Report to the National Cancer Institute

Shipment Number: 53

Date Submitted: March 18, 1988

Quantities Measured:

1. Carcinoembryonic Antigen (CEA, ng/ml)
2. Heat-stable Alkaline Phosphatase (HSAP, u/1)
3. Ferritin (FT, ng/ml)
4. Ratio of Ferritin to Iron (FT:FE)
5. Immunosuppressive Acidic Protein (IAP, ng/ml)
6. Ribonuclease (RNase, u)
7. Sialic Acid (SA, mg/dl)
8. Alkaline Phosphatase Isoenzymes (ALK_PHOS)
9. Carbohydrate Antigen 19-9 (CA 19-9, u/ml)
10. Tissue Polypeptide Antigen (TPA, u/1)
11. TPA*CEA
12. Carcinoembryonic Antigen from the Mayo Clinic (CEA, ng/ml)
13. Overall Evaluation per patient (EVAL)

Number of Vials of Serum Tested: 360

Composition of Panel Included in this Analysis:

Three vials of serum were shipped per patient for each of the following categories:

40 - Colon Cancers, Early Stage 30 - Benign Colons
50 - Healthy Controls
120 Total patients (360 vials)

Criteria Given by Researcher: Screening by a combination of tumor marker assays may be of significance in early cancer detection.

Analyses and Data Presented in this Report

1. Listing of marker values by diagnostic categories (Table 1).
2. Distribution of sex within diagnostic group (Table 2).
3. Histograms and descriptive statistics of age by diagnostic group (Table 3).
4. Frequency plots of assay values by diagnostic group (Figure 1).
5. Histograms and descriptive statistics of assay values by diagnostic group
   (Figure 2)
6. Discrimination based on multiple marker rules (Table 4).
7. Discrimination based on multiple marker rules (DSB Shipment 46) (Table 5).

Specifics:

      Table 1 displays the following data utilized in this analysis: label number, age, gender, marker value, [logl0 ALK PHOS2, logl0 ALK PHOS3, logl0 CA19, logl0 FT, logl0 (FT/FE), logl0 (HASP+1.0), logl0 IAP, logl0 RNase, logl0 SA, logl0 CEA, logl0 TPA, logl0 (CEA*TPA)], and diagnostic group membership. Dr. Kobayashi also provided an Evaluation Code (EVAL) which categorized the patients into four categories; 1 = Healthy, 2 = Benign, 3 = Possible Cancer, 4 = Malignant Cancer. The Evaluation Code was based on information derived from all of the assay values. It should be pointed out that the Data Center generated an exact calculation of CEA*TPA, and a calculated value of FE to be used in the analysis. Also, ALK PHOS1 was dropped from the analysis due to a poor showing in a initial analysis.

      As a first step in this analysis, histograms and descriptive statistics were obtained for all of the raw assay values by diagnostic group. Due to the observed skewness for each assay, the logl0 transformation was applied to each assay. In addition, in order to avoid the singularity of negative infinity for the log transformation, assay values for all subjects were increased by 1.0 whenever any subject displayed a value of zero. In the following text, "pairwise comparison" is used to refer to the comparison of a Case group with the corresponding Benign group, or the Normal Control group. Other comparisons are explicitly stated.

      Table 2 shows the distribution of sex within diagnostic group. Perfect matching was achieved on this panel of data. Table 3 provides a comparison of age by diagnostic group. Standard ANOVA methods were applied for this analysis. The mean age ranged from 61 to 63, with the Healthy Controls being the youngest, and the Colon Cancer subjects, the oldest. The overall F-test was nonsignificant (p=68%), thus attesting to the good age matching achieved by the Mayo Clinic personnel on this serum panel.

      Figure 1 shows the frequency plots of the assays by diagnostic group. Figure 2 shows the histograms and descriptive statistics of the assay values by diagnostic group. Included with the Figure 2 data are the F-tests which test for distributional shifts between the diagnostic groups for each assay. It is seen that CA19-9 (p=69%), HSAP (p=85%) and RNase (p=1.6%) failed to achieve statistical significance at the 5% level. All other assays displayed a statistically significant shift between each diagnostic group.

      Logistic regression analyses were performed to determine whether a combination of the various assay values and other covariates (i.e., sex and age) discriminated between Cases and Controls at an increased level of sensitivity and specificity. The advantage of using this approach is that the assay values and other covariate information can be used to predict the log odds of being in the Case group. A step up procedure which selected variables with the largest efficient score was employed, after first adjusting for any of the following previously selected variables: CEA, TPA, ALK PHOS2, ALK PHOS3, CA19-9, SA, RNase, FT, FE, HSAP, IAP, Age and Sex. All of the previous variables were potential candidates in the model building phase of this analysis. Table 4 presents a summary of this analysis for each of the two pairwise comparisons. Under the column heading of Multiple Logistic Regression Rule, the sensitivity, specificity and the assays selected into the logistic regression model are listed. Also presented in Table 4 are the results of applying the overall patient evaluation rules as provided by Dr. Kobayashi. His combined rules grouped patients into four groups; Normal, Benign, Possible Cancer and Malignant Cancer. In this analysis, the categories of Possible Cancer and Malignant Cancer were pooled.It is seen that the Logistic Regression rule consistently exhibited better specificity than the combined assay rule, as provided by Dr. Kobayashi, while Dr. Kobayashi's rule was more sensitive than the Logistic rule.

Discussion:

      This panel of data (DSB Shipment 53) and a previous panel of data, DSB Shipment 46, were both analyzed by Dr. Tsueno Kobayashi. Both panels utilized the same combination of tumor markers with a similar panel composition. The only difference between the two panels was the addition of Lung Cancer (n=20) and Benign Lung (n=15) samples in DSB Shipment 46. The analysis of DSB Shipment 46 indicated the discrimination ability of the various single markers and the markers in combination with each other. Thus the decision was made to concentrate mainly on the comparison of the discrimination rules determined through Logistic Regression modelling and the use of Dr. Kobayashi's Evaluation codes.

      Table 5 displays a copy of the table, "Discrimination Based on Multiple Marker Rules", from the analyses of DSB Shipment 46. The application of the Logistic Regression rules are comparable for both panels for the Colon Cancer vs. Benign Colon comparison, with estimated sensitivities of 70% (DSB Shipment 46) and 72.5% (present shipment) along with specificities of 68% (DSB Shipment 46) and 76.7% (present shipment). When a test of two proportions was applied to the above pairs of figures, a non-significant (p>5%) result was obtained. Similarly, for the Logistic Regression rule as applied to the Colon Cancer vs. Normal comparison, the sensitivity of 80% (DSB Shipment 46) and 80% (present shipment), along with a specificity of 85% (DSB Shipment 46) and 00% (present shipment), were not significantly different. The same can not be said about Dr. Kobayashi's Evaluation Rule across the two panels. It seems that Dr. Kobayashi has significantly (p<5%) raised the sensitivity from 60% to 87.5% for both Colon Cancer vs. Benign Colon and Colon Cancer vs. Normal Controls, while at the same time paying for this increased sensitivity by a decline (non-significant p>5%) in specificity from 33% to 30% (Colon Cancer vs. Benign Colon) and 80% to 76% (Colon Cancer vs. Normals). This trade off between sensitivity and specificity is a well known phenomena in marker studies.

      Finally, in Table 4, the Logistic Regression rules displayed decreased sensitivity (non-significant p>5%) when compared with Dr. Kobayashi's Evaluation rule for both Colon Cancer vs. Benign Colon, and Colon Cancer vs. Normal comparisons. On the other hand, the Logistic Regression rules exhibited increased specificity for the same two pairwise comparisons. Furthermore, for the Colon Cancer vs. Benign Colon comparison, the specificity of 76.7% is significantly (p<5%) higher than the specificity of 30% as displayed by Dr. Kobayashi's Evaluation rule.

      As noted in the report for DSB Shipment 46, the screening for Early Stage Cancer in the general population is an important, albeit difficult problem. In general, from previous studies it is seen that information obtained from two or three markers usually is as good as the information obtained from a whole battery of markers. It may be worthwhile to utilize the economy realized in only assaying two or three crucial markers with repeated testing on the same patient.