Enhancing Utility of PSA by Multivariate Methods

PSA is a sensitive diagnostic marker for prostate cancer, but its use is hampered by a high rate of falsely elevated values. Approximately 75% of men with a PSA of 4-10 ^g/liter do not have prostate cancer in biopsy. This causes unnecessary anxiety and costs. The proportion of free PSA in serum can be used to reduce the number of unnecessary biopsies by about 20-30% [184, 185]. Many other variables, such as prostate volume, digital rectal examination findings and different molecular forms of PSA have been evaluated in order to reduce the number of unnecessary biopsies. However, for the clinician it is difficult to make an accurate biopsy decision based on many variables. With multivariate methods, that is, by using regression and neural network techniques, the utilization of several variables can be optimized and made easier. PSA-based algorithms have also been developed for staging and prognosis of prostate cancer.

7.1. Multivariate Modeling Techniques

Multivariate algorithms or models can be established using patient data on diagnostic variables and outcome of interest. The goal is to find the best fitting, yet biologically reasonable model to explain the relationship between the diagnostic variables and the outcome variable. Logistic regression has been used extensively for statistical modeling in medicine. The method is similar to linear regression, but the output value is binary [186]. An advantage with logistic regression is that the effect of the diagnostic variables is easy to understand and quantify. This is usually more difficult with neural network techniques [187]. On the other hand, a drawback with logistic regression is that complex relationships between variables are difficult to demonstrate. The type of correlation between the diagnostic variables and the outcome variable must be selected manually by transforming the diagnostic variables. Interactions between variables must also be inserted by hand. For utilizing more complex information the multilayer perceptron (MLP) is better suited. This is an artificial neural network method that can be considered as an extension of logistic regression. About 90% of all artificial neural networks published in clinical medicine are MLPs.

  1. 2. Application of PSA Based Multivariate Algorithms
  2. 2.1. Diagnosis of Prostate Cancer

The most common type of multivariate PSA algorithms predicts the outcome of prostate biopsy [188-194] (Table 2). The PSA range is often limited to 4-10 ,ug/liter, as this is considered a diagnostic gray zone, but there are also algorithms for the PSA range 2-4 ^g/liter and higher than 10 ^g/liter. In most studies, logistic regression was reported to perform as well as artificial neural networks. We have tested algorithms in an external testing group that was derived chronologically in a prospective manner from the study population [194]. Our results indicate that the neural network, despite the use of efficient methods to avoid overtraining, had adapted too well to the training set and did not recognize patients in the testing set very effectively. This is a potential problem for the practical use of neural networks.

7.2.2. Staging of Prostate Cancer

Correct prediction of the spread of prostate cancer on the basis of pre-operative variables would be important, as unnecessary lymph node dissections and radical prostatectomies could be avoided. Tewari et al. [195] constructed MLP models for prediction of margin, seminal vesicle, and lymph node positivity using information on serum PSA, age, race, DRE findings, tumor size, Gleason score, bilaterality, and number of positive biopsy cores. The models had sensitivities of 81-100% at 72-75% specificity. Batuello et al. [196] used information on 6454 patients with clinically localized prostate cancer to train and test an MLP model for prediction of lymph node spread. Clinical stage, Gleason score, and serum PSA were

10 Ways To Fight Off Cancer

10 Ways To Fight Off Cancer

Learning About 10 Ways Fight Off Cancer Can Have Amazing Benefits For Your Life The Best Tips On How To Keep This Killer At Bay Discovering that you or a loved one has cancer can be utterly terrifying. All the same, once you comprehend the causes of cancer and learn how to reverse those causes, you or your loved one may have more than a fighting chance of beating out cancer.

Get My Free Ebook

Post a comment