Summary: reported and anticipated modifiers of treatment effect

COMPANION-002 prespecified three stratification factors that define the subgroups in which effect modification would be examined: stage (locally advanced versus metastatic), anatomic subtype of the primary tumor (intrahepatic cholangiocarcinoma versus other), and Eastern Cooperative Oncology Group performance status (0 or 1). No subgroup analysis by tumor site, prior therapy, or geography has been reported in the sponsor's disclosures of the primary and key secondary endpoints, and the full dataset was scheduled for presentation in October 2026, so subgroup results are not yet available.

The heterogeneity that the sponsor has reported concerns crossover rather than baseline characteristics. Of 57 participants randomized to paclitaxel alone, 31 (54%) crossed over after centrally confirmed progression. Within that control arm, participants who crossed over had a median overall survival of 12.8 months versus 6.1 months for those who did not (hazard ratio 0.54, p=0.04), despite having progressed faster on initial paclitaxel monotherapy (median progression-free survival 1.9 versus 3.6 months, hazard ratio 2.31, p=0.007). In the same 31 participants, median progression-free survival after crossover was 3.5 months compared with 1.9 months before crossover (hazard ratio 0.36, p=0.0016). These are post hoc or within-participant comparisons in a non-randomized subset, and the sponsor states that the rank-preserving structural failure time adjustment for crossover rested on assumptions that were not met and is largely uninterpretable.

No predictive biomarker for DLL4 or VEGF-A blockade has been established in biliary tract cancer. The COMPANION-002 investigators describe DLL4 as highly expressed in biliary tract cancer with up-regulation associated with poor survival, and VEGF-A overexpression as correlating with stage, presence of metastases, and prognosis, but these are prognostic observations and the trial did not select participants by DLL4 or VEGF-A expression. By contrast, the comparator landscape is strongly biomarker-defined: response rates differ from 37.0% with pemigatinib and 42% with futibatinib in FGFR2 fusion-positive disease, to 41·3% with zanidatamab and 46.7% with tucatinib plus trastuzumab in HER2-positive disease, to 61.3% with trastuzumab deruxtecan in centrally confirmed HER2 immunohistochemistry 3+ tumors, to 51% with dabrafenib plus trametinib in BRAF V600E-mutated disease, against 5% with FOLFOX in unselected second-line participants.

Modifiers observed in the comparator trials are mainly clinical rather than molecular. In ABC-06 the survival benefit of FOLFOX was independent of platinum sensitivity, subgroups with suspected poorer prognosis appeared to benefit most, and the benefit appeared weaker in the extrahepatic cholangiocarcinoma subgroup, though the investigators state that the subgroup analyses were insufficiently powered. In SWOG S1815, the addition of nab-paclitaxel improved progression-free survival more in gallbladder carcinoma than in intrahepatic or extrahepatic cholangiocarcinoma (interaction P = .01) without a corresponding overall survival interaction. In the ABC-06 biomarker analysis, median overall survival fell from 8.9 months when no tumor marker was elevated to 3.2 months when carbohydrate antigen 19-9, carcinoembryonic antigen, and cancer antigen 125 were all elevated.

Prespecified stratification factors in COMPANION-002

“Stratification factors include stage (locally advanced vs metastatic), anatomic subtype of the primary tumor (intrahepatic cholangiocarcinoma vs other) and ECOG performance status (0 or 1).” (opens the source at this quote in a new tab)

Crossover subsets of the paclitaxel control arm

“In an analysis of OS in all patients randomized to the paclitaxel control arm (n=57), crossover patients who subsequently received tovecimig demonstrated a statistically significant improvement in median OS of 12.8 months compared to 6.1 months for non-crossover patients who received only paclitaxel (HR=0.54, p=0.04).” (opens the source at this quote in a new tab)
“Another analysis of these same patients randomized to the paclitaxel control arm (n=57) demonstrated that the crossover patients initially progressed faster on paclitaxel monotherapy compared to the non-crossover patients, with a median PFS of 1.9 months versus 3.6 months (HR=2.31, p=0.007).” (opens the source at this quote in a new tab)
“In this subset, tovecimig demonstrated a statistically significant improvement with median PFS2 of 3.5 months after treatment with tovecimig compared to median PFS1 of 1.9 months for paclitaxel (HR=0.36, p=0.0016).” (opens the source at this quote in a new tab)
“Though the RPSFT analysis is intended to adjust for crossover, its validity depends on certain assumptions that were not met in this study and thus its results here are largely uninterpretable.” (opens the source at this quote in a new tab)

DLL4 and VEGF-A expression as prognostic rather than selection markers

“DLL4 is highly expressed in BTC and up-regulation of DLL4 expression is associated with poor survival in patients with BTC” (opens the source at this quote in a new tab)
“VEGF-A overexpression has been reported in BTC, correlating with stage, the presence of metastases and prognosis of the disease” (opens the source at this quote in a new tab)

Statistical power for anatomic subtype effects in the NIFTY trial

“Although our trial was adequately powered for the primary endpoint and had a sample size (n=174) larger than that of the phase 3 ABC-06 trial (n=162), it might be insufficient to fully show the impact of liposomal irinotecan in each subtype of biliary tract cancer.” (opens the source at this quote in a new tab)

Subgroup and ethnicity findings in the pooled NIFTY and NALIRICC analysis

“In the prespecified subgroup analysis, nal-IRI plus 5-FU/LV showed benefit in most subgroups for PFS and OS, with the exception of patients with ECOG performance status 0 and elevated baseline C-reactive protein levels” (opens the source at this quote in a new tab)
“In addition, there was no significant interaction between the two trials for the efficacy of additional nal-IRI (p = 0.12 for PFS and p = 0.09 for OS).” (opens the source at this quote in a new tab)
“Ethnic differences were observed in the AE patterns: gastrointestinal toxicities, such as nausea, vomiting, and diarrhea, were more common and severe in German patients, whereas neutropenia was more frequent in Korean patients.” (opens the source at this quote in a new tab)

Subgroup findings in the second-line chemotherapy comparator

“The ABC-06 study showed how the overall survival and progression-free survival benefit was independent of whether patients were classed as being platinum sensitive or platinum refractory or resistant. In fact, subgroups with suspected poorer prognosis seemed to benefit most from FOLFOX.” (opens the source at this quote in a new tab)
“Our subgroup analysis also showed a weaker overall survival benefit from treatment with FOLFOX in the extrahepatic cholangiocarcinoma subgroup; however, this observation should be viewed with caution owing to insufficient statistical power.” (opens the source at this quote in a new tab)
“The insufficient power for subgroup analyses, which challenges some interpretations, is an acknowledged limitation, especially in such a heterogeneous disease.” (opens the source at this quote in a new tab)

Anatomic subtype as an effect modifier for taxane-containing chemotherapy

“Moreover, GAP versus GC showed greater improvement in PFS among participants with GBC than those with ICC or ECC (interaction P = .01), but not OS (interaction P = .28).” (opens the source at this quote in a new tab)

Tumor markers as prognostic stratifiers in second-line treatment

“In addition, median OS decreased from 8.9 months (no markers elevated) to 3.2 months (all three elevated).” (opens the source at this quote in a new tab)

Absence of a predictive biomarker for first-line immune checkpoint inhibitor choice

“At present, there are no discernible clinical, biochemical or molecular biomarkers, nor any notable differences in efficacy or toxicity, favouring one immune checkpoint inhibitor over the other.” (opens the source at this quote in a new tab)

Molecular covariates of response to futibatinib in FGFR2-altered disease

“Analysis of the molecular profile of the tumors in the 103 patients in the efficacy population indicated that responses did not correlate with FGFR2 fusion-partner status or co-occurring alterations in tumor-suppressor genes or oncogenes (Fig. 2).” (opens the source at this quote in a new tab)
“Numerical differences were noted with respect to median progression-free survival between patients with and those without CDKN2A (4.9 months and 9.7 months, respectively) and those with and those without CDKN2B alterations (4.8 months and 11.0 months).” (opens the source at this quote in a new tab)

HER2 immunohistochemistry score as a modifier of response to zanidatamab

“Although the trial was not designed to distinguish treatment effect by HER2 status, higher response rates were observed in patients with biliary tract cancer with HER2 IHC 3+ than in those with HER2 IHC 2+.” (opens the source at this quote in a new tab)
“For cohort 2, median progression-free survival was 1·9 months (95% CI 1·2–not estimable) and median overall survival was 5·5 months (95% CI 1·2–10·1).” (opens the source at this quote in a new tab)

Absence of molecular correlates of response to dabrafenib plus trametinib

“The findings of our exploratory DNA sequencing analysis did not highlight any specific gene alterations that could be attributed to responders or non-responders, although the small sample size (n=16) did not allow for formal statistical comparisons.” (opens the source at this quote in a new tab)

Tovecimig (CTX-009)

Heterogeneity of treatment effect