Establishment of a new predictive model for the recurrence of upper urinary tract stones

To construct a nomogram for evaluation of the recurrence risk of upper urinary tract stones in patients. We retrospectively reviewed the clinical data of 657 patients with upper urinary tract stones and divided them into stone recurrence group and non-recurrence group. Blood routine, urine routine, biochemical, and urological CT examinations were searched from the electronic medical record, relevant clinical data were collected, including age, BMI, stones number and location, maximum diameter, hyperglycemia, hypertension, and relevant blood and urine parameters. The Wilcoxon rank-sum test, independent sample t test, and Chi-square test were used to preliminarily analyze the data of the two groups, then LASSO and logistic regression analysis were used to find out the significant difference indicators. Finally, R software was used to draw a nomogram to construct the model, and ROC curve was drawn to evaluate the sensitivity and specificity. The results showed that multiple stones (OR: 1.832, 95% CI 1.240–2.706), bilateral stones (OR: 1.779, 95% CI 1.226–2.582), kidney stones (OR: 3.268, 95% CI 1.638–6.518), and kidney ureteral stones (OR: 3.375, 95% CI 1.649–6.906) were high risk factors. And the stone recurrence risk was positively correlated with creatinine (OR: 1.012, 95% CI 1.006–1.018), urine pH (OR: 1.967, 95% CI 1.343–2.883), Apo B (OR: 4.189, 95% CI 1.985–8.841) and negatively correlated with serum phosphorus (OR: 0.282, 95% CI 0.109–0.728). In addition, the sensitivity and specificity of the prediction model were 73.08% and 61.25%, diagnosis values were greater than any single variable. The nomogram model can effectively evaluate the recurrence risk of upper urinary stones, especially suitable for stone postoperative patients, to help reduce the possibility of postoperative stone recurrence.


Introduction
Urinary tract stones is one of the most common urological diseases, and its incidence is closely related to the region and increases year by year, respectively, 7-13% in North America, 5-9% in Europe, 1-5% in Asia, and 6.3% in China [1][2][3]. Urinary stones are divided into upper urinary tract stones and lower urinary tract stones. Upper urinary tract stones refer to kidney stones and ureteral stones, while lower urinary tract stones are bladder stones and urethral stones. Although upper urinary tract stones is a benign disease, it can lead to ureteral obstruction and hydronephrosis, and cause renal colic, hematuria, urinary tract infection, and other discomforts [4]. If appropriate treatment is not taken, serious cases can cause permanent damage to the kidneys [5]. The treatment of upper urinary tract stones includes etiological treatment, medication, and surgery. The diversification of treatment methods increases the stone clearance rate and protects the health of patients, but also brings a large economic burden to patients. It is estimated that it will cost the United States $4.1 billion by 2030 for urolithiasis treatment [6].
Kaiguo Xia, Yuexian Xu, and Qiao Qi contributed equally to this work.
Upper urinary tract stones is usually a lifelong disease that tends to recur, and the recurrence rate was 30%-50% in 5 years for untreated kidney stones [7,8]. The recurrence of upper urinary tract stones is related to the composition and morphology of stones, and cystine stones and uric acid stones are more likely to recur compared with calcium oxalate stones [9,10]. In addition, some research suggests that family history, ethnicity, smoking, diabetes, high triglycerides, age, gender, and obesity are also associated with the recurrence of urolithiasis [11][12][13][14]. Despite recent advances in stone treatment, however, improvements in treatment have not reduced stone recurrence rates and most patients may experience more recurrences [15]. Recurrence means that once again experiencing the pain of the disease and retreatment, not only increases the physical pain of patients, but also increases the medical burden. Therefore, there is no doubt that preventing stone recurrence as important as treating it.
It has been reported that appropriate dietary and pharmacological interventions can help prevent the formation and growth of upper urinary tract stones [16,17]. Given the high recurrence rate of upper urinary tract stones and the fact that effective pharmacological and dietary interventions can appropriately reduce the recurrence rate, it is important to predict and assess in advance those at high risk of stone recurrence and to provide relevant dietary and pharmacological interventions, which can significantly decrease the stone recurrence rate. Therefore, developing a model to predict upper urinary tract stones recurrence is important for focused health guidance of patients with a high risk of stone recurrence. Then we retrospectively studied the clinical data of 657 patients with upper urinary tract stones who were hospitalized in our department from January 2016 to June 2022, classified and collected relevant parameters, selected those with significant differences to construct a nomogram model for predicting the risk of stone recurrence, and explored its predictive value.

Study design and data acquisition
After the Ethics Committee of the First Affiliated Hospital of Anhui Medical University approval (Number: Quick-PJ2022-14-32), we retrospectively analyzed the clinical data of 657 patients with upper urinary tract stones who underwent surgical treatment in the department of urology, the First Affiliated Hospital of Anhui Medical University from January 2016 to June 2022. Blood routine, urine routine, biochemical, urological CT examinations, previous history of stones, and other relevant materials were searched from the electronic medical record. We reviewed the clinical data of patients and evaluated the clinical symptoms, previous history of stones, and imaging data to determine whether the stones were recurrent. Clinical parameters such as age, BMI, stones number and location, stone maximum diameter, hyperglycemia, hypertension, serum urea, serum creatinine (Cr), serum uric acid (UA), serum cystatin C (CysC), serum sodium (Na + ), serum potassium (K + ), serum chlorine (Cl − ), serum calcium (Ca 2+ ), serum phosphorus (P), serum magnesium (Mg 2+ ), urine protein, urine pH, urine red blood cells (urine RBC), urine white blood cells (urine WBC), hemoglobin (Hb), platelet (PLT), cholesterol (TC), triglycerides (TG), apo A, apo B, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and other relevant lipid parameters were collected, and the relationship with stone recurrence was analyzed.
Inclusion criteria: ① adults, age > 18 years old; ② patients with kidney or ureteral stones; exclusion criteria: ① patients with severe heart disease, cerebrovascular disease, renal failure, hyperthyroidism, congenital malformations of the urinary system and pulmonary disease; ② patients with malignant tumor disease; ③ Pregnant women and patients under 18 years old; ④ patients with incomplete case data.

Definition of upper urinary tract stones recurrence
Upper urinary tract stones recurrence is defined as the presence of a new stone at the original site or other positions after the first stone was cleared (as confirmed by KUB or urinary CT) in a patient with a previous history of stones. All patients were followed for 6 years, and some patients may present with renal colic and hematuria, while others are not, and imaging data may indicate the formation of new stones. Both postoperative residual stones and progressive stone growth were not defined as recurrence.

LASSO and logistic regression analysis
We performed LASSO regression by the "glmnet" package in R 4.2.1 software for 657 subjects to preliminarily select clinical parameters without collinearity, to avoid overfitting or overcomplicating the model [18]. Then univariate logistic analysis was performed separately to screen out parameters with statistical differences. The variables obtained from the univariate logistic analysis screening were further performed the multivariate logistic regression analysis [19].

Construction and assessment of nomogram
We created the nomogram by R 4.2.1 software according to the contribution of statistically different variables in multiple logistic regression analysis [20]. After the nomogram was plotted, the consistency C-index of the predictive model was calculated and then the nomogram was assessed by the graph calibration method [21].

Statistical analysis
Statistical analysis was performed using the R software (Version 4.2.1; https:// cran.r-proje ct. org/ bin/ windo ws/ base/) and SPSS 21.0. For measurement data, we used the mean ± standard deviation ( x ± s) for normal distributed data, and the median and interquartile range for skewed distribution. For comparisons between the two groups, t test was used for the measurement data conforming to the normal distribution, Wilcoxon rank-sum test was used for the skewed distribution, and Chi-square test was used for the counting data. All parameters were initially screened using LASSO analysis, followed by univariate logistic regression analysis to screen for statistically significant parameters, then further perform multiple logistic regression analysis. Then draw a nomogram and use the calibration curves to evaluate the accuracy of the model. Lastly, the ROC curve of selected meaningful variables and predictive model was plotted by R software to determine the diagnostic value [22]. P < 0.05 was considered statistically significant.

Patients' clinical characteristics
Among the 657 patients, 208 patients had recurrence of stones, including 125 males and 83 females, with an average age of 51.39 ± 12.01 years old, and 449 patients without recurrence, including 259 males and 190 females, with an average age of 50.55 ± 12.87 years old. Patients' clinical data of age, gender, BMI, stone maximum diameter, urea, Cr, UA, Cystatin C, TC, TG, HDL-C, LDL-C, VLDL, LDH, apo A, apo B, lipoproteins, K + , Na + , Cl − , Ca 2+ , P, Mg 2+ , Glu, GSP, Ne, LYM, RBC, Hb, PLT, urine RBC, urine WBC, urine pH, urine protein, and other relevant parameters were statistically analyzed. The count data were compared between the two groups using Chi-square test, and the measurement data were compared using independent samples t test or Wilcoxon rank-sum test (as shown in Table 1). The results showed that the recurrence of stones was related to the stone number and side. The multiple stones recurrence rate was higher than that of single stone, and the bilateral stones recurrence rate was higher than that of unilateral stones. In addition, the recurrence of stones was closely related to stone position, urine pH, urea, creatinine, cystatin C, apo B, and P. The recurrence probability of kidney stones and kidney ureteral stones were higher than that of ureteral stones alone. The values of creatinine, urea, cystatin C, apo B, and urine pH in stone recurrence group were significantly higher than those in the non-recurrence group, while the serum phosphorus level in the stone recurrence group was lower than that in the non-recurrence group (as shown in Supplementary Fig. 1).

Logistic regression model
The parameters of multiple stones, bilateral stones, stone position, creatinine, urea, cystatin C, P, and urine pH with statistically significant differences were first screened by LASSO regression analysis to exclude collinearity factors (as shown in Fig. 1), then the univariate and multivariate logistic regression analysis was performed (as shown in Table 2). Univariate logistic regression analysis showed that multiple stones, bilateral stones, stone position, urine pH, urea, creatinine, cystatin C, and P were risk factors for stone recurrence. Then multivariate logistic regression analysis showed that the recurrence risk of multiple stones (OR: 1.832, 95% CI 1.240-2.706) was 1.832 times higher than that of single stones, bilateral stones (OR: 1.779, 95% CI 1.226-2.582) was 1.779 times higher than that of unilateral stones, kidney stones (OR: 3.268, 95% CI 1.638-6.518) was 3.268 times than that of ureteral stones, and kidney ureteral stones (OR: 3.375, 95% CI 1.649-6.906) was 3.375 times than that of ureteral stones. Regarding urine pH, we found that within the range of 5 to 7, the risk of stone recurrence increased with the increase of pH values, and the risk of stone recurrence increased by 1.967-fold per 1 unit increase in pH values. At the same time, we found the risk of stone recurrence was positively correlated with creatinine (OR: 1.012, 95% CI 1.006-1.018), urine pH (OR 1.967, 95% CI 1.343-2.883), Apo B (OR: 4.189, 95% CI 1.985-8.841), and negatively correlated with serum phosphorus (OR: 0.282, 95% CI 0.109-0.728).
Then we divided creatinine, serum phosphorus, and apo B into four intervals according to the quartiles and performed subgroup analysis to explore the influence of different interval values on stone recurrence (as shown in Table 3). Compared with creatinine values in the range of 39 umol/L-64 umol/L, the study found that the risk of stone recurrence in patients with creatinine values in the range of 64 umol/L-77 umol/L, 77 umol/L-92 umol/L, and 92 umol/L-466 umol/L increased by 1.828 times, 1.947 times, and 2.615 times, respectively. And compared with the serum phosphorus values in the range of 1.21 mmol/L-2.3 mmol/L, the risk of stone recurrence in the range of 1.07 mmol/L-1.21 mmol/L, 0.96 mmol/L-1.07 mmol/L, and 0.57 mmol/L-0.96 mmol/L was increased 1.126 times, 1.216 times, and 1.909 times, respectively. About apo B, we found that the risk of stone recurrence was greatest in the interval (0.96 g/L-1.9 g/L), more 1.972 times greater than the other intervals.

Construction and assessment of nomogram
Combined with the above analysis, we concluded that the influencing factors of stone recurrence include stone number, stone side, stone position, urine pH, apo B, serum phosphorus, and creatinine. We summarized the seven factors and plotted nomograms by R software based on different standard units (as shown in Fig. 2 and Supplementary Fig. 2). And further plotted the calibration curve (as shown in Fig. 3), which is close to the standard curve. Then the prediction model was validated, and the consistency C-index was calculated to be 0.717, which suggested that the model had good prediction performance.

Nomogram prediction and validation
According to the nomogram, 7 points, 11 points, 17.5 points, and 18.5 points were assigned to the diagnosis of multiple stones, bilateral stones, kidney stones, and kidney ureteral stones, respectively. And urine pH values of 5, 5.5, 6, 6.5, 7 were assigned to points of 0, 5, 7.5, 16, 27.5, respectively. For creatinine, phosphorus, and apo B, we divided them into nine or ten intervals according to the nomogram and assigned the interval average point (Maximum value + minimum value/2) to the data in each interval. Finally, we calculated the total points for each patient. Then the ROC curve of each independent variable  and the ROC curve of the total points for the prediction model was plotted using R software (as shown in Fig. 4).
The results of the study found that the sensitivity and specificity of the prediction model diagnosis were greater than any single variable (as shown in Table 4), which means that the model has a good role in predicting the recurrence of stones in patients. Apo B is not a common parameter to assess stone recurrence, to further determine the role of apo B in the prediction model, we removed the apo B factor and reconstructed the nomogram (Supplementary Fig. 3), and assigned scores to each risk factor and plotted the ROC curve ( Supplementary Fig. 4). The study found that the nomogram without apo B had a diagnostic sensitivity of 65.38% and a specificity of 66.59%, compared with the nomogram include apo B, decreasing the predictive accuracy.

Discussion
In this study, we retrospectively analyzed 657 patients data with upper urinary tract stones, 208 of them developed stone recurrence, with a recurrence rate of 31.7%, including 125 males with a recurrence rate of 32.6% and 83 females with a recurrence rate of 30.4%. The recurrence rate of upper urinary tract stones is high, and the treatment cost is large, the second or multiple recurrences not only increase the patient's pain, but also increase medical expenses, effective prevention of stone recurrence is necessary. However, due to the poor medical compliance, especially for patients recovering from their first stone surgery, they subconsciously believe that they have fully recovered and ignored the possibility of stone recurrence. After discharge from the hospital, the previous inappropriate diet summed to obtain the total points. Different total points correspond to different recurrence risks and lifestyle habits may have continued, accelerating the recurrence of stones. Currently, clinical health education for post-stone surgery is aimed at all stone patients rather than those at high risk of recurrence, which cannot draw the attention of patients, but also increases the workload of medical care. If we can construct a model to assess and predict the risk of stone recurrence in postoperative patients, we can use it to screen out patients at high risk of stone recurrence and provide targeted health education and dietary guidance, and strengthen follow-up after surgery, to minimize the possibility of stone recurrence and prevent them from suffering again and repeating treatment.
Analysis of clinical data from 657 patients in our department showed that the multiple stones recurrence rate was higher than that of single stone, bilateral stones were higher than that of unilateral stone, and the kidney stones and kidney ureteral stones recurrence rate were significantly higher than that of ureteral stones. In addition, it was found that the higher creatinine, apo B, urine pH, the higher likelihood of stone recurrence, and there was a negative correlation between serum phosphorus and stone recurrence, with a high rate of stone recurrence in those with low serum phosphorus levels. We know that creatinine level monitors kidney function, and higher creatinine level means more serious damage to kidney function. It means kidney urination ability decreases, which leads to kidney crystals formation and causes stone recurrence. Apo B is the main structural protein of LDL cholesterol, which is present on the surface of LDL, and directly reflects the level of LDL cholesterol. It is deposited in blood vessels and can cause narrowing and sclerosis of small blood vessels, and has been reported to be closely related to the occurrence of coronary heart disease [23]. Apo B is not a common factor for urinary stone recurrence, and to further determine the effect of apoB on the stone recurrence prediction model, we constructed nomograms (with or without Apo B) and evaluated their diagnostic sensitivity and specificity, and found that apo B improved the accuracy of the model. It may be the elevated apo B that cause lipid metabolism disorders, which further lead to glomerular vasculopathy, affecting renal function and increasing the risk of stone recurrence. In the 5-7 interval, urine pH increases and the risk of stone recurrence increases, which may be related to the imbalance of acid-base substances in the urine. The higher pH increases the recurrence rate of stones, which may be related to the fact that the stone composition of patients in this region is mainly calcium oxalate, magnesium ammonium phosphate, and phosphorus carbonate. Serum phosphorus refers to inorganic phosphorus in the plasma, which is present as phosphate in more than 85% of plasma, while some stone components are dominated by magnesium ammonium phosphate. Therefore, decreased levels of elemental phosphorus in the serum may be associated with the absorption of phosphate into the renal tubules to synthesize magnesium phosphate crystals, and therefore decreased serum phosphorus levels may indicate a risk of stone recurrence.  ROC curves for each variable and for the total points of each predictor variable points were summed. The ROC curve of total points was 0.715 (95% CI 0.678-0.749, P < 0.001), with a diagnostic sensitivity of 73.08% and a specificity of 61.25%. The results showed that the model predicted a higher probability of recurrence of upper urinary tract stones than either individual variable Finally, we summarized the high-risk factors for recurrence of upper urinary tract stones and constructed a predictive model by drawing a nomogram using R software, which was found to have a diagnostic sensitivity of 73.08% and a diagnostic specificity of 61.25%. The prediction model constructed has the following two advantages. First, the parameters for constructing the model do not involve complex and expensive examinations, and the relevant examinations can be completed in primary hospitals, which is convenient and saves medical expenses. Second, the model can screen for people at high risk of stone recurrence from postoperative patients with upper urinary tract stones and provide focused guidance on their diet and lifestyle habits, shorten their review time, and reduce the probability of recurrence. However, the model also has shortcomings, mainly in the following two aspects. First, the main limitations of this work were the retrospective design and the small sample size, which hindered statistical analysis. Second, this study was a retrospective study, and we found that some important clinical parameters that affect stone recurrence, such as stone composition, were incomplete. Among the limited 121 cases stone composition data, we found that the stone composition was dominated by calcium oxalate, phosphorus carbonate, and ammonium magnesium phosphate in our region, accounting for 93.5%. In the future, we will implement a prospective study to specifically analyze the risk factors for recurrence of different component stone and develop new predictive models to further improve the accuracy of prediction of stone recurrence.

Conclusion
This study found that multiple stones, bilateral stones, stones location, urine pH, apo B, creatinine, and serum phosphorus are risk factors for recurrence of upper urinary tract stones, and its constructed upper urinary tract stones recurrence prediction model can better screen out high-risk groups for recurrence from post-stone surgery patients, and then focus on giving guidance on diet and lifestyle habits, and strengthening follow-up to reduce the probability of stone recurrence.