Invited Review Article
Prognostic prediction following radical prostatectomy for prostate cancer using conventional as well as molecular biological approaches
Correspondence: Hideaki Miyake M.D., Division of Urology, Kobe University Graduate School of Medicine, 7-5-1 Kusunoki-cho, Chuo-ku, Kobe 650-0017, Japan. Email: email@example.com
Although radical prostatectomy has been the mainstay of treatment for men with clinically organ-confined prostate cancer, a certain proportion of men undergoing radical prostatectomy fail to achieve a complete cure of this disease; that is, postoperative biochemical recurrence develops in approximately 30% of men, some of whom will ultimately die of disease progression. A number of studies, therefore, have been carried out to identify factors reflecting prognostic outcomes following radical prostatectomy, which would be potentially helpful for properly counseling individual patients undergoing this surgery. Furthermore, various types of model systems using multiple clinicopathological parameters, such as the nomogram, look-up table and artificial neural network, have been shown to have better performance in postoperative prognostic prediction than the opinions of expert clinicians. However, there have not been any standard models uniformly applied to postoperative prognostic prediction, which could be explained, at least in part, by the use of conventional clinicopathological parameters alone, suggesting the need for the additional evaluation of molecular markers simultaneously considering the unique biological features of prostate cancer. In this review, a search of the literature was carried out focusing on the significance of prognostic models following radical prostatectomy, and it is suggested that these models could be promising tools to provide accurate information on the postoperative clinical course of prostate cancer patients. To widely introduce such models into clinical practice, it is necessary to further improve currently available models and develop more reliable, flexible, simple and easily accessible tools by incorporating conventional clinicopathological factors as well as molecular biomarkers.
Abbreviations & Acronyms
= percent free PSA
= biopsy Gleason grade
= biopsy Gleason sum
= biochemical recurrence
= clinical stage
= extracapsular extension
= biochemical recurrence
= heat shock protein
= kallikrein-related peptidase-2
= lymph node invasion
= not available
= prostate cancer
= prostate cancer antigen-3
= % of cancer in positive biopsy core
= % of positive biopsy core
= prostate-specific antigen
= pathological stage
= radical prostatectomy
= radiation therapy
= surgical margin status
= seminal vesicle invasion
= transforming growth factor-β1
PC is the most frequently diagnosed malignancy and second leading cause of cancer-specific death in men in most Western industrialized countries, and the recent incidence of this disease has markedly increased in Asian countries, including Japan, as well. Despite no curative therapeutic options for men with metastatic disease, different treatment options are currently available for those with localized disease, such as RP, external-beam radiation therapy, brachytherapy and active surveillance. Although the ideal therapy for localized PC remains controversial, RP has generally been regarded as the gold standard for men with localized PC.[3, 4] However, despite recent progress in surgical techniques, BR, defined as a persistent increase in the serum value of PSA, occurs in approximately 30% of men undergoing RP,[5, 6] a certain proportion of whom will ultimately die of PC progression.
In the field of clinical oncology, the exact prediction of a clinical course following a definitive treatment is potentially important, as it will result in the improvement of the quality of care, patient expectations and clinical outcomes by facilitating appropriate informed consent regarding a prognostic issue. This would be particularly true in men with PC considering its heterogeneous clinical features, which sometimes make it difficult to select a suitable treatment from multiple therapeutic options. In recent years, however, various model systems that aid in the process of decision-making during the treatment of PC have been developed.[9-12] Although these models have been shown to be more useful for predicting the prognosis of men with PC than clinical judgment based on the experience of each clinician, there has not been any model uniformly introduced into clinical practice as a standard tool for predicting the clinical course of PC.[13, 14] Collectively, these findings suggest the limitations of developing such a model system by using conventional clinicopathological parameters alone; accordingly, special attention has been paid to an additional investigation of molecular biomarkers associated with the progression of PC in order to establish more precise models than those currently available.[15-17]
In the present review, we focused on the significance of prognostic models following RP for clinically organ-confined PC, and provide an overview of the utility of previously reported tools as postoperative prognostic indicators and describe future prospects for further improving the prognostic prediction by considering additional novel parameters, including molecular biomarkers.
Prognostic models following RP
When clinical decision-making relevant to RP for men with organ-confined PC is required, several backgrounds are usually considered, including the experience of clinicians, outcomes of reliable clinical studies and information from decision aids. Of these, predictive models using multiple parameters have been shown to provide more reliable findings regarding the probable clinical course following RP than others;[13, 18, 19] therefore, a number of prognostic model systems for men undergoing RP have been developed, in order to modify therapeutic and/or post-therapeutic strategies for achieving favorable clinical outcomes.[20, 21]
Predictive models can be classified into several types, such as the nomogram, look-up table and artificial neural network, which are comparable from multiple viewpoints as follows: predictive accuracy of the model, generalizability by internal and external validations, model calibration, level of complexity, and clinical implication.[22-24] As these models have been developed based on different methodological approaches, the advantages and limitations of each model should be well recognized. For example, the look-up table is generally regarded as being user-friendly, but with a low accuracy, whereas the nomogram is somewhat complex, but its accuracy is reliable. Collectively, these findings suggest that it should be highly recommended to use an appropriate model according to what the clinician wants to predict, considering the accuracy as well as reproducibility of such models for predicting the clinical course following RP.[9-11]
Preoperative prognostic parameters
It is necessary to consider both cancer control and functional outcomes, including those of continence and erectile function, for men with PC who will be scheduled to be treated with RP; therefore, it is extremely important to preoperatively predict not only the likelihood of BR, but also the pathological stage for planning the surgical procedure.[25, 26] For example, patients with a low probability of non-organ-confined PC are likely to benefit from nerve-sparing surgery compared with those who are assumed to have a relatively advanced disease, whereas more aggressive surgical approaches could be applied to patients who are preoperatively diagnosed as being at high risk of capsular penetration, seminal vesicle invasion and/or lymph node metastasis.
To date, the usefulness of several prognostic factors available before RP have been examined, and, of these, the clinical stage, PSA and biopsy Gleason score are currently regarded as major factors closely associated with postoperative clinical outcomes.[27-29] However, it has been shown to be difficult to predict the disease extension as well as prognosis in men with PC using such a single parameter alone before RP.[30, 31] For example, Cooperberg et al. reported that the clinical stage failed to show a significant correlation with the probability of BR in men included in a study of PC risk assessment. Similarly, the biopsy Gleason score was not identified as an independent predictor of extraprostatic disease extension in Japanese men who were treated with RP in our recently published study. Accordingly, a number of predictive models using multiple major parameters have been developed.[32-40] Of these, Partin tables might be the most widely accepted model for staging prediction, which is the initially developed look-up table for predicting the probabilities of organ-confined disease, extracapsular extension, seminal vesicle invasion and lymph node invasion based on the serum PSA value, clinical stage and biopsy Gleason sum. This table is very simple and could provide satisfactory values of the area under the curve for predicting pathological parameters; however, the outcomes of external validations resulted in conflicting findings; that is, some of them, particularly those targeting a European population, could not show an accuracy similar to that achieved in the original cohort.[41-43]
Consistent with Partin tables, various investigators have reported predictive models for pathological information on RP specimens (Table 1), and, of these tools, some have been intensively validated by other groups. For example, Steuber et al. reported the usefulness of a nomogram predicting side-specific extracapsular extension developed by the data from 1118 men with PC, including the clinical stage, serum PSA level, biopsy Gleason sum, percent of positive biopsy cores and percent of cancer lesions in positive cores, and this nomogram was externally validated in the cohort undergoing robotic RP. Similarly, Gallina et al. developed a nomogram for the preoperative prediction of seminal vesicle invasion based on data from 666 men who underwent RP, and multiple-institutional external validation of this system was carried out in a cohort of 2584 men treated with robotic RP and an excellent discrimination rate was reported. Before the introduction of such tools into clinical practice, however, it should be considered that most of these models were externally validated in a single cohort alone.
Table 1. Predictive models for pathological information in presurgical settings
|Partin et al.||1997||Look-up table||4133||ECE, SVI, LNI||cT, PSA, bGS||72.4%|
|Steuber et al.||2006||Nomogram||1118||Side-specific ECE||cT, PSA, bGS, ppbc, pcpbc||84.0%|
|Gallina et al.||2007||Nomogram||666||SVI||cT, PSA, bGS, ppbc||79.2%|
|Ohori et al.||2004||Nomogram||763||Side-specific ECE||cT, PSA, bGS,a ppbc,a pcpbca||80.6%|
|Koh et al.||2003||Nomogram||763||SVI||cT, PSA, bGS, % of cancer at the base||88.3%|
|Cagiannos et al.||2003||Nomogram||5510||LNI||cT, PSA, bGS||76.0%|
|Briganti et al.||2006||Nomogram||602||LNI||cT, PSA, bGS||76.0%|
|Briganti et al.||2007||Nomogram||565||LNI||cT, PSA, bGS||80.2%|
|Baccala et al.||2007||Nomogram||6740||SVI||cT, PSA, bGS, age||80.0%|
Evidently, prognostic models in a presurgical setting are forced to include only preoperatively available factors; therefore, it has been well documented that postoperative predictive models that can take pathological parameters into account are generally more reliable than preoperative models. To date, however, several presurgery predictive models for BR or PC-specific mortality have been developed (Table 2), and some of these would be helpful in counseling patients with newly diagnosed localized PC.[30, 47-50] For example, D'Amico et al. developed a risk stratification scheme for BR following RP based on data on the clinical stage, PSA and biopsy Gleason sum from 888 men with PC; however, this model was shown to be inferior in its discriminatory ability compared with other models for predicting the postoperative BR.[51, 52] Furthermore, Kattan et al. reported the utility of a nomogram that includes the clinical stage, PSA and both primary and secondary biopsy Gleason scores for predicting the probability of BR at 5 years after RP,[48, 49] and this was subsequently revised to predict BR at 10 years after RP. As for model systems to predict PC-specific mortality, several promising tools have been reported;[53-55] however, the discrimination, as well as calibration, of these systems have not been externally validated.
Table 2. Predictive models for postoperative prognosis in presurgical settings
|Cooperberg et al.||2005||Probability graph||1439||BR||cT, PSA, bGS, ppbc, age||66.0%|
|D'Amico et al.||1998||Probability graph||888||BR||cT, PSA, bGS||NA|
|Kattan et al.||1998||Nomogram||983||BR||cT, PSA, primary and secondary bGG||74.0%|
|Stephenson et al.||2006||Nomogram||1978||BR||cT, PSA, bGS, No of pbc, year of surgery||76.0%|
Pathological factors for prognostic prediction
After surgery, it is possible to obtain variable information on the pathological findings of RP specimens, such as pathological stage, final Gleason sum and surgical margin status, which is a potential advantage when developing predictive models for postoperative prognosis. Accordingly, it is assumed that postoperative models can generally provide more useful prognostic information in men undergoing RP than preoperative models, and a number of predictive models constructed with postoperative factors have been developed (Table 3).
Table 3. Predictive models for postoperative prognosis in postsurgical setting
|D'Amico et al.||1998||Look-up table||862||BR||pT, PSA, GS, SMS||NA|
|Kattan et al.||1999||Nomogram||996||BR||PSA, GS, ECE, SVI, LNI, SMS||89.0%|
|Stephenson et al.||2005||Nomogram||1881||BR||PSA, GS, ECE, SVI, LNI, SMS||86.0%|
|Walz et al.||2009||Nomogram||2911||BR||PSA, GS, ECE, SVI, LNI, SMS||NA|
|Suardi et al.||2008||Nomogram||601||BR||pT, GS, SMS, type of surgery, adjuvant RT||77.2–80.6%|
Initially, D'Amico et al. reported a simple look-up table to predict the probability of 2-year BR, which included data on the serum PSA, pathological stage, final Gleason sum and surgical margin status. Consistent with this model, Kattan et al. reported a postoperative nomogram for 7-year BR predictions using the serum PSA, final Gleason sum, capsular invasion, seminal vesicle invasion, lymph node invasion and surgical margin status obtained from 996 PC patients treated with RP by a single surgeon, which provided 89% accuracy on internal validation. This postoperative nomogram was updated by Stephenson et al. The Stephenson nomogram, using the same parameters as the Kattan nomogram and extending the predictions to 10-year BR, is characterized by adjustment of the predictions according to the disease-free interval. In addition, the Stephenson nomogram was externally validated in two other cohorts, resulting in the achievement of satisfactory discrimination rates. In addition to these models, there still have been nomograms showing excellent accuracy for predicting the probabilities of postoperative BR.[59, 60] For example, Waltz et al. developed a nomogram based on data from 2911 men with PC, which was externally validated using data from 2825 PC patients and showed a discrimination rate of 82.0%. This nomogram is suitable for postoperative BR, particularly that which occurs in the early period following RP. Suardi et al. also developed a nomogram that makes it possible to calculate the probabilities of BR up to 20 years after RP; therefore, this nomogram included adjuvant radiation therapy as one of the variables.
There have been several models using postoperative factors to predict long-term end-points, such as metastatic progression and PC-specific mortality, rather than BR.[61-67] To date, several models for predicting the probability of progression to metastatic disease using postoperative parameters have been reported.[61-64] Of these, Porter et al. created a nomogram for predicting the development of metastatic disease at 5, 10, 15 and 20 years after RP. There have also been postoperative models developed for predicting the likelihood of PC-specific mortality following RP.[65-67] For this objective, PSA kinetics are regarded as key parameters in these models. D'Amico et al. developed a model that enabled them to quantify the risk of postoperative PC-specific death according to the value of PSA velocity in a cohort consisting of 1095 newly diagnosed men with PC, whereas Freedman et al. reported a look-up table based on the data, including the PSA doubling time, final Gleason sum and time from surgery to BR, from 379 men diagnosed with BR for yielding the relative risk of PC-specific mortality. However, it should be considered when applying these models to patients in a clinical setting that no models have been externally validated.
Limitations of existing prognostic models
Although existing prognostic models have been shown to exhibit promising abilities to provide a prognosis for PC patients undergoing RP, a wide variety of limitations for introducing these models into clinical practice have been suggested.[9-11] First, several limitations, which are associated with features of PC and/or RP themselves, have been pointed out.[68-70] One serious problem is the definition of BR, which is the most frequently used prognostic end-point in men with PC after RP; however, uniform criteria on this issue have not been established, which causes the use of different definitions in the literature. This trend has also unfavorably influenced the establishment of a standard prognostic model in men treated with RP. In addition, it has been well recognized that the outcomes of RP could be affected by the skills of surgeons, which might be one of the serious problems inhibiting the standardization of a potential prognostic model after RP. Kattan et al. however, recently reported that incorporating surgeon experience into their previously developed nomograms had no significant impact on the prognostic accuracy.
Next, there have also been several limitations regarding the process for developing predictive models.[58, 60, 72-74] From this viewpoint, it is the most serious limitation to lack proper external validation, which prevents the model from being applied to other cohorts.[72, 73] The absence of updates reflecting changes in the characteristics of a disease, as well as patients, would also be a potential limitation. For example, it would not be suitable to use predictive models based on data from patients undergoing infrequently carried out examinations and treatment in recent years, such as sextant prostate biopsy and neoadjuvant hormonal therapy. Furthermore, it represents another problem that competing risks are not well considered in most existing prognostic models.[58, 60] This is particularly important when using nomograms predicting long-term end-points.
Significance of molecular biomarkers in prognostic prediction following RP
Currently existing prognostic models might not have a sufficient ability to make a postoperative prognosis in an individual PC patient following RP. Considering the intensive efforts made in the field of PC research to clarify the molecular mechanism mediating the progression of PC, it is one promising approach to incorporate useful novel biomarkers that can assist in clinical decision-making to overcome the limitations of existing models. In fact, a number of recent studies have shown the excellent impact of the additional evaluation of molecular biomarkers on the improved ability of conventional models to predict a postoperative prognosis in men with PC (Table 4).
Table 4. Lists of major molecular biomarkers for predicting prognostic outcomes
|Prostate-specific antigen density||PSA-derived|||
|Prostate-specific antigen velocity||PSA-derived|||
|% free Prostate-specific antigen||PSA-derived||[82, 83]|
|Prostate-specific membrane antigen||Prostate-specific|||
|Prostate stem cell antigen||Prostate-specific|||
|Human kallikrein-related peptidase-2||Prostate-specific||[84, 85]|
|Prostate cancer antigen-3||Prostate-specific||[84, 86]|
|Heat shock protein 27||Apoptosis-related||[92, 93]|
|p27||Cell-cycle related||[95, 96]|
|Androgen receptor||Signal transduction-related|||
|Transforming growth factor-β1||Cytokine|||
|Matrix metalloproteinase 2||Cell invasion|||
|Matrix metalloproteinase 9||Cell invasion|||
|Plasminogen activation inhibitor-1||Cell invasion|||
Before evaluating the significance of molecular biomarkers, it is necessary to recognize the limitations of PSA as a prognostic indicator, as PSA is the only biomarker widely used in clinical practice. It is assumed that limitations of the PSA test are mainly as a result of the absence of cancer specificity. For example, elevations of PSA are observed in men with benign enlargement of the prostate gland, such as benign prostatic hyperplasia and those with inflammatory disease in the prostate. Similarly, PSA levels do not have a direct association with an increase in the grade and stage of PC, particularly in men with comparatively low PSA values. This lack of specificity of the PSA test leads to both overdiagnosis and overtreatment of clinically insignificant disease, and also results in confusion for prognostic prediction.
One attractive approach for overcoming the limitations of the PSA test is to measure PSA derivatives, including PSA velocity, PSA density and age-specific PSA intervals, but the significant advantages of these derivatives over the PSA test have not been clearly shown. In addition, of the various molecular forms of PSA, which can be divided into the two major categories of free and complexed forms, the significance of a lower %fPSA as an indicator of a poor prognosis has been intensively analyzed;[82, 83] however, these findings remain controversial. For example, Southwick et al. found %fPSA to be a better predictor of postoperative pathological findings than the Gleason score, whereas Graefen et al. reported that %fPSA has no significant impact on the prediction of BR following RP.
The usefulness of several genes that are specifically expressed in the prostate gland, such as PSA, as possible biomarkers for PC have been examined, and favorable outcomes were reported in studies assessing the significance of some of these candidate genes, such as human KLK2, PCA3, prostate-specific membrane antigen and prostate stem cell antigen. For example, KLK2, a member of the kallikrein gene family of secreted serine proteases showing the greatest abundance in the prostate gland, has been reported to provide precise prognostic information in men undergoing RP over PSA, whereas measurement of the urinary level of PCA3, a non-coding RNA produced exclusively in the prostate gland, particularly in PC cells, has been shown to be a useful tool for predicting pathological features in PC patients.
Numerous studies have shown the involvement of apoptosis-related molecules in the progression of PC. Of these, the Bcl-2 protein family, including anti-apoptotic (such as Bcl-2 and Bcl-xL) and proapoptotic (such as Bax and Bak) genes that have an opposite role in the process of apoptotic cell death, is one of the most intensively investigated molecules as biomarkers of PC. For example, overexpression of Bcl-2 was shown to be closely correlated with the subsequent development of BR in men with PC following RP, whereas altered Bax expression in PC cells, defined as under- or overexpression compared with the staining intensity of non-neoplastic cells, appeared to be an independent predictor of BR.
Recently, changes in expression profiles of a wide variety of genes at various time-points after androgen withdrawal have been precisely characterized using animal model systems mimicking the diverse behavior of human PC, and highlighted genes showing marked changes during progression to castration resistance.[110, 111] Based on these outcomes, special attention has been paid to several genes upregulated after androgen ablation, most of which are characterized by having anti-apoptotic activities and functions like a molecular chaperone.[110, 111] To date, these genes, such as clusterin and HSP27, have been mainly investigated as molecular targets for the treatment of PC; however, it is currently under active investigation whether these genes could be used as biomarkers for predicting the prognosis of PC patients.[90-93, 113-115] In our previous studies, the expression of clusterin, which was significantly increased in RP specimens after neoadjuvant hormonal therapy compared with that in biopsy specimens, was shown to be a useful parameter in predicting BR. Furthermore, we measured serum levels of clusterin in PC patients, and showed that the serum clusterin level and its density in men with PC are closely correlated with disease extension, and that postoperative biochemical recurrence-free survival in patients with an elevated clusterin density was significantly poorer than that in those with a normal density.
Abnormalities in cell-cycle regulation are present in the majority of malignant tumors. In PC as well, a number of studies have shown the association between the outcomes of PC treatment and expression of cell-cycle-related biomarkers, including p16, p21, p27, Aurora-A and Ki-67.[94-99] Of these markers, p27, an endogenous inhibitor of cyclin-dependent kinase, is one of the most widely characterized proteins as a biomarker of PC.[95, 96] For example, Freedland et al. reported that low expression of p27 in biopsy specimens could be used as an independent predictor of BR in men with PC following RP. Furthermore, in our previous study assessing the predictive value of 12 kinds of molecular marker, as well as conventional prognostic parameters for BR in patients undergoing RP, only Ki-67, reflecting a tumor proliferation index, in addition to seminal vesicle invasion and a positive surgical margin, appeared to be independently related to BR on multivariate analysis.
In addition to the molecular biomarkers described above, there have been a number of molecules showing a close association with the prognosis in PC patients following RP, such as signal transduction-related biomarkers, cellular adhesion-related markers, angiogenesis-related markers, epithelial-mesenchymal transition-related markers, cytokines and epigenetic markers.[100-104, 106, 117] For example, Gravdal et al. reported that cadherin switching characterized by low E-cadherin and high N-cadherin expression indicated a strong link to BR in patients after RP, whereas circulating levels of IL-6 and its receptor have been found to be elevated proportionally to features of aggressive PC, such as that with a higher Gleason score, advanced stage and decreased survival.
It has recently been investigated whether a significant improvement in predictive accuracy could be achieved by adding biomarkers to established parameters in the nomogram.[105, 107, 118] In this context, Kattan et al. added preoperative plasma IL-6 soluble receptor and TGF-β1 levels to the standard nomogram for predicting the risk of BR following RP using the pretreatment PSA values, clinical stage and biopsy Gleason sum. They reported an improved prognostic ability of this nomogram with these additional biomarkers; that is, the novel nomogram resulted in an increased predictive accuracy of BR from 75% to 83%. Similarly, Stephenson et al. analyzed gene expression profiling using microarray technology in PC specimens from patients undergoing RP. In this study, models combining conventional parameters and gene expression profile could accurately classify 89% of patients in terms of the development of postoperative BR with a predictive ability superior to the standard nomogram.
It is extremely important to provide an accurate prognosis for patients with clinically localized PC treated with RP; therefore, a number of predictive models have been developed in both pre- and postoperative settings to provide useful prognostic information following RP, which enables each individual patient to select the most suitable treatment from multiple modalities. Although several models have been widely used for predicting the postoperative clinical course in men undergoing RP, no model has been uniformly applied as a standard tool because of several limitations of existing predictive model systems. Considering the biologically heterogeneous nature of PC, it is absolutely necessary to identify molecular biomarkers in order to exactly predict the clinical course of PC, and to integrate such novel markers with conventional clinicopathological variables to produce predictive models showing outcomes superior to the standard predictive system. Collectively, these findings suggest that, despite several limitations to be overcome before the introduction of such predictive models into clinical practice for PC, the combined use of both conventional parameters and novel molecular biomarkers in the development of these models might help provide variable information on clinical decision-making during the treatment of patients with PC undergoing RP.
Conflict of interest