Clinical efficacy of integrated Chinese-Western medicine therapy for metastatic colorectal cancer

Abstract

BACKGROUND

Systemic treatment for metastatic colorectal cancer (mCRC) is limited by toxicity, drug resistance, and variable patient tolerance. Integrated Chinese-Western medicine is widely used in China, but its association with survival in real-world settings requires careful evaluation.

AIM

To evaluate the association between integrated Chinese-Western medicine therapy and survival in patients with mCRC using a target trial emulation-informed observational design.

METHODS

This retrospective multicenter cohort included patients with mCRC treated from January 1, 2017, to January 1, 2026. The components of the target trial emulation were prespecified. Patients were classified according to the observed duration of oral Chinese herbal medicine use (≥ 3 months vs < 3 months). The primary outcome was overall survival (OS), and the secondary outcome was cancer-specific survival. Stabilized inverse probability weighting was used to balance measured baseline covariates. Weighted Cox models were then used to estimate survival associations. Because the groups were defined by treatment duration after baseline, the estimates were interpreted as associations rather than causal effects.

RESULTS

Of 649 screened patients, 627 were eligible: 490 received oral Chinese herbal medicine for ≥ 3 months, and 137 received it for < 3 months. After weighting, 259 patients in the ≥ 3-month group and 111 in the < 3-month group contributed to the analysis. All reported standardized mean differences were < 0.1. Median OS was 37.5 months in the ≥ 3-month group and 19.5 months in the < 3-month group. The < 3-month group had higher observed hazards of cancer-specific death [hazard ratio (HR) = 6.327, 95%CI: 3.692-10.844] and all-cause death (HR = 14.954, 95%CI: 9.054-24.701; P < 0.001). The unusually large HR should not be interpreted as a causal treatment effect. Its magnitude may have been influenced by the duration-based classification, residual confounding, and model specification.

CONCLUSION

In this multicenter real-world cohort, oral Chinese herbal medicine use for ≥ 3 months as part of integrated Chinese-Western medicine therapy was associated with longer OS and cancer-specific survival in patients with mCRC. Prospective studies using landmark and time-varying exposure analyses are needed before causal effects or clinical benefits can be established.

Key Words: Target trial emulation; Integrated Chinese-Western medicine; Metastatic colorectal cancer; Real-world study; Inverse probability weighting

Core Tip: In this multicenter retrospective cohort of patients with metastatic colorectal cancer, oral Chinese herbal medicine use for ≥ 3 months was associated with longer overall and cancer-specific survival after inverse probability weighting. The therapeutic principle was standardized, whereas prescriptions were individualized according to syndrome differentiation. Because exposure was defined by observed treatment duration, immortal time bias and residual confounding remain possible. These hypothesis-generating findings require prospective confirmation using landmark and time-varying exposure analyses.

INTRODUCTION

Colorectal cancer (CRC) is the third most common malignancy and the second leading cause of cancer-related death worldwide[1]. Distant metastasis is the main cause of CRC-related mortality. Approximately 50% of patients with CRC are estimated to present with distant metastasis at diagnosis, and the 5-year survival rate is only 25%-44%[2]. Current treatment options for unresectable metastatic CRC (mCRC) include surgery, local ablation, systemic chemotherapy, targeted therapy, and immunotherapy. Although these treatments have improved survival, major challenges remain. Treatment-related adverse events can markedly reduce quality of life. Patients may develop drug resistance after multiple lines of therapy, and some cannot tolerate standard treatment because of poor physical condition or financial constraints[3]. Therefore, safer, more effective, and more individualized treatment strategies are still needed.

Traditional Chinese medicine (TCM) is used as an adjunct to conventional treatment for CRC in China. A meta-analysis reported improved tumor response when traditional medicines were added to oxaliplatin-based chemotherapy[4]. However, a quality-adjusted meta-analysis of Chinese herbal medicine combined with fluorouracil-based chemotherapy found that most evidence was at high risk of bias and that reliable evidence of clinical benefit remained limited[5]. These findings do not establish a survival benefit for the Jianpi Bushen Jiedu method. Moreover, TCM treatment is individualized through syndrome differentiation, which can complicate evaluation in conventional randomized controlled trials. A rigorous evaluation of this method is therefore needed to characterize its clinical associations.

Real-world studies complement randomized controlled trials by evaluating interventions in complex clinical settings. They can also include large samples, long follow-up periods, and broad patient populations[6]. However, observational studies remain vulnerable to confounding by indication, immortal time bias, and prevalent-user bias[7]. The target trial emulation (TTE) framework specifies the components of a hypothetical randomized trial before observational data are analyzed. This approach makes assumptions more transparent and can reduce avoidable design bias[8]. However, it does not convert an observational association into proof of causality.

Accordingly, this study used a TTE-informed analysis of real-world data to evaluate the association between the Jianpi Bushen Jiedu method and survival in patients with mCRC. The analysis explicitly recognized the residual biases associated with a duration-defined observational exposure.

MATERIALS AND METHODS

Study design

This retrospective multicenter observational study used real-world data to emulate selected features of a target trial. It evaluated the association between the Jianpi Bushen Jiedu method and survival in patients with mCRC. The observed treatment strategies were oral Chinese herbal medicine for < 3 months and for ≥ 3 months. OS was the primary outcome. Results were reported as adjusted associations rather than definitive causal effects.

Data sources

This study expanded the cohort established under the 2022 Capital Health Development Research Key Project (2022-1-4171). It included patients diagnosed with mCRC between January 1, 2017, and January 1, 2026, at three tertiary hospitals: Xiyuan Hospital, China Academy of Chinese Medical Sciences; Peking University Cancer Hospital; and Dongzhimen Hospital, Beijing University of Chinese Medicine. The Ethics Committee of Xiyuan Hospital approved the study, approval No. 2022XLA118-2, and the study was registered under ChiCTR2400081697.

Diagnostic criteria

CRC was diagnosed according to the “Chinese Guidelines for the Diagnosis and Treatment of Colorectal Cancer (2023 Edition)” issued by the National Health Commission[9]. Tumor-node-metastasis staging for colon and rectal cancer was based on the eighth edition of the American Joint Committee on Cancer staging system.

TCM syndrome differentiation was based on GB/T 16751.2-2021, “Clinic terminology of traditional Chinese medical diagnosis and treatment - Part 2: Syndromes/patterns”[10], and the 2002 trial edition of “Guiding Principles for Clinical Research of New Chinese Medicines”, edited by Zheng et al[11]. Deficiency syndromes included spleen-kidney deficiency, liver-kidney deficiency, liver qi stagnation with spleen deficiency, lung-kidney deficiency, and lung-spleen deficiency. Excess syndromes included phlegm-stasis obstruction, stasis-toxin obstruction, phlegm-damp retention, and internal retention of dampness. Syndrome assessments were retrospectively extracted from medical records for the domestic cohort (January 1, 2017, to January 1, 2025). At least two TCM oncologists at the rank of associate chief physician or above independently determined each syndrome classification.

Study methods

Emulation of the target trial: The study design was based on published TTE protocols[12,13]. We used real-world data to emulate a hypothetical randomized trial and to reduce avoidable design bias. The emulated design was aligned with the target trial as closely as the available data allowed. The seven prespecified components were eligibility criteria, treatment strategies, follow-up, assignment procedure, outcomes, causal contrasts, and statistical analysis (Table 1).

Table 1 Emulated target trial framework for Jianpi Bushen Jiedu method in treating metastatic colorectal cancer.

Study process	Real-world study protocol	Target trial emulation protocol
Inclusion criteria	(1) Age ≥ 18 years, any sex; (2) Pathologically confirmed colorectal cancer; (3) TNM stage IV; (4) Received standard Western medicine treatment; (5) First use of Jianpi Bushen Jiedu herbal medicine at baseline; (6) Expected survival ≥ 3 months at baseline; and (7) Complete baseline medical records including TCM four diagnostic methods and syndrome differentiation	Same
Exclusion criteria	(1) Patients with two or more concurrent malignancies; (2) Severe cardiac, hepatic, or renal dysfunction at baseline; (3) Patients who have undergone curative treatment and are tumor-free; and (4) Participation in any interventional clinical trial within 3 months prior to enrollment	Same
Treatment strategy	Treatment group: Oral Jianpi Bushen Jiedu herbal medicine initiated within 30 days after T0, using syndrome-specific core prescriptions with individualized modifications, duration ≥ 3 months	Observed duration-based comparison. The therapeutic principle was standardized, whereas prescriptions were individualized by syndrome. Because duration was determined after T0, this comparison may be affected by immortal time bias and is interpreted as associational
	Control group: The same therapeutic principle and individualized prescribing process, duration < 3 months. Prescription changes and self-reported use were reviewed at follow-up; a validated adherence metric was not available for all patients
Assignment procedure	Randomly assigned to one treatment strategy at baseline	Observed initiation within 30 days after T0; subsequent duration was not assigned at baseline
Treatment initiation	Randomization time	From T0 to within 30 days after T0
Blinding	Open-label	Non-blinded
Time zero	Time of intervention assignment (randomization)	Date of first oral Jianpi Bushen Jiedu prescription; duration-based classification after T0 may still introduce immortal time bias
Follow-up	From baseline until death	From T0 until death
Primary outcome	OS	Same
Causal contrast	Per-protocol effect	Adjusted associational contrast; not a definitive causal effect because treatment duration was observed during follow-up
Statistical analysis plan	Per-protocol analysis	IPW-adjusted observational analysis of measured baseline confounders; landmark and time-varying exposure analyses are needed for a complete assessment

TNM: Tumor-node-metastasis; TCM: Traditional Chinese medicine; IPW: Inverse probability weighting; OS: Overall survival.

Eligibility criteria: The inclusion and exclusion criteria were identical to those specified for the target trial.

Treatment strategies: Conventional Western medicine treatment included chemotherapy with agents such as oxaliplatin, irinotecan, capecitabine, and fluorouracil, as well as targeted therapy with agents such as cetuximab and bevacizumab. Regimens were categorized as first-line, second-line, or later-line according to previous treatment and treatment response. First-line therapy was the preferred standard regimen for treatment-naïve patients. Second-line therapy was given after first-line treatment failed because of disease progression, unacceptable toxicity, or recurrence. Later-line therapy referred to third-line and subsequent treatment.

The core prescriptions used in the integrated treatment group were as follows.

For spleen-kidney deficiency syndrome, the core prescription was modified Sijunzi decoction (Codonopsis Radix 15 g, stir-fried Atractylodis Macrocephalae Rhizoma 12 g, Poria 15 g, and Glycyrrhizae Radix et Rhizoma Praeparata cum Melle 6 g). It was combined with modified Shenqi Pill (Psoraleae Fructus 10 g, Cuscutae Semen 12 g, Ligustri Lucidi Fructus 15 g, and Ecliptae Herba 15 g).

For liver-kidney deficiency syndrome, the core prescription was modified Liuwei Dihuang Pill (Rehmanniae Radix Praeparata 15 g, Corni Fructus 12 g, Dioscoreae Rhizoma 15 g, and Poria 12 g). It was combined with Erzhi Pill (Ligustri Lucidi Fructus 15 g and Ecliptae Herba 15 g).

For liver qi stagnation with spleen deficiency syndrome, the core prescription was Sijunzi decoction (Codonopsis Radix 15 g, stir-fried Atractylodis Macrocephalae Rhizoma 12 g, Poria 15 g, and Glycyrrhizae Radix et Rhizoma Praeparata cum Melle 6 g). It was combined with modified Xiaoyao Powder (Bupleuri Radix 10 g, Angelicae Sinensis Radix 12 g, and Paeoniae Radix Alba 10 g).

For lung-kidney deficiency syndrome, the core prescription was Shengmai Powder (Pseudostellariae Radix 15 g, Ophiopogonis Radix 12 g, and Schisandrae Chinensis Fructus 6 g). It was combined with modified Liuwei Dihuang Pill (Ligustri Lucidi Fructus 15 g, Ecliptae Herba 15 g, and Psoraleae Fructus 10 g).

For lung-spleen deficiency syndrome, the core prescription was Liujunzi decoction (Codonopsis Radix 15 g, stir-fried Atractylodis Macrocephalae Rhizoma 12 g, Poria 15 g, Glycyrrhizae Radix et Rhizoma Praeparata cum Melle 6 g, Citri Reticulatae Pericarpium 10 g, and Pinelliae Rhizoma Praeparatum cum Zingibere et Alumine 9 g). It was combined with modified Yupingfeng Powder (Astragali Radix 20 g and Saposhnikoviae Radix 10 g).

For Phlegm-Stasis Obstruction syndrome, the core prescription was Sijunzi decoction (Codonopsis Radix 15 g, stir-fried Atractylodis Macrocephalae Rhizoma 12 g, Poria 15 g, and Glycyrrhizae Radix et Rhizoma Praeparata cum Melle 6 g). It was combined with modified Taohong Siwu Decoction (Persicae Semen 10 g, Carthami Flos 6 g, Angelicae Sinensis Radix 12 g, and Chuanxiong Rhizoma 10 g), together with herbs that soften hardness and dissipate masses.

For Phlegm-Damp Retention syndrome, the core prescription was Erchen decoction (Citri Reticulatae Pericarpium 10 g, Pinelliae Rhizoma Praeparatum cum Zingibere et Alumine 9 g, Poria 15 g, and Glycyrrhizae Radix et Rhizoma Praeparata cum Melle 6 g). It was combined with modified Pingwei Powder (Atractylodis Rhizoma 10 g and Magnoliae Officinalis Cortex 10 g).

For Internal Retention of Dampness syndrome, the core prescription was modified Shenling Baizhu Powder (Codonopsis Radix 15 g, stir-fried Atractylodis Macrocephalae Rhizoma 12 g, Poria 15 g, stir-fried Coicis Semen 15 g, and stir-fried Lablab Semen Album 12 g). It was combined with modified Huoxiang Zhengqi Powder (Pogostemonis Herba 10 g and Eupatorii Herba 10 g).

For Stasis-Toxin Obstruction syndrome, the core prescription was modified Jiedu Quyu Formula (Euonymi Ramulus 15 g, Salviae Chinensis Herba 15 g, vinegar-processed Curcumae Rhizoma 10 g, Hedyotis Diffusa Herba 20 g, and Scutellariae Barbatae Herba 20 g).

Treatment delivery, individualization, and adherence: The therapeutic principle (Jianpi Bushen Jiedu) was standardized, whereas individual prescriptions were tailored according to syndrome differentiation. The syndrome-specific core prescriptions and typical doses are listed above. During follow-up, physicians added, removed, or adjusted herbs according to changes in syndrome, treatment toxicity, and clinical status. All changes were recorded in the medical record. Survival status, disease progression, prescriptions, and self-reported medicine use were reviewed every 3 months. Adherence was assessed using prescription and refill records together with follow-up documentation. However, a validated quantitative adherence measure was unavailable for all patients, limiting interpretation of the duration-defined exposure. Concurrent systemic treatment was recorded at baseline, including chemotherapy, targeted therapy, immunotherapy when applicable, treatment line, and palliative surgery. Available variables were included among the measured covariates, but residual differences in treatment intensity may have remained between groups.

Time zero and follow-up: In the TTE-informed design, time zero (T₀) was the date of the first oral Jianpi Bushen Jiedu prescription. Eligibility criteria and baseline covariates were assessed at T₀. Follow-up began at T₀ and ended at death, loss to follow-up, or the study end date. Although this common time origin avoided a pre-exposure waiting period, classification based on subsequent treatment duration could introduce immortal time bias. Therefore, we interpreted the primary duration-based analysis as associational. A complete bias assessment requires a 3-month landmark analysis and a time-varying exposure analysis.

Assignment procedure: Patients were classified according to the observed duration of oral Chinese herbal medicine use: ≥ 3 months or < 3 months. Because this classification was determined after baseline, it could not emulate baseline random assignment without additional time-dependent methods. inverse probability weighting (IPW) improved balance in measured baseline covariates, but it could not remove immortal time bias or unmeasured confounding. Accordingly, weighted estimates were presented as adjusted associations rather than causal effects.

Outcomes: OS was the primary outcome and was defined as the time from T₀ to death from any cause. Median OS was calculated from the same time origin. Patients who were alive, lost to follow-up, or event-free at the end of the study were censored at the date of their last confirmed follow-up. Cancer-specific survival (CSS) was the secondary outcome and was defined as the time from T₀ to death attributed to CRC. Deaths from other causes were censored in the cause-specific analysis.

Causal contrast: The primary contrast was the adjusted association between the observed duration of oral Chinese herbal medicine use (≥ 3 months vs < 3 months) and OS. It was not an intention-to-treat effect because the treatment strategy was not assigned at cohort entry and treatment duration was measured during follow-up. The estimate was therefore considered hypothesis-generating.

Statistical analysis

Analyses were performed using R Studio version 4.5.2. Continuous variables were summarized as means and standard deviations and compared using independent-samples t tests. Categorical variables were summarized as counts and percentages and compared using χ² tests. Weighted Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Stabilized IPW was used to balance measured baseline covariates, and weights were truncated at the 1^st and 99^th percentiles. An absolute standardized mean difference < 0.1 indicated acceptable balance. Interaction terms were assessed using likelihood ratio tests. Leave-one-out analyses evaluated the influence of individual observations. The E-value for the treatment HR quantified the minimum association that an unmeasured confounder would need to have with both exposure and outcome to move the observed association to the null. Landmark and time-varying exposure analyses, propensity score matching, and multiple imputation should be completed and reported before any definitive causal interpretation.

Reporting guidelines

This observational study was conducted within the TTE framework. Reporting followed the STROBE statement and the TARGET guideline.

RESULTS

Patient enrollment

Of 649 patients screened, 22 were excluded: 8 had two or more concurrent malignancies, 10 had severe cardiac, hepatic, or renal dysfunction at baseline, and 4 were expected to be unable to complete follow-up.

The final analysis included 627 patients: 490 in the treatment group and 137 in the control group. Of these patients, 525 were treated at Xiyuan Hospital, 62 at Peking University Cancer Hospital, and 40 at Dongzhimen Hospital, Beijing University of Chinese Medicine. Figure 1 shows the screening process. The detailed screening process is shown in Figure 1.

Figure 1 Patient enrollment flowchart. mCRC: Metastatic colorectal cancer; TCM: Traditional Chinese medicine; IPW: Inverse probability weighting.

Before weighting, several baseline characteristics differed between the groups

After IPW, 259 patients in the ≥ 3-month group and 111 in the < 3-month group contributed to the weighted analysis. All reported absolute standardized mean differences were < 0.1 (Table 2), indicating adequate balance in the measured covariates included in the weighting model. However, this finding does not demonstrate balance in unmeasured factors such as tumor burden, treatment tolerance, or changes in systemic treatment.

Table 2 Baseline characteristics before and after inverse probability weighting, n (%)/mean ± SD.

Covariate	Treatment (n = 490)	Control (n = 137)	P value	Treatment (n = 259)	Control (n = 111)	P value
Age	-	-	0.844	-	-	0.905
< 60 years	291 (59.39)	83 (60.58)	-	107 (41.31)	44 (39.64)	-
≥ 60 years	199 (40.61)	54 (39.42)	-	152 (58.69)	67 (60.36)	-
Sex	-	-	0.493	-	-	0.264
Male	201 (41.02)	61 (44.52)	-	156(60.23)	56 (50.45)	-
Female	289 (59.98)	76 (55.48)	-	103 (39.77)	55 (49.55)	-
Tumor location	-	-	0.105	-	-	0.305
Left colon	347 (70.82)	107 (78.10)	-	190 (73.36)	95 (85.59)	-
Right colon	143 (29.18)	30 (21.90)	-	69 (26.64)	16 (14.41)	-
Histology	-	-	0.019	-	-	0.140
Adenocarcinoma	451 (92.04)	117 (85.40)	-	238 (91.89)	92 (82.88)	-
Non-adenocarcinoma	39 (7.96)	20 (14.60)	-	21 (8.11)	19 (17.12)	-
Gene type	-	-	0.752	-	-	0.375
Wild type	146 (29.80)	43 (31.39)	-	83 (32.05)	46 (41.44)	-
Mutant	344 (70.20)	94 (68.62)	-	176 (67.95)	65 (58.56)	-
Prior treatment	-	-	0.528	-	-	0.898
Treatment-naïve	167 (34.08)	35 (25.55)	-	84 (32.43)	33 (29.73)	-
First- or second-line	265 (54.08)	78 (56.93)	-	138 (53.28)	66 (59.46)	-
Third-line or later	58 (11.84)	24 (17.52)	-	37 (14.29)	12 (10.81)	-
Palliative surgery	272 (55.51)	75 (54.74)	0.923	147 (56.76)	58 (52.25)	0.575
Prior targeted therapy	186 (37.96)	44 (32.12)	0.230	92 (35.52)	28 (25.23)	0.272
Comorbidities	148 (30.20)	75 (54.74)	< 0.001	90 (34.75)	54 (48.65)	0.155
TCM deficiency syndromes	-	-	0.187	-	-	0.755
Spleen-kidney deficiency	175 (35.72)	77 (56.20)	-	95 (36.68)	47 (42.35)	-
Liver-kidney deficiency	76 (15.51)	23 (16.79)	-	44 (16.99)	18 (16.22)	-
Liver qi stagnation with spleen deficiency	141 (28.78)	12 (8.76)	-	65 (25.10)	25 (22.52)	-
Lung-kidney deficiency	43 (8.76)	13 (9.49)	-	24 (9.27)	8 (7.20)	-
Lung–spleen deficiency	26 (5.31)	9 (6.57)	-	16 (6.18)	8 (7.20)	-
Others	29 (5.92)	3 (2.19)	-	15 (5.79)	5 (4.51)	-
TCM excess syndromes	-	-	0.001	-	-	0.687
Phlegm-stasis obstruction	236 (48.16)	60 (43.79)	-	144 (55.60)	53 (47.75)	-
Stasis-toxin obstruction	75 (15.31)	11 (8.03)	-	31 (11.97)	17 (15.32)	-
Phlegm-damp retention	156 (31.84)	48 (35.04)	-	80 (30.89)	33 (29.73)	-
Internal retention of dampness	23 (4.69)	18 (13.14)	-	4 (1.54)	9 (7.20)	-
KPS	-	-	< 0.001	-	-	0.372
< 80	295 (60.20)	40 (29.20)		139 (53.67)	67 (60.36)	-
≥ 80	195 (39.80)	97 (70.80)		120 (46.33)	44 (39.64)	-
BMI	23.49 ± 3.91	23.65 ± 3.64	0.769	23.58 ± 3.90	22.96 ± 3.75	0.427
Metastatic sites
Liver	287 (58.57)	84 (61.31)	0.694	147 (56.76)	67 (60.36)	0.383
Lung	182 (37.14)	56 (40.88)	0.488	106 (40.93)	46 (41.44)	0.947
Brain	15 (3.06)	7 (5.11)	0.296	10 (3.86)	3 (2.70)	0.409
Peritoneum	34 (6.93)	10 (7.30)	0.852	12 (4.63)	9 (8.11)	0.734

TCM: Traditional Chinese medicine; KPS: Karnofsky performance status; BMI: Body mass index.

Overall survival

At the end of follow-up, the weighted median OS was 37.5 months in the ≥ 3-month group and 19.5 months in the < 3-month group (P < 0.05). Because group membership depended on treatment duration accumulated after T₀, this difference was descriptive and may have been affected by immortal time bias (Figure 2A).

Figure 2 Survival curves. A: Overall survival curves of the two patient groups; B: Cancer-specific survival curves. IPW: Inverse probability weighting; HR: Hazard ratio; CI: Confidence interval.

CSS analysis

In the CSS analysis, cancer-specific death was treated as the event, whereas non-cancer death was censored. The IPW-weighted Cox model showed a higher observed hazard of cancer-specific death in the < 3-month group than in the ≥ 3-month group (HR = 6.327, 95%CI: 3.692-10.844; P < 0.001). This association should not be interpreted as proof that treatment duration caused the difference in cancer-specific mortality (Figure 2B).

Multivariable Cox regression analysis

In the weighted multivariable Cox model, the observed hazard of all-cause death was higher in the < 3-month group than in the ≥ 3-month group (HR = 14.954, 95%CI: 9.054-24.701; P < 0.001; Table 3). This HR is unusually large for an oncology study and requires caution. Its magnitude may reflect the duration-based exposure definition, residual or time-varying confounding, sparse events in some strata, and model dependence, as well as any true treatment-related association. The E-value was 29.40. Under its assumptions, this value quantifies the minimum strength of association that an unmeasured confounder would need to have with both exposure and outcome to move the estimate to the null. However, the E-value does not address immortal time bias or establish causality.

Table 3 Multivariable Cox proportional hazards regression model.

Covariate	β (coefficient)	HR	E-value	95%CI	P value
Treatment (control)	2.705	14.954	29.40	9.054-24.701	< 0.001a
Sex (male)	0.354	1.424	2.20	1.119-1.813	0.004a
Age (≥ 60 years)	-0.026	0.974	-	0.781-1.215	0.816
Tumor site (right colon)	0.149	1.161	1.59	0.905-1.491	0.241
Histology (non-adenocarcinoma)	0.455	1.576	2.53	1.058-2.347	0.025a
Gene type (mutant)	0.533	1.704	2.80	1.354-2.145	< 0.001a
Prior treatment (first- or second-line vs treatment-naive)	-0.314	0.731	-	0.559-0.955	0.023a
Prior treatment (third-line or later vs treatment-naive)	0.550	1.734	2.86	1.256-2.393	< 0.001a
Palliative surgery	-0.132	0.876	-	0.689-1.115	0.283
Prior targeted therapy	-0.069	0.933	-	0.720-1.210	0.603
Comorbidities	0.176	1.192	1.67	0.950-1.495	0.129
TCM deficiency - spleen-kidney	0.175	1.192	1.67	0.874-1.624	0.267
TCM deficiency - liver-kidney	0.452	1.572	2.52	0.959-2.576	0.073
TCM deficiency - liver qi stagnation with spleen deficiency	0.946	2.573	4.58	1.644-4.028	< 0.001a
TCM deficiency - lung-kidney	-0.038	0.963	-	0.724-1.281	0.795
TCM deficiency - lung-spleen	-0.274	0.760	-	0.477-1.210	0.247
TCM excess - phlegm-stasis obstruction	-0.167	0.846	-	0.632-1.316	0.260
TCM excess - stasis-toxin obstruction	-0.383	0.682	-	0.469-0.992	0.045a
TCM excess - phlegm-damp retention	-0.121	0.886	-	0.675-1.164	0.386
TCM excess - internal retention of dampness	0.341	1.406	2.16	0.890-2.222	0.144
KPS ≥ 80	-1.315	0.269	-	0.101-0.711	0.008a
BMI ≥ 24	-0.009	0.991	-	0.967-1.017	0.504
Liver metastasis	0.037	1.037	-	0.818-1.315	0.762
Lung metastasis	-0.102	0.903	-	0.719-1.134	0.380
Brain metastasis	0.641	1.898	3.20	1.024-3.518	0.042a
Peritoneal metastasis	0.222	1.248	1.80	0.801-1.946	0.327

^aP < 0.05.

HR: Hazard ratio; CI: Confidence interval; TCM: Traditional Chinese medicine; KPS: Karnofsky performance status; BMI: Body mass index.

Subgroup analysis

Interactions were assessed for baseline covariates. Statistically significant or borderline interactions were observed for sex, gene type, prior treatment, TCM deficiency syndrome, and Karnofsky performance status. These subgroup estimates are exploratory because several tests were performed and exposure duration was determined during follow-up. They should not be used to identify patients who will benefit without replication in a prospectively specified analysis (Table 4).

Table 4 Interaction analysis.

Subgroup	HR	95%CI	P value	P for interaction
Sex (male)	0.528	0.296-0.941	0.030	0.003
Sex (female)	1.676	1.014-2.770	0.044	0.003
Gene type (wild type)	0.467	0.236-0.922	0.028	0.005
Gene type (mutant)	1.463	0.989-2.163	0.057	0.005
Prior treatment (third-line or later)	2.175	1.389-3.407	0.001	0.005
TCM deficiency - liver qi stagnation with spleen deficiency	0.465	0.244-0.887	0.020	0.014
KPS < 80	2.492	1.317-4.717	0.005	0.058

HR: Hazard ratio; CI: Confidence interval; TCM: Traditional Chinese medicine; KPS: Karnofsky performance status.

DISCUSSION

This retrospective multicenter study used real-world data and a TTE-informed framework to examine the association between the Jianpi Bushen Jiedu method and survival in patients with mCRC. After IPW balanced measured baseline covariates, the ≥ 3-month group had longer weighted median OS and lower observed cancer-specific mortality than the < 3-month group. These findings are consistent with a potential benefit of integrated Chinese-Western medicine, but they do not establish causality. Patients had to remain alive and sufficiently well to receive at least 3 months of treatment. This requirement may have inflated the duration-based associations. The HR of 14.954 is unusually large and should be interpreted as a model-based association that may have been influenced by selection and residual bias.

Within the TCM theoretical framework, CRC has been described in terms of underlying spleen deficiency together with excess pathogenic factors, including toxic heat. This provides a rationale for combining spleen-strengthening and detoxification approaches[14]. These traditional concepts guide syndrome differentiation but should not be interpreted as established biomedical mechanisms of CRC. In the present study, syndrome classifications were extracted from clinical records and used to guide individualized prescriptions.

The Jianpi Bushen Jiedu method used in this study combined spleen-strengthening, kidney-tonifying, and detoxification principles. Within this treatment rationale, spleen-strengthening and kidney-tonifying prescriptions were intended to address deficiency patterns, while detoxification and related modifications addressed excess patterns. Prescriptions were individualized according to the recorded syndrome and clinical status. A recent review describes a broad range of TCM interventions in CRC, including compound herbal formulas, supportive care, and approaches investigated for use alongside chemotherapy[15]. However, evidence across different formulations and experimental settings cannot establish the efficacy or mechanism of the specific regimen evaluated here. The present survival findings therefore remain observational associations requiring prospective confirmation.

The findings apply only to patients with mCRC and should not be extrapolated to early-stage CRC. Treatment goals, perioperative management, recurrence risk, and opportunities for TCM integration differ by stage. Evidence for early-stage disease requires separate stage-specific cohorts and outcomes such as recurrence-free survival. Within mCRC, associations varied across syndrome subgroups. These exploratory findings may inform future syndrome-stratified studies but cannot independently guide treatment selection.

Integrated Chinese-Western medicine should be considered a complementary component of comprehensive mCRC care rather than a replacement for guideline-directed systemic therapy. Individualized use should be coordinated with oncologic treatment planning and should consider performance status, treatment line, syndrome differentiation, potential herb-drug interactions, and adherence.

TTE provides a structured approach for specifying eligibility criteria, treatment strategies, time zero, follow-up, outcomes, contrasts, and analyses before observational data are evaluated[8]. In this study, the framework improved the transparency of the clinical question and the adjustment for measured covariates. However, use of the TTE label alone does not ensure valid causal inference. Validity depends on the alignment of time zero, treatment assignment, exposure definition, and follow-up. Because use for ≥ 3 months was determined after T₀, the analysis remains susceptible to immortal time bias unless the findings are confirmed using landmark or time-varying methods.

Using the first prescription as T₀ provided a common start for follow-up and avoided assigning exposure before treatment began. However, it did not remove the survival period needed to accumulate 3 months of therapy. Therefore, a prespecified 3-month landmark analysis and a time-varying exposure model are needed. In the landmark analysis, survival should be measured from the landmark among eligible survivors. These analyses would determine whether the observed association persists after this design issue is addressed.

The study evaluated OS and CSS, reported measured-covariate balance, and performed interaction and leave-one-out analyses. These procedures improved descriptive transparency but did not replace analyses of time-dependent exposure. The E-value quantifies sensitivity to unmeasured confounding under specific assumptions. It does not address model misspecification, informative treatment discontinuation, or immortal time bias.

Several limitations should be emphasized. First, IPW balanced only measured baseline variables. Tumor burden, detailed molecular characteristics, changes in performance status and treatment tolerance, socioeconomic factors, and details of concurrent systemic treatment may have been unmeasured or time varying. Second, the duration-based exposure definition may have introduced immortal time bias. Third, adherence and prescription modifications were not fully quantified. Fourth, although data came from three hospitals, 525 of 627 patients were treated at Xiyuan Hospital. This concentration limits generalizability to other regions, practice settings, ethnic groups, and patients with early-stage CRC. Finally, the large HR may be unstable and should be tested using alternative model specifications. Prospective multicenter studies should use standardized data collection and time-dependent analyses.

CONCLUSION

In this multicenter real-world cohort, oral Chinese herbal medicine use for ≥ 3 months was associated with longer OS and CSS in patients with mCRC. These hypothesis-generating results support a possible role for integrated Chinese-Western medicine in comprehensive care. However, they do not permit a causal conclusion because exposure was defined by treatment duration and residual confounding remains. Prospective studies with external validation, landmark analyses, and time-varying exposure models are required.

ACKNOWLEDGEMENTS

We thank the patients and the clinical staff at Xiyuan Hospital, Peking University Cancer Hospital, and Dongzhimen Hospital for their contributions to this study, and colleagues who assisted with data collection, statistical analysis, and figure preparation.