Real-world uptake and effectiveness of triplet versus doublet chemotherapy with bevacizumab in patients with metastatic colorectal cancer in the United States

Abstract

Introduction

Triplet chemotherapy + bevacizumab (TripletBev) demonstrated an overall survival (OS) benefit for patients with newly diagnosed metastatic colorectal cancer in randomized trials. We aimed to evaluate the uptake and estimate the effectiveness of TripletBev versus doublet chemotherapy + bevacizumab (DoubletBev) in a real-world population in the United States.

Methods

We carried out a retrospective cohort study of patients initiating first-line treatment of metastatic colorectal cancer between 23 October 2014 and 31 October 2022 using the Flatiron Health nationwide electronic health record (EHR)-derived de-identified database. The data originated from ∼280 cancer clinics (∼800 sites of care) in the USA. The primary analysis compared OS between patients receiving TripletBev or DoubletBev using a Cox proportional hazards model with adjustment for pre-specified covariates using stabilized inverse probability of treatment weighting. This analysis was also carried out within pre-specified and post hoc subgroups. A secondary analysis used Stürmer-trimming of the propensity score distribution to include patients most likely to be eligible to receive either treatment.

Results

Some 391 patients received TripletBev and 9625 received DoubletBev. There was no association between TripletBev and OS in the primary analysis [hazard ratio (HR) 0.92; 95% confidence interval (CI) 0.75-1.13]. TripletBev was associated with lower hazard of death in patients with synchronous metastases (HR 0.79; 95% CI 0.64-0.98; statistically significant) and in the secondary analysis of patients most likely to be eligible to receive either treatment (HR 0.80; 95% CI 0.63-1.02; non-statistically significant).

Conclusion

Uptake of TripletBev remains low and the primary analysis did not demonstrate effectiveness in a real-world population in the United States. Patients with synchronous metastases and those most likely to be eligible to receive either treatment may be most likely to benefit from TripletBev.

Highlights

• •Triplet chemotherapy + bevacizumab showed improved survival in trials for colorectal cancer.
• •Uptake of this regimen is low in United States patients with metastatic colorectal cancer.
• •Regimen did not show effectiveness versus doublet chemotherapy + bevacizumab in United States patients.
• •Those most likely to be eligible and those with synchronous metastases may benefit.
• •Supports United States guidelines to strongly consider regimen for fit patients.

Introduction

Colon and rectal cancer are expected to account for >150 000 new cancer cases and >50 000 deaths in 2023 and is the second leading cause of cancer death.¹ Amongst patients with metastatic colorectal cancer (mCRC), 5-year overall survival (OS) remains <20%.² For patients with newly diagnosed mCRC, first-line systemic therapy has included doublet chemotherapy (a fluoropyrimidine plus either oxaliplatin or irinotecan) plus a targeted agent such as anti-vascular endothelial growth factor (VEGF) or anti-epidermal growth factor receptor (EGFR) treatment.³ The choice of targeted agent is based on the status of KRAS, NRAS, and BRAF in the tumor and primary tumor sidedness. Current guidelines recommend chemotherapy plus anti-EGFR therapy for patients with left-sided and KRAS, NRAS, and BRAF (RAS/RAF) wild-type disease and chemotherapy plus anti-VEGF for all others.^3,4

In patients in whom bevacizumab (anti-VEGF therapy) is chosen, several randomized controlled trials demonstrated that triplet chemotherapy plus bevacizumab (TripletBev) demonstrated improved progression-free survival (PFS) and OS compared with doublet chemotherapy plus bevacizumab (DoubletBev) in patients with newly diagnosed mCRC, though at the expense of increased rates of neutropenia and diarrhea.^5,6 A subsequent patient-level meta-analysis confirmed the finding that TripletBev was associated with improved OS compared with DoubletBev [hazard ratio (HR) 0.81].⁷ Clinical guidelines recommend first-line TripletBev for the subset of patients with excellent performance status who can withstand the higher toxicity of the triplet regimen, though uptake in the USA has remained low.^3,8

To our knowledge, no studies to date have compared the effectiveness of TripletBev with DoubletBev in a real-world population of patients in the USA. The effectiveness in a real-world setting may differ substantially from that observed in clinical trials which were enriched for younger patients (median age 60 years) and patients with excellent performance status [85%-90% Eastern Cooperative Oncology Group (ECOG) 0].⁹ We aimed to: (i) investigate the change in the proportion of patients receiving TripletBev over time; (ii) compare the effectiveness of TripletBev versus DoubletBev in terms of OS in patients with mCRC; and (iii) evaluate this comparison in pre-specified subgroups and amongst patients most likely to be eligible to receive either treatment defined using Stürmer trimming of the propensity score distribution.

Materials and methods

Study design, data source, and participants

We carried out a retrospective cohort study of patients with a diagnosis of mCRC who started first-line systemic treatment with DoubletBev or TripletBev between 23 October 2014 and 31 October 2022. The study was deemed exempt by the institutional review board (IRB) at the University of Pennsylvania (IRB 852189; 30 September 2022). This study used the Flatiron Health nationwide electronic health record (EHR)-derived de-identified database. This database is a longitudinal database, comprising de-identified patient-level structured and unstructured data, curated via technology-enabled abstraction.^10,11 During the study period, the de-identified data originated from ∼280 cancer clinics (∼800 sites of care). Patients were excluded for an absence of all structured data within 90 days following diagnosis of mCRC due to the potential for misclassified line of therapy data (oncologist-defined, rule-based) due to treatment being received elsewhere.

Exposure, outcome, and covariates

The exposure was first-line systemic treatment with TripletBev or DoubletBev during the study period. Triplet chemotherapy was defined as 5FU, irinotecan, and oxaliplatin administered within 28 days of initiation of systemic therapy for metastatic disease (index date).¹² Doublet chemotherapy regimens were defined as capecitabine and oxaliplatin or 5-fluorouracil (5-FU) plus either irinotecan or oxaliplatin administered within 28 days of the index date.¹² In each treatment arm, patients could be included if bevacizumab was added to doublet or triplet chemotherapy within 2 months after the index date. A landmark time analysis at 2 months after the index date was carried out as a sensitivity analysis to ensure this did not induce immortal time bias. The outcome of interest was OS, defined as the time from first systemic treatment to death, with censoring for last confirmed activity including in-person visit or confirmation of treatment administration.

Covariates of interest were pre-specified based on clinical knowledge of likelihood to confound the relationship between treatment choice and OS. Pre-specified covariates included: age, sex, race/ethnicity, academic or community practice, year of metastatic diagnosis, insurance status, baseline carcinoembryonic antigen (CEA), RAS/RAF alteration status, mismatch repair (MMR)/microsatellite instability (MSI) status, tumor sidedness (Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2024.100087), synchronous or metachronous metastatic disease, ECOG performance status, renal dysfunction, and liver function as assessed by albumin–bilirubin (ALBI) grade. Detailed definitions of covariates are included in the Supplementary material, available at https://doi.org/10.1016/j.esmogo.2024.100087.

Statistical analysis

All statistical analyses were carried out in Stata (Release 16.1; College Station, TX). All statistical testing was two-sided with an α of 0.05. Summary statistics were calculated for baseline covariates in the overall cohort and individually within the DoubletBev and TripletBev treatment groups. The proportion of patients receiving first-line TripletBev compared with all patients receiving first-line doublet or triplet chemotherapy with or without targeted therapy (anti-VEGF or anti-EGFR) was calculated by year of metastatic diagnosis from 2014 to 2022. The denominator of this analysis excluded patients with mCRC who did not receive systemic treatment or received only single-agent chemotherapy, with or without an anti-VEGF or anti-EGFR agent. A non-parametric test for trend was carried out across years of metastatic diagnosis for the proportion of patients receiving TripletBev.¹³ OS was assessed by treatment group in the univariable analysis using the Kaplan–Meier method with log-rank testing.

Missing values from pre-specified covariates were assumed to be missing at random (MAR) and multiple imputation with chained equations with 25 imputations was used to minimize bias related to missing data. For the primary analysis, a propensity score was generated using a logistic regression model with the outcome being TripletBev versus DoubletBev (binary) and the aforementioned, pre-specified set of confounding variables included as the predictors in the model. The propensity score from this model was used to generate stabilized inverse probability of treatment weighting (IPTW) which was used to assess standardized differences in means for continuous variables and standardized differences in proportions for each level of binary, categorical, or ordinal variables using pbalchk in Stata.14, 15, 16 Variables that were persistently imbalanced despite stabilized IPTW, as indicated by a standardized difference of <−0.10 or >0.10, were subsequently included in the Cox proportional hazards outcome model to adjust for this residual imbalance. Cox proportional hazards modeling with OS as the outcome was carried out with stabilized IPTW to estimate the relationship between treatment choice and OS. This modeling was carried out in each of the 25 imputed datasets with the treatment effect estimates combined to obtain the overall treatment effect estimate using Rubin’s rules.¹⁷ The proportional hazards assumption of this Cox proportional hazards model was assessed using Schoenfeld residuals. Predicted median survival and predicted 3-year OS by treatment group were generated from the Cox proportional hazards model with stabilized IPTW using the bsurvci command in Stata with 95% confidence intervals (CIs) generated using bootstrap resampling with 1000 resampling replications, and results were combined using Rubin’s rules over 25 imputed datasets.^17,18

A secondary analysis using Stürmer trimming of the propensity score distribution was carried out to ensure the positivity assumption (the assumption that all patients had the opportunity to receive either treatment) was met and to explore the impact of excluding the patients with treatment strongly contrary to prediction to minimize bias from certain types of unmeasured confounding.^19,20 These included situations where based on baseline characteristics: (i) patients who are expected to receive the treatment of interest do not receive it due to frailty, and (ii) situations where patients who are not expected to receive the treatment of interest receive it as a ‘treatment of last-resort,’ perhaps in the setting of clinically aggressive disease that is not well captured by measured variables. This was carried out by trimming both the TripletBev and DoubletBev observations below the 5th percentile of observed propensity scores in the TripletBev group and above the 95th percentile of observed propensity scores in the DoubletBev group.²⁰ Cox proportional hazards modeling was then carried out as described for the primary analysis.

Pre-specified subgroups that were examined in the same population as the primary analysis included age ≥65 years, age <65 years, CEA > median value (assessed on non-missing values of CEA), CEA ≤ median, RAS/RAF altered, RAS/RAF wild-type, right-sided primary tumor, left-sided primary tumor, synchronous metastases, metachronous metastases, ECOG performance status 0-1, and ECOG performance status ≥2. Subgroups of patients with and without a BRAF V600E alteration, patients with a right-sided primary or left-sided primary that was KRAS, NRAS, or BRAF V600E altered with similar exclusions to the CAIRO5 trial, patients with similar inclusion/exclusion criteria to the TRIBE trial, patients with similar inclusion/exclusion criteria to the TRIBE2 trial, and patients with complete information on KRAS, NRAS, and BRAF V600E were also examined in a post hoc fashion,^5,6,21 Supplementary material, available at https://doi.org/10.1016/j.esmogo.2024.100087. For each subgroup analysis, the subgroup of patients was identified in each imputed dataset following multiple imputation of missing data. The methods used in the primary analysis were then applied to the subgroup of interest as described above, meaning all subgroup analyses were adjusted for pre-specified covariates in the same way as the primary and secondary analyses.

Results

Patient and treatment characteristics

Of patients started on doublet or triplet chemotherapy with or without a targeted agent during the study period, 10 016 patients were eligible for inclusion in the primary analysis with 9625 receiving DoubletBev and 391 receiving TripletBev (Figure 1). In the overall cohort, median age was 63 years, 57% of patients were male, 81% of patients were treated in community oncology practices, 32% of patients had right-sided primary tumors, 66% had synchronous metastatic disease, 17% had RAS/RAF wild-type disease, and 70% had an ECOG performance status of 0-1 (Table 1). The greatest imbalance of baseline characteristics between treatment groups as measured by standardized differences in means or proportions occurred for age, commercial insurance, Medicare, academic versus community practice, left- versus right-sided primary tumor, synchronous versus metachronous disease, year of metastatic diagnosis, MMR/MSI status, RAS/RAF status, and ECOG performance status (Table 1).

Figure 1.

Patient flow diagram indicating number of patients excluded from the study with reasons for exclusion. Anti-EGFR, antibody targeting the epidermal growth factor receptor including panitumumab or cetuximab.

Table 1.

Baseline characteristics of the study population presented by treatment group before multivariable adjustment with the inverse probability of treatment weighting procedure.

	Total	DoubletBev	TripletBev	P value	Standardized differencea
	N = 10 016	n = 9625	n = 391
Age, median (IQR), years	63.0 (54.0-71.0)	63.0 (55.0-71.0)	52.0 (45.0-61.0)	<0.001	0.89
Sex, n (%)				0.91
Female	4266 (43)	4103 (43)	163 (42)		0.11
Male	5749 (57)	5521 (57)	228 (58)		0.11
Missing	1 (0)	1 (0)	0 (0)		0.02
Race/ethnicity, n (%)				0.32
Non-Hispanic White	5840 (58)	5599 (58)	241 (62)		0.07
Non-Hispanic Black	1151 (11)	1115 (12)	36 (9)		0.09
Hispanic or LatinX	895 (9)	857 (9)	38 (10)		0.06
Asian	291 (3)	276 (3)	15 (4)		0.11
Other race	885 (9)	855 (9)	30 (8)		0.13
Unknown	954 (10)	923 (10)	31 (8)		0.03
Insurance status, n (%)				<0.001
Commercial	5661 (57)	5371 (56)	290 (74)		0.37
Medicaid	487 (5)	471 (5)	16 (4)		0.06
Medicare	1205 (12)	1189 (12)	16 (4)		0.32
Other	421 (4)	404 (4)	17 (4)		0.03
No documented insurance	2242 (22)	2190 (23)	52 (13)		0.21
Practice type, n (%)				<0.001
Academic	853 (9)	749 (8)	104 (27)		0.41
Community	8144 (81)	7881 (82)	263 (67)		0.32
Missing	1019 (10)	995 (10)	24 (6)		0.07
Site of primary tumor, n (%)				0.12
Colon	7620 (76)	7339 (76)	281 (72)		0.13
Colorectal NOS	205 (2)	197 (2)	8 (2)		0.00
Rectum	2191 (22)	2089 (22)	102 (26)		0.13
Primary tumor sidedness, n (%)				0.004
Right-sided	3225 (32)	3126 (32)	99 (25)		0.17
Left-sided	5104 (51)	4874 (51)	230 (59)		0.19
Missing	1687 (17)	1625 (17)	62 (16)		0.04
Metastatic disease at diagnosis, n (%)				<0.001
Synchronous	6562 (66)	6236 (65)	326 (83)		0.34
Metachronous	3454 (34)	3389 (35)	65 (17)		0.34
Year of metastatic diagnosis, n (%)				<0.001
2013-2014	410 (4)	406 (4)	4 (1)		0.25
2015-2016	2585 (26)	2522 (26)	63 (16)		0.27
2017-2018	2643 (26)	2572 (27)	71 (18)		0.06
2019-2020	2554 (25)	2446 (25)	108 (28)		0.06
2021-2022	1824 (18)	1679 (17)	145 (37)		0.34
MMR/MSI, n (%)				<0.001
MMRp/MSS	6690 (67)	6361 (66)	329 (84)		0.42
MMRd/MSI-H	279 (3)	267 (3)	12 (3)		0.04
Missing	3047 (30)	2997 (31)	50 (13)		0.45
KRAS status, n (%)				<0.001
Mutation negative	3268 (33)	3084 (32)	184 (47)		0.35
Mutation positive	3618 (36)	3479 (36)	139 (36)		0.04
Missing	3130 (31)	3062 (32)	68 (17)		0.36
NRAS status, n (%)				<0.001
Mutation negative	4968 (50)	4695 (49)	273 (70)		0.39
Mutation positive	238 (2)	221 (2)	17 (4)		0.14
Missing	4810 (48)	4709 (49)	101 (26)		0.46
BRAF status, n (%)				<0.001
Mutation negative	4811 (48)	4559 (47)	252 (64)		0.31
Mutation positive	546 (5)	503 (5)	43 (11)		0.23
Missing	4659 (47)	4563 (47)	96 (25)		0.47
RASRAF, n (%)				<0.001
Mutation negative	1716 (17)	1598 (17)	118 (30)		0.30
Mutation positive	4329 (43)	4137 (43)	192 (49)		0.13
Missing	3971 (40)	3890 (40)	81 (21)		0.43
ECOG performance status, n (%)				0.003
ECOG 0-1	7044 (70)	6742 (70)	302 (77)		0.21
ECOG ≥2	812 (8)	795 (8)	17 (4)		0.16
Missing	2160 (22)	2088 (22)	72 (18)		0.13
Carcinoembryonic antigen (CEA), median (IQR), μg/l	28.5 (6.0-185.8)	28.0 (6.0-182.0)	45.7 (5.3-296.3)	0.062	0.02
Bilirubin, median (IQR), mg/dl	0.4 (0.3-0.6)	0.4 (0.3-0.6)	0.4 (0.3-0.6)	0.12	0.05
Albumin, median (IQR), G/dl	3.9 (3.5-4.2)	3.9 (3.5-4.2)	3.9 (3.6-4.2)	0.69	0.05
Creatinine, median (IQR), mg/dl	0.8 (0.7-1.0)	0.8 (0.7-1.0)	0.8 (0.7-0.9)	<0.001	0.10

DoubletBev, doublet chemotherapy plus bevacizumab; ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; MMR/MSI, mismatch repair/microsatellite instability; MMRd/MSI-H, mismatch repair deficient/microsatellite instability-high; MMRp/MSS, mismatch repair proficient/microsatellite stable; NOS, not otherwise specified; RAS/RAF status, KRAS, NRAS, and BRAF alteration status; TripletBev, triplet chemotherapy plus bevacizumab.

^aAbsolute value of the standardized difference in means or proportions. This is calculated as the difference in means or proportions between treatment groups divided by the pooled standard deviation.

Taking into account all patients with mCRC receiving doublet or triplet chemotherapy with or without targeted agents, the proportion of patients receiving TripletBev was 2.41% over the course of the study. This proportion significantly increased via the non-parametric test for trend from 2014 to 2022 (P < 0.001), though remained small as of 2022 (3.99%) (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2024.100087).

Univariable survival analysis

Amongst patients who were censored, median follow-up was 17.3 months (range: 0.03-95.4 months). In the overall cohort, median OS was 23.6 months (95% CI 23-24.1 months). Median OS in the TripletBev group was 30 months (95% CI 24.7-34.7 months; 3-year OS rate of 41.7%, 95% CI 34.8% to 48.5%) compared with 23.4 months (95% CI 22.8-24.1 months; 3-year OS rate of 31.8%, 95% CI 30.7% to 32.9%) in the DoubletBev group. In the univariable, unadjusted survival analysis, TripletBev was associated with a 28% reduction in hazard of death compared with DoubletBev (HR 0.72; 95% CI 0.62-0.85; P < 0.0001) (Figure 2).

Figure 2.

Unadjusted Kaplan–Meier analysis of overall survival (OS) in patients receiving triplet chemotherapy plus bevacizumab and those receiving doublet chemotherapy plus bevacizumab. DoubletBev, doublet chemotherapy plus bevacizumab; TripletBev, triplet chemotherapy plus bevacizumab.

IPTW-adjusted analyses

With stabilized-IPTW, covariate balance was found to be adequate on all variables (i.e. these variables were adjusted for by the stabilized-IPTW procedure) with the exception of several sub-categories of insurance status and age (Supplementary Figure S2, available at https://doi.org/10.1016/j.esmogo.2024.100087). Insurance status and age were incorporated into the Cox proportional hazards model with stabilized-IPTW to adjust for residual imbalance on these two variables. The proportional hazards assumption and linearity of continuous covariates assumptions were assessed and met for this model. In the stabilized-IPTW analysis, there was no significant difference in hazard of death between the TripletBev group and the DoubletBev group (HR 0.92; 95% CI 0.75-1.13) with predicted median OS [with age set to its mean and insurance status set to the modal value (commercial insurance)] of 24.6 months (95% CI 21.0-29.6 months) and predicted 3-year OS of 33.7% (95% CI 26.6% to 41.2%) in the TripletBev group and predicted median OS of 23.0 months (95% CI 22.3-23.8 months) and predicted 3 year OS of 30.7% (95% CI 29.3% to 32.0%) in the DoubletBev group (Figure 3). Subgroup analyses suggested that patients most likely to benefit from TripletBev were those with synchronous metastatic disease (HR 0.79; 95% CI 0.64-0.98), and this result was statistically significant (Figure 3).

Figure 3.

Forest plot examining the hazard ratio for overall survival for triplet chemotherapy plus bevacizumab compared with doublet chemotherapy plus bevacizumab in the primary analysis, secondary Stürmer trimming analysis, and subgroup analyses. CAIRO5, randomized trial with population of interest including patients with metastatic colorectal cancer with right-sided primary or left-sided primary with KRAS, NRAS, or BRAF V600E alterations (Supplementary material, available at https://doi.org/10.1016/j.esmogo.2024.100087); CEA, carcinoembryonic antigen; DoubletBev, doublet chemotherapy plus bevacizumab; CI, confidence interval; ECOG PS, Eastern Cooperative Oncology Group performance status; HR, hazard ratio; Mets, metastases; RASRAF, KRAS, NRAS, and BRAF status; TRIBE, randomized trial of FOLFIRI/bevacizumab and TripletBev in patients with metastatic colorectal cancer (Supplementary material, available at https://doi.org/10.1016/j.esmogo.2024.100087); TRIBE2, randomized trial of FOLFOX/bevacizumab and TripletBev in patients with metastatic colorectal cancer (Supplementary material, available at https://doi.org/10.1016/j.esmogo.2024.100087); TripletBev, triplet chemotherapy plus bevacizumab. ^aThe sample sizes in each treatment group varied slightly between the 25 imputed data sets for the Stürmer trimming analysis and several of the subgroup analyses. Reported sample size represents the average number of patients in the treatment group over the 25 imputed data sets. ^bThe subgroup of patients with ECOG performance status of ≥2 who received TripletBev was too small (n = 20.4) to adjust for imbalance in pre-specified confounders and the results of this model are not reported. ^cPost hoc subgroup analysis. ^dGiven the small size of the TripletBev group in this analysis, variables with residual imbalance were unable to be adjusted for including NRAS status, albumin-bilirubin grade, insurance status, race/ethnicity, sex, and year of metastatic diagnosis. The remainder of the pre-specified covariates were adjusted for.

The Stürmer trimming analysis (IPTW-adjusted) found that TripletBev was associated with a 20% reduction in the hazard of death compared with DoubletBev in this population, though this result was not statistically significant (HR 0.80; 95% CI 0.63-1.02; P = 0.07) (Figure 3). Predicted median OS (based on the estimated baseline survival function from the Cox proportional hazards model) was 30.1 months (95% CI 24.5-37.0 months) versus 24.6 months (95% CI 23.8-25.4 months) and predicted 3-year OS was 42.4% (95% CI 34.0% to 51.2%) and 34.1% (95% CI 32.6% to 35.6%) for the TripletBev group and DoubletBev groups, respectively.

Sensitivity analyses

The result of the landmark time analysis (IPTW-adjusted) to assess for the possibility of immortal time bias was consistent with the primary analysis (HR 0.91; 95% CI 0.73-1.12). The sensitivity analysis including only patients with complete information on KRAS, NRAS, and BRAF V600E status (IPTW-adjusted) was consistent with the primary analysis (HR 0.93; 95% CI 0.73-1.17).

Discussion

In a real-world population of patients in the United States with newly diagnosed mCRC treated with TripletBev or DoubletBev, we found that TripletBev was associated with better OS in unadjusted analysis, but there was no statistically significant benefit in OS after adjusting for pre-specified confounding variables with stabilized IPTW analysis.

The result from the primary analysis does not rule out that a more narrowly targeted population of patients may derive benefit from TripletBev compared with DoubletBev. The pre-specified exploratory subgroup analyses and the St $ü$ rmer trimming secondary analysis explored this possibility in different ways. The exploratory subgroup analyses carried out the primary analysis within subgroups defined by pre-specified variables. While the subgroup analyses were prespecified (with the exception of several post hoc analyses) and limited, they must be interpreted with caution as no adjustment for multiple testing was carried out.

In the subgroup of patients with synchronous metastases, TripletBev was associated with a significantly lower hazard of death than DoubletBev (Figure 3). In contrast, there was a non-significant trend towards increased hazard of death in patients with metachronous metastatic disease with TripletBev. It is likely that some proportion of patients with metachronous metastases had received prior adjuvant chemotherapy containing oxaliplatin and would be unlikely to benefit, or would be frankly ineligible on the basis of progression or toxicity with oxaliplatin, from receiving TripletBev. The TRIBE2 trial explicitly excluded patients with metachronous metastases who had received prior oxaliplatin-containing adjuvant chemotherapy or recent fluoropyrimidine monotherapy as adjuvant chemotherapy and, thus, had a low proportion of patients with metachronous metastases on the trial (11%).⁶ As we do not have access to information on treatments that patients received in the adjuvant setting, we did not have the ability to directly exclude these patients from our primary analysis in which 34% of patients had metachronous metastases. Thus, there is a possibility that the high proportion of patients with metachronous metastatic disease in the DoubletBev group with a relatively favorable prognosis²² may have biased the results of the primary analysis despite our attempts to attain balance on this variable. The main conclusion from the pre-specified subgroup analyses is that patients with synchronous metastases are the most likely to benefit from TripletBev.

The Stürmer trimming analysis aimed to ensure the positivity assumption was met and to reduce bias due to unmeasured confounding due to ‘frailty’ or ‘treatment of last resort’ by trimming the tails of the propensity score distribution where patients are treated most contrary to their treatment prediction.^19,20 Simulation studies have confirmed that Stürmer trimming consistently reduces bias due to modeled unmeasured confounding at the tails of the propensity score distribution.²⁰ Thus, the Stürmer trimming analysis can be interpreted as including a narrower population of patients most likely to be eligible to receive either TripletBev or DoubletBev based on having a high degree of overlap of clinically important baseline characteristics. This is the population that is of greatest interest to clinicians making recommendations, i.e. the patients for whom either treatment option would be reasonable. In addition to reduced generalizability, one drawback of the Stürmer trimming approach is that the analysis population is difficult to prospectively define. In other words, it is difficult to determine whether an individual patient in a clinical setting would have been included in this Stürmer trimming population. We can get a sense of this analysis population, however, by examining baseline covariates of patients who were included (61% of DoubletBev group; 64% of TripletBev group), and excluded, in the Stürmer trimming analysis (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2024.100087). Following the Stürmer trimming procedure, baseline covariates in this narrower population were remarkably well balanced in both the unadjusted and the IPTW-adjusted analyses, suggesting that Stürmer trimming did an excellent job of identifying patients with similar baseline prognostic features (Supplementary Figure S3, available at https://doi.org/10.1016/j.esmogo.2024.100087). In the Cox proportional hazards model with IPTW adjustment following Stürmer trimming, the hazard ratio for OS and the predicted median OS in each group were remarkably similar to those observed in the patient-level meta-analysis of clinical trials (HR 0.80 versus 0.81; median OS in TripletBev group 30.1 versus 28.9 months; median OS in DoubletBev group 24.6 versus 24.5 months), but the result in our study was not statistically significant.⁷ The interpretation of this analysis is that amongst the population of patients with mCRC treated in a real-world setting in the USA, those most likely to be eligible to receive either regimen may obtain benefit from TripletBev.

One important question to address is how these results should impact clinical practice in the USA going forward, especially with our observation that patients receiving TripletBev currently represent a small proportion of patients undergoing systemic therapy for mCRC in the USA. Notably, the Stürmer trimming population from the secondary analysis included ∼61% of the patients receiving either DoubletBev or TripletBev from the primary analysis. The pre-specified subgroup analysis of patients with synchronous metastatic disease included ∼66% of patients receiving DoubletBev or TripletBev from the primary analysis. In the primary analysis population of patients receiving chemotherapy plus bevacizumab, however, only 4% received TripletBev. This suggests that there is a significant gap between the population of patients receiving TripletBev in clinical practice and those who make up populations who could benefit from TripletBev, such as the Stürmer trimming population and the subgroup of patients with synchronous metastatic disease. These results provide real-world support for clinical guidelines in the USA noting that triplet chemotherapy should be strongly considered for patients with excellent performance status.³

Strengths of the study include that the study population is a national, real-world cohort in the USA incorporating both academic and community settings. To our knowledge, this is the first large study to investigate real-world effectiveness of TripletBev in the United States. This large population along with access to clinically important covariates allowed for a robust adjustment on pre-specified clinical confounders. There are several important limitations to our study. These include that we cannot completely rule out residual, unmeasured confounding. We did, however, explore this possibility using Stürmer trimming which has been shown to reduce bias from unmeasured confounding due to frailty or ‘treatment of last resort’, as discussed above. Additionally, as a low percentage of patients received TripletBev in clinical practice, power to detect clinically meaningful differences in treatment groups is limited, particularly in the Stürmer trimming analysis. We did not have information on outcomes such as the proportion of patients who underwent curative intent surgical resections or PFS. The more distal, patient-important outcome of OS, however, provides important information to guide treatment choices. Additionally, there was significant missingness on clinically important variables including KRAS, NRAS, and BRAF V600E status which we addressed with multiple imputation. We also carried out a sensitivity analysis including only patients with complete information on KRAS, NRAS, and BRAF V600E which was consistent with the primary analysis. Future work may examine different dosing regimens of TripletBev in United States patients and their association with clinical outcomes.

Conclusions

Uptake of TripletBev remains low in the USA. While TripletBev did not demonstrate a benefit in our IPTW-adjusted primary analysis in a real-world population of patients in the United States with newly diagnosed mCRC, we found that patients with synchronous metastases and those most likely to be eligible to receive either treatment may be the most likely to benefit from TripletBev. These results support national guidelines noting that triplet chemotherapy should be strongly considered for patients with excellent performance status.