Development of a new nomogram for predicting the probability of a positive initial prostate biopsy in Japanese patients with serum PSA levels less than 10 ng/mL

Authors


Hiroyoshi Suzuki md phd, Department of Urology, Graduate School of Medicine, Chiba University, 1-8-1 Inohana, Chuo-ku, Chiba-shi, Chiba 260-8670, Japan. Email: hirosuzu@faculty.chiba-u.jp

Abstract

Objectives:  Although several nomograms for prostate cancer detection have been developed for Western populations, the models constructed on Japanese data would be more useful for the Japanese population because of various differences between Western and Asian populations. We previously developed a model for predicting the probability of a positive initial prostate biopsy using clinical and laboratory data from Japanese males. In the present study, a predictive model for Japanese males with a prostate-specific antigen (PSA) < 10 ng/mL was developed to guide decision-making for prostate biopsies.

Methods:  The age, total PSA level, free to total PSA ratio, prostate volume, and the digital rectal examination findings of 1037 Japanese males with a PSA < 10 ng/mL undergoing initial prostate biopsy as part of individual screening were analyzed. For study validation, 20% of these data was randomly reserved. Logistic regression analysis estimated relative risk, 95% confidence intervals, and P-values.

Results:  Age and the independent predictors of a positive biopsy result (elevated PSA, decreased free to total PSA ratio, small prostate volume, and abnormal digital rectal examination findings) were used to develop a predictive nomogram. The area under the receiver operating characteristic curve was significantly higher for the model (73.0%) than for PSA alone (55.0%). If externally validated, the use of this nomogram could reduce unnecessary biopsies by 26% and overall prostate biopsies by 7.8%.

Conclusions:  This predictive nomogram could provide more precise risk-analysis information for individual Japanese patients with PSA levels less than 10 ng/mL and may help to identify patients who need a prostate biopsy.

Introduction

Prostate-specific antigen (PSA) screening is useful to detect early prostate cancer. Recently, the use of serum PSA screening has become widespread in Japan. While clear evidence of a benefit derived from PSA mass screening remains unavailable, recent trends towards a decreased incidence of prostate cancer deaths in the United States and in the Austrian Tyrol region indicate that early detection and intervention are effective.1,2 This is supported by an updated report from a randomized Scandinavian trial that showed an improvement in metastasis-free, prostate cancer-specific survival and overall survival among patients who underwent prostatectomy.3

Serum PSA elevation is caused not only by prostate cancer but also by benign prostatic disease. It is very difficult to distinguish prostate cancer from benign prostate disease, particularly in patients with serum PSA levels less than 10 ng/mL. Therefore, it would be very useful if there were tools that could precisely predict the presence of cancer. This would significantly reduce the number of unnecessary prostate biopsies among males with intermediate serum PSA levels, in whom it is currently difficult to determine whether a prostate biopsy is indicated.

Various strategies have been proposed to improve the detection rate of prostate cancer among patients with intermediate PSA levels. In addition to total PSA, clinical factors associated with the detection of prostate cancer include age, digital rectal examination (DRE) findings, transrectal ultrasound (TRUS), PSA density (PSAD), PSA velocity, transition zone PSAD (PSATZD), free to total PSA ratio (f/T PSA ratio), age-specific PSA, and p53 antibody.4–6 Although the f/T PSA ratio has emerged as the most clinically useful and widespread method of improving the sensitivity and specificity of detecting prostate cancer, the value of the other methods remains a subject of considerable debate.5,6

Several groups have developed models to help predict a positive prostate biopsy among men undergoing evaluation for prostate cancer.4,7–10 These studies demonstrated that, compared with using individual factors alone, either artificial neural networks or logistic regression could improve the accuracy of prediction. Djavan et al. developed an artificial neural network model to determine the risk of a positive biopsy with improved predictive capacity.11

Garzotto et al. developed a nomogram to predict the presence of prostate cancer using age, PSAD, DRE, and TRUS findings in patients with intermediate PSA levels.12 However, there is a possibility that these nomograms based on Western populations might not be accurate for the Japanese population due to the different risks of prostate cancer between Asian and Western populations.

Recently, some nomograms including ethnicity as one of the clinical parameters have been reported, and racial differences have been recognized as an important parameter which influences prostate cancer incidence.13,14 Yanke et al. created the first nomogram to individualize the risk based on whether the patient was African American (AA) or Caucasian; this predictive model could be used to counsel men on the probability of their having cancer at the time of their first biopsy.13 Nam et al. developed a nomogram which includes Asian, White, Black, and other ethnicities as a parameter and the Asian race was evaluated as having a lower risk of prostate cancer.14 It is generally accepted that prostate cancer incidence is affected by environmental factors such as the eating habits of the nation, and the differences of prostate cancer incidence among Asian peoples in the United States has been reported.15 Combined together, it might be inappropriate that we applies these Western nomograms directly to the Japanese population, which has a lower incidence of prostate cancer because of racial, biological, and clinical differences. Therefore, we decided to develop nomograms for the Japanese population in Japan to predict the probability of a positive initial prostate biopsy. Previously, we reported the development of one such nomogram (the Carcinoma Histological Incidence at Biopsy Assistant program: the CHIBA program).16

In the present study, a new nomogram (CHIBA program version 10) was developed based on a Japanese population of patients with PSA levels less than 10 ng/mL. The decision to proceed with a prostate biopsy in such cases is considered difficult. Such a predictive nomogram for the Japanese population could provide more precise risk-analysis information for individual Japanese patients and might be of benefit to Japanese physicians and patients when they are considering a prostate biopsy as the definitive diagnostic tool.

Methods

Study population

Between March 1998 and October 2006, data were prospectively and uniformly collected during the routine clinical care of patients. A total of 1037 patients had all of the necessary data and basically 10 core biopsy specimens obtained during an initial TRUS-guided transperineal prostate biopsy at Asahi General Hospital. The cohort of the present nomogram and that of our previous nomogram16 was partially overlapped. The PSA level before biopsy was 5.57 ± 1.98 ng/mL. A urologist performed a DRE on all patients prior to the TRUS, and the results were classified as normal or cancer-suspect (one or both sides). Prostatic nodules or indurations were considered suspect on DRE. The biopsy results were positive in 236 cases and negative in 801 cases.

Serum total/free PSA assay

A single blood sample drawn before DRE was immediately refrigerated and centrifuged. The serum was separated and stored at −20°C. Free and total PSA were measured using the ARCHITECT PSA kit (Abbott Laboratories, Chicago, IL), according to the manufacturer's instructions.5

Transrectal ultrasonography (TRUS)-guided systematic biopsy

TRUS-guided systematic biopsy of the prostate was performed using a scanner and a 7-MHz transducer. The prostate was scanned in the transverse and sagittal planes with the patient in the lithotomy position. Prostate volume was determined using the formula for a prolate ellipsoid (width × length × height × 0.523).

Random systematic biopsies were performed using an automatic biopsy gun and an 18-gauge needle with TRUS guidance. Basically, the biopsies consisted of eight peripheral and two transition zones. Additional biopsies were obtained if ultrasound images or DRE indicated suspicious areas.

Statistical analysis

To develop the nomogram, 80% of the patients were randomly selected to build the model, while the remaining 20% were reserved to validate the model. The factors we evaluated for the risk of a positive biopsy included age, DRE findings, prostate volume, total PSA level, and the f/T PSA ratio. The significance of each factor was assessed on univariate logistic regression analyses. Multivariate stepwise logistic regression analysis was used to determine which factors were independent predictors of prostate carcinoma in the model-building set. Relative risks (RR) and 95% confidence intervals (95% CI) were also derived. A nomogram for a positive biopsy was developed based on the final logistic regression model. Using the validation data set, receiver operating characteristic (ROC) curves were used to compare the performance of the model with the prediction based on PSA alone. Performance characteristics were examined using calibration plots. We also compared the performance of the model with the previously reported Karakiewicz's nomogram17 using 20% of our dataset to be reserved for validation study. We selected a cut-off value for the predicted probability of prostate cancer based on the ROC curve. The cut-off value provided high sensitivity while reducing the total number of unnecessary biopsies. Logistic regression analyses were performed using R software (http://www.r-project.org/).

Results

Patient data

Table 1 shows the characteristics of the enrolled study population. The mean ± standard deviation of age, serum PSA level, f/T PSA ratio, and prostate volume were 68.86 ± 8.68 y, 5.57 ± 1.98 ng/mL, 21.67% ± 11.33% and 35.20 ± 20.66 cc, respectively. The DRE findings were classified as suspect in 209 (20.2%) of the 1037 patients (Table 2).

Table 1.  Clinical characteristics of 1037 patients with initial biopsy results
VariableTotal patients (n = 1037)Patients with positive biopsy (n = 236)Patients with negative biopsy (n = 801)
  1. PSA, prostate-specific antigen.

Age (years)68.86 ± 8.6869.98 ± 7.8468.52 ± 8.89
Total PSA (ng/mL)5.57 ± 1.985.81 ± 1.905.50 ± 1.99
f/T PSA ratio (%)21.67 ± 11.3317.99 ± 10.6022.75 ± 11.32
Prostate volume (cm3)35.20 ± 20.6627.50 ± 15.6837.47 ± 21.40
Table 2.  Distribution of patients for each variable and the univariate analysis evaluating the risk of a positive biopsy
VariablePatientsPositive biopsyRelative risk95% Confidence intervalP-value
  1. DRE, digital rectal examination; PSA, prostate-specific antigen.

Age (years)
 <6223740 (16.9%)1  
 63–6926564 (24.2%)1.5681.009–2.4380.0457
 70–7425965 (25.1%)1.6501.062–2.5650.0260
 >7527667 (24.3%)1.5791.020–2.4450.0407
Total PSA (ng/mL)
 <4.2725846 (17.8%)1  
 4.27–5.1925860 (23.3%)1.3970.908–2.1470.1280
 5.20–6.8425961 (23.6%)1.4200.925–2.1800.1092
 >6.8426269 (26.3%)1.6481.082–2.5100.0201
f/T PSA ratio (%)
 <12.922177 (34.8%)1  
 13.0–18.025477 (30.3%)0.8140.554–1.1960.2934
 19.0–26.028751 (17.8%)0.4040.268–0.609<0.0001
 >26.127531 (11.3%)0.2380.149–0.378<0.0001
Prostate volume (cm3)
 5.5–21.924799 (40.1%)1  
 22.0–30.526867 (25.0%)0.4980.342–0.7260.0003
 30.6–42.925844 (17.1%)0.3070.204–0.464<0.0001
 43.0–28026426 (9.8%)0.1630.101–0.263<0.0001
DRE findings
 Negative828153 (18.5%)1  
 One side suspect17872 (40.4%)2.9972.118–4.240<0.0001
 Both sides suspect3111 (35.5%)2.4261.139–5.1700.0216

Biopsy results

Adenocarcinoma of the prostate was detected in 22.8% (236 of 1037 patients) of the biopsy specimens. Table 2 shows the patient distribution within each variable and the results of the multivariate analysis that evaluated the risk associated with a positive initial biopsy.

Development of a nomogram

Table 2 shows the significant predictors for a positive prostate biopsy in the order of statistical significance that was determined on univariate logistic regression analysis.

The stepwise multivariate logistic regression analysis showed that the statistical significance of all five risk factors for the detection of prostate carcinoma in the study cohort was <0.05 (Table 3). Independent analyses using forward and backward stepwise procedures yielded identical results. Using the five independent risk factors, a nomogram was developed to diagnose prostate cancer (Fig. 1). As described by Kattan et al.,18 the accumulated number of points for each of the five categories was totaled to calculate the overall likelihood of a positive biopsy.

Table 3.  Multivariate logistic regression model analyzing the predictors for prostate cancer detection on initial prostate biopsy
VariableRelative risk95% Confidence intervalP-value
  1. DRE, digital rectal examination; PSA, prostate-specific antigen.

Age (years)1.0241.002–1.0460.032
Total PSA (ng/mL)1.1761.075–1.287<0.001
f/T PSA ratio (%)0.9710.953–0.9900.002
Prostate volume (cm3)0.9630.949–0.977<0.001
DRE findings
 One side suspect2.9881.980–4.507<0.001
 Both sides suspect3.7321.489–9.3530.005
Figure 1.

Nomogram for detecting prostate cancer on initial biopsy (Carcinoma Histological Incidence at Biopsy Assistant program version 10). Directions: determine age, total prostate-specific antigen (PSA) level, f/T PSA ratio, prostate volume measured by transrectal ultrasound (Echo WT) and digital rectal examination (DRE) findings for an individual patient. Draw a line upwards to the number of points in each category. Total the points, and then draw a line downward to find the probability of a positive biopsy.

For example, according to the model (Fig. 1), a 73-year-old male with a PSA of 5.6 ng/mL, an f/T PSA ratio of 19%, a normal DRE, and a prostate volume of 15 cc had a 32% risk of a positive biopsy, while based on an intermediate PSA level of 4–10 ng/mL, his risk would be 25%. The accuracy of this model was determined by the validation set. The ROC curve evaluated the accuracy of the predicted probability as 73.0% for the model compared with 55.0% based on the PSA alone (Table 4). In addition, the predictive accuracy of this model (AUC = 0.73) was higher than Karakiewicz's nomogram17 (AUC = 0.70) (Table 4). Figure 2 shows the calibration plots with local regression non-parametric smoothing lines of the present nomogram. The calibration plot of our nomogram gave a good agreement between predicted probability and actual probability.

Table 4.  Comparison of the present nomogram, Karakiewicz's nomogram results and other predictors on receiver operating characteristics (ROC) analysis
ModelArea under the curve
  1. DRE, digital rectal examination; PSA, prostate-specific antigen.

Present nomogram0.7302
Karakiewicz's nomogram0.7016
Age0.5276
Total PSA0.5500
f/T PSA ratio0.6461
Prostate volume0.6747
DRE0.6107
Figure 2.

Calibration plots with local regression non-parametric smoothing lines of the present nomogram on the internal validation. X-axes represent nomogram predicted probability of prostate cancer on needle biopsy, and Y axes represent observed rate of prostate cancer diagnosis.

If only those patients with >10% predicted probability of prostate carcinoma (a value that approximates the widely accepted PSA cut-off value of 4.0 ng/mL) were to have a biopsy, then the model would capture 90% of all patients with prostate cancer (sensitivity), while sparing 26% of patients without prostate carcinoma from undergoing an unnecessary procedure (specificity). Overall, the use of this model would reduce the total number of biopsies in the Japanese male population by 7.8%.

Discussion

Various efforts to develop predictive models for prostate cancer based on clinical, laboratory, and ultrasound parameters have been attempted in order to improve the rates of prostate cancer detection.7–11 Logistic regression-based nomograms and artificial neural networks (ANN) represent alternative methodological approaches that can be used to predict the probability of prostate cancer on initial biopsy. Garzotto et al. developed a logistic regression-based nomogram to predict the presence of prostate cancer based on age, PSAD, DRE, and TRUS findings in patients with intermediate PSA levels.12 On the other hand, Djavan et al. made an ANN model for early detection of prostate cancer.11 Chun et al. compared these two models in a head-to-head fashion and found a 4% increase in predictive accuracy in a logistic regression-based nomogram.19 Thus, we consider it would be appropriate that we had developed a logistic regression-based nomogram rather than ANN model.

There appear to be epidemiological and biological differences in prostate cancer in different populations. Since previous epidemiological studies suggested that African-American (AA) males have a significantly higher incidence of and mortality from prostate cancer than Caucasian males, Yanke et al.13 and Nam et al.14 developed nomograms to show that racial difference influences the performance of nomograms about prostate biopsy. Recently Kawakami et al. created nomograms and ANN for patients with PSA levels less than 20 ng/mL using Japanese data.20 They performed external validation using another Japanese institution's data, compared the accuracy of their models with foreign previously published models, and showed superiority of their models by comparison of AUC. In the present study, our models also achieved higher predictive accuracy than Karakiewicz's nomogram.17 Although these results do not directly indicate that a Japanese data based nomogram is more useful for Japanese populations because of the differences of the method of the biopsy and the parameters among the models, the Japanese nomograms might be more accurate for Asian populations. To address this point, various external validations on Japanese and Western data sets will be necessary.

We previously reported the development of a nomogram that could be used to predict the probability of a positive initial prostate biopsy among Japanese patients with PSA levels less than 90 ng/mL.16 However, the usefulness of that nomogram was limited since that did not adequately address the issue in patients with intermediate PSA levels, in whom prostate cancer is the most difficult to identify. Thus, in our present study, the range of PSA levels was limited to less than 10 ng/mL, and a new nomogram was developed to predict the probability of a positive initial prostate biopsy in Japanese males. Using this nomogram, the total number of biopsies in the Japanese male population could be reduced by 7.8%.

The risk factors that are used in such a nomogram should be readily available to urologists in standard clinical practice. Thus, in the present study, 5 independent risk factors that are available to urologists in general practice in Japan were evaluated. Therefore, this nomogram could easily be adopted and used widely in Japan. In our present nomogram, the presence of abnormal TRUS findings was not used as a parameter. Babaian et al. reported that at the prostate biopsy, the positive predictive value of DRE and PSA combination was significantly better than that of TRUS and PSA, but was not different from that of a combination of all three tests.21 Thus, we used DRE finding not TRUS findings as a parameter of the prostate biopsy nomogram.

One of the limitation of this nomogram may be the use of ultrasound guided transperineal 10-core prostate biopsies. The recent increased use of biopsy cores has generated a higher prostate cancer detection rate.22,23 There is no definitive agreement about the method of obtaining a prostate biopsy.24 Recently, Chun et al. tested the accuracy of a previously externally validated sextant biopsy nomogram in males who had ≥10 biopsy cores and found that previously developed predictive models are less accurate in predicting the probability of cancer on initial biopsy.25 To solve this problem, we need to develop a new nomogram including the number of biopsy cores and the method of biopsy (transperineal or transrectal) as one of the factors.

Chun et al. also created an extended repeat biopsy nomogram which could be used as a predictor of prostate cancer on repeat biopsy.26 Remzi et al. noted in a repeat biopsy setting in men with PSA between 4 and 10 ng/mL that total prostate volume has a significant impact on the ability to predict prostate cancer on repeat biopsy.27 Furthermore, Remzi et al. recently reported that the optimal number of biopsy cores should be adjusted for prostate volume and age on repeat biopsy.28 To consider the differences of the significance of the parameters between initial and repeat biopsy, present nomogram on initial biopsy cannot apply to repeat biopsy cases, and further study should be performed in the future.

In our present study, we performed internal validation to assess the validity of the present nomogram and compared the performance of the model with the previously reported Karakiewicz's nomogram.17 To assess the validity of the present nomogram, an additional external validation study would be necessary in the future. In addition, we could not compare the predictive accuracy between the present nomogram and our previous model on internal validation because of the overlap of the cohorts. In the future, we need to perform external validation to assess the validity of our nomogram and the predictive accuracy of various Japanese and foreign nomograms. The predictive nomogram developed for the Japanese population that has been presented in the present paper could provide precise risk-analysis information for individual Japanese patients and might be of benefit to Japanese physicians and patients when they are deciding whether to undergo prostate biopsies as the definitive diagnostic tool. This is the first report dealing with a nomogram that can provide more precise risk-analysis information for Japanese patients with PSA levels less than 10 ng/mL.

Acknowledgments

The present study was supported in part by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sport, Science and Technology of Japan (Nos. 16591582, 16689026 and 19591834), the Japanese Urological Association (2006), the Ministry of Health, Labour and Welfare of Japan (Aid for Cancer Research) and the Prostate Research Fund (2005).

Ancillary