Development and validation of a nomogram for predicting prostate cancer in patients with PSA ≤ 20 ng/mL at initial biopsy.

ABSTRACT: The aim of this study was to construct a nomogram for predicting prostate cancer (PCa) in patients with PSA ≤ 20 ng/mL at initial biopsy.The patients with PSA ≤ 20 ng/mL who underwent prostate biopsy were retrospectively included in this study. The nomogram was developed based on predictors for PCa, which were assessed by multivariable logistic regression analysis. The receiver operating characteristic curve, calibration plots and decision curve analysis (DCA) were used to evaluate the performance of the nomogram.This retrospective study included 691 patients, who were divided into training set (505 patients) and validation set (186 patients). The nomogram was developed based on the multivariable logistic regression model, including age, total PSA, free PSA, and prostate volume. It had a high area under the curve of 0.857, and was well verified in validation set. Calibration plots and DCA further validated its discrimination and potential clinical benefits. Applying the cut-off value of 15%, our nomogram would avoid 42.5% of unnecessary biopsies while miss only 4.4% of PCa patients.The nomogram provided high predictive accuracy for PCa in patients with PSA ≤ 20 ng/mL at initial biopsy, which could be used to avoid the unnecessary biopsies in clinical practice.

Introduction

Prostate cancer (PCa) is the most common malignancy in men worldwide, and its incidence is also rapidly increasing in China. According to the Chinese National Cancer Institute, it is estimated that there is 60,300 new cases of PCa in 2015 in China.[ The prostate biopsy is currently the standard practice for the diagnosis of PCa,[ whereas it could not be used as an extensive screening tool due to its medical cost and invasiveness.[ Therefore, it is more important for urologist to identify the patients with a high risk of PCa for further examinations in clinical practice.

Prostate-specific antigen (PSA) is the most frequently used screening test for PCa, which was first introduced into clinical use in 1980 s and promoted the detection rate of PCa significantly.[ But there has always been controversy over PSA screenings for men since its application. First, PSA is a prostate-specific rather than a PCa-specific marker, and some benign prostate diseases may also lead to the abnormal elevation of PSA, such as benign prostate hyperplasia (BPH) and prostatitis.[ Especially in the range of 4 to 10 ng/mL, there is significant overlap in PSA levels between BPH and PCa patients. Moreover, it is reported that PCa detection rates on initial prostate biopsy ranged between 22.8% and 42.0%, which is partially ascribed to the low specificity of PSA.[ There are still many unnecessary prostate biopsies screened by current PSA test. In addition, it is reported that patients with a PSA level ≤ 4 ng/mL still have the risk of PCa, and the detection rate may reach up to 20%.[

In the past few years, some predictive tools were developed to help identify the PCa risk before biopsy, such as probability table, artificial neural network, and nomogram.[ Comparing with other tools, the nomogram could integrate different risk factors and provide an individualized estimation of PCa probability. Besides, the nomogram can be displayed graphically and easily applied in clinical practices. However, most nomograms were constructed on European or American populations, and fewer models focused on Asian populations.

Therefore, on the basis of the large cohort of Chinese patients who underwent prostate biopsies in our medical center, we developed a nomogram that incorporates several simple information, including age, total PSA, free PSA, and prostate volume to predict the risk for PCa before biopsy.

Methods

Ethical approval

All the procedures followed were in accordance with the Helsinki Declaration (1964, amended in 1975, 1983, 1989, 1996, and 2000) of the World Medical Association. This study was approved by the Ethics Committee of Chinese PLA General Hospital.

Study design and patient population

The study was a retrospective observational study, which was complied with the STROBE statement (Strengthening the Reporting of Observational studies in Epidemiology).

Following institutional review board approval, patients with PSA ≤ 20 ng/mL who had undergone transrectal ultrasound-guided prostate biopsy between July 2009 and March 2018 were initially collected from the Chinese PLA General Hospital. The PLA General Hospital is located in Beijing and is a large tertiary hospital, which provides medical services to people in Beijing area and surrounding provinces. The annual outpatient volume of the PLA General Hospital reaches 8 million. According to the Chinese guideline for diagnosis and treatment of PCa, prostate biopsy should be recommended for patients with the following conditions: PSA is greater than 10 ng/mL; PSA is between 4 and 10 ng/mL, but PSA density (PSAD) or free/total PSA ratio (f/t PSA) results are abnormal; Regardless of PSA level, suspicious digital rectal examination (DRE) or transrectal ultrasonography (TRUS) result. All patients who met the above criteria were recommended for a prostate biopsy. The age, total PSA (tPSA), free PSA (fPSA), prostate volume, and the prostate biopsy result including Gleason score of patients were retrospectively collected. In order to avoid the potential bias, the serum PSA test and TRUS examination were all performed in our hospital. The exclusion criteria contained: We excluded the patients who had prostate surgery history or previous prostate biopsy; Patients who had used 5-α reductase inhibition or hormone deprivation therapy; Patients with prostate volume > 200 mL; and Patients with incomplete clinical data. Overall, 691 patients were enrolled in this study, and the process of patient inclusion and exclusion is shown in Figure 1. The eligible patients were randomly divided into a training set (nomogram development, 505 cases) and a validation set (nomogram validation, 186 cases).

Figure 1

Flow chat of patient exclusion according to the criteria.

Serum PSA concentrations (tPSA and fPSA) were measured before DRE and TRUS by enzyme-linked immunoassay (EIA; Research & Development Systems). The width (W), length (L), and height (H) of prostate were measured on TRUS, and PV (mL) was calculated using the formula: PV = W × L × H × 0.52. TRUS was performed by experienced ultrasonologists.

Nomogram development

To develop a well-calibrated nomogram for predicting the risk of PCa, we performed univariable as well as multivariable logistic regression analyses to evaluate the relationship between the associated factors and PCa. First, the variables that exhibited a statistical significance (P < .05) in univariable logistic analysis were then included in the multivariable logistic regression analysis, which was performed to screen independent predictors for PCa. Independent predictors in multivariable logistic regression analysis were finally included in the nomogram construction.

Evaluating the nomogram performance

Calibration plots

The calibration plot with bootstrapping was used to illustrate the association between the actual probability and the predicted probability, which reflected performance characteristics of the nomogram.[

Receiver operating characteristic (ROC)

The ROC curves of the nomogram in training set and validation set were also compared and area under curve (AUC) was calculated.[ The AUC ranged from 0 to 1, with 1 indicating perfect concordance, 0.5 indicating no better than chance, and 0 indicating discordance. And the ROCs of PSA alone, f/t PSA and PSAD were also measured and their AUCs were compared with that of the nomogram model by the DeLong method respectively.[ And as a screening tool for PCa, the sensitivity of the nomogram should be about 95%.

Decision-curve analysis

The decision curve analysis was performed to further evaluate the performance of the nomogram.[

Internal validation

The AUC of the developed nomogram was verified in the validation set. And the ROC of nomogram was also compared with the tPSA, f/t PSA, and PSAD.

Statistical analysis

The Pearson's Chi-square test was used to evaluate the distribution of categorical variables, and Mann–Whitney U test or Student test was used for distribution of continuous variables. The univariable and multivariable logistic analysis was performed using SPSS software (version 20.0; SPSS Company, Chicago, IL). The development and evaluation of the nomogram were carried out using R software version 3.31 (R Foundation for Statistical Computing, http://www.r-project.org). In this study, all P values were 2-sided with statistical significance at P < .05.

Results

Patient characteristics

Among 1144 patients who underwent prostate biopsy, 691 of them were eligible and included in this study eventually. Five hundred five cases were randomly divided into training set and 186 cases into validation set. The patients’ characteristics are summarized in Table 1. The mean age of the patients was 68.92 ± 9.78 years, mean tPSA was 9.16 ± 4.82 ng/mL, mean fPSA was 1.38 ± 0.93 ng/mL, and mean prostate volume 54.67 ± 30.76 mL. Overall, 262 patients (38%) had positive biopsy results and were diagnosed with PCa. Among them, 103 patients had a Gleason score of 6, 50 of 3+4 = 7, 49 had of 4 + 3 = 7, 35 had of 8, and 25 had of 9 to 10. Age, fPSA, tPSA, and prostate volume were all significantly different between biopsy-positive and negative groups.

Table 1

Baseline clinical characteristics and comparison between patients with positive and negative results on prostate biopsy.

		PCa
Variable	Total	Positive	Negative	P
Total patients	691	262	429
Age, yr
Mean ± SD	68.92 ± 9.78	71.82 ± 9.01	67.14 ± 9.82	<.001
Median	70	73	68
IQR	62–76	66–79	61–74
tPSA, ng/mL
Mean ± SD	9.16 ± 4.82	10.38 ± 4.69	8.42 ± 4.76	<.001
Median	8.61	9.90	7.99
IQR	5.53–12.40	6.73–13.80	5.10–11.40
fPSA, ng/mL
Mean ± SD	1.38 ± 0.93	1.30 ± 0.87	1.43 ± 0.97	.012
Median	1.21	1.19	1.25
IQR	0.70–1.92	0.76–1.77	0.64–2.01
Prostate volume, mL
Mean ± SD	54.67 ± 30.76	40.36 ± 21.94	63.45 ± 32.09	<.001
Median	47.97	34.14	57.30
IQR	31.67–71.57	24.78–49.76	40.67–80.82
f/t PSA
Mean ± SD	0.17 ± 0.12	0.14 ± 0.07	0.18 ± 0.14	<.001
Median	0.15	0.13	0.17
IQR	0.10–0.21	0.09–0.16	0.12–0.24
PSAD
Mean ± SD	0.22 ± 0.18	0.32 ± 0.21	0.15 ± 0.11	<.001
Median	0.16	0.26	0.13
IQR	0.10–0.28	0.16–0.43	0.08–0.19
Gleason score (number)
6		103
3 + 4 = 7		50
4 + 3 = 7		49
8		35
9–10		25

Predictors for positive biopsy results

In univariable logistic regression analysis, patient age, tPSA, fPSA, and prostate volume were significantly related to the presence of PCa in biopsy (Table 1). The significant predictors were then included in the multivariable logistic regression analysis. According to the result, patients with older age [P < .001, odds ratio (OR) = 1.082, 95% confidence interval (95% CI): 1.059–1.105] and elevated tPSA (P < .001, OR = 1.196, 95% CI: 1.137–1.258) were more likely to have positive prostate biopsy results, while patients with higher fPSA (P = .048, OR = 0.695, 95% CI: 0.522–0.924) and bigger prostate volume (P < .001, OR = 0.956, 95% CI: 0.947–0.966) were less likely to have positive results (Table 2). And the Constant listed in Table 2 was obtained by the multivariate logistic regression analysis, which represents the intercept of the equation. Eventually, the age, tPSA, fPSA, and prostate volume were all significantly related to PCa and used for construction of the nomogram.

Table 2

Multivariate logistic regression analysis of factors associated with PCa.

Variable	PCa
	Coefficient	P	OR	95% CI
Age	0.079	<.001	1.082	1.059–1.105
tPSA	0.179	<.001	1.196	1.137–1.258
fPSA	−0.364	.048	0.695	0.522–0.924
Prostate volume	−0.045	<.001	0.956	0.947–0.966
Constant	−4.865	<.001	0.008	–

Construction of the nomogram

Nomograms for predicting PCa (Fig. 2) were developed according to the coefficient of the four significant predictors: age, tPSA, fPSA, and prostate volume. The total point was summed up by the point of each variable (top plotting scale), then was subject to the risk for PCa (bottom plotting scale).

Figure 2

The nomogram for predicting the risk of PCa at initial systematic 12-core prostate biopsy.

Calibration plot and discrimination of the nomogram for PCa detection in the training set

The calibration plot with internally bootstrap sampling (n = 1000) demonstrated that bias-corrected curve was close to the ideal curve (the 45-degree line), which indicated the good calibration of our nomogram (Fig. 3). The ROC was performed to evaluate the predictive accuracy of the nomogram in the training set and validation set respectively. For predicting PCa, the AUC was 0.857 (95% CI: 0.823–0.890) in training set. And the AUC of the nomogram was significantly better than that of PSA (0.664, P < .001), f/t PSA (0.669, P < .001), or PSAD (0.801, P < .05) (Fig. 4, Table 3).

Figure 3

Calibration plot of the nomogram (1000 bootstrap re-samples) for predicting risk of PCa. The y-axis represents actual probability, and x-axis represents the probability estimated by nomogram (Predicted probability).

Figure 4

The ROC of nomogram and other parameters for predicting the risk of PCa at initial biopsy in training set.

Table 3

Comparison of the AUC between the nomogram and PSA-related parameters for predicting PCa at initial biopsy.

	PCa
	AUC	95% CI	P ∗
Nomogram	0.857	0.823–0.890	–
PSA	0.664	0.616–0.712	<.001
f/t PSA	0.669	0.621–0.717	<.001
PSAD	0.801	0.761–0.840	<.05

Internal validation of the nomogram for PCa detection in validation set

The nomogram was validated in the 186 patients of validation set. The acceptable calibration plot result was also observed. The ROC analysis indicated that AUC of the nomogram was 0.834, which was consistent with the AUC of nomogram in training set. Compared with PSA, f/t PSA, or PSAD, the AUC of nomogram was also superior in validation set (Table 3).

Performance of the nomogram for PCa detection

For PCa detection nomogram, the optimal risk cut-off value was 15% with a sensitivity of 95.6%, specificity of 42.5%, positive prediction value (PPV) of 53.0%, and negative prediction value (NPV) of 93.4%. And this cutoff would avoid 42.5% of unnecessary biopsies and 25.3% of total biopsies, with missing only 4.4% of PCa patients. A direct comparison of cut-off performances relative to available alternatives was performed at a 95% sensitivity analysis of nomogram against PSA, f/t PSA, and PSAD (Table 4). And the performance of the nomogram at different cut-off value is shown in Supplementary table 1.

Table 4

Performance comparison of nomogram to alternative models at 95% sensitivity in the training cohort of n = 505 patients.

Method	Cut-off value	Sensitivity	Specificity	PPV	NPV	Predicted negative
Nomogram	15%	95.6%	42.5%	53.0%	93.4%	25.3%
PSA	3.31	95.1%	17.3%	43.8%	83.9%	12.3%
f/t PSA	0.055	95.0%	8.3%	39.5%	46.9%	6.3%
PSAD	0.079	95.1%	25.6%	45.9%	87.8%	16.2%

Decision-curve analysis of the nomogram in clinical practice

The potential clinical benefits of the 2 nomograms were also evaluated by decision curve analysis (Fig. 5). For predicting PCa, the nomogram was shown to achieve more benefits than the intervention-all-patients scheme, the intervention-none scheme, or PSA screening.

Figure 5

Decision curves analysis for the nomogram and the PSA for predicting the risk of PCa in clinical practice.

Discussion

It is always one of the research hotspots that how to improve the predictive accuracy before prostate biopsy. Nomogram is a simple graphical presentation of statistical methods, such as logistic regression and Cox proportional hazards analysis, which is popular in clinical risk or prognosis prediction. In the current study, we develop a nomogram for predicting the risk of PCa using 4 common clinical and laboratory variables: age, tPSA, fPSA, and prostate volume. All these factors could be noninvasively and easily obtained before the initial biopsy. And if a patient is without suspicious DRE or ultrasound but elevated PSA, we can also calculate the positive risk for prostate biopsy using his PSA value, age, and prostate volume. The area under ROC of our nomogram is 0.857, which is significantly higher than PSA (0.664), f/t PSA (0.669), and PSAD (0.801), respectively. Calibration plot and decision curve analysis validate the efficacy and potential benefits of our nomogram. This nomogram is used as a screening tool for prostate biopsy and should avoid missed diagnosis of PCa patients as much as possible. Therefore, we choose 95% sensitivity as the cut off value to ensure that the model has a low false-negative rate. Comparing with f/t PSA and PSAD, the nomogram has higher specificity and NPV, which could avoid 42.5% of unnecessary biopsies and 25.3% of total biopsies with missing only 4.4% (9 patients) of PCa patients. Among them, 3 patients’ gleason score are 6, 2 are 3 + 4 = 7, 2 are 4 + 3 = 7, 1 is 8, and 1 is 9.

Considering sextant biopsy, which was commonly performed in the past, is no longer adequate for detection of PCa, our study enrolled patients who underwent 12-core biopsy according to the current recommendation.[ Moreover, by description analysis of our cohort of patients, we found the positive rate of biopsy for patients with a tPSA level > 20 ng/mL reached 70%, while the rate for patients with a tPSA level 10 to 20 and < 10 ng/mL was only 46% and 32%, respectively. We also tested our nomogram on a larger group of 872 patients that contained patients with a tPSA level >20 ng/mL, and the AUC increased to 0.864 for this group. Although the result was more encouraging with the inclusion of patients with high tPSA values, we limited our included patients to the men with a tPSA level ≤ 20 ng/mL. The clinical usability and value would be certainly questioned and jeopardized with inclusion of patients with high tPSA levels, because most of them will undergo prostate biopsy in clinical practices. In addition, DRE and TRUS results were commonly reported to be risk factors for prostate biopsy, but they have poor sensitivity and high interobserver variability. Therefore, our nomogram was constructed without consideration of DRE and TRUS, which was performed by several different doctors.

Several nomograms have also been developed for predicting PCa before biopsy. Karakiewics et al[ developed a nomogram to estimate the risk of PCa, including age, PSA, f/t PSA, and DRE. However, the included patients had a PSA level ≤ 50 ng/mL, which weaken the usability and benefits of the model in routine clinical practices. Besides, all the enrolled patients had received sextant biopsy, which may underestimate the positive rate and is no longer recommended as the standard biopsy scheme. And Nguyen et al[ had proved that a previous risk model based on sextant biopsy scheme did not perform well in their cohort of men screened for PCa with extended prostate biopsy scheme.

Optenberg et al[ developed another nomogram for PCa enrolled 633 patients with a PSA level ≤ 20 ng/mL, but they did not take the free PSA into consideration. Free PSA is an effective tool to differentiate PCa from BPH, which is widely used in clinical practice. We also tested the contribution of fPSA to our nomogram. The inclusion of the fPSA increased the AUC of our nomogram from 0.816 to 0.857.

Eastham et al[ constructed a model based on a group of 700 patients with an AUC of 0.75. Although they just include the patients with a PSA level <4 ng/mL and with abnormal DRE, their model could not be applied in men with higher PSA level or normal DRE finding. Carlson et al[ included patients with a PSA range of 4 to 10 ng/mL, but they missed the group of patients with lower PSA level (< 4 ng/mL), which constituted significant 14% of our whole biopsy-positive population. More importantly, they did not report a definite AUC of their model.

Race is a significant risk factor for presentation of PCa. However, most nomograms were developed based on European or American population, for example, Zaytoun et al[ constructed a nomogram included the race factor. The genetic and environmental differences may preclude the transfer of existing nomograms from Western studies to Asian populations. Besides, few studies focused on Asian, especially Chinese, population. Zhu et al[ developed a prostate health index (PHI)-based nomogram for predicting PCa based on Chinese population. Their model showed high accuracy (0.839), which included age, prostate volume, and PHI. PHI was calculated using the following formula: (p2PSA/fPSA) × . However, the p2PSA had not been routinely applied in China, which limited its application significantly. Fang et al[ also constructed a nomogram based on Chinese population including MRI information, although the MRI parameters and interpretation criteria are not uniform among different hospitals, which also limits its widely application. In addition, they also included patients with a PSA level < 50 ng/mL, which may jeopardize the model's clinical usability. Another Korean study developed a nomogram that enrolled patients with a PSA level < 10 ng/mL; however, their model had not been internal or external validated.[

Several limitations still exist in our nomogram. The first one is that our study is a single-center retrospective analysis, and the selection bias may exist in our study and interfere with the accuracy of our model. And a future external validation would be required to evaluate the utility of our nomogram by using the data from other institutions or models. Second, the false-negative findings of the prostate biopsy were not taken into consideration in this study, because the data of repeated biopsy in our cohort are scarce. Finally, the MRI tests of our cohort were performed by several different hospitals before coming to our center, and the prostate imaging reporting and data system (PI-RADS) score was not indicated in the record. Considering the potential variability, we did not include the MRI result as a risk factor in our model.

In conclusion, we develop and validate a nomogram for predicting the risk of PCa at initial prostate biopsy in Chinese patients with PSA ≤ 20 ng/mL, which relies on 4 easily obtained factors, including age, tPSA, fPSA, and prostate volume. The nomogram provides a higher prediction accuracy than tPSA, f/t PSA, or PSAD without increase of medical cost or invasiveness, which could be used to avoid the unnecessary biopsies in clinical practice.

Author contributions

Conceptualization: Xiaotao Yin, Jiangping Gao.

Data curation: Qiang Wu, Fanglong Li.

Formal analysis: Qiang Wu, Xiaotao Yin.

Investigation: Qiang Wu, Fanglong Li.

Methodology: Fanglong Li, Xiaotao Yin.

Project administration: Xiaotao Yin, Jiangping Gao.

Supervision: Xiaotao Yin.

Writing – original draft: Qiang Wu.

Writing – review & editing: Jiangping Gao, Xu Zhang.

Correction

The corresponding author has been corrected from Xiaotao Yin to Jianping Gao.