studies have shown that with age, the probability of occurrence of UTI increases significantly because female patients have a broad and short urinary tract, which is close to the anus, leading to bacterial colonization and retrograde infection (Kakde, Redkar & Yelale, 2018). Especially in elderly menopausal female patients, due to the decline of ovarian function, the level of estrogen decreases significantly, which eventually leads to the degeneration of the reproductive system, a decrease in blood supply to the urinary tract, and a reduction in muscle tone that maintains the bladder and urethra (Jung & Brubaker, 2019).

Patients with diabetes and urinary calculi are prone to UTI

Many studies have confirmed that diabetes is closely related to UTI, and the incidence of UTI has positively correlated with the course of diabetes, blood sugar levels and glycosylated haemoglobin (He et al., 2018; Zubair et al., 2019). The reduction of substance metabolism in diabetic patients reduces the body’s synthesis of immunoglobulins, antibodies and complement, thus minimizing the chemotaxis, phagocytosis and sterilization of inflammatory cytokines. Therefore, the body’s cellular immune barrier is destroyed, causing the drop of the immunity system. In addition, patients with a long course of diabetes may be accompanied by peripheral nerve and vascular disease, which may cause urinary tract blood supply disorders and lead to local neurological abnormalities, degrading the genitourinary system and causing infection (Hine et al., 2017). Studies have shown that urinary calculi are closely related to the occurrence of UTI, and patients with urinary calculi (diameter of 2 cm or more) are at higher risk of recurrent UTI (Wang & Li, 2020). The main reason is that bacteria tend to colonize on the irregular surface of urinary calculi, which prevents the antibacterial drugs from reaching the site of action, resulting in poor effects of antibacterial treatment.

Patients who use indwelling catheterization are prone to UTI

Intermittent urinary catheterization is strongly recommended for patients with NLUTD, which is suitable for reducing the incidence of UTI. Still, the indwelling catheterization method had to be used for some severely ill patients (extreme weakness or ureteral reflux patients) (Han et al., 2017). However, as a foreign body, the urinary catheter can cause reactive inflammation when it comes in contact with the urethral mucosa and bladder mucosa. The urethral mucosa is damaged during the catheter operation, which further reduces the urethra physical barrier function and may cause infection after bacteria enter the urethra retrogradely from the urethral orifice to outside the official cavity (Menegueti et al., 2019). For patients with indwelling catheterization, patients should use the closed drainage catheter system (Nicolle et al., 2005), and indwelling catheters should not be replaced systematically but in the event of an obstruction, hematuria, or UTI (Hooton et al., 2010). UTI may be reduced by early anal stretch techniques and pelvic floor muscle training, which could help patientsimprove bladder function recovery, and reduce bladder residual urine output (Ginsberg, 2013).

The correlation between hypoproteinemia and UTI remains to be verified

The study showed that the decrease of serum albumin levels in patients with NLUTD is related to UTI. The reason may be the decreased immunity of patients with hypoalbuminemia (Jiao, 2020). In this study, univariate analysis showed statistical significance between groups (P < 0.05). However, regression analysis showed no statistical difference between hypoproteinemia and UTI. The main reason for this difference may be that hypoalbuminemia in this study mainly occurred in elderly patients. The hypoalbuminemia may be excluded as a confounding factor of this model. In addition, It may also be due to the small sample size of patients with hypoproteinemia (n = 39). Future studies are needed to explore whether hypoproteinemia could be used to predict UTI among patients with NLUTD.

Strengths and limitations of this research

This study attempts to develop and validate an early warning model of UTI among patients with NLUTD compared with previous studies. This model has the characteristics of simple data acquisition and quick evaluation, which can provide purposeful and focused prevention of UTI in people with NLUTD (Autumn et al., 2018). The limitation of this study lies in that it was only verified in one hospital. This study suggests that it still needs to be verified by a large sample, multi-centre, and prospective study before the model is routinely applied to pre-clinical applications. To develop a model that was easy to use in clinical settings, we dichotomized the continuous age variable into a binary variable. This may lose more information on the variable itself.