Biliary tract cancers (BTCs), aggressive tumors originating from the epithelial cells of the bile ducts, are categorized anatomically into gallbladder cancer, perihilar cholangiocarcinoma, distal cholangiocarcinoma, and intrahepatic cholangiocarcinoma. Despite their global rarity—constituting less than 1% of all cancers—the incidence of BTCs is increasing (1). These cancers are highly malignant and prognostically unfavorable, with a 5-year OS rate below 20% (1). BTC typically presents insidiously, with 60-70% of patients diagnosed with unresectable disease. Even those undergoing potentially curative surgery face high recurrence rates, between 70% and 75% (2). Historically, the combination chemotherapy of gemcitabine and cisplatin, validated by the ABC-02 study, served as the standard first-line treatment for advanced BTC (3). However, its efficacy has been unsatisfactory. Recent breakthroughs in immunotherapy have revolutionized first-line treatment options. The TOPAZ-1 study demonstrated that adding durvalumab to the gemcitabine-cisplatin regimen significantly improves OS (4). Similarly, the KEYNOTE-966 study confirmed enhanced survival with the addition of pembrolizumab (5). Despite these advancements, the response to immunotherapy remains limited to a subset of patients. Predictive biomarkers such as programmed death-ligand 1 (PD-L1) expression, microsatellite instability-high, and tumor mutational burden provide some guidance for immunotherapy selection, but their predictive accuracy is limited (6, 7). Challenges such as the invasiveness of obtaining samples, intratumoral heterogeneity, the high cost of tests like immunohistochemistry and next-generation sequencing, and frequently inadequate biopsy specimens restrict their utility. Thus, the development of more convenient, accurate, and cost-effective predictive markers remains a critical need in advancing the treatment of BTC.

The circadian rhythmicity of the immune system is a critical factor that has traditionally been overlooked in cancer immunotherapy. Recent evidence increasingly supports the notion that the efficacy of immunotherapy might vary diurnally. For example, studies have demonstrated that the daytime administration of interferon-β yields a stronger antitumor response in mice compared to evening dosing (8). Clinical evidence also supports this circadian influence. A trial with stage I-II renal cell carcinoma patients revealed that IL-2 administered in the early morning (5:00–13:00) or late night (21:00–5:00) resulted in longer survival compared to afternoon administration (13:00–21:00) (9). In the context of immune checkpoint inhibitors (ICIs), the MEMOIR study pioneered the finding that administering ICIs earlier in the day (with less than 20% of infusions after 16:30) improved prognosis in melanoma patients (10). This suggests that the timing of ICI administration could significantly impact the effectiveness of immunotherapy. Although similar patterns have been observed in retrospective studies on non-small cell lung cancer and renal cell carcinoma (11–13), a pan-cancer study encompassing various solid tumors did not produce significant findings (14). This discrepancy could be attributed to the inclusion of tumors with lower immunogenicity in the pan-cancer study, whereas positive results have primarily been reported in cancers that are more responsive to immunotherapy (10–13, 15–17). Moreover, a recent meta-analysis involving 1,663 patients further supported these findings (18). It demonstrated a significant association between earlier ICI administration and improved OS and PFS, providing additional evidence for the value of chronotherapy in immunotherapy.

BTC is characterized by a poor prognosis, underscoring the urgent need to enhance the efficacy of available treatments, including immunotherapy. Historically regarded as an “immune cold” tumor (19–21), the application of immunotherapy in BTC is a recent development and remains limited in clinical experience. Given this context, it is imperative to explore novel factors that might influence the effectiveness of these treatments. One such factor is the timing of ICI administration. To investigate whether the timing of ICI infusions affects clinical outcomes in advanced BTC, we conducted a single-center retrospective study. We utilized propensity score matching and multivariate Cox proportional hazards regression analyses to minimize potential biases and ensure the robustness of our findings.

From the cohort criteria, 221 patients with ≥2 ICI infusions were included as the pre-matched cohort ( Figure 1 ), of which 178 patients (81%) received ≥3 ICI infusions. The average age was 57.97 ± 10.26 years, with 117 males (53%). ICIs were used as first-line therapy in 137 patients (62%). Among the 221 patients, 172 (78%) were treated with a combination of ICI and chemotherapy, while 19 (9%) received a regimen comprising ICI, chemotherapy, and anti-angiogenic drugs. Five patients (2%) underwent ICI monotherapy, and 21 (10%) were treated with a combination of ICI and anti-angiogenic drugs. A small number of patients followed alternative regimens: one patient (<1%) received ICI, chemotherapy, and targeted therapy (Trastuzumab); another (<1%) was treated with ICI, anti-angiogenic drugs, and targeted therapy (Cetuximab); and two patients (<1%) received ICI and targeted therapy (Trastuzumab). Primary tumor types included intrahepatic bile duct cancer (64%), perihilar bile duct cancer (12%), distal bile duct cancer (7%), and gallbladder cancer (17%). A total of 51 patients (23%) had ≥20% of ICI infusions after 16:30, while 170 (77%) had <20% after this time. Baseline characteristics are detailed in Supplementary Table S1 . Following the identification of influential factors via univariate Cox proportional hazards regression (detailed in Supplementary Table S2 ), a 1:2 propensity score matching yielded a cohort of 139 patients, with 49 (35%) receiving ≥20% of infusions after 16:30 and 90 (65%) receiving <20%. The baseline characteristics of the matched samples are presented in Supplementary Table S1 .

In the pre-matched cohort, patients with ≥20% of ICI infusions after 16:30 had a median follow-up of 8.3 months, compared to 10.45 months for those with <20% late infusions. Kaplan-Meier analysis showed significantly reduced OS for patients with ≥20% late infusions, with a median OS of 9.8 months (95% CI: 7.9–14.1) versus 13.7 months (95% CI: 12.3–17.5) for those with fewer late infusions (HR: 1.68, 95% CI: 1.13–2.51, p = 0.010; Figure 2A ).

After propensity score matching, patients with ≥20% of ICI infusions after 16:30 had significantly shorter OS, with a median of 10.1 months (95% CI: 7.9–14.9) compared to 14.5 months (95% CI: 12.7–19.8) for those with fewer late infusions (HR: 1.80, 95% CI: 1.14–2.84, p = 0.012; Figure 2B ). This association remained significant in the multivariate Cox proportional hazards regression analysis, after adjusting for factors including site of origin (intrahepatic vs. perihilar vs. distal vs. gallbladder), degree of differentiation (poorly vs. moderately-to-well), type of ICI (Anti-PD-1 vs. Anti-PD-L1), line of treatment for ICI (first-line vs. ≥2 lines), pre-treatment CA19-9 ≤ 500 U/mL, pre-treatment CEA ≤ 5 ng/mL, pre-treatment CA125 < 28.65 U/mL, and smoking status. The adjusted HR was 1.69 (95% CI: 1.03-2.80, P=0.040; Table 1 , Figure 2B ). Analysis of the entire unmatched sample through multivariate Cox proportional hazards regression also corroborated these findings, with an HR of 1.61 (95% CI: 1.07-2.42, P=0.021; Figure 2A ).

In the unmatched cohort, patients with ≥20% of ICI infusions after 16:30 had significantly reduced PFS, with a median of 4.9 months (95% CI: 4.2–6.6) compared to 8.1 months (95% CI: 7.0–9.1) for those with fewer late infusions (HR: 1.62, 95% CI: 1.15–2.29, p = 0.006; Figure 3 ). Multivariate Cox regression confirmed this result, adjusting for influential factors (HR: 1.81, 95% CI: 1.26–2.60, p = 0.001; Figure 3 ). No significant differences were found in ORR (χ²=1.53, p = 0.215; Table 2 ) or in AE occurrences, including Grade 3/4 AE, irAEs, and discontinuations due to AE (all p > 0.050; Table 3 ). The most common irAEs observed included thyroid dysfunction (38 patients, 17.2%), rash (14 patients, 6.3%), and cardiac events (11 patients, 5.0%).

We further analyzed the impact of ICI infusion timing using 16:00 and 15:30 as cutoffs. In the propensity score-matched sample, multivariate Cox regression showed no significant association with OS at the 16:00 cutoff (HR: 1.23, 95% CI: 0.82-1.84, p = 0.321) and the 15:30 cutoff (HR: 1.26, 95% CI: 0.86-1.86, p = 0.237; Supplementary Table S3 ). In the unmatched sample, ICI timing was significantly associated with PFS, with HRs of 1.55 (95% CI: 1.11–2.16, p = 0.009) for 16:00 and 1.71 (95% CI: 1.24-2.37, p = 0.001) for 15:30 ( Supplementary Table S4 ) after adjustment in multivariate Cox proportional hazards regression. No significant differences in ORR or irAEs, including Grade 3/4, were observed for either cutoff ( Supplementary Tables S5 , S6 ).

Additional analyses on patients with ≥3 ICI infusions showed that later infusion times were significantly associated with poorer OS after propensity score matching and multivariate Cox regression adjustment, using 16:30, 16:00, and 15:30 as cutoffs ( Supplementary Table S3 ). In the unmatched cohort, late infusion times were also significantly linked to reduced PFS at all three cutoffs after multivariate Cox regression adjustment ( Supplementary Table S4 ). No significant associations were found between infusion timing and ORR, incidence of irAEs, or Grade 3/4 irAEs at any cutoff ( Supplementary Tables S5 – S7 ).

To enhance the clinical relevance of our findings and reduce the impact of inter-patient heterogeneity, we conducted a subgroup analysis focusing on patients receiving first-line ICI plus chemotherapy. This subgroup included 123 patients, of whom 29 (24%) received ≥20% of their ICI infusions after 16:30, while 94 (76%) did not. The baseline characteristics of these patients are summarized in Supplementary Table S8 . In the unmatched cohort, patients with ≥20% of ICI infusions after 16:30 had significantly shorter OS (median OS: 10.1 months, 95% CI: 8.6–not estimable) compared to those with <20% of ICI infusions after this time (median OS: 17.5 months, 95% CI: 14.3–20.2) (HR: 2.59, 95% CI: 1.40–4.76, p = 0.002, Figure 4 ). This difference remained significant after adjustment using multivariate Cox regression (HR: 2.24, 95% CI: 1.14–4.39, p = 0.019). Following propensity score matching, 67 patients were included in the matched cohort, with 24 patients (36%) receiving ≥20% of ICI infusions after 16:30 and 43 patients (64%) receiving <20%. In the matched cohort, patients with later infusions had significantly shorter OS (median OS: 8.7 months, 95% CI: 7.8–not estimable) compared to those with earlier infusions (median OS: 17.6 months, 95% CI: 13.2–not estimable) (HR: 2.74, 95% CI: 1.25–5.99, p = 0.012, Figure 4 ). This finding remained robust after adjustment using multivariate Cox regression (HR: 2.73, 95% CI: 1.18–6.32, p = 0.019). In contrast, PFS showed a nonsignificant trend toward poorer outcomes in patients with ≥20% of infusions after 16:30 in the unmatched cohort (median PFS: 5.0 months, 95% CI: 4.5–9.4) compared to those with <20% (median PFS: 9.4 months, 95% CI: 7.9–10.7) (HR: 1.62, 95% CI: 1.00–2.62, p = 0.052, Figure 5 ). This trend did not reach statistical significance after multivariate Cox regression adjustment (HR: 1.63, 95% CI: 0.96–2.75, p = 0.069). For the alternative cutoff points of 16:00 and 15:30, no significant differences in OS were observed in the matched cohort, nor were there significant differences in PFS in the unmatched cohort ( Supplementary Table S9 ). No significant differences were observed in ORR or irAEs, including Grade 3/4 events, for either cutoff ( Supplementary Tables S10 ). Considering the small sample size, no further subgroup analysis was conducted for patients who received ≥3 ICI infusions.

Our study found that administering ≥20% of ICI infusions after 16:30 was linked to poorer OS and PFS in advanced BTC patients, without significantly affecting ORR or AE incidence. This is the first study to explore the impact of chronotherapy on ICI efficacy in BTC, suggesting that infusion timing may crucially influence outcomes. These findings highlight the potential for optimizing ICI administration timing in clinical practice to enhance prognosis for advanced BTC patients.

Our study aligns with the limited body of research on the chronotherapy of ICIs in cancer patients, which has predominantly focused on malignancies such as melanoma (10, 16), non-small cell lung cancer (12, 15), and renal cell carcinoma (13)—cancers relatively more responsive to immunotherapy. In contrast, BTC, which are highly malignant with limited treatment options, have only recently begun to explore the potential of immunotherapy, with limited experience in its use. The standard first-line treatment for advanced BTC has long been the gemcitabine plus cisplatin regimen, as established by the ABC-02 study (3), with no standard second-line treatments prior to the 2021 ABC-06 study (24), leading to generally poor prognoses. BTC is typically considered an immunologically “cold” tumor (19–21), and the effectiveness of ICIs in this context has been disappointing. Apart from a select group of patients with high microsatellite instability (MSI-H)/mismatch repair deficiency (dMMR) or high PD-L1 expression (25), the efficacy of ICIs remains limited. Data from various small single-arm studies indicate that, for the majority of BTC patients, ORR to ICIs ranges from 3% to 13%, and OS is between 5.2 to 8.1 months (26, 27). It was not until the recent publications of the TOPAZ-1 and KEYNOTE-966 studies that ICIs were formally incorporated into the first-line treatment for advanced BTC, though the improvement in survival has been marginal—durvalumab in the TOPAZ-1 study increased median OS from 11.5 months to 12.8 months, and pembrolizumab in the KEYNOTE-966 study from 10.9 months to 12.7 months (4, 5). Given the limited clinical experience with ICIs in advanced BTC, optimizing treatment modalities and identifying effective prognostic biomarkers or strategies to enhance ICI efficacy remain critical. Our findings indicate that the timing of ICI infusion significantly influences patient outcomes, suggesting that adjusting ICI administration times—a cost-effective intervention—could potentially enhance the effectiveness of immunotherapy in clinical practice for patients with advanced BTC.

BTCs are uncommon (1), and the recent introduction of immunotherapy in this area has limited the available research samples. In our retrospective study, we did not restrict variables like the line of immunotherapy to maximize sample size but used statistical techniques to balance confounding factors. Nevertheless, in the subgroup of patients receiving first-line chemotherapy plus ICI, a positive result was still observed, with ≥20% of ICI infusions after 16:30 being significantly associated with poorer OS. The absence of randomization in retrospective studies may lead to uneven factor distribution, potentially skewing results. To improve robustness, we included a wide range of confounders and identified key variables affecting OS during univariate analysis. Given the small sample size and the risk of reduced testing power and collinearity with multiple covariates in multivariate Cox regression, we used propensity score matching to compare patients across two treatment groups. This method is particularly useful in studies with multiple covariates (28). A 1:2 matching ratio increased the overall sample size. We also explored the effects of different ICI infusion cutoffs (15:30 and 16:00) on survival outcomes; however, not both times revealed significant differences in OS, suggesting that critical periods affecting ICI efficacy need further exploration. The decision to use 16:30 as the cutoff was based on more definitive evidence of its impact (10, 12, 13, 15). Previous studies selected this time not based on theoretical considerations but for practical reasons, such as facilitating reasonable grouping and clinical implementation (10). Similarly, while the MEMOIR study also conducted sensitivity analyses with different time cutoffs, no significant differences in OS were found when using 16:00 as a boundary (10). Although other studies have reported variations with different cutoffs (11, 16, 17), they generally suffer from small sample sizes and lack robust statistical validation. For patients receiving three or more ICI infusions, later infusion timing significantly correlated with reduced OS across all evaluated cutoffs. This discrepancy may be due to fewer observations and more confounders in those with fewer infusions. In fact, the MEMOIR study similarly limited its analysis to patients with at least four infusions. Additionally, categorizing patients solely by infusion timing may not precisely assess ICI administration timing, as it doesn’t account for variations among patients within the same group. Unlike OS, later ICI infusion was consistently linked to poorer PFS across cutoffs and infusion counts, likely due to fewer censored cases and less susceptibility to external confounders. In the subgroup of patients receiving first-line ICI plus chemotherapy, later ICI infusion showed a trend toward poorer PFS, but this did not reach statistical significance, potentially due to the relatively small sample size of this subgroup. No link was found between the timing of ICI infusion and the ORR, the incidence of irAEs or Grade 3/4 irAEs. The low incidence of these outcomes calls for further investigation with larger datasets. Regardless, future prospective trials with randomized infusion timings and larger sample sizes are crucial for validating these findings.

Our study differs from similar research like the MEMOIR study, which required a minimum of four ICI infusions for inclusion (10). We set a lower threshold of two infusions, mainly because in China, patients typically undergo imaging and efficacy evaluations after two to three cycles of systemic treatment. If tumor progression is observed at this stage, treatment plans are often adjusted, and ICI may be discontinued. Excluding patients with fewer than four infusions could omit a significant subset who stop ICI due to apparent poor response, potentially skewing the representation of ICI therapy’s impact in advanced BTC. A four-infusion threshold would also reduce our sample size, weakening the study’s statistical power. Including patients with at least two infusions is thus justified. To enhance robustness, we conducted exploratory analyses on patients receiving at least three infusions, showing consistent positive results across all time points—16:30, 16:00, and 15:30. In contrast, Rousseau et al.’s study (12) on non-small cell lung cancer included the number of ICI infusions as a covariate in their Cox regression and found no correlation between infusion timing and prognosis. However, the number of infusions reflects both treatment efficacy and patient survival; better responses and longer survival often lead to continued ICI, while poor responses result in early cessation. Including infusion number as a covariate may obscure the true effects of infusion timing on outcomes, so we excluded it from our models. Additionally, a pan-cancer retrospective study involving diverse cancers found no significant differences in PFS or OS between patients who received over 20% of their pembrolizumab infusions after 16:30 compared to those who did not (14). This study’s mixed nature, with over 30% rare solid tumors, introduces high variability in tumor immunogenicity, complicating the assessment of ICI infusion timing’s true impact on outcomes across cancers.

Human circadian rhythms may influence ICI efficacy based on treatment timing, paralleling established effects seen in chemotherapy and radiotherapy (29, 30). Studies indicate morning chemotherapy improves prognosis and tolerance (31, 32), and cancer cell radiosensitivity varies by time, with better outcomes from morning treatments (30, 33, 34). These variations likely arise from daily oscillations in metabolic proteins, including those involved in oxidation/reduction reactions and xenobiotic clearance, suggesting optimal treatment timing enhances efficacy and reduces adverse effects (35–37). Immunotherapy timing was largely overlooked until the MEMOIR study highlighted its potential impact (10). Both innate and adaptive immune activities exhibit circadian fluctuations (38, 39). For example, the migration of lymphocytes and dendritic cells to lymph nodes peaks at night, and these cells tend to leave during the day (40–42), and vaccinations later in the day elicit weaker adaptive responses (43, 44). Similarly, the levels of CD4+ and CD8+ T cells are lowest around 16:00, potentially reducing adaptive responses (45, 46). In mice, interferon-β shows stronger antitumor effects during the day (8). IL-2 timing affects survival in renal cell carcinoma patients, favoring morning or nighttime doses (9). A recent Nature study found dendritic cells migrate rhythmically to lymph nodes, impacting CD8+ T cell responses and immunotherapy outcomes (42). Additionally, rhythmic PD-1 expression in tumor-associated macrophages was noted, with peak expression enhancing PD-1 inhibitor efficacy (47). Although ICIs have long half-lives (48), potentially diminishing timing impact, observations suggest a nuanced interaction. A plausible explanation could involve the different pharmacokinetics of ICIs in the blood versus the tumor microenvironment (TME). One reason solid tumors respond less effectively to systemic treatments compared to hematological cancers is the limited direct exposure of drugs within the TME (49). Despite the prolonged half-life of ICIs in the bloodstream, their behavior in the TME can be markedly different. For instance, Cancer cells may secrete immune checkpoint fragments that neutralize ICIs, reducing efficacy (50, 51). If ICIs are given when TME immune cells are inactive, these drugs might be progressively captured by secreted immune checkpoint fragments. By the time immune cells reactivate, the concentration of ICIs may no longer be sufficient to effectively stimulate anti-tumor immunity. Nonetheless, these underlying mechanisms still require further experimental exploration to be definitively understood. For a comprehensive understanding of circadian rhythms and their impact on immunotherapy, recent reviews that detail the underlying biological mechanisms can be highly informative (52–54).

Our study has limitations. Firstly, as a single-center retrospective study, there is an inherent risk of bias, even with propensity score matching to balance confounders. Secondly, the rarity of BTC and the recent use of ICIs limited our sample size. We did not restrict immunotherapy lines or combination regimens, which may have introduced bias, though propensity score matching was used to minimize this. Additionally, we did not analyze factors like genetic mutations, PD-L1 expression, tumor mutational burden, or microsatellite instability due to limited testing. Future large, multicenter randomized trials are needed to validate these findings, alongside basic research to clarify how ICI timing affects efficacy.

In summary, our findings indicate that the timing of ICI infusions significantly affects the efficacy of immunotherapy in patients with advanced BTC. Patients receiving their ICI infusions before 16:30 demonstrated improved survival outcomes compared to those who received treatments later. Given these findings, it may be prudent to consider earlier ICI infusions in the daily treatment schedule for patients with advanced BTC, to potentially enhance therapeutic outcomes.