### Table 1: Evaluation of the best 4 discovered labelings and the original labelings in three data sets. The table reports the composition of the labeling; the max-surprise score, the p-value at the point of max surprise and the number of genes with that p-value; the Sanov score; LOOCV accuracy of predictions the labeling (ignoring control samples); and the Jackard coefficient that measures the similarity of the labeling to the original labeling.

2000

"... In PAGE 7: ...1. Table1 summarizes the scores of the top discovered classifications using the various scoring mechanisms we discussed above and their difference from the original classification of the data. On the leukemia data set we run our search procedure with the additional constraint that it should only examine labeling without control tissues.... In PAGE 7: ... On the leukemia data set we run our search procedure with the additional constraint that it should only examine labeling without control tissues. Peeling found six labelings, the first four of which are shown in Table1 . All six labelings score better than the original labeling by both the max-surprise and Sanov scores, and by the number of significant genes.... In PAGE 7: ... (Clinical risk is evaluated using in- ternational prognostic index, a standard medical index, evaluated at the time the sample was taken.) In Figure 2 we plot survival rates for patients for the four putative DLBCL classifications de- scribed in Table1 . As we can see, some of the classifications, such as the forth one, are not predictive about patient survival.... In PAGE 7: ...8 0.9 1 19 patients 9 deaths 5 patients 2 deaths Figure 2: Kaplan-Meier survival plots for the 4 DLBCL classi- fications described in Table1 . The a4 -axis is the number of years after the samples were taken, and the a85 -axis is the fraction of patients survived so far.... ..."

Cited by 1

### Table 1: Evaluation of the best 4 discovered labelings and the original labelings in three data sets. The table reports the composition of the labeling; the max-surprise score, the p-value at the point of max surprise and the number of genes with that p-value; the Sanov score; LOOCV accuracy of predictions the labeling (ignoring control samples); and the Jackard coefficient that measures the similarity of the labeling to the original labeling.

"... In PAGE 7: ...1. Table1 summarizes the scores of the top discovered classifications using the various scoring mechanisms we discussed above and their difference from the original classification of the data. On the leukemia data set we run our search procedure with the additional constraint that it should only examine labeling without control tissues.... In PAGE 7: ... On the leukemia data set we run our search procedure with the additional constraint that it should only examine labeling without control tissues. Peeling found six labelings, the first four of which are shown in Table1 . All six labelings score better than the original labeling by both the max-surprise and Sanov scores, and by the number of significant genes.... In PAGE 7: ... (Clinical risk is evaluated using in- ternational prognostic index, a standard medical index, evaluated at the time the sample was taken.) In Figure 2 we plot survival rates for patients for the four putative DLBCL classifications de- scribed in Table1 . As we can see, some of the classifications, such as the forth one, are not predictive about patient survival.... In PAGE 7: ...8 0.9 1 19 patients 9 deaths 5 patients 2 deaths Figure 2: Kaplan-Meier survival plots for the 4 DLBCL classi- fications described in Table1 . The DC-axis is the number of years after the samples were taken, and the DD-axis is the fraction of patients survived so far.... ..."

### Table 1. Prognostic factors for coronary heart disease as measured on Czech autoworkers. Source: [29]. The left-hand panel contains the original counts and the right-hand panel the bounds for the margins [ABCE], [ADE], and [BF].

2004

"... In PAGE 6: ... 4.3 Example: Prognostic Risk Factors for Czech Auto Workers The data in Table1 come from a prospective epidemi- ological study of 1841 workers in a Czechoslovakian car factory, as part of an investigation of potential risk factors for coronary thrombosis (see Edwards and Havranek [29]). Prior analyses of these data can be found in Dobra and Fien- berg [19, 20] and Dobra et al.... In PAGE 6: ...erg [19, 20] and Dobra et al. [21]. Here we integrate those results and, in the section that follows, expand upon them focusing in particular on issues of data utility and the risk- utility tradeoff. In left-hand panel of Table1 , A indicates whether or not the worker smokes, B corresponds to strenuous mental work, C corresponds to strenuous physical work, D cor- responds to systolic blood pressure, E corresponds to ra- tio of and lipoproteins, and F represents family anam- nesis of coronary heart disease. Our focus for disclosure limitation is on the three cells in the table with counts of 1 and 2 .... In PAGE 7: ...Table 2. Bounds for Czech auto-workers data from Table1 given the marginals [BF], [BC], [BE],[AB], [AC], [AE], [CE], [DE], [AD] B no yes F E D C A no yes no yes neg lt; 3 lt; 140 no [0,206] [0,167] [0,404] [0,312] yes [0,421] [0,463] [0,119] [0,119] 140 no [0,206] [0,167] [0,404] [0,312] yes [0,416] [0,333] [0,119] [0,119] 3 lt; 140 no [0,181] [0,167] [0,333] [0,339] yes [0,314] [0,344] [0,119] [0,119] 140 no [0,181] [0,167] [0,363] [0,339] yes [0,314] [0,341] [0,119] [0,119] pos lt; 3 lt; 140 no [0,134] [0,134] [0,126] [0,126] yes [0,134] [0,134] [0,119] [0,119] 140 no [0,134] [0,134] [0,126] [0,126] yes [0,134] [0,134] [0,119] [0,119] 3 lt; 140 no [0,134] [0,134] [0,126] [0,126] yes [0,134] [0,134] [0,119] [0,119] 140 no [0,134] [0,134] [0,126] [0,126] yes [0,134] [0,134] [0,119]... In PAGE 8: ... Another Bounds Example. Finally, suppose we consider the release of all 5-way margins of Table1 , the space of ta- bles Q contains only two integer tables: the original table n itself and a second table whose entries are found by adding or subtracting one unit from the corresponding entries in n. Consequently, the feasibility intervals [L(t); U(t)] for all the cells are of length one.... In PAGE 9: ... A somewhat different way to think about assessing risk, related to the calculation of bounds and implementing per- turbations is simply counting the numbers of possible tables instead of putting arbitrary distributions over them. As we saw in the example, for the release of all 5-way margins of Table1 , the space of tables Q contains only two integer tables, one of which is the original table. As we releasing fewer higher order margins the number of possible tables grows, often dramatically, thus making difficult for an in- truder to identify the original table.... ..."

Cited by 2

### Table 2. Prognostic significance of FLT3 mutation status in APL Total WT ITD* P D835/I836 only P Either mutant P

2005

"... In PAGE 5: ...OR, 1.99; 95% CI, 0.94-4.22; P H11005 .09) ( Table2 ). However, no independent effect of FLT3 mutation was observed after adjusting for WBC count (ID: OR, 1.... ..."

### Table 1: Multivariate Cox proportional hazards analysis of (A) standard clinical factors alone, or with (B) the Intrinsic Subtypes in relation to Relapse-Free Survival for the 315-sample combined test set. Size was a binary variable (0 = diameter of 2 cm or less, 1 = greater than 2 cm); node status was a binary variable (0 = no positive nodes, 1 = one or more positive nodes); age was a continuous variable formatted as decade-years. Hazard ratios for Intrinsic Subtypes were calculated relative to the Luminal A subtype. Variables found to be significant (p lt; 0.05) in the Cox proportional hazards model are shown in bold.

"... In PAGE 6: ... In a multivariate Cox pro- portional hazards analysis of the combined test set using these standard clinical parameters, size, grade and ER sta- tus were significant predictors of RFS (Table 1A). To further evaluate the prognostic/predictive value of the intrinsic subtype classification, we performed multivariate Cox proportional hazards analysis of the combined test set using the six intrinsic subtypes/groups defined above and the five standard clinical parameters with RFS, OS, or DSS as the endpoint ( Table1 B shows analysis for RFS). The intrinsic subtypes, when added to the multivariate model containing the standard clinical variables, resulted Kaplan-Meier survival curves of breast tumors classified by intrinsic subtype Figure 3 Kaplan-Meier survival curves of breast tumors classified by intrinsic subtype.... In PAGE 11: ... van apos;t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, et al.: Gene expression profiling predicts clinical outcome of breast can- Additional File 2 Supplemental Table1 . Clinical and microarray information associated with each patient in the 105-sample training dataset.... ..."

### TABLE 3. Insertion Depth of Puncture Needle, Length of Harvested Biopsy Specimen, and Deviation of the Biopsy Trajectory From the Center of the Target Evaluated for Each Biopsy Procedure

2004

### Table 2. Real Data Set - Wisconsin Prognostic Breast Cancer Database. Probabilistic prediction bit score error results and \right/wrong quot; predictive accuracy results using 10-fold cross-validation ( ).

### Tables 1{2 list the surface and upper prognostic quantities which are given every 6 hours; these correspond to instantenous values of the prognostic elds at the synoptic time. The surface (and top of the atmosphere) diagnostic quantities (Tables 3{6) are given every 3 hours over the South American continent. The 3-hourly diagnostics are averages from the previous 3 hours. For example, the 3-hour average evaporation valid at 12Z corresponds to a 3-hour average from 9Z to 12Z.

### Table 4: Marginal Effect on Net Crop Income Allowing for Interaction Effects Unrestricted model Restricted Model

2000

"... In PAGE 14: ... In an effort to make Table 3 more compact, variables with t-ratios below 1 and the commune dummies are not reported.11 Table4 presents the calculated total marginal effects and t-statistics of variables that enter the regression interacted with other variables, evaluated at mean points. Annual land, both irrigated and non-irrigated, and perennial land all have high significant positive overall effects on crop income.... ..."

Cited by 1