Prospective evaluation of operating characteristics of prostate cancer detection biomarkers

The altered levels of adiponectin - leptin as predictive biomarkers to estimate the severity of prostate cancer

Introduction and Aim: Prostate is one of the commonest sites of malignancy affecting elderly male population & is increasingly becoming a significant public health issue especially in countries having aging population. We hypothesized that altered levels of adiponectin-leptins may be an underlying connection between incidence of prostate cancer (PCa) and aged matched males.   Materials and Methods: This study was designed to comparatively corelate circulating serum levels of adiponectin & leptin in 160 elderly patients with PCa to their serum levels in 160 healthy controls. The age and body mass index in all groups were dissimilar in case and control. Based on the Gleason score of 7, =7 >7, patients were further subdivided into low, intermediate, high grades of PCa, respectively.    Results: No significant statistical variance was identified in terms of age, Body mass index (BMI), Radom blood glucose, HDL, LDL, triglycerides, total cholesterol, creatinine, and BUN levels within the compared groups. In PCa patients’ group, concentration levels of serum adiponectin were significantly lower, and levels of serum leptin was significantly greater compared to healthy controls (P<0.001). Statistical analysis revealed a significant positive inverse association between PSA and adiponectin levels (r=0.285, P<0.001) and significant association between serum levels of PSA and leptin (r=0.285, P<0.001). Significant statistical correlation was also evident between BMI, PSA, TG, and leptin were whole group. However, there was no significant association observed between adiponectin or leptin level and grade of the disease.   Conclusion: Evaluation of data in our study suggests that patients of PCa exhibit low concentration of serum adiponectin levels and high concentration of leptin levels. Further, this association was independent of histological grading of disease of disease/disease progression as well as other biochemical parameters.

Prostate health index (PHI) as a reliable biomarker for prostate cancer: a systematic review and meta-analysis

OBJECTIVES

Prostate cancer (PCa) represents the second most common solid cancer in men worldwide. In the last decades, the prostate health index (PHI) emerged as a reliable biomarker for detecting PCa and differentiating between non-aggressive and aggressive forms. However, before introducing it in clinical practice, more evidence is required. Thus, we performed a systematic review and meta-analysis for assessing the diagnostic performance of PHI for PCa and for detecting clinically significant PCa (csPCa).

METHODS

Relevant publications were identified by a systematic literature search on PubMed and Web of Science from inception to January 11, 2022.

RESULTS

Sixty studies, including 14,255 individuals, met the inclusion criteria for our meta-analysis. The pooled sensitivity and specificity of PHI for PCa detection was 0.791 (95%CI 0.739-0.834) and 0.625 (95%CI 0.560-0.686), respectively. The pooled sensitivity and specificity of PHI for csPCa detection was 0.874 (95%CI 0.803-0.923) and 0.569 (95%CI 0.458-0.674), respectively. Additionally, the diagnostic odds ratio was 6.302 and 9.206, respectively, for PCa and csPCa detection, suggesting moderate to good effectiveness of PHI as a diagnostic test.

CONCLUSIONS

PHI has a high accuracy for detecting PCa and discriminating between aggressive and non-aggressive PCa. Thus, it could be useful as a biomarker in predicting patients harbouring more aggressive cancer and guiding biopsy decisions.

The prostate health index and the percentage of [-2]proPSA maintain their diagnostic performance when calculated with total and free PSA from different manufacturers

OBJECTIVES

To evaluate the diagnostic performance of the prostate health index (PHI) and of the percentage of [-2]proPSA (%[-2]proPSA) calculated with total and free PSA from non-Beckman Coulter manufacturers (Roche and Abbott), and compare it with the fully Beckman Coulter [-2]proPSA derivatives.

METHODS

In this study, 237 men (PSA: 2-10 μg/L) scheduled for prostate biopsy were enrolled. %[-2]proPSA and PHI were calculated with total and free PSA from three manufacturers. Beckman Coulter PSA and [-2]proPSA were performed on the Access 2 analyzer (Hybritech calibration). Roche PSA was performed on the cobas e411 and the Abbott PSA on the Architect i2000sr. Statistical analysis was performed, considering prostate cancer (PCa) as the outcome.

RESULTS

Univariate analysis showed that all indices were predictors of cancer, irrespective of the manufacturer (p<0.001). The AUC was similar for all manufacturers, both for %[-2]proPSA (Beckman Coulter: 0.756; Roche: 0.770; Abbott: 0.756) and PHI (Beckman Coulter: 0.776; Roche: 0.785; Abbott: 0.778). When considering the cutoffs that allowed 90% sensitivity, [-2]proPSA derivatives calculated with Roche and Abbott PSA had similar specificities and predictive values when compared to Beckman Coulter. The percentage of missed cancers (8-9%) was the same between manufacturers. The percentage of spared biopsies was significantly higher with Roche's PHI (21.0%) and Abbott's PHI (20.6%) than with Beckman Coulter's PHI (17.2%).

CONCLUSIONS

In the PSA range between 2 and 10 μg/L, [-2]proPSA derivatives maintain their diagnostic performance in PCa detection when calculated with PSA from Roche and Abbott. This can lead to a broader implementation of these indices in clinical laboratories worldwide.

Prostate Cancer Biomarker Development: National Cancer Institute's Early Detection Research Network Prostate Cancer Collaborative Group Review

Prostate cancer remains the most common non-skin cancer and second leading cause of death among men in the United States. Although progress has been made in diagnosis and risk assessment, many clinical questions remain regarding early identification of prostate cancer and management. The early detection of aggressive disease continues to provide high curative rates if diagnosed in a localized state. Unfortunately, prostate cancer displays significant heterogeneity within the prostate organ and between individual patients making detection and treatment strategies complex. Although prostate cancer is common among men, the majority will not die from prostate cancer, introducing the issue of overtreatment as a major concern in clinical management of the disease. The focus of the future is to identify those at highest risk for aggressive prostate cancer and to develop prevention and screening strategies, as well as discerning the difference in malignant potential of diagnosed tumors. The Prostate Cancer Research Group of the National Cancer Institute's Early Detection Research Network has contributed to the progress in addressing these concerns. This summary is an overview of the activities of the group.See all articles in this CEBP Focus section, "NCI Early Detection Research Network: Making Cancer Detection Possible."

Systemic Alterations of Wnt Inhibitors in Patients with Prostate Cancer and Bone Metastases

Purpose

Dickkopf-1 (DKK-1) and sclerostin seem to inhibit osteoblast activity by blocking the Wnt pathway, which leads to progression of metastatic prostate cancer (PC). However, it is unknown whether serum levels of these proteins are altered in PC patients with or without metastasis. The aim of this study was to assess DKK-1 and sclerostin serum levels in PC patients, including patients with bone metastases.

Methods

The study cohort (N = 143) consisted of 53 controls with benign prostatic hyperplasia (BPH), 43 with localized PC (PC cM0), and 47 had PC with metastasis (PC cM1). Serum levels of DKK-1 and sclerostin were measured by enzyme-linked immunosorbent assay. Results were compared using the Kruskal-Wallis tests; post hoc analysis was performed using the Tukey-Kramer test.

Results

Mean DKK-1 levels in patients with BPH (2809.4 pg/ml) (p < 0.001) as well as PC cM1 (2575.5 pg/ml) (p = 0.001) were significantly higher than in patients with PC cN0 cM0 (1551.8 pg/ml). Among PC cM1 patients, median DKK-1 levels were significantly lower in patients with castration-resistant disease compared to those with hormone-sensitive PC (p = 0.02); in contrast, sclerostin concentrations were elevated (p = 0.04). DKK-1 correlated with PSA in the cM1 group (p = 0.03) and sclerostin correlated with PSA in the PC group (0.01).

Conclusions

DKK-1 is involved in the progression of PC. DKK-1-mediated inhibition of osteoblasts, which contributes to tumor progression and osteolytic metastases, may also play a role in the development of metastases with osteoblastic features. The use of DKK-1 antibodies should be considered for studies including metastatic PC patients.

A novel model to predict positive prostate biopsy based on serum androgen level

Circulating levels of prostate-specific antigen (PSA) and testosterone are widely used for the detection of prostate cancer prior to prostate biopsy; however, both remain controversial. Effective screening strategies based on quantitative factors could help avoid unnecessary biopsies. Here, we sought to clarify the predictive value of free testosterone (FT) vs total testosterone (TT) in identifying patients likely to have positive biopsies. This study aims to develop a novel model for predicting positive prostate biopsy based on serum androgen levels. This study included 253 Japanese patients who underwent prostate biopsy at our institution. TT and FT, %FT (=FT/TT), age, PSA, prostate volume (PV) and PSA density (PSAD = PSA/PV) were assessed for association with prostate biopsy findings. The following results were obtained. Of 253 patients, 145 (57.3%) had positive biopsies. Compared to the negative biopsy group, the positive biopsy group demonstrated higher age, PSA and PSAD but lower PV, FT and %FT by univariate analysis. Multivariate logistic regression analysis indicated PSA, PSAD and %FT were independent predictors of cancer detection. We developed a predictive model based on PSAD and %FT, for which the area under the curve was significantly greater than that of PSA (0.82 vs 0.66), a well-known predictor. Applying this analysis to the subset of patients with PSA <10 ng/mL yielded similar results. We confirmed the utility of this model in another independent cohort of 88 patients. In conclusion, lower %FT predicted a positive prostate biopsy. We constructed a predictive model based on %FT and PSAD, which are easily obtained prior to biopsy.

Serial Percent Free Prostate Specific Antigen in Combination with Prostate Specific Antigen for Population Based Early Detection of Prostate Cancer

PURPOSE

We characterized the diagnostic properties of serial percent free prostate specific antigen in relation to prostate specific antigen in a multiethnic, multiracial cohort of healthy men.

MATERIALS AND METHODS

A total of 6,982 percent free prostate specific antigen and prostate specific antigen measurements were obtained from participants in a greater than 12-year Texas screening study comprising 1,625 men who never underwent biopsy, 497 who underwent 1 or more biopsies negative for prostate cancer and 61 diagnosed with prostate cancer. We evaluated the ROC AUC of percent free prostate specific antigen and the proportion of patients with fluctuating values across multiple visits determined according to 2 thresholds (less than 15% vs 25%). The proportion of cancer cases in which percent free prostate specific antigen indicated a positive test before prostate specific antigen greater than 4 ng/ml did and the number of negative biopsies that would have been spared by negative percent free prostate specific antigen test results were calculated.

RESULTS

Percent free prostate specific antigen fluctuated around its threshold of less than 25% (less than 15%) in 38.3% (78.1%), 42.2% (20.9%), and 11.4% (25.7%) of patients never biopsied, and with negative and positive biopsies, respectively. At the same thresholds, percent free prostate specific antigen tested positive earlier than prostate specific antigen in 71.4% and 34.2% of cancer cases, respectively. Among men with multiple negative biopsies and PSA greater than 4 ng/ml, percent free PSA would have tested negative in 31.6% and 65.8%, respectively.

CONCLUSIONS

Percent free prostate specific antigen should accompany prostate specific antigen testing to potentially spare unnecessary biopsies or detect cancer earlier. When near the threshold, both tests should be repeated due to commonly observed fluctuation.

Measurement of serum prostate cancer markers using a nanopore thin film based optofluidic chip

Currently used cancer marker for prostate adenocarcinoma (PC), serum prostate-specific antigen (PSA), greatly overestimates PC population. Patients with high PSA levels have to undergo unnecessary but physically painful and expensive procedure such as prostate biopsies repeatedly. The reliability of PC test can be greatly increased by finding a protein that is secreted selectively by malignant, but not normal, prostate cells. A recently discovered novel protein, referred as neuroendocrine marker (NEM), is secreted only by malignant prostate cells and released in blood circulation. Although NEM seems to be significantly more reliable based on the data obtained from a limited cohort, currently available NEM ELISA is not suitable for undertaking a large study. Therefore, the goal of the present study was to develop an alternative, label-free assay system that can reliably measure NEM and PSA in patient samples. Herein an optofluidic chip that can reliably detect PSA as well as NEM in patient samples has been developed. The optofluidic chip, which consists of arrayed nanopore-based sensors fabricated from anodic aluminum oxide (AAO) thin film, offers improved sensitivity upon the optimization of the concentration of the detector antibodies immobilized on the sensor surface. The results demonstrate that the chip is reliable, extremely sensitive and requires just 1 µl of patient serum (or even less) to measure PSA and NEM even in a non-cancer individual. Compared with the traditional ELISA for PSA, the nanopore-based sensor assay is 50-100 fold more sensitive, and offers many advantages such as elimination of labeled antigen, need for sophisticated equipment and highly trained individuals. These advantages, along with the low cost, should make the technology suitable for point-of-care application to screen elderly male populations for PC and to monitor the progress of patients undergoing PC treatment.

European Randomised Study of Screening for Prostate Cancer (ERSPC) risk calculators significantly outperform the Prostate Cancer Prevention Trial (PCPT) 2.0 in the prediction of prostate cancer: a multi-institutional study

OBJECTIVE

To analyse the performance of the Prostate Cancer Prevention Trial Risk Calculator (PCPT-RC) and two iterations of the European Randomised Study of Screening for Prostate Cancer (ERSPC) Risk Calculator, one of which incorporates prostate volume (ERSPC-RC) and the other of which incorporates prostate volume and the prostate health index (PHI) in a referral population (ERSPC-PHI).

PATIENTS AND METHODS

The risk of prostate cancer (PCa) and significant PCa (Gleason score ≥7) in 2001 patients from six tertiary referral centres was calculated according to the PCPT-RC and ERSPC-RC formulae. The calculators' predictions were analysed using the area under the receiver-operating characteristic curve (AUC), calibration plots, Hosmer-Lemeshow test for goodness of fit and decision-curve analysis. In a subset of 222 patients for whom the PHI score was available, each patient's risk was calculated as per the ERSPC-RC and ERSPC-PHI risk calculators.

RESULTS

The ERSPC-RC outperformed the PCPT-RC in the prediction of PCa, with an AUC of 0.71 compared with 0.64, and also outperformed the PCPT-RC in the prediction of significant PCa (P<0.001), with an AUC of 0.74 compared with 0.69. The ERSPC-RC was found to have improved calibration in this cohort and was associated with a greater net benefit on decision-curve analysis for both PCa and significant PCa. The performance of the ERSPC-RC was further improved through the addition of the PHI score in a subset of 222 patients. The AUCs of the ERSPC-PHI were 0.76 and 0.78 for PCa and significant PCa prediction, respectively, in comparison with AUC values of 0.72 in the prediction of both PCa and significant PCa for the ERSPC-RC (P = 0.12 and P = 0.04, respectively). The ERSPC-PHI risk calculator was well calibrated in this cohort and had an increase in net benefit over that of the ERSPC-RC.

CONCLUSIONS

The performance of the risk calculators in the present cohort shows that the ERSPC-RC is a superior tool in the prediction of PCa; however the performance of the ERSPC-RC in this population does not yet warrant its use in clinical practice. The incorporation of the PHI score into the ERSPC-PHI risk calculator allowed each patient's risk to be more accurately quantified. Individual patient risk calculation using the ERSPC-PHI risk calculator can be undertaken in order to allow a systematic approach to patient risk stratification and to aid in the diagnosis of PCa.

Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests

Many decisions in medicine involve trade-offs, such as between diagnosing patients with disease versus unnecessary additional testing for those who are healthy. Net benefit is an increasingly reported decision analytic measure that puts benefits and harms on the same scale. This is achieved by specifying an exchange rate, a clinical judgment of the relative value of benefits (such as detecting a cancer) and harms (such as unnecessary biopsy) associated with models, markers, and tests. The exchange rate can be derived by asking simple questions, such as the maximum number of patients a doctor would recommend for biopsy to find one cancer. As the answers to these sorts of questions are subjective, it is possible to plot net benefit for a range of reasonable exchange rates in a “decision curve.” For clinical prediction models, the exchange rate is related to the probability threshold to determine whether a patient is classified as being positive or negative for a disease. Net benefit is useful for determining whether basing clinical decisions on a model, marker, or test would do more good than harm. This is in contrast to traditional measures such as sensitivity, specificity, or area under the curve, which are statistical abstractions not directly informative about clinical value. Recent years have seen an increase in practical applications of net benefit analysis to research data. This is a welcome development, since decision analytic techniques are of particular value when the purpose of a model, marker, or test is to help doctors make better clinical decisions.

A case control study of sarcosine as an early prostate cancer detection biomarker

Background

Sarcosine has been investigated as a prostate cancer biomarker with mixed results concerning its predictive power. We performed a case–control evaluation of the predictive value of serum sarcosine for early detection in a population-based cohort of men undergoing prostate-specific antigen (PSA) screening.

Methods

For analysis we used 251 cancer cases and 246 age-matched non-cancer cases from the San Antonio Biomarkers Of Risk (SABOR) screening study. For cancer cases, pre-diagnostic serum was utilized for sarcosine measurement. Controls were defined as men who had been followed at least for 5 years on study with no prostate cancer diagnosis; sarcosine was measured on the initial baseline serum. HPLC-electrospray ionization mass spectrometry was used for serum sarcosine quantification. The association of sarcosine with prostate cancer was assessed using area underneath the receiver-operating characteristic curve (AUC), and logistic regression adjusting for PSA, digital rectal exam, family history, age, race, and history of a prior negative biopsy. Among cancer cases, nominal logistic regression was used for the association of sarcosine with Gleason grade.

Results

Sarcosine levels were overlapping between the prostate cancer cases (median 15.8 uM, range 6.2 to 42.5 uM) and controls (median 16.2 uM, range 6.4 to 53.6 uM). The AUC of sarcosine was not statistically different from random chance either for participants with any PSA value (52.2 %) or those with PSA values in the range of 2 to 10 ng/mL (54.3 %). Sarcosine was not predictive of Gleason score and added no independent predictive power to standard prostate cancer risk factors for detection of prostate cancer (all p-values > 0.05).

Conclusions

Serum sarcosine should not be pursued further as a marker for the early detection of prostate cancer.

Risk calculators and updated tools to select and plan a repeat biopsy for prostate cancer detection

Millions of men each year are faced with a clinical suspicion of prostate cancer (PCa) but the prostate biopsy fails to detect the disease. For the urologists, how to select the appropriate candidate for repeat biopsy is a significant clinical dilemma. Traditional risk-stratification tools in this setting such as prostate-specific antigen (PSA) related markers and histopathology findings have met with limited correlation with cancer diagnosis or with significant disease. Thus, an individualized approach using predictive models such as an online risk calculator (RC) or updated biomarkers is more suitable in counseling men about their risk of harboring clinically significant prostate cancer. This review will focus on the available risk-stratification tools in the population of men with prior negative biopsies and persistent suspicion of PCa. The underlying methodology and platforms of the available tools are reviewed to better understand the development and validation of these models. The index patient is then assessed with different RCs to determine the range of heterogeneity among various RCs. This should allow the urologists to better incorporate these various risk-stratification tools into their clinical practice and improve patient counseling.

Prostate Cancer Prevention Trial risk calculator 2.0 for the prediction of low- vs high-grade prostate cancer

OBJECTIVE

To modify the Prostate Cancer Prevention Trial risk calculator (PCPTRC) to predict low- vs high-grade (Gleason grade≥7) prostate cancer and incorporate percent free-prostate-specific antigen (PSA).

METHODS

Data from 6664 Prostate Cancer Prevention Trial placebo arm biopsies (5826 individuals), where prostate-specific antigen and digital rectal examination results were available within 1 year before the biopsy and PSA was ≤10 ng/mL, were used to develop a nominal logistic regression model to predict the risk of no vs low-grade (Gleason grade<7) vs high-grade cancer (Gleason grade≥7). Percent free-PSA was incorporated into the model based on likelihood ratio analysis of a San Antonio Biomarkers of Risk cohort. Models were externally validated on 10 Prostate Biopsy Collaborative Group cohorts and 1 Early Detection Research Network reference set.

RESULTS

Of all the Prostate Cancer Prevention Trial biopsies, 5468 (82.1%) were negative for prostate cancer, 942 (14.1%) detected low-grade, and 254 (3.8%) detected high-grade disease. Significant predictors were (log base 2) PSA (odds ratio for low-grade vs no cancer, 1.29*; high-grade vs no cancer, 2.02*; high-grade vs low-grade cancer, 1.57*), digital rectal examination (0.96, 1.49*, 1.55*, respectively), age (1.02*, 1.05*, 1.03*, respectively), African American race (1.13, 2.83*, 2.51*, respectively), prior biopsy (0.63*, 0.81, 1.27, respectively), and family history (1.31*, 1.25, 0.95, respectively), where * indicates P value<.05. The new PCPTRC 2.0 either with or without percent free-PSA (also significant by the likelihood ratio method) validated well externally.

CONCLUSION

By differentiating the risk of low- vs high-grade disease on biopsy, PCPTRC 2.0 better enables physician-patient counseling concerning whether to proceed to biopsy.

The diagnostic value of adiponectin multimers in healthy men undergoing screening for prostate cancer

BACKGROUND

Adiponectin has been reported to have a prohibitory effect on prostate cancer. The goal of this study was to evaluate the diagnostic value of adiponectin multimers for prostate cancer.

METHODS

Total adiponectin, high- and low-molecular-weight (HMW, LMW), ratios of these measures, and body mass index (BMI) were compared in a prospective prostate cancer-screened cohort. Multivariable logistic regression was used to assess the association between adiponectin measures, their interaction with BMI, and risk of prostate cancer and Gleason score upgrading from biopsy to prostatectomy.

RESULTS

A total of 228 prostate cancer cases and 239 controls were analyzed: 72 (31.6%) of the cancer cases were high grade (Gleason grade ≥7). Only percent HMW had a statistically significant relationship with prostate cancer (P = 0.04). Among normal and overweight men, the risk of prostate cancer increased as percent HMW increased [OR = 1.24 for a doubling of percent HMW, 95% confidence interval (CI), 0.41-3.75 and OR = 1.81; 95% CI, 1.02-3.20, respectively], whereas among obese men, the risk of prostate cancer decreased (OR = 0.62; 95% CI, 0.32-1.18). Among 97 patients who underwent radical prostatectomy, there was no association between Gleason score upgrading and any of the adiponectin multimers.

CONCLUSION

This study was unable to confirm the utility of total adiponectin as a biomarker for prostate cancer risk. For the adiponectin multimers, only HMW showed increases with prostate cancer but not in all weight classes.

IMPACT

Although adiponectin may play a role in the pathogenesis of prostate cancer, our results do not support adiponectin multimers as biomarkers of detection.

Serum isoform [-2]proPSA derivatives significantly improve prediction of prostate cancer at initial biopsy in a total PSA range of 2-10 ng/ml: a multicentric European study

BACKGROUND

Strategies to reduce prostate-specific antigen (PSA)-driven prostate cancer (PCa) overdiagnosis and overtreatment seem to be necessary.

OBJECTIVE

To test the accuracy of serum isoform [-2]proPSA (p2PSA) and its derivatives, percentage of p2PSA to free PSA (fPSA; %p2PSA) and the Prostate Health Index (PHI)-called index tests-in discriminating between patients with and without PCa.

DESIGN, SETTING, AND PARTICIPANTS

This was an observational, prospective cohort study of patients from five European urologic centers with a total PSA (tPSA) range of 2-10 ng/ml who were subjected to initial prostate biopsy for suspected PCa.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

The primary end point was to evaluate the specificity, sensitivity, and diagnostic accuracy of index tests in determining the presence of PCa at prostate biopsy in comparison to tPSA, fPSA, and percentage of fPSA to tPSA (%fPSA) (standard tests) and the number of prostate biopsies that could be spared using these tests. Multivariable logistic regression models were complemented by predictive accuracy analysis and decision curve analysis.

RESULTS AND LIMITATIONS

Of >646 patients, PCa was diagnosed in 264 (40.1%). Median tPSA (5.7 vs 5.8 ng/ml; p=0.942) and p2PSA (15.0 vs 14.7 pg/ml) did not differ between groups; conversely, median fPSA (0.7 vs 1 ng/ml; p<0.001), %fPSA (0.14 vs 0.17; p<0.001), %p2PSA (2.1 vs 1.6; p<0.001), and PHI (48.2 vs 38; p<0.001) did differ significantly between men with and without PCa. In multivariable logistic regression models, p2PSA, %p2PSA, and PHI significantly increased the accuracy of the base multivariable model by 6.4%, 5.6%, and 6.4%, respectively (all p<0.001). At a PHI cut-off of 27.6, a total of 100 (15.5%) biopsies could have been avoided. The main limitation is that cases were selected on the basis of their initial tPSA values.

CONCLUSIONS

In patients with a tPSA range of 2-10 ng/ml, %p2PSA and PHI are the strongest predictors of PCa at initial biopsy and are significantly more accurate than tPSA and %fPSA.

Comparing ROC curves derived from regression models

In constructing predictive models, investigators frequently assess the incremental value of a predictive marker by comparing the ROC curve generated from the predictive model including the new marker with the ROC curve from the model excluding the new marker. Many commentators have noticed empirically that a test of the two ROC areas often produces a non-significant result when a corresponding Wald test from the underlying regression model is significant. A recent article showed using simulations that the widely used ROC area test produces exceptionally conservative test size and extremely low power. In this article, we demonstrate that both the test statistic and its estimated variance are seriously biased when predictions from nested regression models are used as data inputs for the test, and we examine in detail the reasons for these problems. Although it is possible to create a test reference distribution by resampling that removes these biases, Wald or likelihood ratio tests remain the preferred approach for testing the incremental contribution of a new marker.

Biases introduced by choosing controls to match risk factors of cases in biomarker research

BACKGROUND

Selecting controls that match cases on risk factors for the outcome is a pervasive practice in biomarker research studies. Such matching, however, biases estimates of biomarker prediction performance. The magnitudes of these biases are unknown.

METHODS

We examined the prediction performance of biomarkers and improvements in prediction gained by adding biomarkers to risk factor information. Data simulated from bivariate normal statistical models and data from a study to identify critically ill patients were used. We compared true performance with that estimated from case control studies that do or do not use matching. ROC curves were used to quantify performance. We propose a new statistical method to estimate prediction performance from matched studies for which data on the matching factors are available for subjects in the population.

RESULTS

Performance estimated with standard analyses can be grossly biased by matching, especially when biomarkers are highly correlated with matching risk factors. In our studies, the performance of the biomarker alone was underestimated whereas the improvement in performance gained by adding the marker to risk factors was overestimated by 2-10-fold. We found examples for which the relative ranking of 2 biomarkers for prediction was inappropriately reversed by use of a matched design. The new approach to estimation corrected for bias in matched studies.

CONCLUSIONS

To properly gauge prediction performance in the population or the improvement gained by adding a biomarker to known risk factors, matched case control studies must be supplemented with risk factor information from the population and must be analyzed with nonstandard statistical methods.

Updating risk prediction tools: a case study in prostate cancer

Online risk prediction tools for common cancers are now easily accessible and widely used by patients and doctors for informed decision-making concerning screening and diagnosis. A practical problem is as cancer research moves forward and new biomarkers and risk factors are discovered, there is a need to update the risk algorithms to include them. Typically, the new markers and risk factors cannot be retrospectively measured on the same study participants used to develop the original prediction tool, necessitating the merging of a separate study of different participants, which may be much smaller in sample size and of a different design. Validation of the updated tool on a third independent data set is warranted before the updated tool can go online. This article reports on the application of Bayes rule for updating risk prediction tools to include a set of biomarkers measured in an external study to the original study used to develop the risk prediction tool. The procedure is illustrated in the context of updating the online Prostate Cancer Prevention Trial Risk Calculator to incorporate the new markers %freePSA and [-2]proPSA measured on an external case-control study performed in Texas, U.S.. Recent state-of-the art methods in validation of risk prediction tools and evaluation of the improvement of updated to original tools are implemented using an external validation set provided by the U.S. Early Detection Research Network.