ABSTRACT
Previous studies have demonstrated that 10%–25% of patients receive gene‐matched therapy (GMT) after comprehensive genomic profiling (CGP). However, its real‐world clinical effects remain unclear. This study assessed the feasibility of integrating the Center for Cancer Genomics and Advanced Therapeutics (C‐CAT) repository that documented genomic and clinical data and the quality indicator (QI) dataset that included cancer‐specific data and administered treatment as a model case of real‐world data study of cancer genomic medicine in Japan. We successfully integrated these two datasets and included 1162 patients diagnosed with solid tumors at the National Cancer Center Hospital between 2019 and 2021 who underwent CGP testing. Of these, 432 (37.2%) had druggable mutations, 96 (8.3%) received GMT, and 218 (18.8%) received non‐GMT. Among 314 patients who initiated either GMT or non‐GMT after CGP, the median 2‐year overall survival (OS) was 19.0 and 19.7 months for GMT and non‐GMT, respectively (hazard ratio: 0.87, 95% confidence interval: 0.56–1.35, p = 0.53). Stratified analysis by prior treatment lines (0–1 vs. ≥ 2) demonstrated no significant differences in survival. Sensitivity analyses yielded consistent results. This study demonstrated that integrating the C‐CAT repository and QI datasets enables real‐world comparisons of GMT and non‐GMT outcomes. Unlike previous clinical trials reporting enhanced survival with GMT, our findings indicated no significant OS difference. Potential explanations include differences in cancer type, CGP timing, study population selection, and immortal time bias. Future multicenter studies would clarify the real‐world utility of the CGP and GMT.
Keywords: cancer genomic testing, gene‐matched therapy, non‐gene‐matched therapy, precision medicine, real‐world evidence
We successfully integrated the C‐CAT repository and QI datasets to compare real‐world outcomes of gene‐matched therapy and non‐gene‐matched therapy, showing no significant difference in 2‐year overall survival between the groups. Potential reasons for this discrepancy from previous trials include differences in cancer types, comprehensive genomic profiling timing, study population selection, and immortal time bias.
Abbreviations
- ADL
activities of daily living
- C‐CAT
the Center for Cancer Genomics and Advanced Therapeutics
- CGP
comprehensive genome profiling
- CI
confidence interval
- DPC
Diagnosis Procedure Combination
- EMR
electronic medical record
- GMT
gene‐matched therapy
- HR
hazard ratio
- ICC
intraclass correlation coefficient
- ICD
the International Classification of Diseases
- IPW
inverse probability weighting
- ITT
Intention‐to‐treat
- NCCH
the National Cancer Center Hospital
- OS
overall survival
- QI
quality indicator
1. Introduction
Cancer genomic testing has been implemented as an essential method for identifying genetic mutations and guiding the selection of appropriate therapies, including molecularly targeted agents. Precision medicine aims to enhance prognosis by selecting the most effective treatment while minimizing off‐target effects and enhancing the patient's quality of life. Various international initiatives, such as AACR‐GENIE, NCI‐MATCH, MOSCATO‐01, SCRUM‐Japan, TOP‐GEAR, and K‐MASTER, have been established to advance precision oncology [1].
The detection rate of druggable mutations ranges from 40%–90%, whereas the actual drug delivery rate is approximately 10%–25% [2, 3, 4]. To accurately assess the clinical effectiveness of gene‐matched therapy (GMT), it is crucial to compare patients receiving GMT with those receiving non‐GMT treatments, such as cytotoxic chemotherapy, rather than those ineligible for treatment because of disease progression or poor general condition. However, numerous genomic initiatives lack comprehensive clinical data on post‐genomic testing treatments and outcomes, making robust assessments challenging [1].
In Japan, comprehensive genomic profiling (CGP) tests were covered by the national health insurance system since September 2019, enabling all patients with advanced solid tumors lacking standard treatment options to undergo CGP testing. Additionally, the Center for Cancer Genomics and Advanced Therapeutics (C‐CAT) was established to collect genomic and clinical data as part of insured medical care, with patient consent required for data registration. The reported consent rate was 99.7% [5]. Although clinical data on patients who received GMT are well documented, the extent of data collection, including drug regimens, for patients who did not receive GMT remains unclear.
The National Cancer Center Japan has been conducting a quality indicator (QI) study, a nationwide project to enhance cancer care quality [6]. This study collected hospital‐based cancer registry data from over 800 institutions across Japan [7]. Additionally, for institutions that provided consent, Diagnosis Procedure Combination (DPC) data, including drug administration details similar to fee‐for‐service billing data, were collected. The DPC data systematically document and verify the drug therapies administered to patients.
By integrating the C‐CAT repository with QI data, this study aimed to enhance clinical data on patients who did not receive GMT and compare survival outcomes between the GMT and non‐GMT groups. The feasibility of integrating the C‐CAT and QI databases was assessed, and the overall survival (OS) between the GMT and non‐GMT groups was compared to assess the clinical significance of GMT in a real‐world setting.
2. Materials and Methods
2.1. Patient Selection and Data Sources
This study included patients with solid tumors newly diagnosed at the National Cancer Center Hospital (NCCH) between January 1, 2019, and December 31, 2021, who underwent CGP testing by December 31, 2023. Data were obtained from the QI dataset and C‐CAT repository, exclusively including cases from NCCH.
The QI dataset includes hospital‐based cancer registry and DPC data. The hospital‐based cancer registry is a legally mandated nationwide database that registers all newly diagnosed cancer cases annually [7]. It includes patient demographic data (sex and age), cancer‐related details (date of diagnosis, tumor location, and histology based on the International Classification of Diseases [ICD]‐O‐3), and clinical stage at diagnosis. DPC data—similar to claims‐based datasets—document medical procedures, surgeries, chemotherapy, radiation, and other examinations performed during inpatient and outpatient care. These records span from October of the year before diagnosis to March of the second year after diagnosis (for patients diagnosed in 2021, data were collected from October 2020 to March 2024). Additionally, DPC data include comorbidities coded using ICD‐10 and the Barthel Index as a measure of activities of daily living (ADL) [8]. Additionally, the QI dataset updated the survival status and last confirmed survival date manually from electronic medical records as of February 2024.
The C‐CAT repository registers genomic and clinical data from patients who undergo CGP testing. Additionally, it documents age at diagnosis, cancer type, treatment before and after CGP testing, and survival status [5]. We used clinical data from the C‐CAT repository to determine whether the recommended GMT, based on detected genetic mutations, was administered. Owing to limited data on non‐GMT, treatment data were supplemented using the QI dataset.
Patients who underwent CGP testing at NCCH between 2019 and 2023 were identified from the C‐CAT repository, and only those newly diagnosed between 2019 and 2021 were included. Research‐specific serial numbers were assigned for analysis, and C‐CAT data were linked with the QI dataset. Patients under 18 years of age and those who underwent CGP testing as part of a clinical trial were excluded.
2.2. Definitions
Age at diagnosis was obtained from both the QI dataset and C‐CAT repository. Cancer type was classified based on ICD‐O‐3.2 in the QI dataset and Onco‐Tree classification in the C‐CAT repository (http://oncotree.mskcc.org/#/home). For database comparison, we compared the cancer types classified using OncoTree in the C‐CAT repository with those classified using ICD‐O in the QI dataset. Novel drug therapy initiation after CGP testing was extracted from both the QI dataset and C‐CAT repository. GMT was defined as the administration of expert panel‐recommended novel therapy based on detected genetic mutations. Non‐GMT was defined as any newly initiated therapy after CGP testing that was not based on genetic mutations. The Druggable mutations were defined as a gene alteration for which a specific therapy was recommended by the expert panel, as recorded in the C‐CAT repository. OS was defined as the time from novel drug therapy initiation after CGP testing to either death within 2 years or the last confirmed survival date.
2.3. Outcomes
To assess the data consistency, we compared common variables, including age, sex, and cancer type, between the QI dataset and C‐CAT repository.
To assess the prognostic effect of GMT and non‐GMT, we analyzed survival outcomes in patients who received either therapy after CGP testing.
2.4. Statistical Analysis
To assess data consistency between datasets, we used a Bland–Altman plot for age discrepancies and calculated the intraclass correlation coefficient (ICC) for database consistency. Sex and cancer type agreement was assessed using concordance rates and Cohen's kappa coefficient.
To compare the prognosis of GMT and non‐GMT, we used a target trial emulation approach [9, 10]. We set a hypothetical target trial based on previous clinical research [11]. The target trial was designed for adult patients with unresectable advanced solid tumors who underwent CGP testing. The treatment strategy was defined as GMT if a druggable mutation was present and non‐GMT if no druggable mutation was detected or if no evidence‐based molecularly targeted therapy was available. Follow‐up began at the start of a novel treatment after CGP testing, and OS was the primary outcome. Intention‐to‐treat (ITT) effects were measured.
For target trial emulation, the study population included patients aged ≥ 18 years who underwent CGP testing and subsequently initiated a novel systemic therapy. The treatment strategy was defined as whether the patient received GMT or non‐GMT. In Japan, after CGP testing, C‐CAT provides evidence‐based data on actionable mutations, and the expert panel discussion is conducted at selected facilities. Patients may receive GMT through clinical trials, off‐label use, or approved therapies for their cancer type with detected genetic mutations, whereas non‐GMT is initiated when no actionable mutation is identified. Therefore, in real‐world clinical practice, treatment allocation is determined based on the presence or absence of actionable mutations, similar to the target trial. The follow‐up period and outcome assessment mirrored the target trial, and the ITT effect was assessed.
To adjust for baseline patient characteristics, we used inverse probability weighting (IPW). A logistic regression model was applied to estimate the probability of receiving GMT based on age, sex, cancer type, comorbidities, ADL (Barthel Index: 100 vs. ≤ 95), and the number of prior treatment regimens before CGP testing. Stabilized inverse probability weights were calculated from the estimated probabilities. Group comparisons were performed using the U‐test and chi‐square test for continuous and categorical variables, respectively. Standardized mean differences (SMDs) were used to assess balance before and after IPW adjustment. Kaplan–Meier curves and weighted Cox proportional hazards models were used to compare the OS between GMT and non‐GMT.
2.5. Sensitivity and Subgroup Analyses
For sensitivity analyses, we performed a multivariate Cox regression analysis without IPW adjustment. Additionally, the analysis was repeated using the full dataset, including survival data available up to February 2024, beyond the initial 2‐year OS window.
Because the number of prior regimens affects GMT outcomes, for subgroup analyses, we conducted multivariate Cox regression analyses stratified by patients with 0–1 vs. ≥ 2 prior treatment regimens [12]. Additionally, to directly examine whether the association between GMT and OS differed by cancer type, we performed multivariate Cox regression analyses stratified into three groups: gastrointestinal and hepatobiliary cancers, gynecological cancers, and other cancers. Also, we conducted stratified multivariable Cox regression analyses by cancer rarity (rare vs. non‐rare cancers), based on predefined classifications [13].
All statistical analyses were two‐sided and conducted using Stata v15.0 software (StataCorp LP, College Station, TX, USA). GraphPad Prism 9 software (GraphPad Software Inc., San Diego, CA, USA) was used for graph creation. Statistical significance was set at p < 0.05.
2.6. Study Approval and Data Availability
This observational study was approved by the Institutional Review Board of the National Cancer Center, Japan (approval number 2022‐139) and adhered to the principles of the Declaration of Helsinki. This study used an opt‐out approach, including data only from patients who did not refuse data usage, following Japanese personal data protection laws and ethical guidelines. Data were used with permission granted under privacy protection laws and a data use agreement with the hospitals. The data are not publicly available owing to privacy and ethical restrictions.
3. Results
3.1. Patient Selection
Among the 2461 patients who underwent CGP at NCCH between 2019 and 2023, 1365 were newly diagnosed between 2019 and 2021. The QI dataset and C‐CAT repository were successfully linked for all cases. After excluding patients who underwent CGP testing as part of a clinical trial, 1162 patients were included in the final analysis (Figure 1). Among them, 432 patients (37.2%) had druggable mutations, and 96 patients (8.3%) received GMT (Table 1). Of the 96 patients with GMT, 57 were identified only in the C‐CAT repository, with all treatments administered either as clinical trials or outside the QI data collection period. The remaining 39 cases were detected by both datasets or only by the QI dataset. The drugs used in these cases, listed only in the C‐CAT repository and in some instances in the QI dataset as well, were confirmed via electronic medical records (EMRs) to have been prescribed. In contrast, 218 patients (18.8%) received non‐GMT (Table 1). Among them, 214 were identified only in the QI dataset, whereas four were documented in both the C‐CAT repository and QI dataset with identical treatment records. To evaluate the utility and consistency between the QI dataset and EMRs, we conducted a chart review of 16 patients in the GMT group and 53 patients in the non‐GMT group who were identified as having received molecular targeted therapies or immune checkpoint inhibitors based solely on the QI dataset. In all cases, the date of birth and sex matched between the QI dataset and the EMRs. Furthermore, all drugs listed in the QI dataset were confirmed to have been prescribed in the EMRs. Among 336 patients who did not receive GMT despite having druggable alterations, reasons were available for 328 cases. The most common reasons were clinical decision and/or patient preference (n = 221), followed by ineligibility based on inclusion/exclusion criteria (n = 60) and death or other reasons (n = 47; Table S1).
FIGURE 1.
Flowchart of patient selection. Among 2461 patients who underwent CGP at NCCH from 2019–2023, 1365 were newly diagnosed between 2019 and 2021. After excluding clinical trial cases, 1162 patients were included in the final analysis. The QI dataset and C‐CAT repository were successfully linked for all cases. C‐CAT, the Center for Cancer Genomics and Advanced Therapeutics; CGP, comprehensive genome profiling; NCCH, national cancer center hospital; QI, quality indicator.
TABLE 1.
Patient characteristics.
| Characteristics | Total (n = 1162) |
|---|---|
| Age in years, median (IQR; years) | 58 (48–68) |
| < 65, n (%) | 390 (66.4) |
| ≥ 65, n (%) | 772 (33.6) |
| Sex (female), n (%) | 567 (48.8) |
| Cancer type, n (%) | |
| Hepatobiliary | 252 (21.7) |
| Digestive | 190 (16.4) |
| Gynecologic | 128 (11.0) |
| Skin | 70 (6.0) |
| CNS | 65 (5.6) |
| Genitourinary | 64 (5.5) |
| Lung | 94 (8.1) |
| Bone and soft tissue | 106 (9.1) |
| Breast | 48 (4.1) |
| Other | 145 (12.5) |
| Barthel Index | |
| 100 | 965 (83.1) |
| ≦95 | 197 (16.9) |
| Comorbidity | 174 (15.0) |
| Types of sample | |
| Tissue | 1038 (89.3) |
| Liquid | 124 (10.7) |
| Year of diagnosis, n (%) | |
| 2019 | 360 (31.0) |
| 2020 | 397 (34.1) |
| 2021 | 405 (34.9) |
| Druggable mutation (positive), n (%) | 432 (37.2) |
| Gene‐matched therapy, n (%) | 96 (8.3) |
| Non gene‐matched therapy, n (%) | 218 (18.8) |
Abbreviation: IQR, interquartile range.
3.2. Data Concordance Between the QI Dataset and C‐CAT Repository
The age at diagnosis of the same patients was compared between the two databases (Figure S1), demonstrating an ICC of 0.99. Sex was perfectly concordant (100%) between the datasets (data not shown). The concordance rate for cancer type was 86.7%, with a Cohen's kappa coefficient of 0.85 (Table S2).
3.3. Comparison Between the GMT and Non‐GMT Groups
A total of 314 patients who initiated either GMT or non‐GMT after CGP testing were included in the survival analysis. The 2‐year survival follow‐up rate was 64.3%. Gynecologic, genitourinary, and lung cancers were more prevalent in the GMT group than that in the non‐GMT group (Table 2). In both groups, rare cancers accounted for over 50% (59.5% and 57.8% in the GMT and non‐GMT groups, respectively). The IPW‐adjusted baseline characteristics of the two groups are summarized in Table 2. After applying the IPW, the SMDs for all baseline variables were reduced, indicating that the imbalance between the groups was minimized.
TABLE 2.
Baseline characteristics of patients who underwent GMT or non‐GMT and the results of IPW.
| Characteristics | Total (n = 314) | GMT group (n = 96) | Non‐GMT group (n = 218) | p | SMD before IPW | SMD after IPW |
|---|---|---|---|---|---|---|
| Age in years, median (IQR; years) | 57 (48–67) | 56 (47–64) | 59 (48–68) | 0.21 | −0.23 | 0.02 |
| < 65, n (%) | 216 (68.8) | 73 (76.0) | 143 (65.6) | 0.07 | ||
| ≥ 65, n (%) | 98 (31.2) | 23 (24.0) | 75 (34.4) | |||
| Sex (female), n (%) | 147 (46.8) | 40 (41.7) | 107 (49.1) | 0.23 | −0.15 | −0.01 |
| Cancer type, n (%) | 0.002 | 0.05 | 0.0003 | |||
| Hepatobiliary | 61 (19.4) | 12 (12.5) | 49 (22.5) | |||
| Digestive | 44 (14.0) | 12 (12.5) | 32 (14.7) | |||
| Gynecologic | 38 (12.1) | 19 (19.8) | 19 (8.7) | |||
| Skin | 26 (8.3) | 11 (11.5) | 15 (6.9) | |||
| CNS | 25 (8.0) | 1–3 (2.1) | 23 (10.6) | |||
| Genitourinary | 24 (7.6) | 11 (11.5) | 13 (6.0) | |||
| Lung | 24 (7.6) | 10 (10.4) | 14 (6.4) | |||
| Bone and soft tissue | 15 (4.8) | 1–3 (2.1) | 13 (6.0) | |||
| Breast | 13 (4.1) | 4–7 (6.3) | 7–9 (3.2) | |||
| Other | 44 (14.0) | 11 (11.5) | 33 (15.1) | |||
| Barthel Index, n (%) | 0.50 | 0.08 | 0.008 | |||
| 100 | 268 (85.4) | 80 (83.3) | 188 (86.2) | |||
| ≦95 | 46 (14.7) | 16 (16.7) | 30 (13.4) | |||
| Comorbidity, n (%) | 65 (20.7) | 22 (22.9) | 43 (19.7) | 0.52 | 0.08 | 0.01 |
| Number of pre‐CGP regimens, n (%) | 0.13 | 0.26 | 0.0001 | |||
| 0 | 38 (12.1) | 7 (7.3) | 31 (14.2) | |||
| 1 | 141 (44.9) | 42 (43.8) | 99 (45.4) | |||
| 2 | 65 (20.7) | 19 (19.8) | 46 (21.1) | |||
| ≥ 3 | 70 (22.3) | 28 (29.2) | 42 (19.3) |
Note: Small numbers with fewer than 10 cases have been masked following the Japanese database regulations.
Abbreviations: CGP, comprehensive genome profiling test; CNS, central nervous system; GMT, gene matched therapy; IPW, inverse probability weighting; IQR, interquartile range; SMD, standardized mean differences.
Among the GMT group, 56.3% received insurance‐approved therapies, whereas 42.5% were treated in clinical trials (Figure 2). Pembrolizumab was used in 40% of patients who underwent insurance‐approved therapies, followed by olaparib (17.8%). In the non‐GMT group, the most frequently used agent was tegafur‐gimeracil‐oteracil (13%).
FIGURE 2.
Distribution of treatments in the GMT and non‐GMT groups (N = 314). In the GMT group, 56.3% of patients received insurance‐approved therapies, whereas 42.5% participated in clinical trials. In the non‐GMT group, the most commonly used agent is tegafur‐gimeracil‐oteracil (13%). 5‐FU, fluorouracil; GEM, gemcitabine; GMT, gene‐matched therapy; ICI, immune checkpoint inhibitor; S‐1, tegafur‐gimeracil‐oteracil.
The median 2‐year OS was 19 months in the GMT group and 19.7 months in the non‐GMT group. The hazard ratio (HR) was 0.87 (95% confidence interval [CI]: 0.56–1.35, p = 0.53; Figure 3). Stratification by the number of prior treatment lines before CGP testing (0–1 vs. ≥ 2) demonstrated no significant difference in OS between the GMT and non‐GMT groups (adjusted HR: 0.58, 95% CI: 0.30–1.13, p = 0.11 for 0–1 prior lines; adjusted HR: 1.47, 95% CI: 0.78–2.79, p = 0.24 for ≥ 2 prior lines; Figure S2). There was no statistically significant difference in OS between the GMT and non‐GMT groups in any of the cancer types: gastrointestinal and hepatobiliary cancer (adjusted HR: 1.55, 95% CI: 0.72–3.36, p = 0.27), gynecological cancer (adjusted HR: 1.18, 95% CI: 0.42–3.32, p = 0.75), or other cancer (adjusted HR: 0.77, 95% CI: 0.42–1.40, p = 0.39; Figure S3). In non‐rare cancers (N = 131), prognosis did not differ between the GMT and non‐GMT groups (HR 0.76, 0.37–1.54, p = 0.45). However, in rare cancers (N = 183), the GMT group had a significantly worse prognosis than the non‐GMT group (HR 1.70, 1.01–2.89, p = 0.04; Figure S4).
FIGURE 3.
2‐Year overall survival. Kaplan–Meier survival curve displays crude survival. The hazard ratio using IPW adjusts for age, sex, cancer type, comorbidity, ADL (Barthel Index: 100 vs. ≤ 95), and the number of prior treatment regimens before CGP testing is 0.87. ADL, activities of daily living; CGP, comprehensive genome profiling; IPW, inverse probability weighting.
3.4. Sensitivity Analysis
Multivariate Cox regression analysis without IPW adjustment demonstrated no significant differences between the two groups (adjusted HR: 0.97, 95% CI: 0.63–1.49, p = 0.88; Table S3).
Additionally, in the survival analysis without limiting OS to 2 years, univariable Cox regression with IPW demonstrated consistent results, with no significant difference in prognosis between the two groups (HR: 0.98, 95% CI: 0.62–1.54, p = 0.92).
4. Discussion
This study successfully demonstrated the feasibility of integrating the C‐CAT repository and QI dataset for patients treated at a single institution. The high concordance rates observed for age, sex, and cancer type indicated that these datasets can be effectively merged for analysis. Additionally, while the C‐CAT contains GMT‐related clinical data, data on non‐GMT are limited. This study demonstrated that DPC data collected in the QI dataset can supplement missing treatment data for non‐GMT, enabling a comparative analysis of survival outcomes between GMT and non‐GMT—previously a challenge. Our findings indicated that even after adjusting for age, sex, comorbidities, and overall health status, there was no significant difference in the 2‐year OS between the GMT and non‐GMT groups.
The integration of the C‐CAT repository and QI dataset was feasible, as evidenced by the high concordance rates for age and sex. This indicates that the two datasets were successfully merged for the same patients. However, the cancer type concordance rate was 86.7%. This discrepancy likely stems from differences in classification systems: the QI dataset documents cancer types based on ICD‐O‐3.2, whereas C‐CAT uses Onco‐Tree. Despite these differences, major tumor locations, such as hepatobiliary and lung cancers, were generally consistent. However, bone and soft tissue tumors with less clearly defined locations exhibited lower concordance rates. Additionally, in patients with multiple primary cancers diagnosed in different years, CGP testing may have been performed for an earlier diagnosis rather than the one under study. This may explain why the cancer types documented in the QI dataset differed from those in the C‐CAT repository.
One of the distinctive features of C‐CAT, compared to other genomic medicine databases, is its inclusion of clinical data following CGP testing [5]. However, this study revealed a lack of clinical data on non‐GMT in the C‐CAT repository. This limitation was likely because of the nature of C‐CAT as a genomic medicine database, where physicians primarily focus on registering GMT treatments and may not perceive the necessity of documenting non‐GMT treatments. This study demonstrated that DPC data effectively supplemented the missing data. Additionally, DPC data may capture the majority of GMT treatments conducted as part of standard insurance‐approved medical care. The predominant approved GMT agents identified in this study were pembrolizumab and poly (ADP‐ribose) polymerase inhibitors, consistent with previous studies, including the MCGI study and other research using the C‐CAT repository [5, 14]. These findings indicated that DPC data collection can support the documentation of post‐CGP therapy implementation in the C‐CAT repository.
With the successful integration of these datasets, we compared GMT and non‐GMT regarding their association with 2‐year survival; however, no significant difference was observed between the groups. This result remained consistent even after adjusting for age, sex, comorbidities, and overall health status. Numerous clinical trials have demonstrated a survival benefit for GMT that differs from our findings. This discrepancy can be explained by at least four factors.
First, the definition of Time Zero (Time0) for OS measurement differed among studies. In the GOZILA study [12], Time0 was defined as the date of enrollment, whereas in the MCGI study [14], it was defined as the date of CGP testing. Because there is typically a delay between CGP testing and therapy initiation, these definitions may introduce immortal time bias. To mitigate this, we set Time0 as the date of treatment initiation for either GMT or non‐GMT, ensuring that the start of the survival measurement coincided with the treatment of interest, according to the recommendations by Hernán et al. [15].
Second, differences in cancer types between our study and previous clinical trials may have affected our results. The TRIUMPH study [16] focused on human epidermal growth factor receptor 2‐amplified metastatic colorectal cancer, whereas SCRUM‐MONSTER [11] and GOZILA [12] included cancers with well‐established molecular targets, such as microsatellite instability‐high colorectal cancer and Erb‐B2 Receptor tyrosine kinase 2‐altered gastric cancer. In contrast, our study included patients diagnosed and treated at the NCCH, where nearly half of the cohort had rare cancers. The majority of these cancers had druggable mutations, but effective targeted therapies were still under development. This may explain why our study did not observe an association between GMT and OS. Notably, in SCRUM‐MONSTER, where patients had access to multiple clinical trials for druggable mutations, 5% of CGP‐tested patients were enrolled in clinical trials, whereas in our study, 3% of patients received investigational treatments. Although the proportion of patients enrolled in clinical trials was similar, OS enhancement was not observed in our cohort, indicating that differences in cancer type can be a contributing factor.
Third, differences in CGP testing timing and methodology may have affected survival outcomes. In clinical trials, CGP testing is often conducted early after diagnosis, and studies have demonstrated that patients who receive GMT as first‐line therapy tend to have better survival outcomes [12]. However, in Japan, CGP testing is only approved for patients who lack standard treatment options or are nearing the end of standard therapies. In our study, 43% (135/314) of the patients received at least two prior lines of treatment before CGP testing. This timing difference may partly explain the discrepancy between our findings and those of previous trials. Additionally, although liquid‐based CGP was approved in Japan in August 2021, 89.3% of patients in our cohort underwent tissue‐based CGP because they were diagnosed before this approval. Although the effect of CGP testing on subsequent GMT effectiveness remains unclear, certain studies have demonstrated better outcomes with liquid‐based CGP [17, 18, 19].
Fourth, selection bias may have affected the results. This study included only patients who underwent CGP testing and subsequently initiated novel treatment, potentially excluding those whose disease was very aggressive to start novel therapy. This difference in patient selection may partly explain the discrepancy in survival outcomes between our study and previous reports. Although this selection was necessary to ensure comparable patient backgrounds, it may have resulted in OS overestimation in both the GMT and non‐GMT groups. The median OS in our study was 19 months, whereas studies with comparable or longer follow‐up periods reported median OS values of 13.3–14.8 months in SCRUM‐MONSTER [11] and 9.9–18 months in SCRUM‐GOZILA [12]. We believe that assessing the prognostic impact of GMT specifically in patients who are eligible for treatment after CGP, as done in our study, provides a more clinically relevant comparison.
This study has several limitations. First, as a single‐center study, further confirmation is required to determine whether similar data can be integrated at other institutions. Although institutional characteristics may introduce selection bias, our findings regarding drug delivery and druggable mutation detection rates were consistent with previous studies [20, 21, 22, 23]. Because CGP testing eligibility is strictly regulated under Japan's insurance system, inter‐institutional variation in treatment availability may be less pronounced than that of other therapeutics [5]. Second, as an observational study, we cannot establish causality. However, by applying target trial emulation methods and adjusting for the primary factors affecting OS, we aimed to mitigate confounding bias in assessing the relationship between GMT, non‐GMT, and survival, in line with the framework [24, 25].
This study demonstrated the feasibility of integrating the C‐CAT repository with the QI dataset, enabling a real‐world comparison of GMT and non‐GMT outcomes. Because both the C‐CAT and QI studies collected data from multiple institutions across Japan, future efforts should expand data integration to a multicenter setting to better assess the real‐world effects of GMT on survival outcomes.
Associating the C‐CAT repository with the QI dataset enabled a real‐world comparison of survival outcomes between GMT and non‐GMT. This study demonstrated the potential of integrating insurance claims data with genomic practice data to assess GMT effectiveness. This finding should be further evaluated through multicenter studies in the future.
Author Contributions
Hiroyuki Mano: project administration, supervision, writing – review and editing. Takahiro Higashi: conceptualization, methodology, resources, supervision, writing – review and editing. Yusuke Okuma: funding acquisition, supervision, writing – review and editing. Takashi Kohno: conceptualization, methodology, project administration, resources, supervision, writing – review and editing. Takafumi Koyama: conceptualization, project administration, resources, writing – review and editing. Tatsuya Suzuki: conceptualization, project administration, writing – review and editing. Takashi Yugawa: conceptualization, project administration, resources, writing – review and editing. Taisuke Ishii: conceptualization, data curation, formal analysis, funding acquisition, methodology, project administration, resources, software, writing – original draft, writing – review and editing.
Ethics Statement
This study was approved by the Institutional Review Board of the National Cancer Center, Japan (approval number: 2023‐357) and complied with the Declaration of Helsinki.
Consent
This observational study used an opt‐out approach, using data only from patients who did not refuse data usage, following Japanese personal data protection laws and ethical guidelines.
Conflict of Interest
Takashi Kohno has received research funding from Chugai Pharmaceutical, Sysmex, Konica Minolta, Guardant Health, Eli Lilly Japan, and Eurofins Clinical Genetics. Takashi Kohno is the Editor of Cancer Science. Takafumi Koyama is a consultant/advisor for Amgen. He has received honoraria from Chugai Pharma and Sysmex, as well as research funding from Chugai Pharmaceutical, Daiichi Sankyo RD Novare, Eli Lilly, Novartis, PACT Pharmaceutical, AstraZeneca, Incyte, Janssen Oncology, Boehringer Ingelheim, MSD, Pfizer, Takeda, and Zymeworks. Hiroyuki Mano has received research funding from Chugai Pharmaceutical and Konica Minolta. He has served as a board member of CureGene. The other authors declare no conflicts of interest.
Supporting information
Figure S1. Bland–Altman plot of age at diagnosis based on QI and C‐CAT repository data (N = 1162).
Figure S2. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by the number of treatment regimens before CGP test (N = 314).
Figure S3. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by cancer types (N = 314).
Figure S4. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by rarity of cancer (N = 314).
Table S1. The reasons for non‐administration of gene‐matched therapy among patients with druggable alterations using the C‐CAT repository (N = 336).
Table S2. The concordance of cancer type between the QI database and C‐CAT repository (N = 1162).
Table S3. Factors associated with the overall survival of patients who underwent GMT or non‐GMT (N = 314).
Acknowledgments
We thank all tumor registrars and hospital staff for enabling Hospital‐Based Cancer Registries and participating in our quality‐of‐care project and C‐CAT repository. The authors used ChatGPT (GPT‐4o, OpenAI) to enhance readability and proofread the manuscript. They then reviewed, edited, and took full responsibility for the content of the publication. We would like to thank Editage (www.editage.jp) for English language editing.
Funding: This work was supported by the National Cancer Center Research and Development grant, 2024‐A‐20.
Data Availability Statement
Data were used with special permission under privacy protection laws and a data use agreement with the hospitals. The data are not publicly available because of privacy or ethical restrictions.
References
- 1. Moyers J. T. and Subbiah V., “Think Globally, Act Locally: Globalizing Precision Oncology,” Cancer Discovery 12 (2022): 886–888. [DOI] [PubMed] [Google Scholar]
- 2. Robinson D. R., Wu Y. M., Lonigro R. J., et al., “Integrative Clinical Genomics of Metastatic Cancer,” Nature 548 (2017): 297–303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Tuxen I. V., Rohrberg K. S., Oestrup O., et al., “Copenhagen Prospective Personalized Oncology (CoPPO)‐Clinical Utility of Using Molecular Profiling to Select Patients to Phase I Trials,” Clinical Cancer Research 25 (2019): 1239–1247. [DOI] [PubMed] [Google Scholar]
- 4. Zehir A., Benayed R., Shah R. H., et al., “Mutational Landscape of Metastatic Cancer Revealed From Prospective Clinical Sequencing of 10,000 Patients,” Nature Medicine 23 (2017): 703–713. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Kohno T., Kato M., Kohsaka S., et al., “C‐CAT: The National Datacenter for Cancer Genomic Medicine in Japan,” Cancer Discovery 12 (2022): 2509–2515. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Iwamoto M., Nakamura F., and Higashi T., “Monitoring and Evaluating the Quality of Cancer Care in Japan Using Administrative Claims Data,” Cancer Science 107 (2016): 68–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Higashi T., Nakamura F., Shibata A., Emori Y., and Nishimoto H., “The National Database of Hospital‐Based Cancer Registries: A Nationwide Infrastructure to Support Evidence‐Based Cancer Care and Cancer Control Policy in Japan,” Japanese Journal of Clinical Oncology 44 (2014): 2–8. [DOI] [PubMed] [Google Scholar]
- 8. Katano S., Yano T., Ohori K., et al., “Barthel Index Score Predicts Mortality in Elderly Heart Failure—A Goal of Comprehensive Cardiac Rehabilitation,” Circulation Journal 86 (2021): 70–78. [DOI] [PubMed] [Google Scholar]
- 9. Dickerman B. A., García‐Albéniz X., Logan R. W., Denaxas S., and Hernán M. A., “Avoidable Flaws in Observational Analyses: An Application to Statins and Cancer,” Nature Medicine 25 (2019): 1601–1606. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Petito L. C., García‐Albéniz X., Logan R. W., et al., “Estimates of Overall Survival in Patients With Cancer Receiving Different Treatment Regimens: Emulating Hypothetical Target Trials in the Surveillance, Epidemiology, and End Results (SEER)‐Medicare Linked Database,” JAMA Network Open 3 (2020): e200452. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Hashimoto T., Nakamura Y., Fujisawa T., et al., “The SCRUM‐MONSTAR Cancer‐Omics Ecosystem: Striving for a Quantum Leap in Precision Medicine,” Cancer Discovery 14 (2024): 2243–2261. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Nakamura Y., Ozaki H., Ueno M., et al., “Targeted Therapy Guided by Circulating Tumor DNA Analysis in Advanced Gastrointestinal Tumors,” Nature Medicine 31 (2025): 165–175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Rikitake R., Yatabe Y., Yamamoto Y., et al., “Proposal for a New Classification of Rare Cancers Adopting Updated Histological Tumor Types,” Pathology International 75 (2025): 291–309, 10.1111/pin.70021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Anderson E. C., DiPalazzo J., Lucas F. L., et al., “Genome‐Matched Treatments and Patient Outcomes in the Maine Cancer Genomics Initiative (MCGI),” NPJ Precision Oncology 8 (2024): 67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Hernan M. A., Sauer B. C., Hernandez‐Diaz S., Platt R., and Shrier I., “Specifying a Target Trial Prevents Immortal Time Bias and Other Self‐Inflicted Injuries in Observational Analyses,” Journal of Clinical Epidemiology 79 (2016): 70–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Nakamura Y., Okamoto W., Kato T., et al., “Circulating Tumor DNA‐Guided Treatment With Pertuzumab Plus Trastuzumab for HER2‐Amplified Metastatic Colorectal Cancer: A Phase 2 Trial,” Nature Medicine 27 (2021): 1899–1903. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Jee J., Lebow E. S., Yeh R., et al., “Overall Survival With Circulating Tumor DNA‐Guided Therapy in Advanced Non‐Small‐Cell Lung Cancer,” Nature Medicine 28 (2022): 2353–2363. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Olsen S., Liao J., and Hayashi H., “Real‐World Clinical Outcomes After Genomic Profiling of Circulating Tumor DNA in Patients With Previously Treated Advanced Non‐Small Cell Lung Cancer,” Current Oncology 29 (2022): 4811–4826. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Vidula N., Niemierko A., Malvarosa G., et al., “Tumor Tissue‐ Versus Plasma‐Based Genotyping for Selection of Matched Therapy and Impact on Clinical Outcomes in Patients With Metastatic Breast Cancer,” Clinical Cancer Research 27 (2021): 3404–3413. [DOI] [PubMed] [Google Scholar]
- 20. Beltran H., Eng K., Mosquera J. M., et al., “Whole‐Exome Sequencing of Metastatic Cancer and Biomarkers of Treatment Response,” JAMA Oncology 1 (2015): 466–474. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Ida H., Koyama T., Mizuno T., et al., “Clinical Utility of Comprehensive Genomic Profiling Tests for Advanced or Metastatic Solid Tumor in Clinical Practice,” Cancer Science 113 (2022): 4300–4310. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Meric‐Bernstam F., Brusco L., Shaw K., et al., “Feasibility of Large‐Scale Genomic Testing to Facilitate Enrollment Onto Genomically Matched Clinical Trials,” Journal of Clinical Oncology 33 (2015): 2753–2762. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Radovich M., Kiel P. J., Nance S. M., et al., “Clinical Benefit of a Precision Medicine Based Approach for Guiding Treatment of Refractory Cancers,” Oncotarget 7 (2016): 56491–56500. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Hernan M. A. and Robins J. M., “Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available,” American Journal of Epidemiology 183 (2016): 758–764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Matthews A. A., Danaei G., Islam N., and Kurth T., “Target Trial Emulation: Applying Principles of Randomised Trials to Observational Studies,” BMJ (Clinical Research ed.) 378 (2022): e071108. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1. Bland–Altman plot of age at diagnosis based on QI and C‐CAT repository data (N = 1162).
Figure S2. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by the number of treatment regimens before CGP test (N = 314).
Figure S3. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by cancer types (N = 314).
Figure S4. Subgroup analyses examining 2‐year OS in GMT vs. non‐GMT groups stratified by rarity of cancer (N = 314).
Table S1. The reasons for non‐administration of gene‐matched therapy among patients with druggable alterations using the C‐CAT repository (N = 336).
Table S2. The concordance of cancer type between the QI database and C‐CAT repository (N = 1162).
Table S3. Factors associated with the overall survival of patients who underwent GMT or non‐GMT (N = 314).
Data Availability Statement
Data were used with special permission under privacy protection laws and a data use agreement with the hospitals. The data are not publicly available because of privacy or ethical restrictions.