*Article* **Prognostic Index for Predicting Prostate Cancer Survival in a Randomized Screening Trial: Development and Validation**

**Subas Neupane 1,\*, Jaakko Nevalainen <sup>1</sup> , Jani Raitanen 1,2, Kirsi Talala <sup>3</sup> , Paula Kujala <sup>4</sup> , Kimmo Taari <sup>5</sup> , Teuvo L. J. Tammela <sup>6</sup> , Ewout W. Steyerberg 7,8 and Anssi Auvinen <sup>1</sup>**


**Simple Summary:** A prognostic index for predicting survival of localized prostate cancer (PCa) up to 15 and 20 years was developed. The prognostic index performed well for predicting PCa survival among screened and non-screened men. The performance of the prediction model was superior to the European Association of Urology (EAU) risk groups as well as a modified cancer of prostate risk assessment (CAPRA) risk score. We further constructed a simplified risk score in an unscreened population, using the three most relevant predictors. The simplified risk score was applied to predict PCa survival at 10 years from diagnosis to provide more accurate risk estimation as the basis for decision making.

**Abstract:** We developed and validated a prognostic index to predict survival from prostate cancer (PCa) based on the Finnish randomized screening trial (FinRSPC). Men diagnosed with localized PCa (*N* = 7042) were included. European Association of Urology risk groups were defined. The follow-up was divided into three periods (0–3, 3–9 and 9–20 years) for development and two corresponding validation periods (3–6 and 9–15 years). A multivariable complementary log–log regression model was used to calculate the full prognostic index. Predicted cause-specific survival at 10 years from diagnosis was calculated for the control arm using a simplified risk score at diagnosis. The full prognostic index discriminates well men with PCa with different survival. The area under the curve (AUC) was 0.83 for both the 3–6 year and 9–15 year validation periods. In the simplified risk score, patients with a low risk score at diagnosis had the most favorable survival, while the outcome was poorest for the patients with high risk scores. The prognostic index was able to distinguish well between men with higher and lower survival, and the simplified risk score can be used as a basis for decision making.

**Keywords:** prognostic index; prediction model; prostate cancer; mortality; screening trial

## **1. Introduction**

Prostate cancer (PCa) presents a wide spectrum of behavior, from indolent to highly aggressive [1]. Treatment decisions are required at several phases during the course of the disease [2]. Optimal disease management should avoid both excessively aggressive

Raitanen, J.; Talala, K.; Kujala, P.; Taari, K.; Tammela, T.L.J.; Steyerberg, E.W.; Auvinen, A. Prognostic Index for Predicting Prostate Cancer Survival in a Randomized Screening Trial: Development and Validation. *Cancers* **2021**, *13*, 435. https:// doi.org/10.3390/cancers13030435

**Citation:** Neupane, S.; Nevalainen, J.;

Received: 24 November 2020 Accepted: 20 January 2021 Published: 24 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

treatment in patients who are not at high risk of disease progression and ineffective management of aggressive disease leading to treatment failure and development of metastatic disease. However, the dilemma expressed by Dr. Willet Whitmore persists for PCa: "Is cure possible in those for whom it is necessary—and is cure necessary in those for whom it is possible".

Several prediction methods for the prognosis of localized PCa have been presented as tabulations [3–5], nomograms [6–8], risk groups [9,10] and decision trees [11,12]. However, these methods have mainly divided patients into 3–4 broad risk groups and used biochemical recurrence (BCR) as the end-point rather than PCa death [3,4,9,13,14]; furthermore, few are based on a modern setting with largely prostate-specific antigen (PSA)-detected cases. Prognostic prediction models based on a limited number of relevant clinical characteristics can offer evidence-based input to inform medical practice [15].

We developed and validated a full prognostic index for predicting survival of localized PCa up to 15 and 20 years. We also developed a simplified risk score tool for use at diagnosis and applied it to predict survival at 10 years.

#### **2. Results**

Prognostic factors associated with PCa death included age at diagnosis, trial arm, PSA at diagnosis, European Association of Urology (EAU) risk group, treatment modality, mode of detection and biochemical recurrence (Table 1). All prognostic factors except comorbidity index showed a statistically significant difference between men who died from PCa and others.

Older age at diagnosis was marginally associated with lower PCa mortality at 9–20 years (Table S1). PSA at diagnosis was associated with higher PCa mortality in the first and the last development periods. PCa mortality was higher in the intermediateto high-risk groups compared to the low-risk group in all three follow-up periods. Men treated with radical prostatectomy had the most favorable survival, with the exceptions of radiotherapy and observation in the early follow-up. Biochemical recurrence also predicted increased probability of PCa death, with the largest effect after the first three years.

The distributions of the prognostic index differed markedly across the development periods (Figure 1a–c). The graphs illustrate lower prognostic index (PI) values (indicating worse survival) for men who died from PCa than those who did not die from PCa (cumulative frequency for the former group shown as the dotted blue line above the latter group, shown as the solid red line). PCa mortality increased with increasing values of prognostic index in the initial follow-up, but after 9 years, a clear excess mortality was limited to the two highest quintiles.

The prognostic indices were associated with PCa mortality in all EAU risk groups, though the difference was not obvious in the initial three-year period with low mortality (Figure S1a–c). The prognostic index provided incremental information, especially in the intermediate- and high-risk groups, and its contribution was accentuated with follow-up. Furthermore, the prognostic index also predicted survival within the low-risk group in the longer follow-up.



IQR: Interquartile range. † Includes men alive and deaths due to causes other than PCa. ‡ *p*-values for categorical variables were derived from a chi-square test, whereas for continuous variable, using ANOVA test.

**Figure 1.** Cumulative distribution of prognostic index (PI) values for men who died due to prostate cancer (PCa) and those who did not during (**a**) the development period of 0–3 year (**b**) the development period of 3–9 years and (**c**) the development period of 9–20 years. The distribution of men is described as a cumulative frequency across values of the PI from low (indicating worse survival) to high (indicating favorable survival). **Figure 1.** Cumulative distribution of prognostic index (PI) values for men who died due to prostate cancer (PCa) and those who did not during (**a**) the development period of 0–3 year (**b**) the development period of 3–9 years and (**c**) the development period of 9–20 years. The distribution of men is described as a cumulative frequency across values of the PI from low (indicating worse survival) to high (indicating favorable survival).

> The prognostic indices were associated with PCa mortality in all EAU risk groups, though the difference was not obvious in the initial three-year period with low mortality (Figure S1a–c). The prognostic index provided incremental information, especially in the intermediate- and high-risk groups, and its contribution was accentuated with follow-up. Furthermore, the prognostic index also predicted survival within the low-risk group in The observed mortality matched the expected one very well at all levels of the PI during each development and validation period (Table 2). In all follow-up periods, including the validation periods, the highest quintiles of the PI showed the highest observed and expected PCa mortalities. The differences between the lower quintiles were, however, relatively small.

> the longer follow-up. The observed mortality matched the expected one very well at all levels of the PI during each development and validation period (Table 2). In all follow-up periods, including the validation periods, the highest quintiles of the PI showed the highest observed and expected PCa mortalities. The differences between the lower quintiles were, however, Less than one third of the patients remained in the initial quintile from the 0–3-year development period to the 3–6-year validation period; particularly, progression from Q1 to Q2 and Q4 to Q5 was common (Table S2). The most frequent transition was by one step up, likely due to biochemical recurrence. However, downward transitions also occurred, reflecting changes in the regression coefficients of the variables used in the model.

> relatively small. Less than one third of the patients remained in the initial quintile from the 0–3-year development period to the 3–6-year validation period; particularly, progression from Q1 to Q2 and Q4 to Q5 was common (Table S2). The most frequent transition was by one step up, likely due to biochemical recurrence. However, downward transitions also occurred, reflecting changes in the regression coefficients of the variables used in the model. The predictive ability of the prognostic indices (Figure 2) did not substantially differ The predictive ability of the prognostic indices (Figure 2) did not substantially differ between the development and validation periods: area under the curve (AUC) 0.84 (95% confidence interval, CI 0.77–0.90) for the initial development period (0–3 years) and 0.83 (0.79–0.88) for the corresponding validation period (3–6 years). Similarly, for the second development period, the AUC was 0.84 (0.81–0.88), and it was 0.83 (0.79–0.88)for the subsequent validation period. For the 9–20-year development period, the AUC was 0.83 (0.79–0.86).

> between the development and validation periods: area under the curve (AUC) 0.84 (95% confidence interval, CI 0.77–0.90) for the initial development period (0–3 years) and 0.83 (0.79–0.88) for the corresponding validation period (3–6 years). Similarly, for the second


**Table 2.** Expected and observed probability and number of PCa deaths in the development and validation periods for quintiles of prognostic index.

**Prognostic Index Quintiles** 

**Men** 

**Number of Deaths †**

0.86).

**Figure 2.** (**a**) Receiver operating characteristic (ROC) curves for the development period of 0–3 year and validation period of 3–6 years and the (**b**) development period of 3–9 years and validation period of 9–15 years. **Figure 2.** (**a**) Receiver operating characteristic (ROC) curves for the development period of 0–3 year and validation period of 3–6 years and the (**b**) development period of 3–9 years and validation period of 9–15 years.

**Table 2.** Expected and observed probability and number of PCa deaths in the development and validation periods for quintiles of prognostic index. **Development Period Validation Period Observed Expected Men Observed Expected**  A simplified risk score at diagnosis was calculated among patients in the control arm based on the regression coefficients of three categorical parameters (age at diagnosis, PSA at diagnosis and EAU risk group) to allow easy clinical application. The simplified risk score uses a granular scale of 0 to 100, with higher score indicating increasing risk. The predictive ability of the simplified risk score at diagnosis was 0.68 (0.63–0.73) (Figure S2).

development period, the AUC was 0.84 (0.81–0.88), and it was 0.83 (0.79–0.88) for the subsequent validation period. For the 9–20-year development period, the AUC was 0.83 (0.79–

**Probability ‡ PCa Deaths Probability ‡ PCa Deaths Probability ‡ PCa Deaths Probability ‡ PCa Deaths**  0–3 year 3–6 years Q1 1436 67 0.001 1 0.001 1 1323 0.001 1 0.001 1 Q2 1446 87 0.003 4 0.002 3 1352 0.005 7 0.003 4 The full prognostic model displayed superior discrimination (*p <* 0.001) compared to the EAU risk group alone in all three periods (AUC for EAU risk group: 0.61, 0.53 and 0.39 during the follow-up periods of 0–3, 3–9 and 9–20 years, respectively). The simplified risk score showed superior discrimination only in the 9–20-year period (*p* < 0.001).

Q3 1438 44 0.003 4 0.003 4 1354 0.003 4 0.004 5 Q4 1436 66 0.002 3 0.006 9 1299 0.010 13 0.008 10 Q5 1465 175 0.031 46 0.028 41 1396 0.044 61 0.044 62 The simplified risk score at diagnosis (Table 3) was used to calculate the predicted PCa survival at 10 years (Figure 3). Overall, men with a high-risk score at diagnosis had poorer survival.

Total 7221 439 0.008 58 0.008 58 6724 0.013 86 0.012 82

Q1 1319 118 0.002 3 0.002 3 1133 0.002 2 0.001 1 Q2 1353 165 0.007 9 0.004 6 1155 0.005 6 0.003 4 Q3 1356 150 0.005 7 0.007 10 1147 0.004 4 0.004 5 Q4 1319 176 0.018 24 0.014 19 1152 0.009 10 0.011 13 Q5 1377 95 0.074 102 0.075 103 1248 0.054 69 0.047 61 Total 6724 704 0.022 145 0.021 141 5875 0.016 91 0.014 84

3–9 years 9–15 years

9–20 years

Q1 1122 26 0.002 2 0.002 2


**Table 3.** Scoring rules for constructing the simplified risk score at diagnosis for PCa survival based on the complementary log–log regression model among PCa-diagnosed cases in the control arm. Q2 1148 55 0.004 5 0.004 5 Q3 1133 93 0.007 8 0.006 7

*Cancers* **2021**, *12*, x 6 of 12

† Regression coefficients from complementary log–log model. § Risk score at baseline calculated by dividing each beta coefficient by the sum of the highest beta coefficient of each variable and then multiplied by 100 (scores are rounded). The simplified risk score at diagnosis (Table 3) was used to calculate the predicted PCa survival at 10 years (Figure 3). Overall, men with a high-risk score at diagnosis had poorer survival.

**Figure 3.** Predicted probability of PCa survival at 10 years from diagnosis among controls in the Finnish randomized screening trial. **Figure 3.** Predicted probability of PCa survival at 10 years from diagnosis among controls in the Finnish randomized screening trial.

We calculated the risk score at diagnosis and 10-year survival probability among patients in both study arms, as well as performing a complete case analysis in the control arm only as a sensitivity analysis (Tables S3 and S4 and Figure S3), with no substantial difference in the results. We calculated the risk score at diagnosis and 10-year survival probability among patients in both study arms, as well as performing a complete case analysis in the control arm only as a sensitivity analysis (Tables S3 and S4 and Figure S3), with no substantial difference in the results.

**Table 3.** Scoring rules for constructing the simplified risk score at diagnosis for PCa survival based on the complementary log–log regression model among PCa-diagnosed cases in the control arm.

**Characteristics Categories β † Risk Score §**

 61–70 1.00 35 71–75 0.72 25

 Low Ref 0 Intermediate 0.75 30 High 1.46 50 **†** Regression coefficients from complementary log–log model. **§** Risk score at baseline calculated by dividing each beta coefficient by the sum of the highest beta coefficient of each variable and then

≤60 1.12 40

≥76 Ref 0

≤19.9 Ref 0 ≥20.0 0.22 10

*Cancers* **2021**, *12*, x 7 of 12

Age at diagnosis (years)

PSA at diagnosis (ng/mL)

EAU risk group

multiplied by 100 (scores are rounded).

The decision curve analysis for simplified risk score at diagnosis is presented in Figure 4. The graph gives the expected net benefit per patient relative to no PCa mortality in any patient (Treat None). The risk prediction model is of benefit for a reasonable range of 3–25%: the curve diverges only at the threshold probability of about 3%. However, the net benefit of the model is about the same as the net benefit of Treat All below 3%. The decision curve analysis for simplified risk score at diagnosis is presented in Figure 4. The graph gives the expected net benefit per patient relative to no PCa mortality in any patient (Treat None). The risk prediction model is of benefit for a reasonable range of 3–25%: the curve diverges only at the threshold probability of about 3%. However, the net benefit of the model is about the same as the net benefit of Treat All below 3%.

**Figure 4.** Decision curve analysis for the simplified risk score at diagnosis. The dotted green line is the simplified risk score (prediction model), the blue solid line assumes PCa mortality in all patients and red solid line assume no patient deaths due to PCa. Threshold probability on the *x*-axis is the level of diagnostic certainty above which the patient would choose to be treated. **Figure 4.** Decision curve analysis for the simplified risk score at diagnosis. The dotted green line is the simplified risk score (prediction model), the blue solid line assumes PCa mortality in all patients and red solid line assume no patient deaths due to PCa. Threshold probability on the *x*-axis is the level of diagnostic certainty above which the patient would choose to be treated.

#### **3. Discussion**

The full prognostic index with seven variables predicted PCa mortality with a performance superior to that of the EAU risk group (AUC 0.83–0.84 vs. 0.61). The robustness of the results was confirmed by sensitivity analyses including both trial arms and omitting patients with missing data.

Our model correctly predicted the 3-year survival of 99% for the patients in the lowest quintile and 97% for those in the highest quintile of the prognostic index. We divided the follow-up time into several segments due to lack of proportionality across the entire follow-up. In the second development period, 6-year survival was 99% among men in the lowest quintile, while it was 89% among men in the highest quintile.

Primary treatment predicted PCa mortality already in the early follow-up. The effect of biochemical recurrence increased with follow-up. Other factors did not show clear changes over the follow-up. Similar to earlier findings [16,17], we found no strong impact of comorbidity at baseline on PCa-specific survival. No earlier PCa survival prediction models have utilized the context of a randomized screening trial. Our approach enhances the applicability of the prognostic index to the current setting with widespread PSA testing.

A simplified risk score at diagnosis was developed using three predictors selected based on their importance and interpretability in the prognostic index model. The simplified risk score is based on a granular scale ranging from 0 to 100 with three categorical variables and can be adopted in daily clinical practice with minimal data entry.

PSA was used as a component of our prognostic index, despite being a part of the EAU risk group, because the analysis revealed that its impact was not fully captured in the EAU classification.

Our findings are mainly in line with earlier prediction models, although patient populations, outcomes and methodological approaches differ between studies. The performance of our simplified risk score tool at diagnosis was superior to that of the D'Amico risk classification and EAU risk group (AUC 0.68 vs. 0.59 and 0.61, respectively). The simplified risk score also outperformed an abridged version of the cancer of prostate risk assessment (CAPRA) risk score (AUC 0.59), though we were unable to incorporate percentage cancer in biopsy for estimating the CAPRA score in our analysis due to lack of data. Furthermore, all patients in our study were aged > 50 years at diagnosis [7].

An earlier study presented a clinical–genomic risk group classification for localized prostate cancer that showed a 10-year rate for distant metastases of 3.5% for a low-risk group, while it was 58% in a high-risk group in the training cohort, and the corresponding values for the validation cohort were 0% and 63%, respectively [9]. That risk group required extensive genomic data, restricting its applicability. Peters et al. [18] developed a prediction model for recurrent disease with three categories, which showed 60% biochemical diseasefree survival and 40% composite end-point-free survival at 4 years for a low-risk group, while the corresponding figures for a high-risk group were 7% and 0%, respectively.

Decision curve analysis shows the benefit of use of the prediction model (simplified risk score at diagnosis). A net benefit was found for a reasonable range of 3–25%: the curve diverged only at the threshold probability of about 3%.

Our study had also some limitations. The patients were treated during a period spanning from the 1990s into the 2010s and treatment modalities have evolved over time. However, long follow-up is required due to the favorable prognosis to capture the full natural course of the disease and accrue a sufficient number of PCa deaths. Completeness of data was high, with the highest proportion of missing data for PSA, at 3%. However, we used imputation in the main analysis, and sensitivity analyses of complete cases yielded comparable results, suggesting that this did not affect our findings. We incorporated biochemical relapse in the prognostic index, even though in the clinical setting, it is not available at diagnosis. On the other hand, its inclusion enhances the applicability of our results in prognostic prediction after the initial phase and post-primary treatment.

#### **4. Material and Methods**

We used data from the Finnish Randomized Study of Screening for Prostate Cancer (FinRSPC). The trial protocol and main results have been described elsewhere [19]. In brief, a random sample of 8000 men aged 55–67 years were allocated to the screening arm (SA) annually in 1996–1999 and the remaining men (48,278 in total) formed the control arm that received no intervention. Men in the screening arm were invited for screening based on serum PSA. Screen-positive men (defined as those with PSA ≥ 4.0 ng/mL or PSA 3.0–3.9 ng/mL with free–total PSA ratio < 0.16) were referred to a local urological clinic for diagnostic examinations including transrectal ultrasound-guided biopsy. The second screening round was conducted four years later, and the final one after 8 years. Men aged > 71 years, those diagnosed with prostate cancer and men who had emigrated from the study area were no longer invited.

All men diagnosed with localized prostate cancer between randomization and the end of 2015 were included in this analysis (N = 7042). The follow-up for the primary analysis started at diagnosis and ended at death, emigration or the common closing (31 December 2015). Death from prostate cancer was the end-point in the analysis, with underlying causes of death obtained from Statistics Finland.

Information on tumor, lymph node, and metastasis (TNM) stage and Gleason score were abstracted from medical records. For previous cases, Gleason scores were revised according to the 2002 system by two pathologists. PSA at diagnosis was used for all men. Information on biochemical recurrence was obtained from laboratory databases.

Biochemical recurrence (BCR) was defined as PSA reaching at least 0.2 ng/mL in two measurements after prostatectomy, while BCR after radiotherapy was defined as a rise in PSA by at least 2.0 ng/mL above the lowest level (nadir). A modified version of the Charlson comorbidity index [20] was constructed based on hospital inpatient episodes obtained from the nationwide hospital discharge registry and categorized into no versus any comorbidity (score 0 versus 1 to 8) [21]. Prognostic risk group for PCa survival at diagnosis was classified as low, moderate and high, according to the European Association of Urology (EAU) criteria [10]. Low-risk PCa was defined as stage T1–T2a with Gleason score < 7 and PSA < 10 ng/mL; intermediate risk as T1–T2b with either Gleason 7 or PSA 10–20; high risk was stage T1–T2c with either Gleason > 7 or PSA 20–100, or T2c.

Primary treatment was retrieved from medical records and classified as radical prostatectomy, curative radiation therapy (external beam or brachytherapy), endocrine therapy (luteinizing hormone-releasing hormone agonist/antagonist, anti-androgen, or both or surgical castration), observation (watchful waiting or active surveillance) or no treatment.

#### *4.1. Ethical Issues*

Helsinki and Tampere University Hospital Ethics committees reviewed the study protocol (tracking number R10167). Cancer registry data were obtained with permission from the National Institute for Health and Welfare (Dnro THL/1601/5.05.00/2015). Written informed consent was obtained from the men participating in the screening arm.

#### *4.2. Statistical Analysis*

For the preliminary investigation, the proportional hazards assumption of the factors in the Cox regression was evaluated by graphical examination in log–log plots. These plots formed approximate parallel straight lines as required, except those for primary treatment which crossed each other. For this reason, we divided the follow-up time into three periods (0–3, 3–9 and 9–20 years) to model the effect of the full prognostic index separately during each period.

We used a same set of variables (age at diagnosis, study arm, PSA at diagnosis, EAU risk group, comorbidity index, primary treatment and biochemical recurrence) in a complementary log–log regression model, identified by a stepwise forward selection with *p* = 0.10 as the cut-off, in each of the three periods. Only statistically significant interaction terms (5% level) were included in the final model. The prognostic index (PI) was then derived as a linear combination of the variables, including interaction terms and their coefficients from the regression model. We generated prognostic indices separately for the three follow-up periods, hereafter called development periods, using regression coefficients estimated from the complementary log–log models. The probabilities of PCa death during each development period using prognostic indices were calculated.

Missing values (3.1% in PSA and 2.6% in the EAU risk group) were imputed using a multiple imputation by chained equations (MICE) algorithm, assigning multiple likely values from a predicted distribution based on association with other variables [22]. Multiple imputation creates multiple copies of the dataset, which are analyzed separately. Finally, the results were appropriately combined [23].

To avoid overfitting and overestimation of the predictive ability, we validated the results by applying them to a subsequent follow-up period for the first two development periods (i.e., prognostic index derived from the development period of 0–3 years from diagnosis to predict survival during the validation period of 3–6 years, and the index derived from the development period of 3–9 years to predict survival during years 9–15) [24]. Expected and observed probabilities and numbers of PCa deaths were calculated for each development and validation period. Expected probability of PCa death for the validation period was calculated as the inverse of the complementary log–log transformation.

Reclassification probabilities of men in the quintiles of prognostic index for the first development period (0–3 years) and validation period (3–6 years) were calculated. Moreover, we presented the distribution and the mean values of full prognostic indices by risk group

for all development periods. Cumulative distribution of prognostic index values according to the survival status of the patients for all three development periods was plotted.

Receiver operating characteristic (ROC) was calculated for the test and validation periods to illustrate sensitivity and specificity. The area under the curve (AUC) was calculated to assess the discriminative power of the prediction models. We further developed a simplified risk score at diagnosis using the information at diagnosis (age at diagnosis, PSA at diagnosis and EAU risk group) for the control arm only to avoid lead-time by screening. Only three variables were selected based on their importance and interpretability in the prognostic index model. The simplified risk score was then used to calculate the predicted probability of 10-year PCa survival and was presented graphically. We further developed the decision curve analysis to determine the clinical usefulness of a simplified risk score at diagnosis by quantifying the net benefits at different threshold probabilities.

As a sensitivity analysis, we calculated the risk score at diagnosis based on data in both study arms. Furthermore, we performed a complete case analysis in the control arm by calculating the risk score and predicted 10-year PCa survival to examine the potential influence of imputation.

Analyses were performed using Stata Statistical Software version 16.0 (StataCorp, College Station, TX, USA) and IBM SPSS Statistics 23 (IBM Corp., Armonk, NY, USA).

#### **5. Conclusion**

The prognostic index accurately predicted prostate cancer survival at follow-up reaching 20 years. A simplified risk score at diagnosis using the three most relevant parameters to predict the survival at 10 years can be helpful for providing more accurate risk estimation as the basis for decision making. However, our prediction model requires further external validation.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-6 694/13/3/435/s1. Table S1: Prognostic factors for prostate cancer mortality from the prediction model. Estimates from the complementary log–log regression models with their 95% confidence intervals in different time periods. Table S2: Distribution of prognostic index quintiles for the development period 0–3 years reclassified in new prognostic index quintiles for the validation period 3–6 years. Table S3: Risk score at diagnosis for PCa survival based on the complementary log–log regression model among all men with PCa diagnosis. Table S4: Risk score at diagnosis for PCa survival based on the complementary log–log regression model among men with PCa diagnosis in the control arm (complete case analysis). Figure S1: Box plots showing the distribution of the prognostic index by EAU risk groups and by period (a) 0 to 3 years, (b) 3 to 9 years and (c) 9 to 20 years. Figure S2: ROC curve for the simplified risk score at diagnosis. Figure S3: Predicted probability of PCa survival at 10-year from diagnosis among controls in a Finnish randomized screening trial (complete case analysis).

**Author Contributions:** Study concept: A.A., J.N. and S.N. Study design: A.A., T.L.J.T. and K.T. (Kimmo Taari). Data acquisition: J.R., J.N. and S.N. Quality control of data and algorithms: S.N., J.N., J.R. and K.T. (Kirsi Talala). Data analysis and interpretation: S.N., J.R., J.N. and A.A. Statistical analysis: S.N. and J.R. Article preparation: S.N., A.A. and J.N. Article editing: S.N., A.A., K.T. (Kirsi Talala), J.R., J.N., E.W.S. and P.K. Article review: All authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was funded in part by competitive state research funding administered by Expert Responsibility area of Tampere University Hospital, grants 9N064 and 9R002, the Finnish Academy (grant number #260931), and the Cancer Society of Finland grant to Prof. Anssi Auvinen.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of Helsinki and Tampere University Hospital (protocol code R10167 and date of approval 18.04.1995).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Conflicts of Interest:** The authors have declared no conflict of interest.

## **References**


**Takahiro Kimura 1,\*, Shun Sato <sup>2</sup> , Hiroyuki Takahashi <sup>2</sup> and Shin Egawa <sup>1</sup>**


**Simple Summary:** The incidence of prostate cancer (PC) is statistically biased due to the increase in prostate-specific antigen (PSA) screening and the accuracy of national cancer registration systems. However, studies on latent PC provide less biased information. This comprehensive review included studies evaluating latent PC in several countries. The prevalence of latent PC has been stable since 1950 in Western countries, but it has increased over time in Asian countries. Latent PC in Asian men has increased in prevalence and is higher in grade. This increase occurred not only due to the increase in PSA screening, but also due to increasing adoption of a Westernized lifestyle. Racial differences between Caucasian and Asian men may also explain the tumor location of latent PC. The autopsy findings in patients with latent PC included a significant proportion of high grade and stage cancers, suggesting a need to reconsider the definition of clinically insignificant PC.

**Abstract:** The incidence of prostate cancer (PC) has been increasing in Asian countries, where it was previously low. Although the adoption of a Westernized lifestyle is a possible explanation, the incidence is statistically biased due to the increase in prostate-specific antigen (PSA) screening and the accuracy of national cancer registration systems. Studies on latent PC provide less biased information. This review included studies evaluating latent PC in several countries after excluding studies using random or single-section evaluations and those that did not mention section thickness. The findings showed that latent PC prevalence has been stable since 1950 in Western countries, but has increased over time in Asian countries. Latent PC in Asian men has increased in both prevalence and number of high-grade cases. Racial differences between Caucasian and Asian men may explain the tumor location of latent PC. In conclusion, the recent increase in latent PC in Asian men is consistent with an increase in clinical PC. Evidence suggests that this increase is caused not only by the increase in PSA screening, but also by the adoption of a more Westernized lifestyle. Autopsy findings suggest the need to reconsider the definition of clinically insignificant PC.

**Keywords:** latent cancer; prostate cancer; autopsy

## **1. Introduction**

The incidence of prostate cancer (PC) has been increasing globally in recent years. It is the second most frequently diagnosed cancer and the fifth leading cause of cancer-related deaths among men worldwide [1]. The incidence of PC in recent decades has been heavily influenced by the emergence of prostate-specific antigen (PSA) testing. The availability of PSA testing from the middle to the late 1980s led to the intensive use of the test for screening, with a subsequent rapid increase in the incidence rate in Western countries. This trend has also been growing in Asian countries, where the incidence of PC was previously low [1,2]. The cause of this increase in Asian countries is thought to be multifactorial. Although the spread of PSA screening may be a major cause, changes in lifestyle due to more Westernized diets might be another [3,4]. The accuracy of national cancer registration systems may also influence the incidence, as national cancer registration has not been developed in some Asian countries. However, PC mortality has been decreasing in many Western countries,

**Citation:** Kimura, T.; Sato, S.; Takahashi, H.; Egawa, S. Global Trends of Latent Prostate Cancer in Autopsy Studies. *Cancers* **2021**, *13*, 359. https://doi.org/10.3390/ cancers13020359

Received: 26 December 2020 Accepted: 17 January 2021 Published: 19 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

possibly linked to earlier diagnosis due to PSA screening and improved treatment. In contrast, PC mortality is increasing in several Asian and developing countries [1]. These reports may support the influence of changes in risk factors due to more Westernized lifestyles in such countries.

Studies on latent PC provide less biased information about PC incidence compared to studies on clinical PC. Latent PC is defined as PC that is first detected in autopsy without any clinical signs of PC during the patient's lifetime. Since Mintz and Smith first reported latent PC in 13% of 100 autopsied cases in 1934 [5], many studies have been reported globally. A recent meta-analysis of 29 studies from 1948 to 2013 indicated that while the prevalence of latent PC significantly increases with age, there is no obvious time trend [6]. However, the time trend of latent PC prevalence might differ among countries. For example, a more recent study indicated that the prevalence of latent PC in Japanese men has been increasing [7]. In addition, a prospective study comparing latent PC in Asian and Caucasian men indicated that the prevalence in Asian men did not differ significantly from that in Caucasian men [8]. These results suggest that not only recent efforts for early detection, such as PSA screening, but also the change to Westernized diets and lifestyles may have influenced the increase in PC in Asian countries.

Information from studies on latent PC provides important insights from a different viewpoint. This review comprehensively discusses the results of latent PC studies in Western and Asian countries.

#### **2. Potential Biases in Methodology in Latent PC Studies**

As the methodology for latent PC studies has not yet been standardized and there are several biases between the studies, careful evaluation of their methods is required for precise interpretation. First, study populations, subject sources, and inclusion criteria differed among the studies. While most of the studies involved autopsies performed in hospitals, other studies assessed forensic autopsies. In addition, some studies have analyzed databases of institutional autopsy records or national or regional autopsy registries. However, a meta-analysis indicated that the source of subjects (population vs. hospital-based) was not significantly associated with the prevalence of latent PC [6]. Age was significantly associated with latent PC prevalence, which increased with each decade of age [6]. Thus, the inclusion and exclusion criteria of age and its distribution significantly influenced the prevalence of latent PC. Race is another major factor that affects the prevalence of latent PC. These results must be presented separately in studies that include various races. The methods of sample preparation also differed among the studies. The time elapsed from death to autopsy, step-sectioning versus random and/or single-section evaluation, and the interval between sections in step sectioning can also influence the prevalence of latent PC. The prevalence was reportedly higher in step-sectioned tissues than in randomly divided tissues [9], whereas there was no evidence that the section thickness or delay of autopsy affect PC prevalence [6]. However, information regarding the delay of autopsies is limited in most studies. The methods of diagnosis such as central review or not and use of immunohistochemical evaluation may also differ among studies, although a metaanalysis concluded that the use of immunohistochemistry was not associated with PC prevalence [6].

#### **3. Prevalence of Latent PC in Western Countries**

The most important topic in autopsy studies was the prevalence of latent PC. After the first report by Mintz and Smith in 1934, many such studies have been conducted in Western countries [5]. Studies evaluating latent PC by step-sectioning of the prostate in the US and Europe are listed in Table 1 [8,10–33]. Studies using random or single-section evaluations and those that did not mention section thickness were excluded. Cohorts of different nationalities or races are listed separately even if they were reported within the same study. The prevalence of latent PC varied from 9.6% to 58.6% between the studies. The ranges and distributions of age also varied between the studies. For example, some

studies included men under 20 years of age [25,29], while another study included only men older than 70 years [15]. Age influenced the prevalence of latent PC, as it was one of the most significant factors associated with prevalence [6,34]. A recent meta-analysis of 29 studies reported an estimated mean cancer prevalence at age < 30 years of 5% (95% confidence interval (CI): 3–8%), which increased nonlinearly to 59% (95% CI: 48–71%) by age > 79 years [6]. Race is another factor that affects prevalence. Six studies in the US reported the prevalence of latent PC in Caucasian and Black men separately [15,22,23,25,26,33], with reported prevalence rates of 25.9–58.6% and 19.4–43.3%, respectively. All four studies that conducted statistical analyses on the prevalence of latent PC in Caucasian and Black men concluded that racial differences did not exist. However, these results require careful interpretation, as age distributions may differ between these races, especially in forensic studies. A recent review of 19 studies including 6,024 autopsies suggested a racial difference in latent PC prevalence between Caucasian and Black men (35.7% vs. 50.5%), but did not conduct a statistical analysis [34].

**Table 1.** Studies evaluating latent PC by step-sectioning of prostate in the US and Europe.


NA: not available.

Figure 1 shows the prevalence of latent PC in studies of Caucasians in the US and Europe published after 1950 by year of publication. The size of each point was proportional to the number of men included in each study. The analytic linear approximation line of Haas 2007

US

the datapoints indicated that the latent PC prevalence was stable over time. The spread of PSA screening programs is thought to have increased the diagnosis of insignificant PC and decreased the prevalence of latent PC. However, few studies have examined the changes in the prevalence of latent PC before and after the PSA era. A retrospective study using an autopsy record database from a single institution in the US reported that the prevalence of latent PC decreased three-fold with the widespread use of PSA screening [35]. In this study, the prevalence was 4.8% in men older than 40 years between 1955 and 1960, compared to 1.2% between 1991 and 2001. However, this study was limited by the lack of whole-mount sections to examine the prostate, which might lead to a lower prevalence compared to those in other autopsy studies using step-sectioning. However, the prevalence of latent PC in Japan has increased despite the spread of PSA screening, although the exposure rate of PSA testing in Asian countries is still lower than that in Western countries [2,7]. The trends in Asian countries are discussed in Section 4. spread of PSA screening programs is thought to have increased the diagnosis of insignificant PC and decreased the prevalence of latent PC. However, few studies have examined the changes in the prevalence of latent PC before and after the PSA era. A retrospective study using an autopsy record database from a single institution in the US reported that the prevalence of latent PC decreased three-fold with the widespread use of PSA screening [35]. In this study, the prevalence was 4.8% in men older than 40 years between 1955 and 1960, compared to 1.2% between 1991 and 2001. However, this study was limited by the lack of whole-mount sections to examine the prostate, which might lead to a lower prevalence compared to those in other autopsy studies using step-sectioning. However, the prevalence of latent PC in Japan has increased despite the spread of PSA screening, although the exposure rate of PSA testing in Asian countries is still lower than that in Western countries [2,7]. The trends in Asian countries are discussed in the next section.

(92% Caucasian) NA Hospital 164 Median, 64 47 28.7 4 [31]

Figure 1 shows the prevalence of latent PC in studies of Caucasians in the US and Europe published after 1950 by year of publication. The size of each point was proportional to the number of men included in each study. The analytic linear approximation line of the datapoints indicated that the latent PC prevalence was stable over time. The

*Cancers* **2021**, *13*, x 4 of 11

Stamtiou 2007 Greece 2002–2004 Hospital 212 ≥ 30 40 18.8 4 [30]

Polat 2009 Turkey NA Hospital 114 Mean, 55 11 9.6 4 [32] Powell 2010 US/Black 1993–2004 Forensic 630 20–79 NA 35.1 2.5 [33] Powell 2010 US/Caucasian 1993–2004 Forensic 426 20–79 NA 48.1 2.5 [33] Zlotta 2013 Russia 2008–2011 Hospital 220 Mean, 62.5 82 37.3 4 [8] NA: not available.

**Figure 1.** Prevalence of latent PC in studies of Caucasians in the US and Europe. **Figure 1.** Prevalence of latent PC in studies of Caucasians in the US and Europe.

The prevalence of latent PC in studies of US and European Caucasians published after 1950 by year of publication. The size of each point was proportional to the number of men included in each study. The prevalence of latent PC in studies of US and European Caucasians published after 1950 by year of publication. The size of each point was proportional to the number of men included in each study.

#### **4. The prevalence of Latent PC in Asian and Other Countries 4. The prevalence of Latent PC in Asian and Other Countries**

Studies investigating latent PC are fewer in Asian countries than in Western countries. In 1937, Yotsuyanagi et al. first reported a 3% prevalence of latent PC in Japanese men in a domestic journal [36]. Among the literature published in international journals, in 1961, Karube first reported a latent PC prevalence of 10.9% in Japanese men older than Studies investigating latent PC are fewer in Asian countries than in Western countries. In 1937, Yotsuyanagi et al. first reported a 3% prevalence of latent PC in Japanese men in a domestic journal [36]. Among the literature published in international journals, in 1961, Karube first reported a latent PC prevalence of 10.9% in Japanese men older than 40 years by step-sectioning [36]. Studies on latent PC in Asia and Africa are listed in Table 2 [7,8,19,20,23,36–44]. Studies using random or single-section evaluations that were not published in English were excluded. Some results are part of a multinational study. Most studies in Asia were from Japan [7,8,19,23,36,38–40], with the exception of two studies from Singapore [20,37] and one each from China [41], Hong Kong [20], and Iran [42]. Reports from other regions include Jamaica in Latin America and Uganda in Africa as part of a multinational study in 1977 [20]. The prevalence of latent PC in Jamaica and Uganda was 32.7% and 24.0%, respectively, which were higher than those in Asian countries in the same reports (15.0% and 14.5% in Hong Kong and Singapore, respectively). In addition, the mean age of men in Uganda was 58.3 years, which was 5 years younger than of those in other countries (64.5 in Singapore, 63.4 in Hong Kong, and 63.2 in Jamaica). Updated data for prevalence in Latin America and Africa are required.


**Table 2.** Studies evaluating latent PC by step-sectioning of prostate in Asian and other countries.

Figure 2 shows the prevalence of latent PC in studies of Asian countries published after 1950. Data from the study conducted by Takahashi et al. in Japan were excluded because they focused on men older than 90 years of age [40]. The size of each point is proportional to the number of men in the study. The analytic linear approximation line of the datapoints indicated that the prevalence of latent PC in Asia increased over time compared to that in the US and Europe (Figure 1). The prevalence was 8.3–27.2% from the 1960s to 1970s, 5.5–34.6% from the 1980s to the 1990s, and 9.4–39.0% after 2000. Two Japanese studies directly compared the time trends in latent PC within the same institutions. Yatani et al. compared the latent PC prevalence within the same institution between men from 1965 to 1979 and from 1982 to 1986, both in pre-PSA era periods. The prevalence increased significantly from 22.5% to 34.6% [39]. More recently, Kimura et al. compared the prevalence of latent PC between Japanese men in pre- and post-PSA eras. The prevalence in men was 20.8% in 1983–1987 and 43.3% in 2008–2013 [7]. Both studies indicated a significant increase in higher-grade and larger cancers. Yatani et al. reported a higher rate of infiltrative tumors in the cohort in 1965–1979 than in 1982–1986, at 9.8% and 17.8%, respectively [39]. Kimura et al. reported a significantly larger index cancer volume in men in 2008–2013 compared to that in 1983–1987 [7].

The increased prevalence of latent PC in Asian men is consistent with the increased prevalence of clinical PC [45]. A major explanation for the increase in latent PC in Asian countries may be lifestyle changes due to more Westernized diets. The incidence of clinical PC in US men of Japanese ancestry in 1973–1986 was between that of Caucasians in the US and Japanese men born in Japan within the same period, suggesting the influence of both genetic and lifestyle factors on PC incidence [46]. In contrast, a comparative study published in 1973 showed that the age-adjusted prevalence of latent PC did not differ significantly between Japanese men in Japan and those in Hawaii (20.5% and 26.7%, respec-

tively). However, the age-adjusted prevalence of the proliferative type of latent PC was higher in Japanese Hawaiians than in native Japanese (19.1% and 8.7%, respectively) [19]. of infiltrative tumors in the cohort in 1965–1979 than in 1982–1986, at 9.8% and 17.8%, respectively [39]. Kimura et al. reported a significantly larger index cancer volume in men in 2008–2013 compared to that in 1983–1987 [7].

the datapoints indicated that the prevalence of latent PC in Asia increased over time compared to that in the US and Europe (Figure 1). The prevalence was 8.3–27.2% from the 1960s to 1970s, 5.5–34.6% from the 1980s to the 1990s, and 9.4−39.0% after 2000. Two Japanese studies directly compared the time trends in latent PC within the same institutions. Yatani et al. compared the latent PC prevalence within the same institution between men from 1965 to 1979 and from 1982 to 1986, both in pre-PSA era periods. The prevalence increased significantly from 22.5% to 34.6% [39]. More recently, Kimura et al. compared the prevalence of latent PC between Japanese men in pre- and post-PSA eras. The prevalence in men was 20.8% in 1983–1987 and 43.3% in 2008–2013 [7]. Both studies indicated a significant increase in higher-grade and larger cancers. Yatani et al. reported a higher rate

*Cancers* **2021**, *13*, x 6 of 11

**Figure 2.** Prevalence of latent PC in the studies on Asian populations. Prevalence of latent PC in studies of Asian countries published after 1950 by year of publication. The size of each point was proportional to the number of men in the study. **Figure 2.** Prevalence of latent PC in the studies on Asian populations. Prevalence of latent PC in studies of Asian countries published after 1950 by year of publication. The size of each point was proportional to the number of men in the study.

The increased prevalence of latent PC in Asian men is consistent with the increased prevalence of clinical PC [45]. A major explanation for the increase in latent PC in Asian countries may be lifestyle changes due to more Westernized diets. The incidence of clinical PC in US men of Japanese ancestry in 1973–1986 was between that of Caucasians in the US and Japanese men born in Japan within the same period, suggesting the influence of both genetic and lifestyle factors on PC incidence [46]. In contrast, a comparative study Zlotta et al. prospectively compared the prevalence of latent PC in 100 Japanese men and 220 Russian men [8]. The prevalence was 35.0% and 37.3% in Japanese and Russian men, respectively, and did not differ significantly. However, Japanese men had a greater probability of having a PC Gleason score (GS) of 7 than Russian men after adjusting for age and prostate weight. These results suggest the increasing prevalence and grade of latent PC in Asian men over the past few decades.

published in 1973 showed that the age-adjusted prevalence of latent PC did not differ

#### significantly between Japanese men in Japan and those in Hawaii (20.5% and 26.7%, re-**5. Pathological Findings from Latent PC**

spectively). However, the age-adjusted prevalence of the proliferative type of latent PC was higher in Japanese Hawaiians than in native Japanese (19.1% and 8.7%, respectively) [19]. Zlotta et al. prospectively compared the prevalence of latent PC in 100 Japanese men and 220 Russian men [8]. The prevalence was 35.0% and 37.3% in Japanese and Russian men, respectively, and did not differ significantly. However, Japanese men had a greater probability of having a PC Gleason score (GS) of 7 than Russian men after adjusting for age and prostate weight. These results suggest the increasing prevalence and grade of latent PC in Asian men over the past few decades. **5. Pathological Findings from Latent PC**  Although latent PC does not cause clinical symptoms and is not generally detected during the lifetime, most studies showed that a significant proportion of latent PC had high-grade, capsular, or seminal vesicle invasion. In studies in the US and Europe, 5.43% Although latent PC does not cause clinical symptoms and is not generally detected during the lifetime, most studies showed that a significant proportion of latent PC had high-grade, capsular, or seminal vesicle invasion. In studies in the US and Europe, 5.43% of cancers were GS 7 or greater and 11–13% were pT3 or greater [8,27,29–31]. In contrast, in Asian studies, 35.7–51.4% of latent PC was GS 7 or higher, with proportions higher than those in Western reports, although the proportion of cancers with pT3 or greater was similar (11.5–12.7%) [7,8,42]. Consequently, these cases of latent PC included clinically significant cancer as defined by Epstein (the presence of T3 or greater and/or index tumor volume of 500 mm<sup>3</sup> or greater and/or GS <sup>≥</sup> 7 [47]). A prospective comparative study by Zlotta et al. reported that 29.3% and 51.4% of latent PC cases were clinically significant in Russian and Japanese men, respectively [8]. A comparative study of contemporary latent PC and historical controls in Japan reported an index cancer volume of 500 mm<sup>3</sup> in 9.6% of cancers in men in 1983–1987 and 25.5% in 2008–2013, a significant difference [7]. The increase in proportion of significant cancer in latent PC notwithstanding the spread of PSA screening might suggest an increase in high-grade cancer in Asian countries, especially in Japan. These results also suggest the need to reexamine the definition of clinically insignificant PC. Stamey et al. defined clinically significant PC as organ-confined tumors of <0.5 cm<sup>3</sup> , GS 3+3 with no grade 4 or 5 [48]. However, it can also be defined as a cancer that does not affect the patient during the natural course of his lifetime. The requirements of Stamey's definition may be too stringent.

Investigating the tumor location of latent PC could improve our understanding of the origin of PC and how it grows [49]. Racial differences between Caucasian and Asian men have been suggested to affect not only the prevalence, but also the tumor location of PC. A comparative study of radical prostatectomy specimens reported that 35.5% and 0.6% of PCs originated in the transition zone (TZ) in Japanese men and US men, respectively [50].

Studies that categorized tumor location into anterior or posterior regions reported that anterior cancer was more prevalent in Asian men than in Caucasian men [50–56]. However, most studies evaluating tumor location in the prostate have analyzed only prostatectomy specimens. The tumor location of the prostatectomy specimens may overestimate the prevalence in the peripheral zone (PZ) or posterior cancer because of its higher detectability by digital rectal examination and transrectal prostate biopsy compared to in the TZ or anterior cancer. In this sense, the tumor location in latent PC may be less biased. There are limited reports regarding tumor location in latent PC. An autopsy study in Hungary including 139 men aged 18–95 years reported a latent PC prevalence of 38.8%; among the 64 tumor foci, 82.8% and 18.9% were present in the PZ and TZ, respectively [29]. Another study in the US including 164 men aged 54–73 years reported that latent PC was present in 29% of the cases, with 62% and 36% of PCs located in the posterior and anterior regions, respectively, and 77% and 16%—in the PZ and central zone in the prostate, respectively [31].

Reports on tumor locations of latent PC in Asian men are limited. A report of 149 autopsies of Iranian men over 50 years of age detected invasive adenocarcinoma in 14 (9.4%) cases, including nine cases (64%) in the posterior region, one case (7%) in the anterior region, and four cases (29%) in both lobes of the prostate [42]. A report of 182 Japanese men observed latent PC in 39.0% of cases, occurring in the TZ, PZ, or without dominance in 38.0%, 57.8%, and 4.2% of cases, respectively [43]. The tumors were located in the anterior and posterior regions in 49.3% and 40.8% of the cases, respectively. Approximately 40% of the tumors were located in the TZ and anterior region of the prostate, a rate higher than that reported in Western studies. The age distribution also differed between TZ and PZ cancers. In elderly men, cancer is more frequently diagnosed in the PZ than in the TZ [43]. This was consistent with the report by Takahashi et al. on autopsies in men over 90 years of age, which revealed that all latent PCs were localized in the PZ of the prostate [40]. An autopsy study in the US reported that most TZ cancers showed a different pathological pattern from that of PZ cancers, with lower GS and less aggressiveness [57]. However, a Japanese study reported that the pathological features did not differ between the TZ and PZ and between anterior and posterior cancers in terms of GS, tumor volume, or prevalence of clinically significant cancer; however, there is variation in the pT stage—PZ cancer has a significantly higher pT stage than TZ cancer. Several anatomical explanations have been proposed to explain this difference. For example, the TZ is separated from the surrounding area by fibromuscular tissue, whereas no such structure exists in the PZ. Moreover, the TZ contacts with the prostate capsule from the outside to the back of the PZ, and T3b cases of the TZ are few because of the anatomical position [43]. However, a prospective comparative study of Japanese and Russian autopsy cases reported similar tumor locations between cohorts, in which latent PC was located in the TZ in 25.9% and 20.7% and in the anterior region in 20.0% and 21.9% of the cases in Japanese and Russian men, respectively. Further investigation is required to determine whether there is a racial difference in tumor location and whether the location in Asian men has changed due to Western diet and lifestyle.

Few studies have investigated tumor location in the vertical direction. In their international multicenter study investigating 1327 autopsies from seven counties or regions, Breslow et al. reported latent PC in 350 cases. In the vertical direction, more tumors were present at the middle and apex levels than at the base. However, the evaluation method to describe the tumor distribution has not yet been standardized, and further studies are warranted.

Most latent PC cases represent the less aggressive forms of PC. Thus, comparing molecular markers or genomic aberrations between latent and clinical PCs is an ideal method to investigate their effects. Igawa et al. reported a significantly higher nm23-HI gene expression level in clinical PC than in normal prostatic tissues, latent PC, and clinical PC [58]. Watanabe et al. reported that *Ras* gene mutations in latent PC varied among ethnic groups and that the frequency in Japanese men was higher than that in US Black or Caucasian men [59]. Alipov et al. compared the expression of the ETS1 proto-oncogene in latent PC, benign prostatic hyperplasia, normal prostatic tissues, and clinical PC [60], reporting negative expression in benign tissues and higher levels in clinical PC than in latent PC. Maekawa et al. investigated the TMPRSS2 Met160Val polymorphism in Japanese men, including 518 men with sporadic PC, 433 healthy controls, and 154 men with latent PC [61]. The TMPRSS2 Met160Val polymorphism is a genetic risk factor for sporadic PC but not for latent PC in the Japanese population. However, molecular studies using latent PC are limited, possibly because of the limited quality and availability of latent PC specimens.

#### **6. Limitations of Autopsy Studies**

Latent PC has a unique cancer status compared to the malignancies of other origins. Although it has been investigated for a long time, several problems remain to be solved. Section 5 described variations in the methodologies used for sample preparation and diagnosis. Although step-sectioning of the whole prostate is a standard method, the duration from death to autopsy is difficult to control. A central review for diagnosis is mandatory because of inter- and intra-observer variability for the pathological diagnosis of PC [62]. A fundamental bias also exists in autopsy studies. As the subjects were men who died in the hospital, their backgrounds differed from those of healthy men. Inaba et al. reported that 106 of the 182 autopsy cases (58.2%) had been performed due to death of malignancy other than PC (unpublished data).

The available data on latent PC were provided by a limited number of countries and regions. Most studies were from North America, Western Europe, and Japan, whereas data from Africa and Latin America are limited. More importantly, the number of autopsies has been steadily declining over the past 30–40 years worldwide [63]. After the 2000s, the autopsy rate was only 7–9% in the US, compared to approximately 25–35% in the mid-1960s and 50% of all hospital deaths in the 1940s and 1950s [64,65]. In Japan, more than 40,000 autopsies were performed in 1985, but this number had gradually decreased to approximately 10,000 by 2018 [66].

One explanation for the limited number of studies evaluating molecular markers in latent PC is the low quality of specimens from autopsies. RNA and proteins were extracted during the time between death and autopsy. To overcome such limitations in autopsy studies, rapid autopsies have emerged [67]. In this new methodology, tissues are collected as soon as possible after the patient's death. Ideally, the quality of a rapid autopsy tissue can be considered comparable to the quality of a fresh surgical biopsy tissue.

#### **7. Learning from Latent PC and Future Directions**

While latent PC studies have a long history, the available evidence remains limited. Latent PC studies have revealed a larger prevalence of insignificant PC than the incidence of clinical PC. PC prevalence increases with age and more than half of both Caucasian and Asian men over 80 years of age have indolent PC. The recent increase in latent PC in Asian men is consistent with an increase in clinical PC in Asian countries. These findings suggest that this increase in clinical PC in Asian countries is due not only to the spread of PSA screening, but also to the adoption of Westernized lifestyles.

In addition, the results of autopsy studies suggest the need to reconsider the definition of clinically insignificant PC, which is thought to be an ideal candidate for active surveillance. The present definition might be too strict, as latent PC included a significant proportion of cancer cases thought to be life-threatening, such as with GS ≥ 7 and pT3 or greater. Cancer volume and the percentage of high-grade cancer cases also increased with age. However, the individuals lived without the influence of PC throughout their lives. Molecular analyses are required in latent PC studies to distinguish between indolent and life-threatening PC. Methodologies such as rapid autopsies have opened the door for new studies of latent PC.

**Author Contributions:** Writing—original draft preparation, T.K.; writing—review and editing, S.S., H.T., and S.E. Authorship must be limited to those who have contributed substantially to the work reported. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** T.K. is a paid consultant/advisor to Astellas, Bayer, Janssen, and Sanofi. S.E. is a paid consultant/advisor to Takeda, Astellas, AstraZeneca, Sanofi, Janssen, and Pfizer. The funders had no role in the study design; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

#### **References**

