INTRODUCTION
Prostate-specific antigen (PSA) testing is a basic screening tool for prostate cancer, but it has low specificity, which can lead to overdiagnosis and overtreatment.1-3 PSA velocity and density have been recommended as adjunctive tests for PSA,4, 5 but PSA velocity is difficult to measure because it requires at least 3 separate measurements over a minimum of 18 months.6 PSA measurements also fluctuate in between recordings. Analyzing PSA fluctuation may differentiate patients with benign and malignant tumors.7 Larger changes in PSA taken at 6- to 12-month intervals are associated with a lower likelihood for prostate cancer.
The National Comprehensive Cancer Network (NCCN) guidelines recommend multiparametric magnetic resonance imaging (MRI) and biomarker testing prior to prostate biopsy. A prebiopsy MRI predicts clinically significant prostate cancer and prevents unnecessary prostate biopsies in patients with no radiographic lesions. Additional strategies are required to make prebiopsy MRI more cost-effective.
This study examined the role of short-term PSA variation in predicting prostate cancer among patients with PSA ≤20 ng/mL who underwent prebiopsy MRI. PSA was measured at 2 time points: when a prostate biopsy was recommended (1st PSA) and when the prostate biopsy was performed (2nd PSA). This study also compared the accuracy of a PSA variation prediction model with different prostate cancer prediction models based on conventional factors, such as age, prostate volume, PSA, PSA density, or number of target lesions. Therefore, we attempted to analyze PSA variation through short-term PSA follow-up and to confirm whether this could be a new predictor of prostate cancer.
MATERIALS AND METHODS
1. Study Population
This study was approved by the ethics committee of our institution (Yonsei University Health System, Seoul, Korea, 3-2019-0368), and the requirement for written informed consent was waived because of the study's retrospective nature. The study design complied with the principles of the 1946 Declaration of Helsinki and its 2008 amendment.
This study examined the data of 281 consecutive patients with PSA ≤20 ng/mL who underwent prebiopsy MRI and a prostate biopsy between September 2019 and February 2021. When patients underwent the 1st PSA measurement, they were subjected to urine analysis and urine culture. If they had pyuria or a positive urine culture, appropriate oral or intravenous antibiotics were administered until the pyuria disappeared and the urine culture was confirmed to be negative. Twenty-three patients, including 15 with prebiopsy antibiotic administration, 6 with no 1st PSA measurement and 2 with a history of radiotherapy, were excluded from the final analysis. Clinicopathologic data, such as age, history of prostate biopsy, prostate volume, digital rectal examination (DRE) findings, family history of prostate cancer, 1st and 2nd PSA measurements, PSA density, number of positive biopsy cores, and MRI findings, were documented. A positive family history was defined as a father and/or one or more brothers with prostate cancer. Patients whose family history could not be determined, such as those with no brothers or whose father died early, were considered to have a negative family history. PSA density was calculated by dividing the 1st PSA measurement by the prostate volume.
2. Indication for Prebiopsy MRI and Prostate Biopsy
Prebiopsy MRI is covered by the national reimbursement program when it is requested in patients with elevated PSA >100 ng/mL, palpable nodule/s on DRE, and/or with hypoechoic lesions that are suspicious for prostate cancer. A prostate biopsy was requested based on the decision of 3 clinicians and in consideration of patient variables, such as a positive family history, palpable nodule/s on DRE, findings on MRI, elevated or rising PSA during the follow-up period, and/or PSA density.
3. MRI Protocol and Interpretation
MRI was performed using a 3.0 Tesla MRI system (Intera Achieva 3.0 T; Phillips Medical System, Best, Netherlands) equipped with a 6-channel phased array coil. The prostate MRI protocol involved diffusion-weighted and T2-weighted imaging. T2-weighted turbo spin-echo MRI was acquired in the axial, sagittal, and coronal planes. MRI datasets were obtained at identical slice locations with a slice thickness of 3 mm and no intersection gap. Two b-values (0-1400) were used, and diffusion restriction was quantified with apparent diffusion coefficient mapping. Dynamic contrast-enhanced MRI was also performed.
Uro-radiologists identified suspicious regions of interest on the apparent diffusion coefficient maps on a Digital Imaging and Communications in Medicine workstation. The Prostate Imaging-Reporting and Data System (PI-RADS) was used to describe the MRI findings.8 A PI-RADS score ≥3 was given in the presence of visible lesions.
4. Prostate Biopsy Technique
Twelve-core prostate biopsies were performed in all patients, and all biopsies were performed by a single urologist. Each visible lesion was sampled with 4 MRI-targeted cores prior to the 12-core biopsy. MRI-targeted biopsy was performed with an MRI-target biopsy protocol that utilized the embedded side-fire method of the BK3000 ultrasound system (BK Medical, Peabody, MA, USA) and an image-based fusion program (BioJet; GeoScan, Lakewood Ranch, FL, USA).
5. Study Endpoints
The primary endpoint was the impact of PSA variation on the prostate cancer diagnosis of patients with PSA ≤20 ng/mL who underwent prebiopsy MRI. The secondary endpoint was the comparative accuracy of the conventional prostate cancer prediction models and our model on PSA variation.
6. Statistical Analysis
Continuous variables are expressed as median (interquartile range) or mean±standard deviation and categorical variables as the number of occurrences and frequency. Pearson χ2 test was utilized to compare continuous and categorical variables. Clinically significant prostate cancer was defined as prostate cancer with a Gleason grade group ≥2. Positive inflammation on biopsy was defined as inflammation confined to more than one core biopsy. Univariable logistic regression analyses were performed to identify potential parameters, such as age, family history, prostate biopsy history, DRE findings, prostate volume, 1st PSA measurement, PSA density, number of target lesions, and PSA variation. Considering the multicollinearity of prostate volume, PSA, and PSA density, we developed 4 multiple logistic regression prediction models. Model 1 included age, prostate volume, prostate biopsy history, and number of target lesions; model 2 included age, prostate volume, prostate biopsy history, number of target lesions, and PSA variation; model 3 included age, PSA density, prostate biopsy history, and number of target lesions; and model 4 included age, PSA density, prostate biopsy history, number of target lesions, and PSA variation. Receiver operating characteristic (ROC) curves and area under the ROC curves (AUCs) were used to calculate the performance of PSA variation as an independent predictor of prostate cancer. The AUCs of the 4 models were compared, and pairwise comparisons of the ROC curves were utilized to evaluate the predictive performance of individual and combined parameters. The optimal cutoff value for PSA variation was derived from these analyses. The optimal cutoff values were based on predefined values and analyzed according to the Youden index (sensitivity+specificity-1). All statistical comparisons were performed on IBM SPSS Statistics ver. 26.0 (IBM Co., Armonk, NY, USA) and MedCalc version 11.6 (MedCalc Software, Acacialaan, Oostende, Belgium). A p-value of < 0.05 was considered statistically significant.
RESULTS
1. Patient Demographics
Baseline characteristics of the study population are shown in Table 1. Of the 258 patients, 166 (64.3%) were diagnosed with prostate cancer. The prostate cancer (+) group included older patients (70.2 years vs. 64.7 years, p < 0.001), with higher PSA density, higher 1st PSA values, and lower prostate volumes than the prostate cancer (−) group. The prostate cancer (−) group was more likely to have no target lesions than the prostate cancer (+) group (16 [9.6%] vs. 49 [53.3%], p < 0.001) and were received less number of core biopsies (14.0±2.3 vs. 16.3±2.3, p < 0.001).
Table 1.
Baseline patient characteristics
| Characteristic | Prostate cancer (−) | Prostate cancer (+) | p-value |
|---|---|---|---|
| No. of patients | 92 (35.7) | 166 (64.3) | |
| Age (yr) | 64.7±8.4 | 70.2±7.7 | <0.001 |
| 1st PSA∗ (ng/mL) | 6.17 (4.63-9.03) | 6.53 (5.10-9.65) | 0.341 |
| PSA <10 ng/mL | 76 (82.6) | 129 (77.7) | 0.353 |
| Prostate volume (cm3) | 41.2 (31.6-52.8) | 28.1 (22.3-39) | <0.001 |
| PSA density (ng/mL/cm3) | 0.15 (0.11-0.21) | 0.23 (0.14-0.39) | <0.001 |
| Abnormal digital rectal examination findings, yes | 5 (5.4) | 8 (4.8) | 0.702 |
| Family history, yes | 27 (29.3) | 45 (27.1) | 0.829 |
| Prostate biopsy history, yes | 22/92 (23.9) | 22/166 (13.3) | 0.042 |
| 2nd PSA† (ng/mL) | 5.83 (4.07-8.4) | 6.39 (4.51-9.85) | 0.032 |
| Duration between 1st and 2nd PSA (day) | 20.5 (13.3-30) | 20 (13-33) | 0.142 |
| PSA variation‡ (ng/mL) | −0.27 (−1.27 to 0.40) | −0.09 (−0.87 to 0.79) | 0.006 |
| PSA variation‡ (−0.54 to 0.05 ng/mL) | 9 (9.8) | 40 (24.1) | 0.002 |
| %PSA variation§ (−8.23% to 6.47%) | 13 (14.1) | 37 (22.3) | 0.117 |
| No. of patients have target lesions | <0.001 | ||
| 0 | 49 (53.3) | 16 (9.6) | |
| 1 | 39 (42.4) | 127 (76.5) | |
| ≥2 | 4 (4.3) | 23 (14.9) | |
| Inflammation on biopsy, yes | 21 (22.6) | 28 (16.9) | 0.076 |
| Positive systemic biopsy | 0 (0) | 155 (93.4) | <0.001 |
| Gleason grade group | <0.001 | ||
| 1 | - | 32 (20.6) | |
| ≥2 | - | 123 (79.4) | |
| No. of positive biopsy cores | - | 3.9±2.8 | <0.001 |
| No. of target lesions | 72 (32.6) | 149 (67.4) | <0.001 |
| PI-RADS | |||
| 3 | 33 (63.5) | 19 (36.5) | |
| 4 | 35 (29.4) | 84 (70.6) | |
| 5 | 4 (8.0) | 46 (92.0) | |
| Gleason grade group | <0.001 | ||
| 1 | 0 (0) | 32 (21.5) | |
| ≥2 | 0 (0) | 117 (78.5) | |
| No. of total biopsy cores | 14.0±2.3 | 16.3±2.3 | <0.001 |
The prostate cancer (+) group had less PSA variation than the prostate cancer (−) group (−0.09 ng/mL vs. −0.27 ng/mL, p=0.006). The optimal cutoff of PSA variation was within −0.54 to 0.05 ng/mL. A higher proportion of patients with PSA variation had prostate cancer than no prostate cancer (40 [24.1%] vs. 9 [9.8%], p=0.002). There was no significant difference in the median duration between the 1st and 2nd PSA measurements between the groups (20.5 days vs. 20 days, p=0.142).
2. Identifying the Potential Predictors of Prostate Cancer
Age (odds ratio [OR], 1.09; p < 0.001), prostate volume (OR, 0.95; p < 0.001), PSA density (OR, 1.60; p < 0.001), prostate biopsy history (OR, 0.49; p=0.031), number of target lesions (1 vs. 0: OR, 9.98; p < 0.001; ≥2 vs. 0: OR, 17.61; p < 0.001), and PSA variation (−0.54 to 0.05 ng/mL) (OR, 2.93; p=0.007) were identified as potential predictors of prostate cancer (Table 2). The 1st PSA measurement, abnormal findings on DRE, and a positive family history of prostate cancer were not risk factors for a prostate cancer diagnosis.
Table 2.
Univariate analysis of the potential predictors of prostate cancer
| Variable | OR (95% CI) | p-value |
|---|---|---|
| Age | 1.09 (1.051-1.127) | <0.001 |
| 1st PSA∗ | 1.03 (0.965-1.108) | 0.341 |
| Prostate volume | 0.95 (0.934-0.969) | <0.001 |
| PSA density (0.1 unit) | 1.60 (1.286-2.002) | <0.001 |
| Abnormal digital rectal examination findings | 0.90 (0.509-1.574) | 0.701 |
| Family history | 0.88 (0.280-2.776) | 0.829 |
| Prostate biopsy history | 0.49 (0.252-0.937) | 0.031 |
| No. of target lesions | <0.001 | |
| 1 vs. 0 | 9.98 (5.109-19.465) | <0.001 |
| ≥2 vs. 0 | 17.61 (5.291-58.608) | <0.001 |
| PSA variation† (−0.54 to 0.05 ng/mL) | 2.93 (1.350-6.351) | 0.007 |
| %PSA variation‡ (−8.23% to 6.47%) | 1.89 (1.101-4.759) | 0.238 |
3. Comparison of 4 Prostate Cancer Prediction Models Regardless of PSA Variation
Four models were constructed based on the potential variables identified above (Table 3). Models 2 and 4 analyzed conventional variables and PSA variation, but PSA variation remained the significant factor for predicting prostate cancer. The AUC for prostate cancer was the highest in model 4 (0.870, 95% confidence interval [CI], 0.827-0.913) (Fig. 1). The AUC of model 4 was not significantly different from that of models 1 (0.862; 95% CI, 0.816-0.907) and 2 (0.867; 95% CI, 0.823-0.911) but was significantly different from that of model 3 (0.853; 95% CI, 0.806-0.899; p=0.009).
Table 3.
Multivariable analysis of the predictors of prostate cancer by model
| Variable | Model 1 | Model 2 | Model 3 | Model 4 | ||||
|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | p-value | OR (95% CI) | p-value | OR (95% CI) | p-value | OR (95% CI) | p-value | |
| Age | 1.09 (1.041-1.139) | <0.001 | 1.09 (1.041-1.139) | <0.001 | 1.07 (1.029-1.121) | 0.001 | 1.08 (1.031-1.126) | 0.001 |
| Prostate volume | 0.94 (0.923-0.964) | <0.001 | 0.95 (0.925-0.967) | <0.001 | - | - | - | - |
| PSA density (0.1 unit) | - | - | - | - | 1.80 (1.363-2.372) | <0.001 | 1.93 (1.442-2.573) | <0.001 |
| Prostate biopsy history | 0.41 (0.176-0.951) | 0.038 | 0.42 (0.178-0.992) | 0.048 | 0.31 (0.137-0.699) | 0.005 | 0.30 (0.130-0.704) | 0.006 |
| No. of target lesions | <0.001 | <0.001 | <0.001 | <0.001 | ||||
| 1 vs. 0 | 8.59 (4.007-18.421) | <0.001 | 8.81 (4.035-19.223) | <0.001 | 8.55 (3.967-18.416) | <0.001 | 9.02 (3.989-20.404) | <0.001 |
| ≥2 vs. 0 | 10.70 (2.948-38.818) | <0.001 | 12.43 (3.346-46.147) | <0.001 | 13.526 (3.719-49.196) | <0.001 | 18.02 (4.674-69.444) | <0.001 |
| PSA variation∗ (−0.54 to 0.05 ng/mL) | - | - | 2.60 (1.039-6.509) | 0.041 | - | - | 4.95 (1.937-12.643) | <0.001 |
Fig. 1.
Area under the curve of 4 prostate cancer prediction models. Model 1, age, prostate volume, prostate biopsy history, and number of target lesions; model 2, model 1 plus prostate-specific antigen (PSA) variation; model 3, age, PSA density, prostate biopsy history, and number of target lesions; model 4, model 3 plus PSA variation.
4. Predictors of Prostate Cancer Among Patients With PSA < 10 ng/mL
We performed a subgroup analysis of patients with PSA < 10 ng/mL and considered age, prostate volume, PSA density, number of target lesions, and PSA variation as potential predictors of prostate cancer (Table 4). Four models were constructed for the subgroup analysis (Table 5). Models 2 and model 4 analyzed PSA variation (−0.54 to 0.05 ng/mL) as the main predictor of prostate cancer, in addition to the conventional variables. The AUC for prostate cancer was the highest in model 4 (0.860; 95% CI, 0.809-0.911) (Fig. 2). There was no significant difference in the AUC values between models 2 (0.856; 95% CI, 0.803-0.909) and 4, but there were significant differences in the AUC values between models 1 (0.844; 95% CI, 0.789-0.900; p=0.003) and 3 (0.841; 95% CI, 0.789-0.895; p=0.001) and model 4.
Table 4.
Univariate analysis of the potential predictors of prostate cancer in patients with PSA <10 ng/mL
| Variable | OR (95% CI) | p-value |
|---|---|---|
| Age | 1.10 (1.055-1.142) | <0.001 |
| 1st PSA∗ | 1.01 (0.879-1.150) | 0.940 |
| Prostate volume | 0.95 (0.929-0.971) | <0.001 |
| PSA density (0.1 unit) | 2.21 (1.503-3.237) | <0.001 |
| Abnormal digital rectal examination findings | 0.95 (0.507-1.780) | 0.873 |
| Family history | 0.88 (0.240-3.216) | 0.884 |
| Prostate biopsy history | 0.69 (0.310-1.518) | 0.353 |
| No. of target lesions | <0.001 | |
| 1 vs. 0 | 8.02 (3.873-16.605) | <0.001 |
| ≥2 vs. 0 | 13.93 (4.051-47.894) | <0.001 |
| PSA variation† (−0.54 to 0.05 ng/mL) | 3.11 (1.408-6.864) | 0.005 |
| %PSA variation‡ (−8.23% to 6.47%) | 1.33 (0.931-3.416) | 0.468 |
Table 5.
Multivariable analysis of the predictors of prostate cancer in patients with PSA <10 ng/mL by model
| Variable | Model 1 | Model 2 | Model 3 | Model 4 | ||||
|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | p-value | OR (95% CI) | p-value | OR (95% CI) | p-value | OR (95% CI) | p-value | |
| Age | 1.08 (1.029-1.132) | 0.002 | 1.08 (1.029-1.132) | 0.002 | 1.09 (1.036-1.140) | 0.001 | 1.09 (1.037-1.143) | 0.001 |
| Prostate volume | 0.94 (0.915-0.964) | <0.001 | 0.94 (0.919-0.968) | <0.001 | - | - | - | - |
| PSA density (0.1 unit) | - | - | - | - | 3.18 (1.891-5.344) | <0.001 | 3.35 (1.946-5.759) | <0.001 |
| No. of target lesions | <0.001 | <0.001 | <0.001 | <0.001 | ||||
| 1 vs. 0 | 6.27 (2.729-14.393) | <0.001 | 6.37 (2.711-14.981) | <0.001 | 6.60 (2.822-15.451) | <0.001 | 6.55 (2.690-15.937) | <0.001 |
| ≥2 vs. 0 | 9.37 (2.469-35.575) | 0.001 | 11.16 (2.857-43.564) | 0.001 | 12.05 (3.070-47.292) | 0.001 | 15.76 (3.805-65.311) | <0.001 |
| PSA variation∗ (−0.54 to 0.05 ng/mL) | - | - | 2.77 (1.097-7.005) | 0.031 | - | - | 4.29 (1.640-11.206) | 0.003 |
Fig. 2.
Area under the curve of 4 prostate cancer prediction models for patients with prostate-specific antigen <10 ng/mL. Model 1, age, prostate volume, prostate biopsy history, and number of target lesions; model 2, model 1 plus prostate-specific antigen (PSA) variation; model 3, age, PSA density, prostate biopsy history, and number of target lesions; model 4, model 3 plus PSA variation.
DISCUSSION
Age, family history, DRE findings, prostate volume, PSA level, and PSA density are significant predictors of a prostate cancer diagnosis.9 While elevated PSA findings are associated with a prostate cancer diagnosis, it is not a cancer-specific marker. Most patients with elevated PSA levels do not have prostate cancer because other conditions, such as prostatic inflammation, perineal trauma, sexual activity, and, rarely, other tumors, present with elevated PSA.3, 10 Carter et al.11, 12 proposed the benefits of utilizing PSA velocity in prostate cancer prediction; however, this study assessed PSA over an interval of several years, which resulted in fluctuating PSA values. Clinicians had difficulty reproducing these results when PSA levels were measured at 1- and 2-year intervals.13 Interestingly, Park et al.7 reported that greater PSA fluctuation decreased the detection of prostate cancer.
Prebiopsy MRI and the standardization of MRI interpretation through the PI-RADS increased the detection of clinically significant prostate cancer,14, 15 but identifying biomarkers for prostate cancer may make the diagnostic method more cost-effective.16-18 To the best of our knowledge, this is the first study that fully investigated the role of PSA fluctuation in differentiating benign and malignant tumors among patients with prebiopsy MRI. This study collected PSA data at the time when a prostate biopsy was recommended and performed (median, 20 days). Our data suggested that a prostate cancer diagnosis was less likely with large PSA fluctuations, which correlated with previous research that examined PSA fluctuations in 6-month intervals.7 We evaluated PSA fluctuations over a short-term 1-month follow-up period. Our data may have been more susceptible to greater variations from confounding factors that could abruptly increase PSA levels. Based on our results, clinicians should recommend another PSA measurement within 1 month of the 1st measurement among patients with PSA ≤20 ng/mL who underwent prebiopsy MRI to identify which patients can benefit best from a biopsy.
Chronic inflammation results in PSA variation. The role of antibiotic treatment in decreasing PSA levels prior to a prostate biopsy has been reported. A study demonstrated that antibiotic therapy decreased PSA levels in patients with elevated PSA19; however, antibiotic treatment is not part of the standard of care prior to prostate biopsies, because it does not decrease PSA levels to the normal range. In contrast, a systematic review proposed that antibiotic therapy did not result in changes in PSA levels. Decreased PSA levels after antibiotic treatment did not reduce the risk for prostate cancer.20 Unnecessary antibiotic use contributes to the problem of antibiotic overuse and abuse and increases the possibility of complications caused by resistant strains. The practical use of prophylactic antibiotics in prostate cancer is limited.
According to the NCCN guidelines, the detection rate of prostate cancer among patients with PSA < 10 ng/mL is 30%-35%.21 In our study, the detection rate in patients with PSA < 10.0 and ≤20 ng/mL was 62.9% (129 of 205) and 64.3% (166 of 258), respectively. Our results were higher than those of previous studies, which may be due to the availability of prebiopsy MRI data. MRI findings in our study were classified according to the PI-RADS,8 and utilizing PI-RADS more often increases prostate cancer detection. In our study, 36.5% (19 of 52), 70.6% (84 of 119), and 92% (46 of 50) of patients with PI-RADS 3, PI-RADS-4, and PI-RADS 5 scores, respectively, had prostate cancer, similar to the findings of a previous study.22 However, this study did not examine the association between the PI-RADS score and diagnosis of clinically significant prostate cancer but examined the possible predictors of a prostate cancer diagnosis. Therefore, the number of target lesions visible on prebiopsy MRI was used as potential variable in our analysis.
Several prostate cancer risk calculators, such as the European Randomized Study of Screening for Prostate Cancer and Prostate Cancer Prevention Trial risk calculators, have been proposed. These consider easily available clinical parameters, such as prostate volume, PSA, and DRE and prostate ultrasound findings, to further improve cancer detection and reduce unnecessary biopsies.23, 24 Several calculators have also examined the role of MRI findings in predicting prostate cancer among biopsy-naïve and biopsy-negative patients. Lee et al.25 developed a risk prediction model based on age, PSA density, primary biopsy findings, and PI-RADS scores from bi-parametric MRI. The AUC value for prostate cancer in this cohort was 0.870. Among the 4 models in our study of patients with PSA ≤20 ng/mL, the model that was developed based on age, PSA density, prostate biopsy, number of target lesions on multiparametric MRI, and PSA variation demonstrated the highest performance (0.870) and corresponded with previous literature.
Although this study demonstrated clinically useful results, it has some limitations. First, prediction tools require external validation through a multicenter study to assess their wider applicability. This study was a retrospective study that examined a small sample size. Our findings need to be validated in studies with more patients. Second, PSA is a known predictor of prostate cancer, but it was not included as a potential variable in this study. The authors theorized that there were 11 patients (6.6%) who demonstrated positive prostate biopsy results but negative systemic biopsy results because PSA was excluded as a potential variable. Including PSA in prostate cancer prediction models may result in different findings.
CONCLUSIONS
Conventional variables such as PSA, PSA density, and PSA velocity, have limitations in prostate cancer diagnosis. Prostate MRI is a powerful predictor of prostate cancer, however, it is not cost-effective. This study demonstrated that 2 PSA level measurements within 1 month can predict prostate cancer diagnosis among patients who underwent prebiopsy MRI. Our study may contribute to the selection of patients who require prostate MRI or prostate biopsy.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
