cs.LGFeb 25, 2026

Discovering Subgroups with Exceptional Survival Characteristics

Authors: Mhd Jawad Al Rahwanji, Sascha Xu, Nils Philipp Walter, Jilles Vreeken

Organizations: CISPA Helmholtz Center for Information Security Saarbrücken, Germany

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

In many applications, it is important to identify subpopulations that survive longer or shorter than the rest of the population. In medicine, for example, it allows determining which patients benefit from treatment, and in predictive maintenance, which components are more likely to fail. Existing methods for discovering subgroups with exceptional survival characteristics rely on restrictive assumptions about the survival model (e.g. proportional hazards), require pre-discretized features, and, as they compare average statistics, tend to overlook individual heterogeneity. In this paper, we propose Sysurv, a non-parametric, fully differentiable method that discovers human-readable rules selecting subgroups with exceptional survival characteristics. Empirical evaluation on a wide range of datasets and settings, including a case study on cancer data, shows that Sysurv reveals insightful and actionable survival subgroups, outperforming the state of the art.

Figures & tables

Figure 1: Survival subgroups . Sysurv finds patients suffering from a therapy-resistant tumor (Left) , and people with exceptionally long (1) resp. short (2) durations until re-employment (Right) .
Figure 2: (Left) Individual survival functions S^(t∣xi)\hat{S}(t|\mathbf{x}_{i}) are more informative than their group-level estimate S^(t)\hat{S}(t) . (Right) The absolute difference between group and reference survival (Pop.) is effective in the absence of crossing ( S^1\hat{S}_{1} and S^2\hat{S}_{2} ) and in its presence ( S^3\hat{S}_{3} ), unlike the signed difference.
Figure 3: Soft rules. The soft condition approaches a hard interval with decreasing temperature τ→0\tau\to 0 (a) . Multiple soft conditions combine to form a soft rule depicted as a hyper box in the covariate space (b) . Adapted from Xu et al. (2024) .
Figure 4: Synthetic setting. Comparison of Sysurv and each of RuleKit , EsmamDs , and Fibers in terms of F1F_{1} -scores recovering planted subgroups with increasingly large dataset dimensionalities (Left) , increasingly many censored subjects (Center) , and increasingly large planted subgroups (Right) . EsmamDs is the closest competitor to Sysurv closely followed by RuleKit . Higher is better. The shaded areas show ±1 standard error over 10 runs.
 Our objectiveLogrankMean-shift
 OursRKEDFibOursRKEDFibOursRKEDFib
UnempDur6.43.83.83.819.3226.5125.0125.02.02.12.12.1
Nwtco1011.6681.0681.0208.6374.2281.331.40.01162.0688.2688.20.6
Rott2559.2326.2487.6487.6267.093.071.7169.331.510.025.225.2
Rdata259.2213.8213.8158.5113.4107.814.643.31116.6933.5933.5705.5
Aids266.660.554.034.70.21.74.30.022.761.261.40.7
Table 1: Real-world setting. Exceptionality of subgroups discovered by Sysurv ( Ours ), RuleKit ( RK ), EsmamDs ( ED ), and Fibers ( Fib ), measured using 3 metrics (our objective, logrank, and mean-shift). Sysurv ranks first across metrics. Greater is better. Highest values marked in bold.
Figure 5: Real-world setting. Survival subgroups discovered in the unemployment ( a ) and heart attack ( b ) datasets using Sysurv and RuleKit resp. EsmamDs . Sysurv learns more exceptional subsets of the subgroups discovered by baselines. The shaded areas show 95% confidence intervals.
Figure 6: Case study. Sysurv discovers, overall, sets of diverse subgroups in the primary cohort ( a ), and two subgroups of poor responders in the postoperative cohort ( b ) of HNSCC data. The shaded areas show 95% confidence intervals.
Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 7: Histogram of the exceptionality scores obtained from 1000 runs of BuildDFD on the postoperative cohort from our case study, with a fitted normal distribution. The mean and standard deviation of the fitted normal distribution are μ=16.9\mu=16.9 and η=2.77\eta=2.77 , respectively.
Figure 8: Synthetic setting. Comparison of Sysurv and each of RuleKit , EsmamDs , and Fibers in terms of runtime recovering planted subgroups with increasingly large dataset dimensionalities (Left) . Lower is better. Also, in terms of F1F_{1} -score, comparisons between Sysurv and baselines under varying population-subgroup hazard ratios (Center) , where the hazard ratio indicates the rate by which the population outsurvives the subgroup, or vice versa, and increasingly large datasets (Right) . Higher is better. EsmamDs is the closest competitor to Sysurv closely followed by RuleKit . The shaded areas show ±1 standard error over 10 runs.
 nnSysurvRuleKitEsmamDsFibers
UnempDur3241112.0 (3.5%)1620.5 (50.0%)1620.5 (50.0%)1585.6 (48.9%)
Nwtco4028168.0 (4.2%)2014.0 (50.0%)1490.0 (37.0%)4027.0 (100.0%)
Rott22982320.0 (10.7%)1825.7 (61.2%)1007.2 (33.8%)1210.6 (40.6%)
Rdata1040140.0 (13.5%)520.0 (50.0%)268.8 (25.8%)503.2 (48.4%)
Aids22839306.0 (10.8%)1342.6 (47.3%)503.7 (17.7%)2834.0 (99.8%)
Dialysis68052665.0 (39.2%)4095.7 (60.2%)2636.8 (38.7%)2568.6 (37.7%)
Appendix
Table 2: Average subgroup sizes (with coverage percentages) per method across real-world datasets.
Table 11
Original rulePruned ruleChanges
“ ACTN1∈[−3.3,−0.9]∧AKT1∈[−3.6,−1.9]∧ALDH3A1∈[−8.9,−1.4]∧ANLN∈[−4.3,−2.2]∧BCL2L1∈[−7.2,−5.3]∧BNIP3L∈[−3.0,−0.6]∧CAV1∈[−4.4,−1.1]∧CDKN3∈[−4.5,−2.5]∧CXCL12∈[−8.2,−2.6]∧DCBLD1∈[−6.9,−4.7]∧ERBB2∈[−1.9,0.7]∧ERBB3∈[−7.0,−2.9]∧GNAI1∈[−5.2,−0.0]∧IGF1R∈[−3.5,−0.9]∧LOX∈[−6.4,−3.3]∧MTOR∈[−3.2,−2.3]∧NOTCH1∈[−4.9,−2.6]∧PLAU∈[−4.5,−1.2]∧RB1∈[−3.8,−2.4]∧RELA∈[−3.2,−1.8]∧RFC4∈[−4.6,−1.6]∧RPA2∈[−4.0,−2.1]∧SNAI1∈[−6.9,−3.3]∧TCF3∈[−3.2,−1.3]∧TGFB1∈[−2.6,−1.0]∧TPI1∈[−7.2,−5.3]∧XPA∈[−4.7,−1.6]∧XRCC1∈[−4.2,−2.6]\mathit{ACTN1}\in[-3.3,-0.9]\land\mathit{AKT1}\in[-3.6,-1.9]\land\mathit{ALDH3A1}\in[-8.9,-1.4]\land\mathit{ANLN}\in[-4.3,-2.2]\land\mathit{BCL2L1}\in[-7.2,-5.3]\land\mathit{BNIP3L}\in[-3.0,-0.6]\land\mathit{CAV1}\in[-4.4,-1.1]\land\mathit{CDKN3}\in[-4.5,-2.5]\land\mathit{CXCL12}\in[-8.2,-2.6]\land\mathit{DCBLD1}\in[-6.9,-4.7]\land\mathit{ERBB2}\in[-1.9,0.7]\land\mathit{ERBB3}\in[-7.0,-2.9]\land\mathit{GNAI1}\in[-5.2,-0.0]\land\mathit{IGF1R}\in[-3.5,-0.9]\land\mathit{LOX}\in[-6.4,-3.3]\land\mathit{MTOR}\in[-3.2,-2.3]\land\mathit{NOTCH1}\in[-4.9,-2.6]\land\mathit{PLAU}\in[-4.5,-1.2]\land\mathit{RB1}\in[-3.8,-2.4]\land\mathit{RELA}\in[-3.2,-1.8]\land\mathit{RFC4}\in[-4.6,-1.6]\land\mathit{RPA2}\in[-4.0,-2.1]\land\mathit{SNAI1}\in[-6.9,-3.3]\land\mathit{TCF3}\in[-3.2,-1.3]\land\mathit{TGFB1}\in[-2.6,-1.0]\land\mathit{TPI1}\in[-7.2,-5.3]\land\mathit{XPA}\in[-4.7,-1.6]\land\mathit{XRCC1}\in[-4.2,-2.6] ”“ CDKN3∈[−4.5,−2.5]∧ERBB2∈[−1.9,0.7]∧LOX∈[−6.4,−3.3]∧RB1∈[−3.8,−2.4]∧RPA2∈[−4.0,−2.1]∧TGFB1∈[−2.6,−1.0]\mathit{CDKN3}\in[-4.5,-2.5]\land\mathit{ERBB2}\in[-1.9,0.7]\land\mathit{LOX}\in[-6.4,-3.3]\land\mathit{RB1}\in[-3.8,-2.4]\land\mathit{RPA2}\in[-4.0,-2.1]\land\mathit{TGFB1}\in[-2.6,-1.0] ”Subgroup size: 6→66\to 6 Exceptionality: 19.1051→19.103719.1051\to 19.1037
“ AKT1∈[−3.2,−1.2]∧ALDH1A1∈[−8.1,−2.3]∧ASS1∈[−4.4,−0.4]∧CENPK∈[−6.1,−4.4]∧CHEK2∈[−5.5,−3.8]∧CXCR4∈[−6.6,−2.8]∧CYP1B1∈[−11.3,−4.5]∧ERBB3∈[−5.5,−3.5]∧ERBB4∈[−10.6,−7.6]∧FGFR3∈[−7.1,−3.1]∧FOSL1∈[−5.3,−0.7]∧GNAI1∈[−3.2,−0.4]∧GPI∈[−2.0,0.2]∧HIF1A∈[−6.8,−4.7]∧KDR∈[−5.5,−3.5]∧LDHA∈[−6.9,−4.3]∧LGALS1∈[−2.6,1.0]∧LIMD1∈[−6.2,−4.7]∧LOX∈[−4.1,−0.9]∧MCM6∈[−5.9,−3.7]∧MET∈[−5.3,−2.8]∧MMP7∈[−9.2,−3.4]∧MYC∈[−2.8,0.3]∧NBN∈[−3.7,−2.7]∧PDK1∈[−4.9,−3.5]∧PFKFB3∈[−3.9,−0.7]∧RASSF6∈[−7.6,−5.0]∧RB1∈[−4.4,−2.5]∧SYK∈[−6.0,−3.9]∧SYNGR3∈[−8.9,−4.7]∧TPI1∈[−7.2,−4.8]∧XRCC1∈[−5.4,−3.4]∧CD44∈[−0.1,2.1]\mathit{AKT1}\in[-3.2,-1.2]\land\mathit{ALDH1A1}\in[-8.1,-2.3]\land\mathit{ASS1}\in[-4.4,-0.4]\land\mathit{CENPK}\in[-6.1,-4.4]\land\mathit{CHEK2}\in[-5.5,-3.8]\land\mathit{CXCR4}\in[-6.6,-2.8]\land\mathit{CYP1B1}\in[-11.3,-4.5]\land\mathit{ERBB3}\in[-5.5,-3.5]\land\mathit{ERBB4}\in[-10.6,-7.6]\land\mathit{FGFR3}\in[-7.1,-3.1]\land\mathit{FOSL1}\in[-5.3,-0.7]\land\mathit{GNAI1}\in[-3.2,-0.4]\land\mathit{GPI}\in[-2.0,0.2]\land\mathit{HIF1A}\in[-6.8,-4.7]\land\mathit{KDR}\in[-5.5,-3.5]\land\mathit{LDHA}\in[-6.9,-4.3]\land\mathit{LGALS1}\in[-2.6,1.0]\land\mathit{LIMD1}\in[-6.2,-4.7]\land\mathit{LOX}\in[-4.1,-0.9]\land\mathit{MCM6}\in[-5.9,-3.7]\land\mathit{MET}\in[-5.3,-2.8]\land\mathit{MMP7}\in[-9.2,-3.4]\land\mathit{MYC}\in[-2.8,0.3]\land\mathit{NBN}\in[-3.7,-2.7]\land\mathit{PDK1}\in[-4.9,-3.5]\land\mathit{PFKFB3}\in[-3.9,-0.7]\land\mathit{RASSF6}\in[-7.6,-5.0]\land\mathit{RB1}\in[-4.4,-2.5]\land\mathit{SYK}\in[-6.0,-3.9]\land\mathit{SYNGR3}\in[-8.9,-4.7]\land\mathit{TPI1}\in[-7.2,-4.8]\land\mathit{XRCC1}\in[-5.4,-3.4]\land\mathit{CD44}\in[-0.1,2.1] ”“ ALDH1A1∈[−8.1,−2.3]∧ASS1∈[−4.4,−0.4]∧CYP1B1∈[−11.3,−4.5]∧ERBB3∈[−5.5,−3.5]∧ERBB4∈[−10.6,−7.6]∧FOSL1∈[−5.3,−0.7]∧GNAI1∈[−3.2,−0.4]∧GPI∈[−2.0,0.2]∧HIF1A∈[−6.8,−4.7]∧KDR∈[−5.5,−3.5]∧LDHA∈[−6.9,−4.3]∧LGALS1∈[−2.6,1.0]∧LIMD1∈[−6.2,−4.7]∧LOX∈[−4.1,−0.9]∧MCM6∈[−5.9,−3.7]∧MET∈[−5.3,−2.8]∧MMP7∈[−9.2,−3.4]∧MYC∈[−2.8,0.3]∧NBN∈[−3.7,−2.7]∧PDK1∈[−4.9,−3.5]∧PFKFB3∈[−3.9,−0.7]∧RASSF6∈[−7.6,−5.0]∧RB1∈[−4.4,−2.5]∧SYK∈[−6.0,−3.9]∧SYNGR3∈[−8.9,−4.7]∧TPI1∈[−7.2,−4.8]∧XRCC1∈[−5.4,−3.4]∧CD44∈[−0.1,2.1]\mathit{ALDH1A1}\in[-8.1,-2.3]\land\mathit{ASS1}\in[-4.4,-0.4]\land\mathit{CYP1B1}\in[-11.3,-4.5]\land\mathit{ERBB3}\in[-5.5,-3.5]\land\mathit{ERBB4}\in[-10.6,-7.6]\land\mathit{FOSL1}\in[-5.3,-0.7]\land\mathit{GNAI1}\in[-3.2,-0.4]\land\mathit{GPI}\in[-2.0,0.2]\land\mathit{HIF1A}\in[-6.8,-4.7]\land\mathit{KDR}\in[-5.5,-3.5]\land\mathit{LDHA}\in[-6.9,-4.3]\land\mathit{LGALS1}\in[-2.6,1.0]\land\mathit{LIMD1}\in[-6.2,-4.7]\land\mathit{LOX}\in[-4.1,-0.9]\land\mathit{MCM6}\in[-5.9,-3.7]\land\mathit{MET}\in[-5.3,-2.8]\land\mathit{MMP7}\in[-9.2,-3.4]\land\mathit{MYC}\in[-2.8,0.3]\land\mathit{NBN}\in[-3.7,-2.7]\land\mathit{PDK1}\in[-4.9,-3.5]\land\mathit{PFKFB3}\in[-3.9,-0.7]\land\mathit{RASSF6}\in[-7.6,-5.0]\land\mathit{RB1}\in[-4.4,-2.5]\land\mathit{SYK}\in[-6.0,-3.9]\land\mathit{SYNGR3}\in[-8.9,-4.7]\land\mathit{TPI1}\in[-7.2,-4.8]\land\mathit{XRCC1}\in[-5.4,-3.4]\land\mathit{CD44}\in[-0.1,2.1] ”Subgroup size: 12→1212\to 12 Exceptionality: 16.6102→16.607916.6102\to 16.6079
“ ADM∈[−5.3,−0.8]∧AKT1∈[−3.6,−1.7]∧ALDH3A1∈[−8.9,−0.5]∧ANLN∈[−4.3,−2.8]∧ASS1∈[−4.9,−0.0]∧ATP5G3∈[−1.3,1.4]∧ATR∈[−5.9,−3.8]∧BCL2L1∈[−7.8,−4.9]∧BIRC5∈[−4.7,−1.0]∧CAV1∈[−4.4,0.3]∧CDKN3∈[−5.9,−2.5]∧CLDN4∈[−5.0,−0.6]∧CXCL12∈[−8.2,−3.2]∧ERBB3∈[−6.3,−3.2]∧ERCC5∈[−5.2,−3.6]∧FAM83B∈[−7.4,−1.7]∧FLT1∈[−5.7,−3.1]∧GNAI1∈[−5.2,−0.2]∧IGF1R∈[−3.5,−0.6]∧INHBA∈[−4.4,1.9]∧KDR∈[−5.5,−2.8]∧LDHA∈[−7.5,−5.7]∧LGALS1∈[−2.1,1.0]∧LOX∈[−6.4,−1.4]∧MDM2∈[−4.6,−2.2]∧MMP13∈[−9.3,1.3]∧MRE11A∈[−6.1,−3.6]∧MTOR∈[−3.7,−2.3]∧NOTCH1∈[−5.9,−2.6]∧P4HA2∈[−4.7,−1.4]∧PSMD9∈[−4.2,−2.1]∧PTEN∈[−7.3,−5.6]∧RAD23B∈[−0.9,0.8]∧RB1∈[−3.8,−2.3]∧RPA2∈[−4.9,−2.1]∧SERPINB2∈[−6.6,−1.2]∧TCF3∈[−3.5,−1.0]∧TGFB1∈[−2.6,−0.3]∧TPI1∈[−7.2,−5.0]∧XRCC1∈[−5.3,−2.6]∧XRCC5∈[−2.5,−0.8]∧CD44∈[−0.8,2.5]\mathit{ADM}\in[-5.3,-0.8]\land\mathit{AKT1}\in[-3.6,-1.7]\land\mathit{ALDH3A1}\in[-8.9,-0.5]\land\mathit{ANLN}\in[-4.3,-2.8]\land\mathit{ASS1}\in[-4.9,-0.0]\land\mathit{ATP5G3}\in[-1.3,1.4]\land\mathit{ATR}\in[-5.9,-3.8]\land\mathit{BCL2L1}\in[-7.8,-4.9]\land\mathit{BIRC5}\in[-4.7,-1.0]\land\mathit{CAV1}\in[-4.4,0.3]\land\mathit{CDKN3}\in[-5.9,-2.5]\land\mathit{CLDN4}\in[-5.0,-0.6]\land\mathit{CXCL12}\in[-8.2,-3.2]\land\mathit{ERBB3}\in[-6.3,-3.2]\land\mathit{ERCC5}\in[-5.2,-3.6]\land\mathit{FAM83B}\in[-7.4,-1.7]\land\mathit{FLT1}\in[-5.7,-3.1]\land\mathit{GNAI1}\in[-5.2,-0.2]\land\mathit{IGF1R}\in[-3.5,-0.6]\land\mathit{INHBA}\in[-4.4,1.9]\land\mathit{KDR}\in[-5.5,-2.8]\land\mathit{LDHA}\in[-7.5,-5.7]\land\mathit{LGALS1}\in[-2.1,1.0]\land\mathit{LOX}\in[-6.4,-1.4]\land\mathit{MDM2}\in[-4.6,-2.2]\land\mathit{MMP13}\in[-9.3,1.3]\land\mathit{MRE11A}\in[-6.1,-3.6]\land\mathit{MTOR}\in[-3.7,-2.3]\land\mathit{NOTCH1}\in[-5.9,-2.6]\land\mathit{P4HA2}\in[-4.7,-1.4]\land\mathit{PSMD9}\in[-4.2,-2.1]\land\mathit{PTEN}\in[-7.3,-5.6]\land\mathit{RAD23B}\in[-0.9,0.8]\land\mathit{RB1}\in[-3.8,-2.3]\land\mathit{RPA2}\in[-4.9,-2.1]\land\mathit{SERPINB2}\in[-6.6,-1.2]\land\mathit{TCF3}\in[-3.5,-1.0]\land\mathit{TGFB1}\in[-2.6,-0.3]\land\mathit{TPI1}\in[-7.2,-5.0]\land\mathit{XRCC1}\in[-5.3,-2.6]\land\mathit{XRCC5}\in[-2.5,-0.8]\land\mathit{CD44}\in[-0.8,2.5] ”“ ANLN∈[−4.3,−2.8]∧ATR∈[−5.9,−3.8]∧CLDN4∈[−5.0,−0.6]∧CXCL12∈[−8.2,−3.2]∧ERBB3∈[−6.3,−3.2]∧ERCC5∈[−5.2,−3.6]∧LDHA∈[−7.5,−5.7]∧LGALS1∈[−2.1,1.0]∧MDM2∈[−4.6,−2.2]∧MTOR∈[−3.7,−2.3]∧PTEN∈[−7.3,−5.6]∧RAD23B∈[−0.9,0.8]∧RB1∈[−3.8,−2.3]∧SERPINB2∈[−6.6,−1.2]∧TCF3∈[−3.5,−1.0]∧TGFB1∈[−2.6,−0.3]∧TPI1∈[−7.2,−5.0]\mathit{ANLN}\in[-4.3,-2.8]\land\mathit{ATR}\in[-5.9,-3.8]\land\mathit{CLDN4}\in[-5.0,-0.6]\land\mathit{CXCL12}\in[-8.2,-3.2]\land\mathit{ERBB3}\in[-6.3,-3.2]\land\mathit{ERCC5}\in[-5.2,-3.6]\land\mathit{LDHA}\in[-7.5,-5.7]\land\mathit{LGALS1}\in[-2.1,1.0]\land\mathit{MDM2}\in[-4.6,-2.2]\land\mathit{MTOR}\in[-3.7,-2.3]\land\mathit{PTEN}\in[-7.3,-5.6]\land\mathit{RAD23B}\in[-0.9,0.8]\land\mathit{RB1}\in[-3.8,-2.3]\land\mathit{SERPINB2}\in[-6.6,-1.2]\land\mathit{TCF3}\in[-3.5,-1.0]\land\mathit{TGFB1}\in[-2.6,-0.3]\land\mathit{TPI1}\in[-7.2,-5.0] ”Subgroup size: 28→2928\to 29 Exceptionality: 16.0098→15.793416.0098\to 15.7934
“ AKT1∈[−3.6,−1.5]∧ALDH3A1∈[−8.9,−0.5]∧ANLN∈[−4.5,−2.6]∧ASS1∈[−4.9,−0.0]∧ATP5G3∈[−1.3,1.7]∧ATR∈[−5.9,−3.7]∧BCL2L1∈[−7.7,−5.0]∧BIRC5∈[−5.3,−1.3]∧BNIP3L∈[−3.1,−0.6]∧CAV1∈[−4.4,0.4]∧CDKN3∈[−5.8,−2.5]∧CLDN4∈[−5.0,−0.7]∧CXCL12∈[−8.2,−2.2]∧ERBB3∈[−6.2,−3.2]∧ERBB4∈[−10.6,−7.4]∧ERCC5∈[−5.2,−3.6]∧FAM83B∈[−6.8,−1.6]∧GNAI1∈[−5.2,−0.2]∧HIF1A∈[−7.7,−4.7]∧IGF1R∈[−3.5,−0.6]∧INHBA∈[−4.8,1.8]∧KDR∈[−5.5,−3.6]∧LDHA∈[−7.5,−5.3]∧LGALS1∈[−2.6,1.0]∧LOX∈[−6.4,−1.7]∧MCM6∈[−5.9,−3.0]∧MMP13∈[−9.3,1.3]∧MRGBP∈[−5.0,−2.8]∧MTOR∈[−3.8,−2.3]∧P4HA2∈[−4.7,−1.4]∧PDK1∈[−4.9,−2.3]∧PGAM1∈[−2.6,−0.7]∧PSMD9∈[−4.2,−2.0]∧PTEN∈[−7.3,−5.5]∧RAD23B∈[−0.8,0.8]∧RB1∈[−3.9,−2.3]∧TCF3∈[−3.4,−1.4]∧TGFB1∈[−2.6,−0.0]∧TP53∈[−4.3,−1.0]∧TPI1∈[−7.2,−4.9]∧CD44∈[−0.4,2.1]\mathit{AKT1}\in[-3.6,-1.5]\land\mathit{ALDH3A1}\in[-8.9,-0.5]\land\mathit{ANLN}\in[-4.5,-2.6]\land\mathit{ASS1}\in[-4.9,-0.0]\land\mathit{ATP5G3}\in[-1.3,1.7]\land\mathit{ATR}\in[-5.9,-3.7]\land\mathit{BCL2L1}\in[-7.7,-5.0]\land\mathit{BIRC5}\in[-5.3,-1.3]\land\mathit{BNIP3L}\in[-3.1,-0.6]\land\mathit{CAV1}\in[-4.4,0.4]\land\mathit{CDKN3}\in[-5.8,-2.5]\land\mathit{CLDN4}\in[-5.0,-0.7]\land\mathit{CXCL12}\in[-8.2,-2.2]\land\mathit{ERBB3}\in[-6.2,-3.2]\land\mathit{ERBB4}\in[-10.6,-7.4]\land\mathit{ERCC5}\in[-5.2,-3.6]\land\mathit{FAM83B}\in[-6.8,-1.6]\land\mathit{GNAI1}\in[-5.2,-0.2]\land\mathit{HIF1A}\in[-7.7,-4.7]\land\mathit{IGF1R}\in[-3.5,-0.6]\land\mathit{INHBA}\in[-4.8,1.8]\land\mathit{KDR}\in[-5.5,-3.6]\land\mathit{LDHA}\in[-7.5,-5.3]\land\mathit{LGALS1}\in[-2.6,1.0]\land\mathit{LOX}\in[-6.4,-1.7]\land\mathit{MCM6}\in[-5.9,-3.0]\land\mathit{MMP13}\in[-9.3,1.3]\land\mathit{MRGBP}\in[-5.0,-2.8]\land\mathit{MTOR}\in[-3.8,-2.3]\land\mathit{P4HA2}\in[-4.7,-1.4]\land\mathit{PDK1}\in[-4.9,-2.3]\land\mathit{PGAM1}\in[-2.6,-0.7]\land\mathit{PSMD9}\in[-4.2,-2.0]\land\mathit{PTEN}\in[-7.3,-5.5]\land\mathit{RAD23B}\in[-0.8,0.8]\land\mathit{RB1}\in[-3.9,-2.3]\land\mathit{TCF3}\in[-3.4,-1.4]\land\mathit{TGFB1}\in[-2.6,-0.0]\land\mathit{TP53}\in[-4.3,-1.0]\land\mathit{TPI1}\in[-7.2,-4.9]\land\mathit{CD44}\in[-0.4,2.1] ”“ ALDH3A1∈[−8.9,−0.5]∧CLDN4∈[−5.0,−0.7]∧ERBB3∈[−6.2,−3.2]∧ERBB4∈[−10.6,−7.4]∧HIF1A∈[−7.7,−4.7]∧LDHA∈[−7.5,−5.3]∧MMP13∈[−9.3,1.3]∧MRGBP∈[−5.0,−2.8]∧P4HA2∈[−4.7,−1.4]∧PDK1∈[−4.9,−2.3]∧PGAM1∈[−2.6,−0.7]∧PTEN∈[−7.3,−5.5]∧RAD23B∈[−0.8,0.8]∧TP53∈[−4.3,−1.0]∧TPI1∈[−7.2,−4.9]\mathit{ALDH3A1}\in[-8.9,-0.5]\land\mathit{CLDN4}\in[-5.0,-0.7]\land\mathit{ERBB3}\in[-6.2,-3.2]\land\mathit{ERBB4}\in[-10.6,-7.4]\land\mathit{HIF1A}\in[-7.7,-4.7]\land\mathit{LDHA}\in[-7.5,-5.3]\land\mathit{MMP13}\in[-9.3,1.3]\land\mathit{MRGBP}\in[-5.0,-2.8]\land\mathit{P4HA2}\in[-4.7,-1.4]\land\mathit{PDK1}\in[-4.9,-2.3]\land\mathit{PGAM1}\in[-2.6,-0.7]\land\mathit{PTEN}\in[-7.3,-5.5]\land\mathit{RAD23B}\in[-0.8,0.8]\land\mathit{TP53}\in[-4.3,-1.0]\land\mathit{TPI1}\in[-7.2,-4.9] ”Subgroup size: 25→2625\to 26 Exceptionality: 15.8073→15.206915.8073\to 15.2069
Appendix
Table 5: Rules learned on the primary RCT cohort before and after pruning using a Jaccard similarity threshold of 0.950.95 for membership indicators along with the respective changes in subgroup sizes and exceptionality.
Original rulePruned ruleChanges
“ ALDH1A1∈[−7.0,−3.2]∧BCL2L1∈[−6.9,−5.5]∧BIRC5∈[−4.6,−1.6]∧BNIP3L∈[−2.5,−1.1]∧CA9∈[−9.0,−4.4]∧CBX4∈[−5.3,−3.7]∧CLDN4∈[−5.3,−1.3]∧CXCL12∈[−5.2,−1.1]∧EGLN3∈[−4.4,−0.9]∧ENO1∈[−0.2,1.2]∧ENO2∈[−5.8,−2.4]∧EPHA1∈[−7.2,−4.6]∧GPI∈[−1.7,−0.2]∧HSPB1∈[−4.0,−2.0]∧ITGB2∈[−4.7,−2.3]∧KCTD11∈[−2.3,0.1]∧MAP2K2∈[−4.3,−3.3]∧MCM6∈[−5.8,−4.0]∧MMP13∈[−1.9,1.7]∧MMP9∈[−2.7,0.6]∧MPRS17∈[−5.0,−1.2]∧PFKFB3∈[−2.7,−0.2]∧PTEN∈[−6.6,−5.1]∧RAD23B∈[−1.0,0.9]∧RB1∈[−5.0,−2.5]∧RELA∈[−2.9,−1.6]∧SERPINB2∈[−7.6,−2.6]∧SLC5A1∈[−9.9,−3.6]∧TCF3∈[−3.8,−2.3]∧XRCC1∈[−5.2,−3.2]∧XRCC4∈[−6.1,−4.2]∧XRCC5∈[−2.1,−0.6]\mathit{ALDH1A1}\in[-7.0,-3.2]\land\mathit{BCL2L1}\in[-6.9,-5.5]\land\mathit{BIRC5}\in[-4.6,-1.6]\land\mathit{BNIP3L}\in[-2.5,-1.1]\land\mathit{CA9}\in[-9.0,-4.4]\land\mathit{CBX4}\in[-5.3,-3.7]\land\mathit{CLDN4}\in[-5.3,-1.3]\land\mathit{CXCL12}\in[-5.2,-1.1]\land\mathit{EGLN3}\in[-4.4,-0.9]\land\mathit{ENO1}\in[-0.2,1.2]\land\mathit{ENO2}\in[-5.8,-2.4]\land\mathit{EPHA1}\in[-7.2,-4.6]\land\mathit{GPI}\in[-1.7,-0.2]\land\mathit{HSPB1}\in[-4.0,-2.0]\land\mathit{ITGB2}\in[-4.7,-2.3]\land\mathit{KCTD11}\in[-2.3,0.1]\land\mathit{MAP2K2}\in[-4.3,-3.3]\land\mathit{MCM6}\in[-5.8,-4.0]\land\mathit{MMP13}\in[-1.9,1.7]\land\mathit{MMP9}\in[-2.7,0.6]\land\mathit{MPRS17}\in[-5.0,-1.2]\land\mathit{PFKFB3}\in[-2.7,-0.2]\land\mathit{PTEN}\in[-6.6,-5.1]\land\mathit{RAD23B}\in[-1.0,0.9]\land\mathit{RB1}\in[-5.0,-2.5]\land\mathit{RELA}\in[-2.9,-1.6]\land\mathit{SERPINB2}\in[-7.6,-2.6]\land\mathit{SLC5A1}\in[-9.9,-3.6]\land\mathit{TCF3}\in[-3.8,-2.3]\land\mathit{XRCC1}\in[-5.2,-3.2]\land\mathit{XRCC4}\in[-6.1,-4.2]\land\mathit{XRCC5}\in[-2.1,-0.6] ”“ EPHA1∈[−7.2,−4.6]∧GPI∈[−1.7,−0.2]∧KCTD11∈[−2.3,0.1]∧MAP2K2∈[−4.3,−3.3]∧MMP13∈[−1.9,1.7]∧MMP9∈[−2.7,0.6]∧XRCC5∈[−2.1,−0.6]\mathit{EPHA1}\in[-7.2,-4.6]\land\mathit{GPI}\in[-1.7,-0.2]\land\mathit{KCTD11}\in[-2.3,0.1]\land\mathit{MAP2K2}\in[-4.3,-3.3]\land\mathit{MMP13}\in[-1.9,1.7]\land\mathit{MMP9}\in[-2.7,0.6]\land\mathit{XRCC5}\in[-2.1,-0.6] ”Subgroup size: 12→1212\to 12 Exceptionality: 27.8189→27.743527.8189\to 27.7435
“ ANXA5∈[−3.0,−0.9]∧ATM∈[−3.8,−1.6]∧ATP5G3∈[−1.1,1.0]∧BNIP3∈[−6.1,−2.6]∧BSG∈[−9.4,−5.4]∧CD24∈[−8.8,−4.5]∧CXCL12∈[−6.2,−1.3]∧DCBLD1∈[−7.5,−4.5]∧DKK3∈[−5.3,−1.4]∧FANCA∈[−6.7,−3.1]∧FN1∈[−4.6,1.8]∧HIF1A∈[−7.1,−4.6]∧HK2∈[−6.7,−2.6]∧HSPA4∈[−3.9,−2.2]∧INHBA∈[−7.9,−1.3]∧ITGB1∈[−2.5,−0.1]∧KRT17∈[−3.2,2.8]∧LGALS1∈[−3.4,−0.2]∧LOXL2∈[−6.4,−3.0]∧MCM6∈[−5.3,−2.5]∧MDM2∈[−4.3,−0.7]∧MMP10∈[−9.7,−1.8]∧MMP13∈[−10.4,−2.8]∧MMP2∈[−5.0,−0.5]∧MTOR∈[−3.9,−2.3]∧MYC∈[−3.7,−0.6]∧MYNN∈[−5.1,−3.0]∧NR1D2∈[−4.9,−2.3]∧P4HA2∈[−5.1,−2.4]∧PSMD9∈[−4.3,−2.2]∧RMI2∈[−5.1,−2.9]∧RPA2∈[−4.5,−2.1]∧SFN∈[−1.6,3.1]∧SLC3A2∈[−4.3,−2.3]∧SMDT1∈[−3.9,−2.2]∧SNAI1∈[−7.6,−3.8]∧SPP1∈[−9.0,−3.9]∧TCF3∈[−4.0,−2.1]∧TP53∈[−3.8,−1.0]∧XRCC4∈[−5.2,−3.5]∧XRCC5∈[−2.8,−0.9]∧YAP1∈[−3.6,−0.2]\mathit{ANXA5}\in[-3.0,-0.9]\land\mathit{ATM}\in[-3.8,-1.6]\land\mathit{ATP5G3}\in[-1.1,1.0]\land\mathit{BNIP3}\in[-6.1,-2.6]\land\mathit{BSG}\in[-9.4,-5.4]\land\mathit{CD24}\in[-8.8,-4.5]\land\mathit{CXCL12}\in[-6.2,-1.3]\land\mathit{DCBLD1}\in[-7.5,-4.5]\land\mathit{DKK3}\in[-5.3,-1.4]\land\mathit{FANCA}\in[-6.7,-3.1]\land\mathit{FN1}\in[-4.6,1.8]\land\mathit{HIF1A}\in[-7.1,-4.6]\land\mathit{HK2}\in[-6.7,-2.6]\land\mathit{HSPA4}\in[-3.9,-2.2]\land\mathit{INHBA}\in[-7.9,-1.3]\land\mathit{ITGB1}\in[-2.5,-0.1]\land\mathit{KRT17}\in[-3.2,2.8]\land\mathit{LGALS1}\in[-3.4,-0.2]\land\mathit{LOXL2}\in[-6.4,-3.0]\land\mathit{MCM6}\in[-5.3,-2.5]\land\mathit{MDM2}\in[-4.3,-0.7]\land\mathit{MMP10}\in[-9.7,-1.8]\land\mathit{MMP13}\in[-10.4,-2.8]\land\mathit{MMP2}\in[-5.0,-0.5]\land\mathit{MTOR}\in[-3.9,-2.3]\land\mathit{MYC}\in[-3.7,-0.6]\land\mathit{MYNN}\in[-5.1,-3.0]\land\mathit{NR1D2}\in[-4.9,-2.3]\land\mathit{P4HA2}\in[-5.1,-2.4]\land\mathit{PSMD9}\in[-4.3,-2.2]\land\mathit{RMI2}\in[-5.1,-2.9]\land\mathit{RPA2}\in[-4.5,-2.1]\land\mathit{SFN}\in[-1.6,3.1]\land\mathit{SLC3A2}\in[-4.3,-2.3]\land\mathit{SMDT1}\in[-3.9,-2.2]\land\mathit{SNAI1}\in[-7.6,-3.8]\land\mathit{SPP1}\in[-9.0,-3.9]\land\mathit{TCF3}\in[-4.0,-2.1]\land\mathit{TP53}\in[-3.8,-1.0]\land\mathit{XRCC4}\in[-5.2,-3.5]\land\mathit{XRCC5}\in[-2.8,-0.9]\land\mathit{YAP1}\in[-3.6,-0.2] ”“ ATM∈[−3.8,−1.6]∧HIF1A∈[−7.1,−4.6]∧KRT17∈[−3.2,2.8]∧MMP13∈[−10.4,−2.8]∧MMP2∈[−5.0,−0.5]∧PSMD9∈[−4.3,−2.2]∧RMI2∈[−5.1,−2.9]∧SNAI1∈[−7.6,−3.8]∧XRCC4∈[−5.2,−3.5]∧YAP1∈[−3.6,−0.2]\mathit{ATM}\in[-3.8,-1.6]\land\mathit{HIF1A}\in[-7.1,-4.6]\land\mathit{KRT17}\in[-3.2,2.8]\land\mathit{MMP13}\in[-10.4,-2.8]\land\mathit{MMP2}\in[-5.0,-0.5]\land\mathit{PSMD9}\in[-4.3,-2.2]\land\mathit{RMI2}\in[-5.1,-2.9]\land\mathit{SNAI1}\in[-7.6,-3.8]\land\mathit{XRCC4}\in[-5.2,-3.5]\land\mathit{YAP1}\in[-3.6,-0.2] ”Subgroup size: 47→4547\to 45 Exceptionality: 12.0529→11.926412.0529\to 11.9264
“ ALDH1A1∈[−7.0,−3.4]∧BSG∈[−8.9,−3.9]∧CA9∈[−9.5,−4.5]∧CBX4∈[−5.3,−3.7]∧CLDN4∈[−4.9,−0.4]∧CXCL12∈[−4.3,−1.5]∧ENO1∈[−0.3,1.2]∧ENO2∈[−5.8,−2.4]∧EPHA1∈[−7.2,−4.6]∧EPOR∈[−10.2,−6.6]∧ERCC5∈[−5.2,−3.0]∧HIF1A∈[−5.7,−3.7]∧LOX∈[−2.9,−1.0]∧MMP9∈[−2.8,0.6]∧MYNN∈[−5.2,−2.8]∧PGK1∈[−2.0,−0.2]∧PTEN∈[−6.5,−5.1]∧RAD23B∈[−1.0,1.1]∧RAD50∈[−3.6,−1.6]∧RB1∈[−5.0,−2.4]∧SLC5A1∈[−10.0,−3.3]∧TCF3∈[−4.0,−2.1]∧XRCC1∈[−5.2,−4.1]∧XRCC4∈[−6.2,−4.1]∧XRCC5∈[−2.2,−1.0]\mathit{ALDH1A1}\in[-7.0,-3.4]\land\mathit{BSG}\in[-8.9,-3.9]\land\mathit{CA9}\in[-9.5,-4.5]\land\mathit{CBX4}\in[-5.3,-3.7]\land\mathit{CLDN4}\in[-4.9,-0.4]\land\mathit{CXCL12}\in[-4.3,-1.5]\land\mathit{ENO1}\in[-0.3,1.2]\land\mathit{ENO2}\in[-5.8,-2.4]\land\mathit{EPHA1}\in[-7.2,-4.6]\land\mathit{EPOR}\in[-10.2,-6.6]\land\mathit{ERCC5}\in[-5.2,-3.0]\land\mathit{HIF1A}\in[-5.7,-3.7]\land\mathit{LOX}\in[-2.9,-1.0]\land\mathit{MMP9}\in[-2.8,0.6]\land\mathit{MYNN}\in[-5.2,-2.8]\land\mathit{PGK1}\in[-2.0,-0.2]\land\mathit{PTEN}\in[-6.5,-5.1]\land\mathit{RAD23B}\in[-1.0,1.1]\land\mathit{RAD50}\in[-3.6,-1.6]\land\mathit{RB1}\in[-5.0,-2.4]\land\mathit{SLC5A1}\in[-10.0,-3.3]\land\mathit{TCF3}\in[-4.0,-2.1]\land\mathit{XRCC1}\in[-5.2,-4.1]\land\mathit{XRCC4}\in[-6.2,-4.1]\land\mathit{XRCC5}\in[-2.2,-1.0] ”“ ALDH1A1∈[−7.0,−3.4]∧CA9∈[−9.5,−4.5]∧CBX4∈[−5.3,−3.7]∧ENO1∈[−0.3,1.2]∧EPHA1∈[−7.2,−4.6]∧LOX∈[−2.9,−1.0]∧MMP9∈[−2.8,0.6]∧MYNN∈[−5.2,−2.8]∧SLC5A1∈[−10.0,−3.3]∧TCF3∈[−4.0,−2.1]∧XRCC1∈[−5.2,−4.1]\mathit{ALDH1A1}\in[-7.0,-3.4]\land\mathit{CA9}\in[-9.5,-4.5]\land\mathit{CBX4}\in[-5.3,-3.7]\land\mathit{ENO1}\in[-0.3,1.2]\land\mathit{EPHA1}\in[-7.2,-4.6]\land\mathit{LOX}\in[-2.9,-1.0]\land\mathit{MMP9}\in[-2.8,0.6]\land\mathit{MYNN}\in[-5.2,-2.8]\land\mathit{SLC5A1}\in[-10.0,-3.3]\land\mathit{TCF3}\in[-4.0,-2.1]\land\mathit{XRCC1}\in[-5.2,-4.1] ”Subgroup size: 12→1212\to 12 Exceptionality: 27.4009→26.071827.4009\to 26.0718
“ ATM∈[−4.7,−1.6]∧BIRC5∈[−4.2,−1.0]∧BNIP3L∈[−3.2,−1.2]∧BSG∈[−9.2,−3.9]∧CA9∈[−10.7,−2.9]∧CBX4∈[−5.3,−3.1]∧CD24∈[−8.3,−4.5]∧CDKN3∈[−6.1,−2.7]∧CXCL12∈[−7.5,−2.0]∧DCBLD1∈[−7.5,−4.5]∧ENO2∈[−6.2,−2.4]∧ERCC5∈[−5.7,−3.0]∧HK2∈[−6.7,−2.2]∧HSPA4∈[−4.0,−2.2]∧KDR∈[−6.0,−2.4]∧KRT17∈[−3.0,5.1]∧LOX∈[−7.1,−1.5]∧LOXL2∈[−6.4,−1.8]∧MCM6∈[−4.8,−2.5]∧MME∈[−9.6,−4.0]∧MRE11A∈[−6.1,−4.0]∧MTOR∈[−4.2,−2.3]∧MYC∈[−3.7,−0.5]∧MYNN∈[−5.0,−2.3]∧NBN∈[−3.9,−1.6]∧NOTCH1∈[−6.3,−1.9]∧RAD23B∈[−1.4,1.1]∧RAD50∈[−4.0,−1.6]∧RASSF6∈[−8.1,−1.6]∧RB1∈[−5.0,−2.3]∧RELA∈[−3.5,−1.4]∧SDHA∈[−6.2,−2.9]∧SERPINB2∈[−8.0,−0.6]∧TCF3∈[−3.9,−2.1]∧TP53∈[−5.0,−1.0]∧XPA∈[−5.3,−2.6]∧XPC∈[−5.7,−3.7]∧XRCC1∈[−5.2,−2.5]∧XRCC4∈[−5.8,−3.5]∧XRCC5∈[−2.8,−0.8]\mathit{ATM}\in[-4.7,-1.6]\land\mathit{BIRC5}\in[-4.2,-1.0]\land\mathit{BNIP3L}\in[-3.2,-1.2]\land\mathit{BSG}\in[-9.2,-3.9]\land\mathit{CA9}\in[-10.7,-2.9]\land\mathit{CBX4}\in[-5.3,-3.1]\land\mathit{CD24}\in[-8.3,-4.5]\land\mathit{CDKN3}\in[-6.1,-2.7]\land\mathit{CXCL12}\in[-7.5,-2.0]\land\mathit{DCBLD1}\in[-7.5,-4.5]\land\mathit{ENO2}\in[-6.2,-2.4]\land\mathit{ERCC5}\in[-5.7,-3.0]\land\mathit{HK2}\in[-6.7,-2.2]\land\mathit{HSPA4}\in[-4.0,-2.2]\land\mathit{KDR}\in[-6.0,-2.4]\land\mathit{KRT17}\in[-3.0,5.1]\land\mathit{LOX}\in[-7.1,-1.5]\land\mathit{LOXL2}\in[-6.4,-1.8]\land\mathit{MCM6}\in[-4.8,-2.5]\land\mathit{MME}\in[-9.6,-4.0]\land\mathit{MRE11A}\in[-6.1,-4.0]\land\mathit{MTOR}\in[-4.2,-2.3]\land\mathit{MYC}\in[-3.7,-0.5]\land\mathit{MYNN}\in[-5.0,-2.3]\land\mathit{NBN}\in[-3.9,-1.6]\land\mathit{NOTCH1}\in[-6.3,-1.9]\land\mathit{RAD23B}\in[-1.4,1.1]\land\mathit{RAD50}\in[-4.0,-1.6]\land\mathit{RASSF6}\in[-8.1,-1.6]\land\mathit{RB1}\in[-5.0,-2.3]\land\mathit{RELA}\in[-3.5,-1.4]\land\mathit{SDHA}\in[-6.2,-2.9]\land\mathit{SERPINB2}\in[-8.0,-0.6]\land\mathit{TCF3}\in[-3.9,-2.1]\land\mathit{TP53}\in[-5.0,-1.0]\land\mathit{XPA}\in[-5.3,-2.6]\land\mathit{XPC}\in[-5.7,-3.7]\land\mathit{XRCC1}\in[-5.2,-2.5]\land\mathit{XRCC4}\in[-5.8,-3.5]\land\mathit{XRCC5}\in[-2.8,-0.8] ”“ BIRC5∈[−4.2,−1.0]∧BSG∈[−9.2,−3.9]∧CA9∈[−10.7,−2.9]∧CD24∈[−8.3,−4.5]∧DCBLD1∈[−7.5,−4.5]∧LOX∈[−7.1,−1.5]∧LOXL2∈[−6.4,−1.8]∧MCM6∈[−4.8,−2.5]∧MME∈[−9.6,−4.0]∧MTOR∈[−4.2,−2.3]∧MYC∈[−3.7,−0.5]∧MYNN∈[−5.0,−2.3]∧RB1∈[−5.0,−2.3]∧TCF3∈[−3.9,−2.1]∧XPC∈[−5.7,−3.7]∧XRCC1∈[−5.2,−2.5]\mathit{BIRC5}\in[-4.2,-1.0]\land\mathit{BSG}\in[-9.2,-3.9]\land\mathit{CA9}\in[-10.7,-2.9]\land\mathit{CD24}\in[-8.3,-4.5]\land\mathit{DCBLD1}\in[-7.5,-4.5]\land\mathit{LOX}\in[-7.1,-1.5]\land\mathit{LOXL2}\in[-6.4,-1.8]\land\mathit{MCM6}\in[-4.8,-2.5]\land\mathit{MME}\in[-9.6,-4.0]\land\mathit{MTOR}\in[-4.2,-2.3]\land\mathit{MYC}\in[-3.7,-0.5]\land\mathit{MYNN}\in[-5.0,-2.3]\land\mathit{RB1}\in[-5.0,-2.3]\land\mathit{TCF3}\in[-3.9,-2.1]\land\mathit{XPC}\in[-5.7,-3.7]\land\mathit{XRCC1}\in[-5.2,-2.5] ”Subgroup size: 85→8785\to 87 Exceptionality: 11.2782→11.126011.2782\to 11.1260
Appendix
Table 6: Rules learned on the postoperative RCT cohort before and after pruning using a Jaccard similarity threshold of 0.950.95 for membership indicators along with the respective changes in subgroup sizes and exceptionality.
More in this topic

Survival Analysis

Survival analysis models when events such as disease progression, failure, or death occur, accounting for incomplete follow-up and censoring.

See all 106 papers

Explore similar work

CardsList
Apr 30, 2026

Differential Subgroup Discovery: Characterizing Where Two Populations Differ, and Why

We study the problem of understanding where two populations differ within a feature space, which we formalize in the concept of a differential subgroup: a subset of individuals from both populations who, despite sharing similar characteristics, exhibit exceptional differences in a target outcome. Differential subgroups reveal the regions of the feature space where population-level gaps are most pronounced and can help practitioners identify the covariate combinations that are structurally responsible for these differences, e.g.~in clinical analysis, model diagnostics, or treatment-effect studies. We introduce a general optimization objective for discovering differential subgroups and establish conditions under which the resulting subgroups admit a causal interpretation of population differences. We propose DiffSub, a gradient-based approach that discovers interpretable differential subgroups in tabular data. Across synthetic benchmarks, medical case studies, model-error analyses, and treatment-effect settings, DiffSub identifies informative subgroups that reveal where population differences arise and why.

Sascha Xu, Jilles Vreeken

CISPA Helmholtz Center for Information Security Saarbrücken, Germany

Topic pending
AskPDF
Jun 3, 2026

SurvPFN: Towards Foundation Models for Survival Predictions

Tabular foundation models (TFMs) have made rapid progress in standard classification and regression, but time-to-event survival prediction tasks have remained largely untouched. Unlike in standard regression tasks, survival prediction models must account for censored data. Standard TFMs cannot handle natively censored data, leading to biased and inaccurate predictions, making them unsuitable for real-world applications. To overcome this fundamental limitation, we propose SurvPFN, a prior-data fitted network (PFN), for survival prediction tasks. We pretrain SurvPFN on millions of synthetic survival prediction tasks to learn survival via distributional regression that accounts for censored data. SurvPFN works by (1) generating data with Weibull event times and a non-informative censoring mechanism; (2) integrating a censored event indicator; and (3) minimizing a censored negative log-likelihood. On SurvSet, a collection of real-world survival tasks, SurvPFN is highly competitive with classical and deep survival baselines without per-dataset fitting, a survival-specific architecture, or feature engineering. We show that survival can be treated as a continuous-time distributional regression problem with censored loss, unlocking the power of PFNs for time-to-event predictions.

Samuel Böhm, Lennart Purucker, Frank Hutter +1

Institute of Epidemiology and Prevention, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany · Institute of Epidemiology and Prevention, Medical Center – University of Freiburg, Freiburg, Germany · Department of Computer Science, University of Freiburg, Freiburg, Germany +7

Tabular Foundation ModelsSurvival AnalysisPrior-Data Fitted NetworksMortality Prediction
AskPDF
May 5, 2026

TabSurv: Adapting Modern Tabular Neural Networks to Survival Analysis

Survival analysis on tabular data is a well-studied problem. However, existing deep learning methods are often highly task-specific, which can limit the transfer of new approaches from other domains and introduce constraints that may affect performance. We propose TabSurv, an approach that adapts modern tabular architectures to survival analysis using either the Weibull distribution or non-parametric survival prediction. TabSurv optimizes SurvHL, a novel histogram loss function supporting censored data. In addition to a baseline feed-forward network, we implement deep ensembles of MLPs for survival analysis within TabSurv. In contrast to prior work, the ensemble components are trained in parallel, optimizing survival distribution parameters before averaging, which promotes diversity across ensemble component predictions. We perform a comprehensive empirical evaluation of different proposed architectures on 10 diverse real-world survival datasets. Our results show that TabSurv consistently outperforms on average established classical and deep learning baselines, such as RSF, DeepSurv, DeepHit, SurvTRACE. Notably, deep ensembles with Weibull parametrization instead of non-parametric models achieve the highest average rank by C-index. Overall, our study clarifies how modern tabular neural networks can be adapted and trained to tackle survival analysis problems, offering a strong and reliable approach. The TabSurv implementation is publicly available.

Stanislav Kirpichenko, Andrei Konstantinov, Lev Utkin

Higher School of Artificial Intelligence Technologies, Peter the Great St. Petersburg Polytechnic University, St. Petersburg, Russia

Survival AnalysisDeep EnsemblesTabular ML
AskPDF