Systems and methods for improving disease diagnosis using measured analytes - Patents.com

Abstract

A system and method for diagnosing diseases such as prostate cancer, breast cancer, lung cancer, and ovarian cancer and their stages are provided. In some embodiments, the disclosed systems and methods collect patient samples, calculate biomarker concentrations and proximity scores, and use these calculations to create a training set model that is used to correlate biomarker concentrations and proximity scores with disease diagnosis and disease state (e.g., cancer stage). In some embodiments, correlation techniques used include simple regression, ROC curve area maximization, topological stabilization, or spatial proximity correlation analysis. [Selected Figure] Figure 1

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Application No. 62 / 542,865, filed Aug. 9, 2017, which is incorporated by reference in its entirety.

[0002] Related patent application WO 2014 / 000041, filed March 13, 2014, which is incorporated herein by reference in its entirety, describes a method for improving disease prediction using an independent variable for association analysis that is not a directly measured analyte concentration, but a calculated value called a "proximity score" that is calculated from the concentration but is also normalized for some age (or other physiological parameter) to remove age drift and non-linearities in how concentration values ​​drift or shift with physiological parameters (e.g., age, menopausal status, etc.) as a disease state moves from non-diseased to diseased.

[0003] The present invention relates to methods for improving the accuracy of disease diagnosis and related diagnostic tests that involve correlation of measured analytes with binary outcomes (e.g., non-disease or disease) and higher order outcomes (e.g., one of several disease stages). [Background technology]

[0004] Correlation methods in which three or more independent variables are used to correlate a binary outcome (such as the presence or absence of a given disease) commonly use spatial proximity correlation methods (also called cluster or neighborhood search methods), regression methods, and wavelet methods. In the case of disease prediction, common components of blood or serum are measured and correlations are performed using these concentrations as independent variables for the prediction of various disease states. In the case of a given disease state, where the outcome is either "disease" or "non-disease", logistic regression methods are commonly used. Other techniques include, for example, genetic algorithms. The predictive power of these methods depends heavily on the constituent analytes selected for the method. The skilled artisan is aware that many analytes and parameters that appear to have predictive power do not actually improve diagnostic and analytical power.

[0005] Regression methods use the trend of the independent variable to correlate with the outcome. Linear methods are based on linear trends, whereas logistic regression is based on logarithmic trends. In biological disease prediction, logistic regression is most commonly used to determine outcomes.

[0006] The group spatial proximity method investigates the topology of variables correlation for grouping similar results. The spatial proximity method has the advantage that the trend is not continuous but the topology shows local reversals in the trend. However, this method is highly nonlinear and prone to very local variable results even with small measurement errors which are highly predictive in biological applications. In addition, both above methods can be combined with the spatial proximity method applied on a small scale to become a full regression method by integration.

[0007] However, some independent variables that appear to be logically correlated do not actually show predictive trends. What is needed, therefore, is an approach to improve diagnostic accuracy by utilizing patient- and population-specific variables that have not previously contributed useful information to the diagnosis of a disease state.

[0008] Much research has been done to discover biomarkers that, alone or in combination, can predict disease status with sufficient reproducibility and predictive power for clinical use. This research has been limited or unsuccessful. High abundance proteins (HAPs) have been extensively studied to discover a single protein that can make this prediction. Although numerous examples have been found, none have low enough false negative levels to allow the marker to screen patients for disease.

[0009] As a result, such single biomarkers are used only for treatment monitoring, with the exception of PSA (Prostate Specific Antigen) for prostate cancer. This test requires that the concentration indicating a biopsy has an adequate gradient to reduce false negatives and, therefore, very high false positive levels. As many as 80% of men who are indicated for a biopsy are actually negative for prostate cancer. DNA markers have also proven very useful in some cases for cancer subtypes, but are not suitable for screening for the same reasons as HAPs above.

[0010] Proteomic approaches are also being investigated using a large number of proteins. This work is again centered on HAPs, or high level effector proteins. Multiplex protein measurement methods such as immunoassays, chips, and mass spectrometry dominate this work. Very early studies have shown some success in ovarian cancer. However, the problem with all these methods is that many of the proteins selected do not have a strong correlation with the progression from health to disease (and many do not have a known biological relationship with the disease state, as is typical for mass spectrometry, for example). Furthermore, the fact that the whole serum sample is queried by the spectrophotometer for protein levels creates a severe oversampling problem in mass spectrometry, making it difficult to train correlation algorithms. In the case of mass spectrometry, a whole serum sample may have over 200 proteins and 10,000 mass spectrometry peaks.

[0011] What is also needed in the diagnostic area are analytical techniques that analyze low-abundance biomarkers, along with techniques that take advantage of low-abundance proteins that are more useful for diagnostic purposes than HAPS.

[0012] Diagnostics for a simple and accurate serum-based blood test to detect cancer and whether it is serious or dormant have been researched for many years. For example, the current test, Prostate Specific Antigen (PSA) for prostate cancer, has a very high false positive rate and a high false negative rate of 1 in 10 men. This test has a predictive power of about 57%. Furthermore, men diagnosed with low-grade prostate cancer may not require treatment for many years or for the rest of their lives. This diagnosis can only be obtained accurately today with a PCa biopsy. With the current PSA test, all men with PSA levels of about 4.0ng / ml (90%, 1 in 10 men not affected) are sent for a biopsy, and only about 20% have any PCa, regardless of Gleason score. Additionally, men with low-grade PCa are at risk of converting to high-grade disease over the rest of their lives, and the only sure way to accurately diagnose this is to perform more biopsies. Additional biopsies for monitoring are unacceptable to the medical community due to the cost and to patients due to pain and side effects. Therefore, ongoing monitoring of men with low-grade PCa is performed by regular PSA testing with digital rectal exam (DRE) and sometimes CT scans. Often, prophylactic removal of the prostate is performed when not medically necessary. From this patient population, new serum-based tests are taught that can distinguish men with no PCa from high-grade and detect men with low-grade PCa that will transform later in life. Additionally, blood-based tests are taught that can distinguish early solid tumor cancers or cancer stages such as lung or breast cancer.

[0013] The current PSA screening test was approved in the mid-1980s and is now off patent. The new so-called 4K score test, offered by OPKO as a "self-prepared test," has not received regulatory approval. It is intended to detect men with high-grade PCa and separate this condition from low-grade PCa. Generally, high-grade PCa is considered to have a Gleason score (obtained by biopsy) of 7 (4+3) or higher (8,9,10), while low-grade is considered to be 7 (3+4) or lower. The PSA test has a predictive power of about 57% for detecting men with all grades of PCa, or for a sensitivity of 90%, the false positive rate is about 80% (1 in 4 positives is actually negative). The 4K score test has a predictive power of about 64%. Thus, for a false negative rate of 1 in 10, the false positive rate is about 50%, i.e., about 5 in 10 men are actually negative. This is the current state of PCa diagnostic testing in modern medicine. Summary of the Invention [Problem to be solved by the invention]

[0014] Currently, there are no regulatory approved methods to detect diseases such as lung and breast cancer with a simple blood test. Furthermore, these diseases can only be graded for severity by biopsy. We also propose additional tests to grade tumors using these proteins in the tumor microenvironment using serum as a surrogate for active cytokines. [Means for solving the problem]

[0015] To address these and other shortcomings in the art, a new test is described in which serum serves as a surrogate for these proteins, using active cytokines as an example in the tumor microenvironment. [Brief description of the drawings]

[0016] A more complete understanding of the present invention and many of the attendant advantages thereof will become more readily appreciated by reference to the following detailed description when considered in conjunction with the accompanying drawings, in which: [Figure 1] FIG. 1 shows the variation of biomarker concentrations by Gleason score for prostate cancer. [Diagram 2] FIG. 1 shows the variation of biomarker concentrations by Gleason score for lung cancer. [Diagram 3] FIG. 1 displays the average upregulation of biomarker concentrations corresponding to breast cancer stage. [Figure 4] FIG. 1 displays VEGF receiver operating characteristic ("ROC") curves for advanced prostate cancer and non-cancer. [Diagram 5] FIG. 1 displays TNFα ROC curves for aggressive prostate cancer and non-cancer. [Figure 6] FIG. 1 displays PSA ROC curves for advanced prostate cancer and non-cancer. [Figure 7] FIG. 13 displays IL6 ROC curves for advanced prostate cancer and non-cancer. [Figure 8] FIG. 1 shows IL10 ROC curves for late-stage and early-stage lung cancer. [Figure 9] FIG. 1 shows IL6 ROC curves for late-stage and early-stage lung cancer. [Figure 10] FIG. 1 displays VEGF ROC curves for late-stage and early-stage lung cancer. [Figure 11] FIG. 13 illustrates blind test results with two samples failing the topology instability test and corrected with a discrepancy algorithm according to an embodiment of the disclosed diagnostic method. [Figure 12] FIG. 1 shows the results of a clinical study on breast cancer according to an embodiment of the disclosed diagnostic method, in this case training set cancer scores for training set model I using 10 bimarker planes. [Figure 13]FIG. 1 shows the results of a clinical study on breast cancer according to an embodiment of the disclosed diagnostic method, in this case training set cancer scores for training set model II using 105 bimarker planes. [Figure 14] FIG. 1 shows actual diagnostic results for blind sample runs in a clinical study according to an embodiment of the disclosed diagnostic method. [Figure 15] FIG. 1 shows a deciles biomarker plane, which shows the proximity scores of two of the biomarkers used by an embodiment of the disclosed diagnostic method. [Figure 16] FIG. 1 illustrates a bimarker plane containing training set data points according to an embodiment of the disclosed diagnostic method. [Figure 17] FIG. 13 illustrates a bimarker plane without training set data points, according to an embodiment of the disclosed diagnostic method. [Figure 18] FIG. 13 shows a bimarker plane including shaded areas of reduced influence for immune system response according to an embodiment of the disclosed diagnostic method. [Figure 19] FIG. 13 illustrates a bimarker plane including shaded areas with reduced influence for topology stability issues, according to an embodiment of the disclosed diagnostic method. [Figure 20] FIG. 13 illustrates a bimarker plane including shaded areas with reduced influence for known analyte measurement uncertainty, according to an embodiment of the disclosed diagnostic method. [Figure 21] FIG. 13 illustrates blind test results including two samples that failed the topology instability test and were corrected with a discrepancy algorithm according to an embodiment of the disclosed diagnostic method. [Figure 22] 3 is a flow diagram illustrating the general logic path followed by the software of the present invention, in accordance with an illustrative embodiment. [Diagram 23] 1 is a flow diagram illustrating a process for constructing a training set model (or diagnostic model) that generates a diagnostic score for a blind sample that determines the risk of having a diseased or non-disease state. [Figure 24]A typical population distribution is shown for the cytokine interleukin 6 (IL6). [Diagram 25] FIG. 1 shows the conversion of biomarker concentrations to proximity scores (a type of pseudo-concentration). [Figure 26] 1 shows a representative diagram of the hardware used to execute the software of the present invention, according to an exemplary embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] Without being limited to the above, in a preferred embodiment, the present invention relates to improving the predictive power and diagnostic accuracy of methods for predicting disease states using multivariate correlation methods, including proteomics, metabolomics and other techniques that involve the determination of the levels of various biomarkers found in bodily fluid and tissue samples.

[0018] Various embodiments devised by the inventors and described in this application include the use of meta-variables that specifically employ methods to adjust the impact of measured biomarker analytes on the correlation score. Such meta-variables can be identified based on expert knowledge of immune system responses and knowledge of possible measurement errors. These methods can be applied either to building a training set model or to blind samples of subjects.

[0019] In one aspect, the present invention relates to a method for diagnosing a disease, comprising the steps of: (a) determining the concentrations of at least three predetermined analytes in a blind sample from a subject; (b) selecting one or more meta-variables associated with the subject, the meta-variables varying for each member of the subject and the associated population, where the members of the population are known to have or not have the disease; (c) calculating a proximity score representing each analyte by transforming the concentrations of the analytes according to one or more population distribution characteristics and the one or more meta-variables; (d) comparing the proximity scores to a training set model of the proximity scores determined for the members of the population known to have or not have the disease; and (e) determining whether the comparison indicates that the subject has the disease. It is contemplated that the step (a) of determining the concentrations (or levels) of the predetermined analytes is performed at a time and place separate from the remaining steps of the method. Similarly, the other steps of the method may be performed in whole or in part at separate times and places. Thus, the inventors consider the method including only steps (b)-(f) to be their invention.

[0020] In describing the preferred embodiments of the invention as illustrated in the drawings, specific terminology will be relied upon for the sake of clarity. However, the invention is not intended to be limited to the specific terms so selected, and it should be understood that each specific term includes all technical equivalents which function similarly to achieve the same purpose. Several preferred embodiments of the invention have been described for purposes of illustration, and it should be understood that the invention may be embodied in other forms not specifically shown in the drawings.

[0021] For purposes of this application, specific terminology will be used to better describe the preferred embodiments of the invention, and is defined as follows:

[0022] "Analytical sensitivity" is defined as three standard deviations above the zero calibrator. Diagnostic indications are not considered accurate for concentrations below this level. Thus, clinically relevant concentrations below this level are not considered accurate and are not used for diagnostic purposes in clinical laboratories.

[0023] An "individual baseline analyte measurement" is a set of measurements of the biomarker of interest for an individual patient's transition from a non-disease state to a disease state measured for one individual multiple times over a period of time. The baseline analyte measurements for the non-disease state are measured when the individual patient does not have the disease, whereas the baseline analyte measurements for the disease state are determined when the individual patient has the disease. These baseline measurements are considered unique to the individual patient and may be useful in diagnosing the transition from non-disease to disease for this individual patient. The baseline analyte measurements for the disease state may be useful in diagnosing the disease for a second or more severe occurrence in this individual.

[0024] "Biological sample" means a tissue or bodily fluid, such as blood or plasma, taken from a subject, from which the concentration or level of an analyte (also called a marker or biomarker) that is informative in a diagnostic can be determined.

[0025] "Biomarker" or "marker" refers to a biological component of a subject's biological sample, typically a protein or metabolomic analyte measured in a bodily fluid such as serum proteins. Examples include cytokines, tumor markers, and others. Other indicators such as "biomarkers" and "markers" are also contemplated by the present invention, including, but not limited to, height, eye color, geographic factors, environmental factors, and the like. Generally, such indicators include any measurement or attribute that remains measurable, determinable, or observable even as it varies in a population.

[0026] A "blind sample" is a biological sample taken from a subject who has no known diagnosis of a given disease and for whom a prediction of the presence or absence of the disease is desired.

[0027] A "disease-related function" is a property of a biomarker that is either an effect of the disease to persist or worsen, or an effect of the body to stop the progression of the disease. In the case of cancer, the tumor acts on the body by requiring the maintenance or promotion of increased blood circulation, and the immune system enhances pro-inflammatory effects to kill the tumor. These biomarkers are in contrast to tumor markers that have no disease-related function, as they detach and enter the circulation and can be measured. An example of a functional biomarker would be interleukin 6, which triggers the immune system's action, or VEGF, which is secreted by tumors to cause localized blood vessel growth. On the other hand, an example of a non-functional one would be CA125, a structural protein found in the eye and female reproductive system that has no effect on the body to kill the tumor, nor on the tumor to help the tumor grow.

[0028] The "limit of detection" (LOD) is defined as two standard deviations of concentration value above the value of the "zero" concentration calibrator. Typically, the zero calibrator is performed in 20 or more replicates to obtain an accurate indication of the standard deviation of the measurement. Concentration determinations below this level are considered to be zero, or nonexistent, for example for viral or bacterial detection. For purposes of the present invention, a standard deviation of 1.5 is used when samples are performed in duplicate replicates, although the use of 20 replicates is preferred. Diagnostic indications requiring a single concentration number generally do not fall below this level. Measurements of the limit of detection level are statistically at a 95% confidence level. Prediction of disease status using the above method is not based on a single concentration, and it has been found that measurements at levels below the concentration-based LOD are possible.

[0029] A "low abundance protein" is a protein that is present in serum at very low levels. Although the definition of this level is not explicitly given in the literature, as used herein, this level would be less than about 1 picogram / milliliter in serum or plasma and other bodily fluids from which the sample is taken.

[0030] "Meta-variable" means information, other than analyte and biomarker concentrations or levels, that is specific to a given subject but not necessarily unique or unique to that subject. Examples of meta-variables include, but are not limited to, the subject's age, menopausal status (pre-menopausal, peri-menopausal, post-menopausal), and other conditions and characteristics such as pubertal stage, weight, geographic location or region of patient residence, geographic source of the biological sample, body fat percentage, age, race or race mix, or era.

[0031] "Population distribution" refers to the concentration range of a particular analyte in biological samples of a given subject population. A particular "population" refers, but is not limited to, to people selected from a geographic area, a particular race, or a particular gender. And the population distribution characteristics selected for use as described in this application further contemplate the use of two separate subpopulations within this larger definition of a population that are members of the population diagnosed as having a given disease state (disease subpopulation) or not having the disease state (non-disease subpopulation). The population can be any population for which disease prediction is desired. It is also contemplated that a suitable population includes subjects with a disease that has progressed to a particular clinical stage rather than other disease progression stages.

[0032] "Population distribution characteristics" can be determined in the population distribution of a biomarker, such as the mean or median concentration value or dynamic range of concentrations of a particular analyte, or how the population distribution separates into groups that are recognizable as distinct peaks as the onset and progression of disease affects the degree of up- and down-regulation of various biomarkers and meta-variables of interest as a patient undergoes a biological transition or progression from a non-disease to a disease state.

[0033] "Predictive power" means the average sensitivity and specificity of a diagnostic analyte or test, or the total number of false predictions (both false negatives and false positives) divided by the total number of samples minus one.

[0034] "Proximity score" refers to a surrogate or replacement value for the concentration of a measured biomarker, and is in effect a new independent variable that can be used in diagnostic correlation analysis. Proximity scores are calculated from and relate to the concentration of a measured biomarker analyte, such analyte having predictive power for a given disease state. As disclosed in WO2017 / 127822 and WO2014 / 158287, proximity scores are calculated using the population distribution characteristics of interest adjusted with meta-variables to change the actual measured concentration of a predictive biomarker for a given patient for which a diagnosis is desired. "Proximity score" and "pseudo concentration" have the same definition and can be used interchangeably.

[0035] "Specificity" is the true-false-positive rate of a test. Mathematically, it is one minus the number of false-positive measurements of a test divided by the total number of true-negative samples measured.

[0036] The "disjointed training set model" (or "secondary algorithm") is a secondary training set model that uses a different phenomenological data reduction method so that individual points of the grid in the bimarker plane are not destabilized in both the primary correlated training set model and this secondary algorithm.

[0037] "Spatial Proximity Correlation Method" (or Proximity Search or Cluster Analysis) is a method for determining correlation between independent variables and a binary outcome where the independent variables are plotted on orthogonal axes. Prediction of blind samples is based on proximity to a number of so-called "training set" data points (3, 4, 5 or more) for which the outcome is known. Binary outcome scoring is based on the total distance calculated from the multidimensional blind points to the training set points of the opposite outcome. The shortest distance determines the score of each blind data point. This same analysis can be performed in a bimarker plane that is a cut of a multidimensional lattice where each bimarker plane score is combined with the scores of the other planes and summed. This cut, or use of a two-dimensional orthogonal projection in space, can reduce computation time.

[0038] A "training set" is a group of patients (typically 200 or more to obtain statistical significance) with known biomarker concentrations, known metavariable values, and known diagnoses. The training set is used to determine the axis values ​​of the "bimarker" plane, "proximity scores," along with the score grid points from the spatial proximity analysis that are used to score the individual blind samples.

[0039] A "training set model" is an algorithm or group of algorithms constructed from a training set that allows the adjudication of blind samples with respect to a predicted outcome regarding the probability that a subject (or patient) has or does not have a disease. The "training set model" is then used to calculate a score for the blind sample for clinical or diagnostic purposes. For this purpose, a score is assigned across any range that indicates the percent likelihood of disease or non-disease, or some other predefined indicator reading value that is preferred by the healthcare provider providing the diagnosis to the patient.

[0040] A "Receiver Operating Characteristic (ROC) curve" is a graphical method of illustrating the performance of a signaling method used to make decisions when a trade-off is made between the rate of false positives and false negatives and the strength of the detection signal. In this graphical representation, the ordinate of the diagram contains the sensitivity of the testing method and the abscissa represents the false positive rate. For a biomarker (or signal) that includes an upward movement toward a disease trip point, the curve lies above the 45° null line that begins at the origin of the diagram (0,0) and ends at the top right of the diagram (1.0,1.0). The area under the curve indicates how useful the biomarker is in making predictions.

[0041] The "ROC curve 'area under the curve' (AUC)" is the area under the biomarker characteristic curve and the abscissa. For a completely useless biomarker, the AUC is 0.5, the area below the 45° null line above. A perfect test has an AUC of 1.0, extending from the origin to the 100% sensitivity point on the ordinate, then intersecting the ROC curve at the 1.0,1.0 point on the upper right.

[0042] The "tumor microenvironment" is the cellular environment in which a tumor resides, including the surrounding blood vessels, immune cells, fibroblasts, bone marrow-derived inflammatory cells, lymphocytes, signaling molecules, and extracellular matrix, bathed in tumor interstitial fluid (TIF).

[0043] "Tumor markers" are protein markers shed from the TME or blood supply that have no apparent function, either to stimulate tumor growth through tumor secretions or to suppress tumors through the immune system.

[0044] In recent years, cancer immunotherapy research has increasingly focused on the tumor microenvironment (TME), which represents a potentially vast repository of diagnostic content and an ideal R&D platform for the development and advancement of novel therapeutics. The TME, bathed in tumor interstitial fluid (TIF), is the cellular milieu in which tumors reside, including surrounding blood vessels, immune cells, fibroblasts, bone marrow-derived inflammatory cells, lymphocytes, signaling molecules, and the extracellular matrix.

[0045] TIF is important because it is also the transport fluid that connects the tumor (and TME) to the blood supply and is the "battlefield messenger" for active proteins that the immune system uses to try to suppress the tumor or that the tumor puts out to promote its growth. These competing proteins or cytokines, which are constantly at war with each other, are included in several functional categories of low-level signaling proteins: pro- and anti-inflammatory, anti-tumorigenic (or cell apoptosis), angiogenic, and neovascular.

[0046] Although recognized as a potential source of rich diagnostic information, the development of TIF analysis as a cancer screening modality has not progressed because sampling this fluid is very difficult and to do this implies knowing the location of the tumor and whether a tumor is already present. Even more challenging is detecting the presence of TME / TIF and malignancy without this knowledge. This requires a more accessible fluid for clinical diagnosis such as serum coupled with the analysis of a large number of proteins, known as proteomics, that can be presumed to correlate with the presence or absence of disease. Some problems in this regard arise with serum, as it is more confounded by the patient's physical symptoms than a direct route to detect the presence of active TME (and therefore tumor).

[0047] In this disclosure, we describe a method to analyze specific cytokines present in serum as accurate surrogates for proteins active in TME and TIF. This method requires several steps, including two dedicated processes called proteomic noise suppression and multidimensional (or spatial) correlation. The method we describe derives accurate surrogates for the activity of proteins found in TIF, and is thus useful for detecting the presence of active TME in organisms, and thus tumors. In essence, this method isolates traces of TME in serum, indicating the presence (or absence) of active TME, which indicates the presence of an active tumor. In addition, this method measures the modulation of these proteins, which provides useful information about the state of the tumor, the extent of its progression, and the stage, as well as the progress of tumor suppression by the immune system.

[0048] Biomarkers of interest The biomarkers of interest in this disclosure are circulating growth factors such as pro-inflammatory (interleukin 6, IL6, etc.), anti-inflammatory (interleukin 10, IL10, etc.), anti-tumor or tumoricidal cytokines (tumor necrosis factor alpha, TNFα, etc.), angiogenic (interleukin 8, IL8, etc.) and angiogenic cytokines (vascular endothelial growth factor, VEGF, etc.). These are cytokines with functions directly related to the immune system's response to the tumor or the tumor's effect on the body. The angiogenic factor VEGF is the tumor's effect of growing circulatory system in the growing tumor mass. The anti-tumorogenic factor TNFα is the immune system's effect of eradicating the tumor (apoptosis), and the pro-inflammatory factor IL6 is a mediator of the overall immune system effect. The anti-inflammatory IL10 is secreted by tumors into tumor interstitial fluid to suppress the immune system. And finally, angiogenic factors such as IL8 are secreted by tumors to promote angiogenesis in the surrounding tissues.

[0049] Generally, cancer is an inflammation-induced disease in which factors such as IL-6 are upregulated. However, in the three examples described in this application, advanced tumors secrete anti-inflammatory cytokines into the tumor interstitial fluid (and therefore blood). This effect is found to occur at advanced stages of cancer, stage 3 or 4 in lung and breast cancer, and at high Gleason scores (Gleason 8, 9, 10) in prostate cancer. At this point, the anti-inflammatory effect tends to downregulate the pro-inflammatory response of the organism's immune system. In some cases, in breast cancer, the angiogenic response is also suppressed at advanced stages. Finally, the tumor size increases, and the tumor angiogenic effect increases, as would be expected at stage 3 or 4 in breast and lung cancer, and at Gleason scores 8, 9, 10 in prostate cancer. All these effects occurring in the tumor microenvironment can be determined by sampling the organism's serum and applying the methods described in the referenced International Publication Nos. 2017 / 127822 and 2014 / 158287.

[0050] In certain cancer cases, recent therapeutic research has focused to a high extent on the so-called "tumor microenvironment" (TME) for the development of treatment modalities. Inhibition or promotion of regulation of proteins acting in the tumor interstitial fluid (TIF) found within the TME is considered to be a useful development route for these treatments. Proteins in TIF that have proven to be good indicators are generally from five functional cytokine groups: pro- or anti-inflammatory, anti-tumorigenic (or cell apoptosis), angiogenic, and neovascular.

[0051] Measuring the activity of these proteins can provide insight into tumor activity and therapeutic efficacy. For example, treatment modalities that promote or inhibit protein activity can be monitored in TIF to determine efficacy. Although appropriate for therapeutic applications where cancer is known to be present, sampling of TIF for diagnostic purposes has not been pursued. Its use as a diagnostic tool is meaningless, since the presence of TIF (and thus of the TME) by definition means that the patient has an active tumor at a known site. Additionally, accessing these proteins for diagnostics when present in other bodily fluids such as serum or urine has not been explored, as to date these have been unusable due to proteomic noise issues.

[0052] In this application, we describe a system and method to generate accurate surrogates for TME activity using active proteins in TIF using an easily accessible surrogate fluid - in this case serum (other fluids such as urine are possible). It should be noted that serum is a mixture of whole organism conditions (called "proteomic noise") and is not specific to tumors. Our proposed methodology removes the proteomic noise to allow for accurate determination of patient condition.

[0053] In general, the systems and methods disclosed herein require 1) selecting active TIF proteins indicative of a state within the TME, 2) measuring these proteins in serum surrogates, 3) suppressing proteomic noise to specifically identify cancer-associated activities within the proteins, 4) implementing correlation methods to amplify the effects of these proteins in a multidimensional matrix, and 5) scoring protein activities to indicate the presence or absence of cancer and, if present, its stage of progression. This is first done to create a training set of examples that represent the entire population and act as a standard against which individual samples are compared to determine their state - diseased or non-diseased.

[0054] Effect of biomarker combinations These cytokine biomarkers are very active in high-grade prostate cancer and are significantly up- or down-regulated compared to levels in "healthy" men, and therefore are very good indicators of disease state. Note also that they are active in lung and breast cancer. Figures 1, 2, and 3 show their effect as tumors progress. Note that in non-small cell lung and prostate cancers shown in Figures 1 and 2, IL6 down-regulates late-stage cancers or high Gleason scores 8, 9, and 10. Note also that in both cases, the transition from low-grade lung cancer or low Gleason score prostate cancer results in the down-regulation of IL6, as a result of increased interleukin 10 secreted by the tumor. Note also that secretion of IL10 into tumor interstitial fluid, and therefore into the blood, is associated with poor patient prognosis. This usually means that late-stage lung cancer is present. Thus, the combination of IL6 and IL10 in disease state correlation analysis is improved by using a combination of pro- and anti-inflammatory cytokines. Furthermore, it is noted that as tumors become late stage or more advanced, angiogenic cytokines generally continue to be upregulated.

[0055] Three of these biomarkers have unique ROC curve characteristics that are not common to tumor biomarkers. They have a flat portion at 100% sensitivity for some low level of biomarker concentration. They also have fairly wide areas under the curve (AUC), indicating that they are very good biomarkers for this disease, namely aggressive prostate cancer (PCa) and non-PCa. One of them has a linear vertical segment ascending the ordinate from [0,0], indicating that samples in this signal range should have a zero false positive rate for PCa.

[0056] In the scientific literature, there is limited publication of some of the best biomarkers mentioned herein, and nothing is found regarding the detection of high-grade PCa versus the general population, i.e., patients without PCa. In the case of VEGF, there is literature showing the upregulation of the biomarker, but nothing is written regarding the stage of the cancer, especially the Gleason score. Most of the literature is limited to the use of VEGF as a prognostic factor for the treatment of men already diagnosed with PCa. TNFα is also not related to differences in biomarker action regarding the stage of the tumor, especially the Gleason score. The same information is obtained from scientific investigations on interleukin 6. In fact, in low-grade PCa, the literature points out a slight upregulation of the biomarker. Our measurements do not support this, since in low-grade PCa we see a slight downregulation, but in high Gleason score PCa we see a very strong downregulation, and this cytokine, along with others, is a strong indicator of the presence of high-grade PCa. Much of the literature calls for the use of these biomarkers as prognostic factors in men with PCa and the in vitro expression of proteins from PCa cell lines, and for investigating methods to inhibit expression (especially VEGF) for therapeutic purposes.

[0057] ROC curve for prostate cancer VEGF The ROC curve for aggressive VEGF (Gleason score 7 (4+3), 8, 9, 10) is shown in Figure 4. Note the large flat portion of the ROC at the top where the sensitivity is 100%. This level of sensitivity or concentration levels of VEGF below about 50 pg / ml did not detect any high-grade PCa. The AUC of this biomarker for disease / non-disease is 0.87. In addition, the unique shape of the non-false positives below the 50 pg / ml level makes it a very good candidate for a high-grade "PCa" vs. "non-PCa" biomarker since no PCa is seen below concentration levels of about 50 pg / ml.

[0058] TNFα As shown in Figure 5, the comments for TNFα are the same as for the characteristics of the ROC curve for aggressive (Gleason scores 7 (4+3), 8, 9, 10). In this case, the AUC is also high at 0.85, and the trip point is the same as for non-false-negative results below about 6.5 pg / ml. TNFα also shows a portion of the curve with zero false-positive rate (abscissa) for samples above about 9.85 pg / ml. No false-positive results occur in this region.

[0059] PSA The comments for PSA, as shown in Figure 6, are the same as the ROC curve characteristics for aggressive (Gleason scores 7 (4+3), 8, 9, 10). In this case, the AUC is also high at 0.85, and the trip point is the same as for non-false-negative results below about 2 ng / ml. The ROC curve for the PSA analyte common to all Gleason scores of PCa is shown in green for reference (labeled "All PCa").

[0060] IL6 In contrast, IL6 shows strong downregulation in advanced (Gleason scores 7(4+3), 8, 9, 10) and as shown in Figure 7, the AUC is about twice that of the current PSA for detection of PCa in the general population (the curve would have to be inverted to account for this downregulation). One could speculate that high-grade PCa is probably more effective at suppressing the immune system, but that is not the point of this discussion. The fact is that this biomarker shows strong downregulation. Limited literature points to some upregulation in general PCa. Our measurements of PCa of all Gleason scores show some downregulation. Nevertheless, at higher Gleason scores, the cytokine shows strong downregulation. The overall PCa population is about 80% low grade, so at the time of sampling for PCa, low grade will dominate the characteristics of the cohort. This downregulation may be caused by the secretion of anti-inflammatory cytokines (IL10) when tumors progress to an aggressive Gleason score of 7 (4+3) or higher.

[0061] ROC curve for lung cancer IL10 Figure 8 shows the ROC curve for Interleukin 10 in separating low-grade (stages 1 and 2) from late-stage stage 3 and 4 non-small cell lung cancer. Note the upregulation in the transition from early (1 and 2) to late stage (3 and 4). This corresponds to the downregulation of Interleukin 6 and is caused by the anti-inflammatory effect of the tumor secreting IL10 into the tumor microenvironment and subsequently into the bloodstream.

[0062] IL6 The ROC curves of IL6 for early (1 and 2) and late (3 and 4) stage non-small cell lung cancer are shown in Figure 9. As demonstrated in Figure 9, this effect of IL6 is suppressed by the anti-inflammatory effect of the tumor.

[0063] VEGF The ROC curve for VEGF is shown in Figure 10 and demonstrates the upregulation of angiogenic factors seen in other cancers as tumors grow and progress to end stage.

[0064] Testing for advanced or terminal and non-advanced or early stage cancer These biomarkers can be combined to develop a very simple proteomic algorithm to monitor men with low-grade Gleason scores 5, 6, 7 (3+4) prostate cancer for progression to high-grade PCa with high-grade Gleason scores 7 (4+3), 8, 9, 10. These biomarkers can also distinguish early cancer stage 1 or 2 from stage 3 or 4. The combination of IL6 and IL10 with the opposing effects (by simple correlation methods such as logistic regression) yields a predictive power of 80%. The addition of proteomic noise suppression and spatial proximity correlation methods yields a predictive power of 90%. The addition of VEGF effects to the biomarker panel improves the predictive power to over 95%.

[0065] Testing Men with Advanced Prostate Cancer and Those without Cancer In fact, VEGF by itself gives a test with 76% predictive power and 100% sensitivity and 76% specificity (24% false positive rate). This simplified model simply excludes non-PCa in the concentration range excluded from the ROC curve and re-includes PCa in those zones included in the ROC curve. It then uses simple trip point counts and counts of positive and negative scoring for each biomarker that are not within the exclusion or inclusion criteria. The count must be greater than 3 out of 4 for those not pre-excluded or included. This simplified model gives a 100% representative sample set of 100 high Gleason score (defined as 7(4+3) or greater) PCa and 100 non-PCa samples. The combination of VEGF, IL6, TNFα, and PSA gives a predictive power of 90%. Additionally, the test predicts "non-cancer" in men with elevated PSA but no elevated cytokines. These men have benign prostatic hyperplasia or another non-malignant prostate condition and constitute a large population of false-positive results for the current PSA screening test for prostate cancer. A test incorporating these cytokines would solve this false-positive problem.

[0066] Cancer stage prediction Data from a breast cancer study supervised by the Gertsen Institute in Moscow made it possible to predict breast cancer stage with high accuracy using the instruments and reagents described below. 189 breast samples were obtained with stage information (0 to 4). Measurements were for the tumor marker PSA and the four cytokines mentioned above: proinflammatory (IL6), antitumorigenic (TNFα), angiogenic (IL8) and angiogenic (VEGF). In this case, the goal was to score each sample for the same staging information from the biopsy. The correlation methods are all binary in nature and cannot score the four different results without some manipulation. Therefore, the stage groups were combined with a binary group representing all stage groups: 1 plus 2,3,4, 2 plus 1,3,4, 3 plus 1,2,4 and 4 plus 1,2,3. All four groups were modeled and scored using age normalization, noise suppression, and spatial proximity correlation methods described in WO2017 / 127822 and WO2014 / 158287. The individual group scores for each sample were then used to calculate scores for each individual sample, adding a weight based on each's contribution to its group (1 or 1 / 3). This model has an accuracy of 99%.

[0067] Several methods are described herein to improve the predictive power of traditional proteomic correlation methods for diagnosing disease. These include 1) using metavariables and proximity scores for correlations, and 2) using special knowledge of topological stability and analyte measurement properties to adjust the influence of the bimarker plane in the training set model. Also described is a method for using mismatched training set models to find and correct blind sample stability issues unique to a particular training set model. Additionally, a method is described for finding and correcting corresponding non-disease conditions that are partially similar to the training set model for a given disease state. All of these methods are complementary and can be used in conjunction. For example, adjusting the training set model for areas of high probability of instability does not completely eliminate this issue from the blind sample prediction calculation, and therefore both methods can be used to improve predictive power. The inventors have found that combining these methods can provide predictive power of over 95%, and in the breast cancer study described in Example 1 below, a predictive power of over 98% (sensitivity 100%, specificity 97.5%).

[0068] Example 1: Clinical study evaluating a breast cancer blood test The performance of the OTraces BC Sera Dx test kit and the OTraces CDx immunochemistry instrument system (www.otraces.com) was evaluated in an experiment to determine the risk of the presence of breast cancer. The test kit measures the concentration of five very low level cytokines and tissue markers and uses the training set model developed above to calculate scores CSl and CSq to determine the risk of breast cancer. The proteins measured were IL-6, IL-8, VEGF, TNFα, and PSA. The experiment consisted of the measurement of approximately 300 patient samples roughly divided into 50% of the cases of breast cancer diagnosed by biopsy and 50% from patients presumptively considered to be non-diseased (or not breast cancer in this case). Within this group, the biopsy results of 200 samples were precisely divided into 50% non-diseased and 50% with breast cancer disease, and each group was further subdivided into the specified age groupings.

[0069] The sample analysis results were used to develop a training set model to predict disease status. The remaining samples (approximately 110) were then run as blind samples through the training set model to result in numerical cancer risk scores, and these scores were released to the host clinical center. These blind sample scores were then analyzed by the clinical center to determine the clinical accuracy of the results.

[0070] Two diagnostic models were developed for this experiment, referred to herein as Algorithm I and Algorithm II. A spatial proximity analysis method was used for both algorithms. The subject's age was used as a meta-variable rather than an independent variable, and the measured concentrations were converted into new independent variables, referred to herein as proximity scores, which were used directly in the correlation analysis. The difference between Algorithm I and Algorithm II is the number of new independent variables used in the correlation. Algorithm I uses five proximity score variables in a 10-dimensional cluster space. The lower limit of Algorithm I is two-dimensional, based on the fact that the correlation is performed, not on a particular method. Correlation inherently requires more than one dimension. The upper limit of Algorithm I is theoretically infinite, but is practically limited by computation time and power. The cluster space can be seen by the human eye through projection, or by cutting this multidimensional space to see a two-dimensional bimarker plane. In an exemplary embodiment of Algorithm I, there are 10 such planes.

[0071] Algorithm II uses 10 times more independent variables, so there are about 100 bimarker planes. It is expected that 200 samples in the training set model are sufficient to model the general population with a reasonable approximation. A secondary or mismatched training set model was developed from the same 200 sample training set data set. The training set model is the primary scoring method used to explain the results herein. The mismatched training set model is used to process cancer scores from the primary training set model that are considered to be unstable, i.e., scores that lie in areas of topological instability. Although the mismatched training set model is somewhat less accurate with blind samples, the predictive power can still be improved by processing the primary training set model.

[0072] The spatial proximity analysis method described above has a significant advantage over logistic regression in that it can accommodate highly nonlinear trends in the independent variables used to determine the calculated outcome. The outcome is disease or non-disease (cancer or non-cancer in this case) based on the proximity scores of the training set model calculations. A drawback of this method is the highly nonlinear areas associated with very steep topological gradients. Thus, we see the effect of unknown (or blind) samples being located at steep peaks or deep, pointed valleys, which amplify small errors in the calculated proximity scores. We determined the stability of the calculated scores with a proprietary stability test, and then used Algorithm II to mediate Algorithm I for cases that showed stability.

[0073] Figures 11, 12 and 13 show the training set results for Algorithm I. The model itself consists of 10 bimarker planes with 40,000 topological points each scored for non-disease and disease (here breast cancer) by the spatial proximity method. The model's ability to separate the two sets, non-cancer and cancer, is shown in these figures. The model must be built from very close to one of the two outcome states, or preferably exactly 50%-50%, 50%-50%. This method also uses age as a transformation metavariable. The training set samples had samples distributed across all age groups of interest. The model for Algorithm I (Figure 12) was built with 100 healthy women and 98 women with breast cancer. The tabulation in Figure 12 shows the numerical results, where N=198 is the number of samples. CI is the correct sample, FI is the incorrect sample, and 4 samples were considered indeterminate. To distinguish the four uncertain samples resulting from the use of the primary training set model, a secondary training set model was developed. This model is the non-matching training set model. This secondary model uses the same training set data as the primary. Figure 13 shows the results of the non-matching training set model calculation. Algorithm II shows 100% separation with more than 60 separation points.

[0074] Blind sample test results in breast cancer research Figure 14 shows the results of blind samples evaluated in a clinical study. The results show a sensitivity of 100% and a specificity of 97.5%. The oncologists at the clinical research center set the diagnostic transition value so that all breast cancer positive samples were correctly identified. Thus, two non-disease samples were marked as cancer positive. This is clinically correct, since all samples marked as positive undergo the next diagnostic step of imaging mammography. Many women do not undergo imaging mammography because they do not live near facilities with medical equipment. However, blood can be drawn at a remote location of the clinical laboratory and shipped refrigerated to a laboratory in a large city.

[0075] Example 2: Using the metavariable "age" to improve diagnostic accuracy

[0076] Table 1 (below) shows the aggregate results of a clinical breast cancer study with a sample of 868 subjects.

[0077] [Table 1] Table 1: Summary of diagnostic accuracy for breast cancer

[0078] Table 2 (below) shows a comparison of various methods for calculating correlations. The standard method, logistic regression, shows a predictive power of only 82%. A refinement of this, standard spatial proximity analysis, gives a predictive power of about 88% in linear form and 90% in logarithmic form. The method described here, using metavariable and weighting approaches, topological stability adjustments, immune system reaction grouping, and weighting adjustments for analytical performance - when combined with blind sample instability testing and discrepancy algorithm corrections - gives a predictive power of greater than 97%.

[0079] [Table 2] Table 2: Comparison of predictive power of disease correlation calculations

[0080] Example 3: Using the metavariable "age" to improve diagnostic accuracy in ovarian cancer research

[0081] Table 3 (below) shows the results of a study of 107 women with and without ovarian cancer using the metavariable methods described in embodiments of the present application. Although the study did not use all of the predictive power improvements described herein, it achieved relatively good predictive power of approximately 95%.

[0082] [Table 3] Table 3: Summary of ovarian cancer diagnostic accuracy

[0083] Example 4: Using the metavariable "age" to improve the accuracy of prostate cancer diagnosis Prostate cancer

[0084] Table 4 (below) shows the results of a study of 259 men with either prostate cancer or benign prostatic hyperplasia (BPH) using the metavariable method described herein. This study also did not use all of the predictive power improvements described herein, but still achieved a relatively good predictive power of about 94%. Note that BPH is the most common condition that produces false positive results with current prostate cancer PSA tests. Men with BPH have about 4 out of 5 positive cases in traditional prostate cancer diagnoses, resulting in the majority of prostate cancer biopsies being negative. The metavariable method can correct these incorrect diagnoses as described above.

[0085] [Table 4] Table 4: Summary of diagnostic accuracy for prostate cancer

[0086] The results above in Examples 3 and 4 (for ovarian and prostate cancer, respectively) did not use meta-variable or effect adjustment methods (LOD, subpopulation grouping, and instability), nor blind sample stability methods.

[0087] To further improve predictive power, these age and grouping adjusted concentrations are normalized to reduce or eliminate spacing bias (also known as spatial bias) of clustering in marker plots by multidimensional grouping for spatial proximity analysis. For example, see FIG. 15 where the bimarker planes for IL6 and VEGF are seen. Ten of these planes are for a five-biomarker breast cancer testing panel. In this case, the calculated proximity score values ​​are normalized and shifted to obtain arbitrary values ​​between zero and 20, and outliers with high upregulated concentrations are highly compressed.

[0088] Each of the bimarker projections onto the multidimensional marker plane at the same normalized interval across concentrations by age / grouping analysis is compressed and normalized to the age-adjusted means along with the age (or total population) adjusted subgroupings.

[0089] Improving the predictive power of training set models using adjustable bimarker plane influence levels Typically, the bimarker plane is scored with a binary number for non-disease and disease (e.g., +1 and -1). The proximity scores described herein can receive further refinement in predictive power by selectively adjusting the influence levels of these two binary numbers. The following method is developed in a training set model and anchored to the model.

[0090] Figures 16 and 17 show the projection of one bimarker plane for the case of a five biomarker used to predict the presence of a disease state, in this case breast cancer, using five markers: IL-6, IL-8, TNFα, VEGF, and PSA. Figure 16 shows the training set model with the data used to score the grid points on the map by the spatial proximity analysis method. Figure 17 shows the training set model without the data. This constitutes the training set model. There is no training set data used to create the model, since each of the 40,000 grid points is scored and blind samples are scored according to where they are located on the grid. The topology shows positive for cancer in red and negative for cancer in blue. In this case, non-disease grid points are set to +1 and disease (cancer) grid points are set to -1 when calculating the overall score. Each bimarker in this five biomarker example is analyzed in five orthogonal spaces, of which Figure 16 is one of the two-dimensional projections. In this figure, the topology of the various subgroupings of the immune system response is shown. In this case, all grid spots (2000 x 2000 or 40,000 in this case) are scored in the usual way, with values ​​assigned being -1 for positive disease status (breast cancer) and +1 for non-disease. This bimarker plane is normalized by the proximity score interval and for the meta-variable age as above.

[0091] FIG. 18 shows the same bimarker model, plus immune response grouping within the grey area (see FIG. 24). The influence of the grey area has been adjusted to reflect the fact that each grey block area has a somewhat different impact on the probability that a patient is non-diseased or diseased. This adjustment can be done by human estimation with validation of the training set, or by sophisticated computational multivariate incremental analysis. These adjustments improve the training set model. Two separate bimarker planes are created for the two outcomes, diseased and non-diseased states. In this case, the blind data points in immune response group IV are much more likely to be diseased, and their influence is increased slightly (absolute value) (e.g., by changing the score from -1 to -1.1). The actual amount of this increment would preferably be determined by computational analysis, and possibly by sophisticated manual methods. This method can also be implemented for spatial proximity (also known as pseudoconcentration) correlation analysis methods, although other means could be used to the same effect. These methods of weighting the influence of disease associations provide an improvement in predictive power of about 1%. With a predictive power of over 95%, this is highly significant.

[0092] FIG. 19 shows the same bimarker plane again with grey areas circled by complex areas of nonlinear and rapidly changing disease and non-disease topologies. Such areas can be identified by inserting the test blind sample values ​​of the projection noise (i.e., + / - 10%) into the model and then introducing the amount of noise from the measurements. Most of these blind points will not change substantially in the disease (here cancer) score. However, some grid points are found that change dramatically from non-disease to disease scores after this kind of noise adjustment. These are areas where most or all of the bimarker planes have rapidly changing topologies that overlap with the entire multidimensional bimarker plane. By carefully reducing the influence in these areas, the weighting can be increased in a few relevant bimarker planes where the noise data lies in a broad plane without changing the results at the neighboring boundaries. This method has been shown to correct erroneous predictions. In the above case, the influence of the red cancer areas would be, for example, a downshift (absolute value) from -1.0 to -0.9. Alternatively, the blue non-disease areas would be shifted down from +1.0 to -0.9. The optimal level of shifting can be determined by precise computational analysis.

[0093] Analytical noise can affect the accuracy of correlation analysis. This noise is particularly problematic below the detection limit of the analyte. This noise can also be eliminated by reducing the influence of measurement points for individual biomarkers that are in these unstable zones. Figure 20 again shows the bimarker planes for PSA and IL-6 for a breast cancer panel. All areas within the grey rectangular area at the bottom left of the figure are below the traditional assay limit of detection (LOD) of the analyte. Traditionally, the LOD is defined as two standard deviations of the 20 zero calibrators plus the average of the values ​​of the 20 zero calibrators. The statistical certainty of values ​​at this level is 95% within two standard deviations, and of course the measurement certainty decreases as the measurement sample falls below the LOD. The data still contains useful information, but should be applied to a less influential analysis. In this case, the influence of the blind sample data points in the grey area is reduced, for example from +1.0 to -0.9 for the grid points of the training set model in the grey area. This increases the influence of data points for this test sample that are above the detection limit in the other bimarker planes. The above methods are complementary and can be implemented synergistically.

[0094] A method to improve predictive power by testing blind samples for instability Once the training set model is complete and fixed, it is used to calculate the cancer score for the blind patient sample. We use two preferred methods to generate the cancer score. The first method, called the linear method (CSl), multiplies the topological location scores (+1 or -1) by the predictive power of this bimarker plane. These are then added, scaled, or shifted to obtain a score between 0 and 200. The second score, called the q score (CSq), is calculated by taking the square root of the sum of the squares of these same values. This second method highlights the differences in individual bimarker scores, which is beneficial to the physician's overall final diagnosis.

[0095] Due to the highly nonlinear nature of the spatial proximity correlation method, topological instability remains in the bimarker plane and is never completely eliminated. According to another aspect of the invention, a stability test and a technique involving projected noise can be applied to the blind data set. A mismatched training set model can then be used to reconcile or modify the cancer score. In this aspect of the invention, a certain level of noise (e.g., plus or minus 10%) is introduced for each blind patient data set. If the blind sample set is about 100 patients, an actual training set model computer implementation would be for a set of 300 samples, each tripled (raw data plus noise and minus noise). The resulting tripled data set is tested for stability (a is -10%, b is +10%, c is raw data). Table 5 (below) shows the results of the stability test for data from a clinical study. Note that three samples show very high instability in the cancer score. Samples 138, 207, 34, and 29 all show very high figures of merit. The figure of merit (lower is better) should include both the degree of score shift and, specifically, whether the score shifts for predicting healthy vs. cancer or vice versa. In these data sets from blinded samples, there is a high risk of the diagnostic prediction being wrong.

[0096] [Table 5-1] Table 5-1: Topology instability checks [Table 5-2] Table 5-2: Topology instability checks

[0097] The unmatched training set model can be used to process the "risk" patient sample data sets that failed the performance noise test. These points are at risk due to unavoidable measurement noise, either random or systematic, combined with extreme topological instability caused by the fact that most, if not all, of the blind sample data points are located on very steep slopes of the bimarker plane such that small perturbations will cause large swings in the scores. Table 5 shows samples with noise introduced. Each sample has three values: 1) plus noise, 2) minus noise, and 3) raw data without noise. These samples show cancer scores that change from diseased to non-disease and also from non-disease to disease with the introduction of noise of +-10%. In this case, these sample data are deemed unstable. The level of instability is not precisely defined and adjustments can be made for various levels of noise introduction. In this case, they are corrected for noise of +-10% and stability scores greater than 2000 (note that stability scores and cancer scores are two separate numbers with different meanings).

[0098] Measurement noise can be reconciled with this discordant second algorithm (algorithm II). The discordant algorithm used for reconciliation improves the chances of being a correct score, so it can be used to correct these "risk" patient sample sets, even if it has a slightly lower predictive power than the main algorithm. In this case, two were corrected (see Figure 21). Sample 138 had a non-disease score of 85, which was corrected to 195 with the discordant algorithm (this score remained stable with algorithm I), and sample 34 had a score of 102 (linear method), which was also corrected to 198 with algorithm II. Sample 29,207 was not changed by the discordant algorithm.

[0099] The mismatched training set model (Algorithm II) uses 105 bimarker planes and is mismatched with the primary training set model (Algorithm I) in that these same samples show stability as the stability test of Algorithm II. The test of the mismatched training set model is done exactly the same as the primary training set model. Note that a logistic regression method could also be used to calculate these sample scores. Algorithm II was used because it has high predictive power. The processed training set model could be used even if it has lower predictive power than the main algorithm (preferably at least 50% predictive power) as long as it has a likely correct result without instability. Note that the correction is significant for the blind samples in question that failed the noise test. These samples were actually all high scoring cancers. Eight of the 10 bimarker planes for these samples were on topologies with very high instability lattice points. Thus the scores were at risk and indeed inaccurate (one was wrong and one was inconclusive with a score of 100 / 120). In this case, one sample was corrected to increase the predictive power from 97% to 98%, resulting in a very high error reduction (50%). One sample was indeterminate but changed to cancer and was also corrected.

[0100] A method for improving the predictive power of disease-status-adjusted binary outcomes by excluding isolated conditions that partially resemble one of the outcome conditions of the primary disease analysis. Spatial proximity analysis usually uses three or more independent variables, usually serum protein concentrations of the patient. Although correlation algorithms only work on a binary outcome of non-disease or disease, a continuous scoring is performed that more closely relates to the probability that the actual outcome is the two binary states. In some cases, other conditions that are nominally classified as non-disease are found that partially resemble disease states within the population distribution of the biomarkers used. In some of these cases, this non-disease "look-alike" condition produces a false positive result of the correlation analysis. The way to resolve this type of false positive result is to create an additional new correlation analysis that is completely separate from the non-disease or disease analysis. This new correlation analysis preferably uses the exact same biomarker measurement data for the non-disease or disease correlation, or uses some or all of the different biomarkers. This new correlation analysis provides a "non-disease look-alike" or "disease" outcome, or at least a score that allows a judgment to be made about the patient's actual situation. Indeterminate or approximate transition scores for non-disease or disease analyses combined with very low or high scores for non-disease similarity or disease correlation can help practitioners improve disease status determination and reduce false positive scores.

[0101] An example of this situation where a non-disease state closely resembles a disease state is the non-malignant condition benign prostatic hyperplasia (BPH). This condition usually exhibits elevated levels of at least one biomarker used to diagnose prostate cancer. For example, the biomarker prostate specific antigen (PSA) is elevated in men with both BHP and prostate cancer. Table 4 shows that this additive correlation analysis method distinguishes between men with BHP and prostate cancer, and similarly, using the same biomarkers but a different training set model, it can distinguish between men with a presumed non-disease state and men with a disease state and confirmed prostate cancer. A small percentage of men will have false positive results, which are distinguished by the training set models for non-disease and cancer. In these cases, two scores, one for presumed non-disease and cancer, and one for BHP and cancer, can help a physician or other medical professional determine the next diagnostic step. For example, with a total scoring of 0 to 200 for both models (either CSl or CSq), a score of 110 for "non-prostate or prostate cancer" would be a low score for cancer positive, but considering a secondary score of 30 for BPH or cancer would indicate to the practitioner that BPH is more likely than cancer. The practitioner would use this additional information along with other medical information and the patient history to determine the next diagnostic steps.

[0102] Detailed description of each method Analysis steps / algorithms The process for developing an analytical model in accordance with the present invention generally follows the following logic path, as illustrated in FIG.

[0103] In step 2200 "Collect Patient Samples," the software collects a large group of known non-disease and disease patient samples. The samples are generally not screened for any other unrelated conditions (non-malignant in the case of cancer) and are collected so that the sample set appears statistically to the general population.

[0104] In step 2202 "Measure Biomarker Concentration," the software measures the biomarker parameter concentration using methods and devices well known in the art.

[0105] In step 2204 "Calculate Proximity Score for Each Biomarker", the software calculates the proximity score curve for each biomarker and sets the zones for each, as shown in FIG.

[0106] In step 2206, "Score Sample as Cancer or Non-Cancer," the software runs the model program to score the sample using a spatial proximity correlation method. The model uses a unique compression or renormalization equation for each of the four zones (see Equation 1 below).

[0107] In step 2208 "Check and correct scoring", the software checks each sample for topological stability and corrects those that fail the discrepancy algorithm. First, all cancer scores are checked for topological stability in the usual way by introducing plus or minus noise to the measured intensity levels, calculating dithered proximity scores, and applying these to the primary spatial proximity model. If these dithered cancer scores shift beyond a predefined limit, the cancer scores calculated using the primary model are rejected. The original intensity levels of the failed tests are then moved to new proximity scores using the secondary or discrepancy model. These new proximity scores for these failed samples are then applied to the spatial proximity correlation model. These new cancer scores are then checked for stability with the secondary model in the same way. If the samples pass the stability check, they are reported as analyzed by the discrepancy model. If both the primary and secondary models are unstable, the samples are reported as indeterminate.

[0108] Finally, in step 2210, "Classify diseased or non-diseased with training set model", the software outputs the results.

[0109] Instrumentation and reagents used in breast cancer confirmatory studies: description of testing platform OTraces CDx Instrumentation System The test data contained below and many of the runs noted above were measured on equipment with the reagents listed below. Data was processed on an OTraces LIMS system and in some cases calculations were completed on PC based software. All calculation software was written and validated by OTraces Inc. It will be readily apparent to one of ordinary skill in the art that other equivalent hardware, equipment and reagents could be used to achieve similar results.

[0110] The CDx instrumentation system is based on the Hamilton MicroLab Starlet system, which has been customized with programming to transfer the OTraces immunoassays to the Hamilton Rapid ELISA Robot. Hamilton Company is a reputable company that sells automated liquid handling systems worldwide, including the MicroLab Starlet. The unit has been customized for the OTraces by Hamilton Company to provide full automation. The OTraces CDx system includes an integrated Microplate Washer System and Reader. These two additional pieces of equipment allow the system to complete a run of all five immunoassays in the test panel in one shift without operator intervention after initial setup. The configured system completes 40 cancer scores daily. Modifications include software to run one analyte of interest at a time. Thus, it is necessary to be able to rerun a particular test if an error occurs during the run of the entire test.

[0111] BC Sera DX Test Kit The test kit includes all the reagents and disposable equipment to perform the 120 cancer test scores, including all buffers, blocking solutions, washes, antibodies, and calibrators. The only modification required to fully commercialize the test kit is the addition of two control samples. These controls provide independent confirmation that the appropriate cancer scores are obtained from the "blind" test samples. The two controls are designed to produce close scores of 50 and 150, respectively. The LIMS system (see below) QC program verifies that these controls are correct and validates each test run at the site. The test kit is manufactured in a GMP facility and is CE marked. The microtiter plates are pre-coated at the facility with the capture antibody and protein blocking solution.

[0112] Laboratory Information Management System (LIMS) For example, all clinical chemistry systems currently marketed by Roche and Abbott include a graphical interface with sufficient software to manage patient data, quality control the instrument and chemistry operations, and facilitate test sample identification and loading into the testing system. These menus are integrated with the delivered chemistry system. Autolaces' business model involves these functions on Autolaces' computer servers located at Autolaces' US facilities, and uses cloud computing to seamlessly connect the CDx instruments to these servers via the Internet. This provides several significant advantages: 1) The LIMS software incorporates FDA-compliant archiving software so that data from all tests performed by each CDx system deployed in the field is stored on the Autolaces server. Applying feedback from the installed base, input from the central facility on patient results allows Autolaces to collect FDA-compliant data for US-based FDA market clearance submissions. 2) The instrument and LIMS can be connected to all QC test results from the factory QC tests, preferably with barcoded reagent packaging. These data are available in real time as tests are performed on-site for further confirmation of on-site test results. 3) The CDx system is run only against AutoLacease confirmatory reagents and therefore does not allow for testing using non-AutoLacease reagents. The system presents the operator with a typical user interface appearance with all functions performed in real time and patient reports are available as soon as test runs are completed.

[0113] The step-by-step process for developing a training set model and calculating a risk score is shown in the flow chart of Figure 23. This process may be implemented in software in an embodiment of the invention. The construction of a training set model is performed first, and the final product allows for the production of diagnostic results for unknown patient samples, called blind samples, since the correct diagnosis is not known at the time of analysis for these blind samples. In general, the present invention provides a risk score to a caregiver who considers the risk score along with other patient factors to make a clinical decision about the presence or absence of a given disease state.

[0114] Steps 2302 through 2318 outline the process by which a training set model is created. In step 2302, the software defines the training set sample requirements in terms of diagnostic needs, which are predefined criteria established by one skilled in the art. For example, these criteria could be diseased versus non-disease states, and more specifically, for example, in breast cancer, a comparison of samples known to be breast cancer positive versus non-breast cancer.

[0115] In step 2304, the software defines the independent variables (ie, biomarkers) that will be measured along with the meta-variables that will be calculated.

[0116] In step 2306, the software collects training set samples according to the parameters set in steps 2302 to 2304. In step 2308, the software uses the appropriate medical equipment for each training set sample to determine the correct disease diagnosis associated with these results along with the independent and meta-variables from the measurements. In step 2310, the software calculates the bimarker topology for each of the training set samples. In step 2312, the software calculates the optimal bimarker topology weighting or influence adjustment for: (1) samples determined to be below the detection limit uncertainty, e.g., conventional detection limit; (2) extreme topological instability, determined, e.g., by the method described in paragraph

[0070] and with respect to the topological stability described above. In step 2314, the calculations are considered complete and the primary training set model is frozen for the diagnosis of the disease (e.g., cancer). In step 2316, the software develops a secondary training set model using essentially non-matching correlation modeling (see, e.g., FIG. 10). In step 2318, the computations are considered complete and frozen as a secondary training set model for the diagnosis of the disease state is created. In this manner, a set of training set models for the diagnosis of the disease is created.

[0117] Steps 2320 through 2338 describe how the software of the present invention uses the training set model developed to diagnose a disease such as cancer. In step 2320, the software measures blind sample independent variables such as biomarkers using medical instruments similar to those used in the development of the training set model. In step 2322, the software obtains or measures and calculates meta-variable data for each blind sample. In step 2324, the software uses this data to calculate an initial disease status risk score for the blind sample using the primary training set model. In step 2326, the software determines the topological stability of the blind patient sample scores. In step 2328, the software checks whether the score passes the topological stability test. The pass / fail criteria include how large the instability induced error is and, most importantly, determining whether the score changes from disease positive to negative or vice versa. In step 2330, if the score is found to pass the stability test, the diagnosis report and risk score are output and / or published. If the score does not pass, then in step 2332 the software further calculates a secondary disease state risk score using the discrepancy method algorithm (Algorithm II) described above. In step 2334 the software again checks whether the score passes the topological stability check. If in step 2336 it is found that the score passes the stability check then a diagnostic report and risk score is output and / or published. If the score still does not pass, then in step 2338 the software prepares a diagnostic report and output and / or publishes the result as indeterminate as to whether the disease state is present.

[0118] The primary score calculated here has several unique properties. In one embodiment, the mean value of the protein is embedded in a logarithmic compression as a ratio to the actual measured concentration for patients of that age. Essentially, this method creates a fan of similar equations, for example, each unique to the age of the patient population. Each unknown sample is given its own equation for the age of the sample.

[0119] A relationship can be used that includes age-adjusted mean values ​​for non-diseased and diseased and actual patient sample concentrations of the following form:

[0120] equation 1 Proximity score = (K)*ln((C i / C (c or h) )-(C h / C c )) 2 K=constant of proportionality C i = measured concentration of actual patient analyte C (c or h) = Patient age adjusted concentration of this patient analyte. Values ​​are adjusted for whether the patient is non-diseased or diseased. C h = patient age-adjusted mean concentration of analyte in non-diseased patients C c = patient age adjusted mean concentration of analyte in diseased patients.

[0121] This equation 1 is designed to adjust compression and expansion depending on the upregulation grouping zone, as shown in Figure 25. The above formula for the proximity score meets this requirement. However, as will be apparent to those skilled in the art, many other forms of this equation can be implemented. For example, C i ,C h ,C c would be the actual concentration or concentration distance by the mean, mean, distance from the subgroup median or edge of the dynamic range, as above. Other variations of this calculation are reproduced below as Equations 2 and 3.

[0122] equation 2 Proximity score = K*ln(((concentration of unknown sample) / (mean concentration of cancer at unknown sample age))-((mean concentration of non-cancer at unknown sample age) / (mean concentration of cancer at unknown sample age))) 2 .

[0123] From this equation, negative infinity (the natural logarithm of zero) is obtained when the unknown sample is equal to the non-cancer mean for the unknown sample's age. This is overwritten by the actual detailed equation to a set value, e.g., 2, as shown in Figure 25. In other words, values ​​outside the preset range are checked by the software and reset to the limits of the preset range.

[0124] equation 3 Proximity score = K*ln(((concentration of unknown sample) / (mean concentration of non-cancer at unknown sample age))-((mean concentration of cancer at unknown sample age) / (mean concentration of non-cancer at unknown sample age))) 2 .

[0125] From Equation 3, we obtain the natural logarithm of zero, which is negative infinity when the unknown sample is equal to the cancer mean for the unknown sample's age. This embodiment of the equation is used when the unknown sample is above the midpoint concentration between the non-cancer and cancer mean for the unknown sample's age (presumed cancer). In this situation, the entire equation is inverted to positive infinity when the unknown sample is the mean cancer concentration for that age. This infinity is overwritten to a set value, e.g., 18, in the actual detailed equation. The graph in FIG. 25 shows the set of equations that are obtained for the age range of interest. The equations work independently in each of the four zones shown in FIG. 25. The zones are: 1) below the non-disease population mean, (2) above the non-disease mean and below the midpoint derived between the non-disease and disease mean (non-disease / disease transition), (3) between the midpoint derived between the non-disease / disease mean and the population disease mean, and (4) above the mean for the disease state. Note that these zones do not indicate that samples located within the zones are diseased or non-disease. The true diagnosis of an individual sample is either, or its location if it is "incorrect" due to another condition affecting the biomarker. We call this proteomic noise. The zones simply refer to how the individual samples relate to the mean, providing a scaffold for compression or renormalization, as illustratively done by Equation 1. Note that each equation represents only one age value, and that the entire set comprises many equations, each representing a single age value. The entire set of equations is devised to set the proximity score value to the same predefined value for all ages when the actual concentration is exactly equal to the mean value. The ages shown are 35, 50, and 65. The entire set looks like a fan, with one equation for each unknown sample age.

[0126] Proximity scores (not concentrations or levels, since they are absolute numbers) are illustratively calculated as described above and used in spatial proximity correlation multidimensional plots for analysis. All plots are also normalized to a common characteristic of the population distributions, namely the age means, medians, or dynamic ranges of subgroupings of non-diseased and diseased (age adjusted or not). These methods can provide predictive power improvements of 5 or more percentage points.

[0127] The above exemplary embodiments of the system and method of the present invention may be implemented in software that functions in both networked and non-networked hardware. An exemplary embodiment of hardware used to implement the invention is described in connection with FIG. 26. In the exemplary system 2600, one or more peripheral devices 2610 are connected to one or more computers 2620 through a network 2630. Examples of peripheral devices 2610 include smartphones, smart watches, tablets, wearable electronic devices, medical devices such as EKS and blood pressure monitors, and other devices that collect biomarker data as known in the art. The network 2630 may be a wide area network such as the Internet or a local area network such as an intranet. Because of the network 2630, the physical location of the peripheral device 2610 and the computer 2620 does not affect the function of the invention. Both implementations are described herein and, although not specified, it is contemplated that the peripheral device 2610 and the computer 2620 may be in the same or different physical locations. Communication between the hardware components of the system may be achieved using a number of well-known techniques, such as network connection components such as modems and Ethernet adapters. Both the peripheral device 2610 and the computer 2620 include or are attached to communications equipment. Communications are expected to occur over industry standard protocols such as HTTP.

[0128] Each computer 2620 comprises a central processing unit 2622, a storage medium 2624, a user input device 2626, and a display 2628. Examples of computers that may be used are commercially available personal computers, open source computing devices (e.g., Raspberry Pi), commercially available servers, and commercially available portable devices (e.g., smartphones, smart watches, tablets). In one embodiment, each of the peripheral devices 2610 and each of the computers 2620 of the system have software associated with the system installed thereon. In such an embodiment, biomarker data may be stored locally on the networked computers 2620 or on one or more remote servers 2640 accessible to any of the networked computers 2620 through the network 2630. In an alternative embodiment, the software is implemented as an application on the peripheral device 2610.

[0129] While certain features of the described implementations have been illustrated as described herein, many variations, substitutions, changes, and equivalents will occur to those skilled in the art. It is therefore intended that the present invention is not limited to the particular embodiments or arrangements disclosed, but rather to cover any changes, modifications, or variations that are included within the scope and spirit of the invention as defined in the appended claims. It is expressly intended that all references cited in this application, including patents and applications, are incorporated in their entirety by reference. [Appendix 1] 1. A computer-implemented method for diagnosing a disease, comprising: (a) receiving a first set of one or more concentration values ​​of a first biomarker from a first patient sample, the first patient sample comprising a non-disease diagnosis; (b) receiving a second set of concentration values ​​of one or more of the first biomarkers from a second patient sample, the second patient sample comprising a disease diagnosis; (c) calculating a first set of proximity scores from the first set of density values ​​and a second set of proximity scores from the second set of density values; (d) calculating a correlation between the first marker and a disease diagnosis from the first and second sets of concentration values ​​and the first and second sets of proximity score values, the correlation being one of simple regression, ROC curve area maximization, topological stabilization, or spatial proximity analysis; A computer-based method comprising the steps of: [Appendix 2] The computer-based method of claim 1, wherein steps (a)-(d) are repeated for up to five biomarkers. [Appendix 3] 2. The computer-implemented method of claim 1, wherein the association is a combination of two or more of the simple regression, the ROC curve area maximization, the topology stabilization, and the spatial proximity analysis. [Appendix 4] 2. The computer-implemented method of claim 1, wherein the first and second patient samples comprise at least one of a blood sample, a urine sample, or a tissue sample. [Appendix 5] 2. The computer-implemented method of claim 1, wherein the disease being diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer. [Appendix 6] 6. The computer-implemented method of claim 5, wherein the disease diagnosed is the stage of the prostate, breast, lung, or ovarian cancer based on Gleason score. [Appendix 7] 7. The computer-implemented method of claim 6, wherein the first and second patient samples comprise cancer stage data, and the cancer stage data is classified into a plurality of binary groups. [Appendix 8] 8. The computer-implemented method of claim 7, wherein each of the binary groups is scored. [Appendix 9] 2. The computer-based method of claim 1, wherein the biomarkers are selected from functional groups of cytokines, and the functional groups of the cytokines are at least three of pro-inflammatory, anti-inflammatory, anti-tumorigenic, cell apoptosis, and angiogenesis. [Appendix 10] 2. The computer-based method of claim 1, wherein the first biomarker is VEGF. [Appendix 11] (a) receiving a first set of one or more concentration values ​​of a first biomarker from a first patient sample, the first patient sample comprising a non-disease diagnosis; (b) receiving a second set of concentration values ​​of one or more of the first biomarkers from a second patient sample, the second patient sample comprising a disease diagnosis; and (c) calculating a first set of proximity scores from the first set of density values ​​and a second set of proximity scores from the second set of density values; (d) calculating a correlation between the first biomarker and a disease diagnosis from the first and second sets of concentration values ​​and the first and second sets of proximity score values, wherein the correlation is one of simple regression, ROC curve area maximization, topological stabilization, and spatial proximity analysis; and A non-transitory computer-readable medium storing a program that causes a computer to execute a process including the steps of: [Appendix 12] 12. The non-transitory computer-readable medium of claim 11, wherein steps (a)-(d) are repeated for up to five biomarkers. [Appendix 13] 12. The non-transitory computer-readable medium of claim 11, wherein the correlation is a combination of two or more of the simple regression, the ROC curve area maximization, the topology stabilization, and the spatial proximity analysis. [Appendix 14] 12. The non-transitory computer-readable medium of claim 11, wherein the first and second patient samples comprise at least one of a blood sample, a urine sample, or a tissue sample. [Appendix 15] 12. The non-transitory computer-readable medium of claim 11, wherein the disease diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer. [Appendix 16] 16. The non-transitory computer-readable medium of claim 15, wherein the disease diagnosed is the stage of the prostate cancer, breast cancer, lung cancer, or ovarian cancer based on Gleason score. [Appendix 17] 17. The non-transitory computer-readable medium of claim 16, wherein the first and second patient samples comprise cancer stage data, and the cancer stage data is classified into a plurality of binary groups. [Appendix 18] 8. The non-transitory computer-readable medium of claim 7, wherein each of the binary groups is scored. [Appendix 19] 12. The non-transitory computer readable medium of claim 11, wherein the biomarkers are selected from functional groups of cytokines, and the functional groups of the cytokines are at least three of pro-inflammatory, anti-inflammatory, anti-tumorigenic, cell apoptosis, and angiogenesis. [Appendix 20] 12. The non-transitory computer readable medium of claim 11, wherein the first biomarker is VEGF. [Explanation of symbols]

[0130] 2600 System Example 2610 Peripheral Devices 2620 Computer 2622 Central Processing Unit 2624 Storage medium 2626 User Input Devices 2628 Display 2630 Network 2640 Remote Server

Claims

1. A computer-implemented method comprising: (a) receiving a first set of one or more concentration values ​​of a first biomarker present in tumor interstitial fluid from a first patient sample, the first biomarker being collected from a surrogate fluid of the first patient, the first patient sample comprising a non-disease diagnosis; (b) receiving a second set of concentration values ​​of one or more of the first biomarkers present in tumor interstitial fluid from a second patient sample, the first biomarkers being collected from a surrogate fluid of the second patient, the second patient sample comprising a disease diagnosis; (c) calculating a first set of proximity scores from the first set of density values ​​and a second set of proximity scores from the second set of density values; (d) calculating a correlation between the first biomarker and a disease diagnosis from the first set of concentration values ​​and the second set of concentration values ​​and the first set of proximity scores and the second set of proximity scores, the correlation comprising topology stabilization, the correlation being calculated as an initial disease state risk score; (e) validating the initial disease state risk score by using a topology stability test to determine whether an instability-induced error associated with the initial disease state risk score exceeds an acceptable range; The method includes:

2. 2. The method of claim 1, wherein steps (a) through (d) are repeated for up to five biomarkers.

3. 2. The method of claim 1, wherein the first patient sample and the second patient sample comprise at least one of a blood sample, a urine sample, or a tissue sample, or the disease being diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer.

4. 4. The method of claim 3, wherein the disease diagnosed is the stage of the prostate, breast, lung, or ovarian cancer based on Gleason score.

5. 5. The method of claim 4, wherein the first patient sample and the second patient sample comprise cancer stage data, and the cancer stage data is classified into a plurality of binary groups.

6. The method of claim 5 , wherein each of the binary groups is scored.

7. 2. The method of claim 1, wherein the first biomarker is selected from a functional group of cytokines, the functional groups of the cytokines being at least three of pro-inflammatory, anti-inflammatory, anti-tumorigenic, cell apoptosis, and angiogenesis, or the first biomarker is VEGF.

8. (a) receiving a first set of one or more concentration values ​​of a first biomarker present in tumor interstitial fluid from a first patient sample, the first biomarker being collected from a surrogate fluid of the first patient, the first patient sample comprising a non-disease diagnosis; (b) receiving a second set of one or more concentration values ​​of the first biomarkers present in tumor interstitial fluid from a second patient sample, the first biomarkers being collected from a surrogate fluid of the second patient, the second patient sample comprising a disease diagnosis; (c) calculating a first set of proximity scores from the first set of concentration values ​​and a second set of proximity scores from the second set of concentration values; (d) calculating a correlation between the first biomarker and a disease diagnosis from the first set of concentration values ​​and the second set of concentration values ​​and the first set of proximity scores and the second set of proximity scores, the correlation comprising topology stabilization, the correlation being calculated as an initial disease state risk score; (e) validating the initial disease state risk score by using a topology stability test to determine whether an instability-induced error associated with the initial disease state risk score exceeds an acceptable range; and A non-transitory computer-readable medium storing a program for causing a computer to execute a process including the steps of:

9. 9. The non-transitory computer readable medium of claim 8, wherein steps (a) through (d) are repeated for up to five biomarkers.

10. 9. The non-transitory computer-readable medium of claim 8, wherein the first patient sample and the second patient sample include at least one of a blood sample, a urine sample, or a tissue sample, or the disease being diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer.

11. 11. The non-transitory computer readable medium of claim 10, wherein the disease diagnosed is a stage of the prostate, breast, lung, or ovarian cancer based on a Gleason score.

12. 12. The non-transitory computer readable medium of claim 11, wherein the first patient sample and the second patient sample comprise cancer stage data, and the cancer stage data is classified into a plurality of binary groups.

13. The non-transitory computer-readable medium of claim 12 , wherein each of the binary groups is scored.

14. 9. The non-transitory computer readable medium of claim 8, wherein the first biomarker is selected from a functional group of cytokines, the functional groups of the cytokines being at least three of pro-inflammatory, anti-inflammatory, anti-tumorigenic, cell apoptosis, and angiogenesis, or the first biomarker is VEGF.

Images