Overall survival of patients with localized soft tissue sarcoma treated with or without adjuvant chemotherapy (DEEPSARC): a real world nationwide retrospective cohort study in France

Summary

Background

Whether adjuvant chemotherapy (AC) improves overall survival (OS) after complete resection in localized soft tissue sarcoma (STS) remains unproven and optional in guidelines. We measured OS in patients with operated non-metastatic STS treated with or without AC in the nationwide exhaustive NETSARC+ registry.

Methods

The DEEPSARC project linked the NETSARC+ registry and the French nationwide health data system (SNDS) into a retrospective cohort. Adults with non-metastatic, undergoing primary surgical resection between 2012 and 2017 were included, excluding those with incomplete linkage data or prior neoadjuvant chemotherapy. We analysed the entire series of 8331 patients with successful linkage and a complete-case subset of 4296 patients. Survival with or without AC were compared using crude and multivariate Cox proportional hazards regression models, including adjustment for confounders and inverse probability of treatment weighting (IPTW). Sensitivity analyses and subgroups analysis in high-risk patients were performed.

Findings

In the complete-case series, the commonest histotypes were liposarcoma (30%), undifferentiated pleomorphic sarcoma (23%) and leiomyosarcoma (17%). Overall, 389 (9.1%) patients received AC and, compared with those not receiving AC, had fewer comorbidities and presented more high-risk tumour features. With 5.8 years median follow-up, 5-yr OS was 61% (95%CI: 60–63) in the entire series, and 64% (95%CI: 63–66) in the complete-case analysis. 5-yr OS was inferior in patients receiving AC, both in the entire series (50% [47%–54%] vs. 63% [61%–64%]) and in the complete-case analysis (53% [48–59%] vs. 65% [64–67%]). After adjustment (HR = 1.46 [1.12–1.93]) or IPTW (HR = 1.56 [1.17–2.08]), AC remained associated with inferior OS. Findings were consistent across sensitivity and subgroup analyses.

Interpretation

In this real-world nationwide study, AC was not associated with superior OS, but rather with an inferior OS in patients operated of a localized STS and did not compensate for the adverse prognostic profile of treated patients. Future research should therefore focus on identifying predictive biomarkers in order to better identify patients who are likely to derive real benefit from systemic treatment.

Funding

This work was supported by the Health Data Hub; the Institut National Du Cancer and the French Ministry of Health (Direction Générale de l’Offre de Soins, DGOS) for the NETSARC, RRePS and RESOS networks; the INTERSARC+ program; the Agence Nationale de la Recherche; the LYriCAN+ program; the Ligue Nationale contre le Cancer; the Ligue contre le Cancer (Comité de l’Ain); the Fondation ARC; and the European Reference Network for Rare Adult Solid Cancers.

Research in context.

Evidence before this study

Soft-tissue sarcomas (STS) are rare and heterogeneous malignancies, accounting for approximately 1% of adult cancers and comprising more than 100 histological subtypes. Surgery and perioperative radiotherapy are the cornerstone of treatment for localized disease; however, around half of patients relapse and 5-year overall survival remains approximately 50–60%. The role of adjuvant chemotherapy (AC) after complete resection has been debated for more than five decades.

We searched PubMed from January 1st 1974 inception to January 1st 2026 using the terms “sarcoma”, “adjuvant chemotherapy”, and “overall survival”, and “randomized clinical trial” without language restrictions. We identified 22 randomized trials and 3 meta-analyses comparing anthracycline-based adjuvant chemotherapy with observation. The oldest meta-analyses reported reductions in local or distant recurrence but no statistically significant overall survival benefit. A More recent meta-analysis suggested a modest improvement in recurrence-free and overall survival, particularly with anthracycline–ifosfamide combinations. However, the largest randomized trial (EORTC 62931) did not demonstrate improvement in relapse-free or overall survival. Retrospective and post-hoc subgroup analyses have suggested possible benefit in selected high-risk groups (high-grade tumours, R1 resections, or high-risk profiles defined by prognostic nomograms such as Sarculator), but these findings were purely exploratory.

Current international guidelines therefore do not consider AC a standard treatment after resection of localized STS and leave its use to individualized decision-making in selected high-risk patients.

Added value of this study

Using the nationwide NETSARC+ registry linked with the French National Health Data System (SNDS), the DEEPSARC study provides the largest real-world evaluation to date of adjuvant chemotherapy in adults with resected localized soft-tissue sarcoma. Among 8331 patients fewer than 10% received AC. Patients treated with AC had more adverse prognostic features, including larger and higher-grade tumours.

Across multiple analytic strategies, including multivariable Cox models, propensity score weighting, and instrumental variable analyses, AC was not associated with improved overall survival. In adjusted analyses, the overall survival of patients receiving AC remains lower than the survival of those not receiving it (5-year adjusted HR 1.46 [1.12–1.93]; IPTW HR 1.56 [1.17–2.08]). No predefined clinical or histological subgroup showed a statistically robust survival benefit, including high-grade tumours larger than 5 cm and patients classified as high-risk according to Sarculator.

Implications of all the available evidence

Our findings do not support a survival benefit of AC in the broad population of adults with resected localized soft-tissue sarcoma. The absence of observed benefit in high-risk subgroups challenges the assumption that conventional anthracycline-based regimens meaningfully improve long-term survival outside clinical trials. Future research should therefore focus on identifying predictive biomarkers in order to better identify patients who are likely to derive real benefit from systemic treatment.

Introduction

Whether adjuvant chemotherapy (AC) reduces relapse and improves overall survival (OS) of patients with resected soft tissue sarcomas (STS) has been studied in a limited number of low powered, heterogeneous randomized clinical trials. These trials are inconclusive and AC remains debated.1, 2, 3, 4 A first meta-analysis of randomized clinical trials (RCTs) comparing AC to no chemotherapy showed a reduction in local relapse but no significant improvement in OS,⁵ whereas a subsequent meta-analysis on published data reported improvements in relapse-free survival (RFS) and OS.⁶ Subsequently, the largest randomized study exploring AC vs. no treatment did not demonstrate improvement for RFS nor OS.^7,8

Hence, guideline recommendations after resection of localized STS report that AC is not a standard treatment in any group.5, 6, 7, 8, 9 Subgroups of patients who may benefit from AC are not identified, but retrospective analyses from RCTs suggest benefit from AC in men, R1 resections, high-risk patients as defined by tools like SARCULATOR [score over 60%],^8,9 or as defined by inclusion criteria of RCTs.8, 9, 10 Thus, evidence for AC remains limited and mainly from small size retrospective or subgroup analyses.

The rarity and heterogeneity of sarcomas^11,12 challenges the feasibility and interpretability of RCTs. Yet, identifying histological subtypes in which AC would be beneficial is an important objective for the sarcoma community, as shown for gastrointestinal stromal tumour (GIST) with imatinib¹³ and attempts conducted in other subtypes such as angiosarcomas.13, 14, 15 Real-world data may provide additional insight on this question still unsolved 50 years after doxorubicin.

We linked two real-life nationwide databases. The NETSARC+ database¹⁶ gathers since 2010 prospectively collected sarcoma observations from all patients treated in France, with exhaustivity achieved since 2013.16, 17, 18, 19, 20 All cases included are reviewed and validated by pathology sarcoma specialists ensuring quality of diagnosis for these registries.16, 17, 18 In 2019, the DEEPSARC project was launched to merge this database with the National Health Data System (Système National des données de Santé, SNDS) which includes health records of nearly all French citizens covered by Social Security. This pilot project was conducted as a partnership between the French Sarcoma Group and the French Health Data Hub.

The primary objective of DEEPSARC was to establish whether the OS of patients with localized resected sarcoma receiving AC would be superior to that of those not receiving AC, using the largest dataset ever reported on this question. Secondary objectives included identifying potential subgroups benefiting from AC among clinical or histological factors.

Methods

Study design and data sources

DEEPSARC is a nationwide retrospective cohort study aiming to estimate OS of adult patients diagnosed with non-metastatic STS who underwent surgical resection between 2010 and 2017 in France and to compare the outcome of patients receiving or not AC.

Data linkage

Linkage between NETSARC+ with SNDS allowed longitudinal follow-up on survival, recurrence, and healthcare utilization at the individual patient level.

To comply with national requirements of general data protection regulation (GDPR), the NETSARC+ database lacks national identification numbers, thus merging using a unique identifier was impossible. Probabilistic matching methodology used in the present project has been reported in a specific article.^21,22 In short, linkage was conducted with a two-step indirect deterministic matching and a per-patient matching quality metric was calculated.

Eligibility criteria

Inclusion criteria were adults (≥18 years) with non-metastatic STS confirmed by NETSARC+, treated with surgical resection as part of primary treatment between 01/01/2012 and 31/12/2017. Study period began in 2012 because mapping of pathologies in SNDS is available from 2012 onwards. Exclusion criteria were patients with incomplete or erroneous linkage data and patients who received neo-adjuvant chemotherapy. Two populations were analysed: the entire series of successfully linked patients and a complete-case subset of patients for whom all key confounders were documented (age, sex, histological subtypes, tumour grade, site, size, depth, margins, comorbidities and social deprivation index).

Variables

Exposure was administration of AC identified in the SNDS. Covariates included patient demographics (age, sex, social deprivation index²³), comorbidity (Mortality-Related Morbidity Index or MRMI)^24,25 derived or calculated using data from the SNDS. Tumour-specific variables (histological subtypes, size, depth, and grade according to the French Federation of Cancer Centres (FNCLCC)²⁶), were sourced from NETSARC+ database. These variables were standardised to the time of surgery (T0). Data were collected at diagnosis, post-surgery and during follow-up. Survival outcomes were assessed using vital status and dates of death recorded in the SNDS.

Statistical analysis

OS was calculated from date of surgery to date of last follow-up or death. Baseline and treatment-related characteristics were compared between groups using χ² tests (including tests for trends) or Fisher’s exact tests for categorical variables, and Student’s t-tests or Wilcoxon rank-sum tests for continuous variables, as appropriate. Survival curves were estimated using the Kaplan–Meier method. Proportional hazards assumption was assessed via Schoenfeld residuals and addressed using time-dependent covariates when unverified.

Complete-case analyses used Cox proportional hazards models adjusted for confounders («Adjustment model») and inverse probability of treatment weighting (IPTW) using a propensity score («PS model») including age, sex, histological subtypes, tumour grade, site, size, depth, comorbidities and social deprivation index (Statistical model specifications available in Supplementary Methods S3). Adequate covariate balance was achieved (Supplementary Fig. S1 and Table S1).

Subgroup analyses included pre-planned analyses by age, sex, histological subtypes, tumour grade, site, size, surgical margins, re-excision status and matching quality metric,²¹ as well as post-hoc analyses focusing on treatment period (2012–2014 vs. 2015–2017), grade 3 tumours >5 cm and predictions from the SARCULATOR nomogram.²⁷ Survival comparisons between subgroups were done through Cox univariate analyses.

Additional post-hoc sensitivity analyses included adjustments using cardiovascular comorbidity instead of MRMI, classifying missingness, and instrumental variable analyses (Two-stage residual inclusion - 2SRI, Two-stage least square - 2SLS, G-Estimation)28, 29, 30, 31 with recommended instruments32, 33, 34: residence-centre distance, type of care centre (comprehensive cancer centre vs. others) and median regional chemotherapy usage.

All tests were two-sided and performed using R version 3.6.3.

Governance

A restricted core steering group comprising sarcoma clinicians, biostatistician, methodologists, and Deepsarc project manager oversaw study design, data quality, and methodological and analytical decisions. A broader multidisciplinary committee including the core steering group, project managers, Health economics specialists, technical experts from the Health Data Hub and NETSARC+ met regularly to address technical, regulatory, and clinical aspects of data linkage, analyses, and interpretation.

Ethics

The study was conducted in accordance with the principles of the Declaration of Helsinki.

Informed oral consent was obtained for patient participation in Netsarc+ registry.16, 17, 18 In accordance with French legislation and as approved by the ethics authorities, the requirement for written informed consent was waived due to the registry-based and non-interventional design of the study.

In accordance with current French legislation, a waiver of informed consent was granted for this registry-based study. Patients included in the NETSARC+ registry are informed through an individual information notice at the time of inclusion, in accordance with regulatory requirements. For the DEEPSARC project, a complementary collective information notice was implemented, as approved by the French Data Protection Authority (Commission Nationale de l’Informatique et des Libertés, CNIL), reflecting the reuse of existing health data without direct patient contact.

Ethics approval was obtained from the Comité éthique et scientifique pour les recherches, les études et les évaluations dans le domaine de la santé (CESREES; approval granted on November 21, 2019), and authorization was obtained from the French Data Protection Authority (Commission Nationale de l’Informatique et des Libertés, CNIL; Authorisation No. 920271V1), granted on November 20, 2020.

In accordance with French regulations, ethnicity was not recorded in this study, as no specific authorization to collect such data was requested.

Transparency and reproductibility

The dataset is available through the French Health Data Hub (HDH) catalogue since February 7, 2025.³⁵

Access to the data by an independent expert mandated by a scientific publisher shall be carried out under the conditions set out in Deliberation No. 2018–155 of May 3, 2018, approving the French reference methodology for the processing of personal data implemented in the context of research not involving human participants, studies, and evaluations in the field of health (MR-004).³⁶

Role of the funding source

The funders had no involvement in study design, data analyses, data interpretation, or the writing of the report.

The Health Data Hub facilitated access to data and supported data extraction in collaboration, but had no role in the scientific conduct of the study or in the decision to submit the manuscript for publication.

Results

24,539 patients were paired between NETSARC+ and SNDS. 14,808 patients met the eligibility criteria (Fig. 1) and 6477 were excluded because of missing tumour location, primary visceral locations (including GIST), bone sarcomas or age < 18. The entire series consisted of 8331 STS patients (Supplementary Table S2). Among these, 3596 patients had at least one missing confounding variable and 439 patients received neoadjuvant chemotherapy. The complete-case subset included therefore 4296 patients (Fig. 1) and the most common histological subtypes were liposarcoma (LPS) (any subtype), undifferentiated pleomorphic sarcoma (UPS) and leiomyosarcoma (LMS) (Table 1).

Fig. 1.

Study profile of the Deepsarc cohort, based on the ESMO-GROW flowchart for real-world evidence studies in oncology, France, 2010–2017. The figure shows data sources Netsarc+ and French National Health Data System (SNDS), linkage process, eligibility criteria, exclusions, final data cleaning, and the complete-case analysis population. N = number of patients; SNDS = Système National des Données de Santé/French National Health Data System.

Table 1.

Baseline and treatment characteristics by adjuvant chemotherapy in the complete-case analysis population (N = 4296), Deepsarc study (2012–2017), France.

	Adjuvant chemotherapy		Total (N = 4296)	P-valuec
	No (N = 3907)	Yes (N = 389)
Year of sarcoma diagnosis				0.006
2012	562 (87.7%)	79 (12.3%)	641 (100.0%)
2013	701 (90.0%)	78 (10.0%)	779 (100.0%)
2014	617 (90.6%)	64 (9.4%)	681 (100.0%)
2015	611 (91.3%)	58 (8.7%)	669 (100.0%)
2016	658 (93.3%)	47 (6.7%)	705 (100.0%)
2017	758 (92.3%)	63 (7.7%)	821 (100.0%)
Age at diagnosis				<0.001
N	3907	389	4296
Median (Q1, Q3)	66.0 (53.0, 76.0)	56.0 (42.0, 66.0)	65.0 (52.0, 75.0)
Q5–Q95	33.0, 86.0	23.0, 79.6	32.0, 86.0
Sex (SNDS)				0.291
Female	1854 (91.5%)	173 (8.5%)	2027 (100.0%)
Male	2053 (90.5%)	216 (9.5%)	2269 (100.0%)
Distance between patient’s location and treatment centre (km)				0.341
N	3669	374	4043
N Obs not available	238	15	253
Median (Q1, Q3)	22.5 (7.1, 68.2)	21.6 (6.7, 58.0)	22.5 (7.1, 67.1)
Q5–Q95	0.0, 161.6	0.0, 176.9	0.0, 163.7
Social deprivation index quintiles (FDep) (SNDS)				0.203
N	3907	389	4296
Median (Q1, Q3)	3.0 (2.0, 4.0)	3.0 (2.0, 4.0)	3.0 (2.0, 4.0)
Q5–Q95	1.0, 5.0	1.0, 5.0	1.0, 5.0
Social deprivation index quintiles (FDep) (SNDS)				0.041
1	791 (91.8%)	71 (8.2%)	862 (100.0%)
2	802 (89.2%)	97 (10.8%)	899 (100.0%)
3	848 (90.3%)	91 (9.7%)	939 (100.0%)
4	797 (90.6%)	83 (9.4%)	880 (100.0%)
5	669 (93.4%)	47 (6.6%)	716 (100.0%)
Morbidity index (MRMIp) (SNDS)				0.151
N	3907	389	4296
Median (Q1, Q3)	0.0 (0.0, 1.0)	0.0 (0.0, 1.0)	0.0 (0.0, 1.0)
Q5–Q95	0.0, 2.0	0.0, 2.0	0.0, 2.0
Morbidity index (MRMIp) (SNDS)				0.335
0	2541 (90.5%)	266 (9.5%)	2807 (100.0%)
1	834 (91.3%)	79 (8.7%)	913 (100.0%)
≥2	532 (92.4%)	44 (7.6%)	576 (100.0%)
Cardiovascular comorbidity				0.024
No	3052 (90.4%)	323 (9.6%)	3375 (100.0%)
Yes	855 (92.8%)	66 (7.2%)	921 (100.0%)
Most common histological subtypes				<0.001
Liposarcoma	1218 (93.8%)	80 (6.2%)	1298 (100.0%)
Undifferentiated sarcoma	886 (91.0%)	88 (9.0%)	974 (100.0%)
Leiomyosarcoma	677 (92.9%)	52 (7.1%)	729 (100.0%)
Miscellaneous sarcomas	372 (85.7%)	62 (14.3%)	434 (100.0%)
Myxofibrosarcoma	387 (93.5%)	27 (6.5%)	414 (100.0%)
Synovial sarcoma	109 (83.8%)	21 (16.2%)	130 (100.0%)
Malignant peripheral nerve sheath tumour	95 (86.4%)	15 (13.6%)	110 (100.0%)
Rhabdomyosarcoma	62 (68.9%)	28 (31.1%)	90 (100.0%)
Primary sites				0.009
Head and neck	330 (89.9%)	37 (10.1%)	367 (100.0%)
Internal trunk	1025 (89.8%)	116 (10.2%)	1141 (100.0%)
Lower limb	1353 (92.5%)	110 (7.5%)	1463 (100.0%)
Trunk wall	731 (88.8%)	92 (11.2%)	823 (100.0%)
Upper limb	458 (93.1%)	34 (6.9%)	492 (100.0%)
Size of tumour (cm)				0.049
Median (Q1, Q3)	7.1 (4.1, 13.1)	8.1 (5.1, 12.1)	7.1 (4.1, 13.1)
Q5–Q95	1.6, 22.3	2.4, 18.9	1.6, 22.3
Size of tumour				<0.001
≤5	1253 (93.7%)	84 (6.3%)	1337 (100.0%)
>5	2654 (89.7%)	305 (10.3%)	2959 (100.0%)
Depth of tumour				<0.001
Deep	2507 (89.7%)	288 (10.3%)	2795 (100.0%)
Superficial	964 (94.2%)	59 (5.8%)	1023 (100.0%)
Superficial and deep	434 (91.2%)	42 (8.8%)	476 (100.0%)
Grade of tumour				<0.001
Grade I	656 (97.2%)	19 (2.8%)	675 (100.0%)
Grade II	1356 (95.1%)	70 (4.9%)	1426 (100.0%)
Grade III	1626 (85.3%)	280 (14.7%)	1906 (100.0%)
Non gradable	269 (93.1%)	20 (6.9%)	289 (100.0%)
Tumour in irradiated fields				0.034
No	3745 (91.1%)	365 (8.9%)	4110 (100.0%)
Yes	153 (86.4%)	24 (13.6%)	177 (100.0%)
Treatment
Time interval between date of diagnosis and date of surgery (months)				<0.001
Median (Q1, Q3)	1.1 (0.1, 2.4)	0.3 (0.1, 1.4)	1.0 (0.1, 2.3)
Q5–Q95	0.0, 5.4	0.0, 3.5	0.0, 5.3
Amputation surgery				0.976
No	3603 (90.7%)	370 (9.3%)	3973 (100.0%)
Surgery quality				<0.001
R0 margin	2025 (93.3%)	145 (6.7%)	2170 (100.0%)
R1 margin	890 (90.2%)	97 (9.8%)	987 (100.0%)
R2 margin	165 (79.3%)	43 (20.7%)	208 (100.0%)
N Obs not available	827	104	931
Re-excision				0.798
No	1856 (91.7%)	168 (8.3%)	2024 (100.0%)
Yes	863 (91.4%)	81 (8.6%)	944 (100.0%)
N Obs not available	1188	140	1328
Tumour spillage				<0.001
No	2150 (93.3%)	155 (6.7%)	2305 (100.0%)
Yes	221 (79.2%)	58 (20.8%)	279 (100.0%)
Not available	1536	176	1712
Peri-operative radiotherapy (neo-adjuvant and/or adjuvant)				<0.01
No	2134 (92.1%)	182 (7.9%)	2316 (100.0%)
Yes	1773 (89.5%)	207 (10.5%)	1980 (100.0%)
Neo-adjuvant radiotherapy				<0.001
No	3576 (90.4%)	380 (9.6%)	3956 (100.0%)
Yes	<11 so not disseminablea		331 (100.0%)
Adjuvant radiotherapy (9 months delayb)				<0.001
No	2450 (92.8%)	190 (7.2%)	2640 (100.0%)
Yes	1457 (88.0%)	199 (12.0%)	1656 (100.0%)
Delay between date of surgery and date of adjuvant radiotherapy (months)				<0.001
Median (Q1, Q3)	2.7 (1.9, 3.9)	5.5 (4.0, 7.0)	2.9 (2.0, 4.5)
Q5–Q95	1.2, 6.9	1.3, 8.6	1.2, 7.5

cm = centimetres; FDep = French Deprivation Index; km = kilometres; Q1 = first quartile; Q3 = third quartile; Q5–Q95 = 5th–95th percentiles. N = number of observations. N Obs not available = number of missing observations for the variable; MRMIp = Mortality-Related Morbidity Index. Version noted MRMIp when indices are computed using exclusively condition-related weights; R0/R1/R2 = surgical margins (R0: complete resection with negative margins; R1: microscopic residual disease; R2: macroscopic residual disease); SNDS = Système National des Données de Santé (French National Health Data System).

Bold P-values indicate statistically significant differences between groups.

^aWhen using the National Health Data System (Système National des données de Santé, SNDS), any tabulated results, statistics, or data that refer to fewer than 11 individuals are not published in order to ensure patient privacy.

^bWe take into account adjuvant radiotherapy started within 9 months after surgery.

^cCategorical variables were compared using Chi-square tests, and continuous variables using Wilcoxon rank-sum tests.

Proportion of patients receiving AC decreased between 2012 and 2017 (P = 0.006) (Table 1). Patients receiving AC were significantly younger (P < 0.001), had better social deprivation index (P = 0.04), less cardiovascular morbidity (P = 0.02), and had significantly less primary limb sarcomas (vs. trunk wall and head and neck) (P = 0.009) (Table 1). A larger proportion of tumour size over 5 cm (P < 0.001), a smaller proportion of patients with superficial sarcomas (P < 0.0001), and a larger proportion of grade 3 sarcomas (P < 0.0001) were observed in the group receiving AC (Table 1). Sex, morbidity indexes and distance from treatment centre were not significantly different (Table 1).

Patients receiving AC had shorter delay between diagnosis and surgery (P < 0.001) (Table 1). There were no difference in amputation rates (P = 0.976) (Table 1). Surgery was more commonly inadequate (R2 or not documented) in the subgroup receiving AC, with 11% (43/389) vs. 4% (165/3907) of R2 resections, and 26% (104/389) vs. 21% (827/3907) undocumented quality of resection in the AC vs. no AC groups (P < 0.001) (Table 1). Conversely, there was no difference in the proportion of patients with re-excision between groups. There was also significantly more reported tumour spillage in the AC group (58/213 [27.2%] vs. 221/2371 [9.3%], P < 0.001) (Table 1). Adjuvant radiotherapy was more commonly administered in patients having received AC (P < 0.001) (Table 1).

Median follow-up was 5.8 years [5.7–5.9] and median survival were 7.2 years [95% CI 6.9–7.4] in the entire series and 7.5 years [95%CI 7.2–7.9] in the complete case subset (Fig. 2A and B). 5-year OS of patients with localized STS was about 60% without plateau on both series. Patients who received AC had a significantly inferior survival (Fig. 2C and D).

Fig. 2.

Kaplan–Meier curves for overall survival in the entire series and complete-case analysis populations, overall and according to adjuvant chemotherapy, Deepsarc study, France, 2012–2017. Panels show: entire series overall survival (A), complete-case analysis overall survival (B), survival according to adjuvant chemotherapy in the entire series (C), and survival according to adjuvant chemotherapy in the complete-case analysis population (D). Overall survival was estimated from the date of surgery to death or last follow-up. Survival probabilities are reported at 1, 3, and 5 years with 95% confidence intervals. Median survival and Survival probabilities are presented with 95% confidence intervals (CI). No covariate adjustment is applied in these curves (unadjusted descriptive analysis). N = number of patients; Event N = number of deaths; CI = confidence interval.

Methods used to correct for confounding biases (Adjustment model and PS model) showed similar results (Tables 2 and 3). In analyses adjusted on classical prognostic factors (Table 2), as well as on surgery quality (R0, R1, R2, R unknown; Table 3), the administration of AC was still associated with an inferior OS.

Table 2.

Global survival multivariate analysis (risk of death) according to adjuvant chemotherapy adjusted on age, sex, social deprivation index, morbidity index, histology category, size, depth and grade of tumour (N = 4296), Deepsarc study (2012–2017), France.

	Without adjustment model		Adjustment model			Propensity Score model
	HRa	95% CI	HRb	95% CI	P-value	HRc	95% CI	P-value
Adjuvant chemotherapy (reference = No)
Yes					<0.001			<0.001
1 year	1.79	1.48; 2.17	1.83	1.49; 2.23		1.63	1.31; 2.03
3 years	1.38	1.18; 1.62	1.64	1.38; 1.94		1.59	1.33; 1.91
5 years	1.06	0.82; 1.38	1.46	1.12; 1.93		1.56	1.17; 2.08

HR = Hazard Ratio, CI = Confidence Interval; P-value obtained using Likelihood ratio test.

Bold P-values indicate statistically significant differences between groups.

^aSignificant time-dependent coefficient for variable Adjuvant chemotherapy.

^bSignificant time-dependent coefficients for variables Age, Size and Grade of tumor.

^cNo significant time-dependent coefficient.

Table 3.

Global survival multivariate analysis (risk of death) according to adjuvant chemotherapy adjusted on age, sex, social deprivation index, morbidity index, histology category, size, depth, grade of tumour and margin quality, re-excision and tumor spillage (N = 2263), Deepsarc study (2012–2017), France.

	Without adjustment model		Adjustment model			Propensity Score model
	HRa	95% CI	HRb	95% CI	P-value	HRc	95% CI	P-value
Adjuvant chemotherapy (reference = No)
Yes					0.01			0.04
1 year	1.89	1.41; 2.55	1.58	1.15; 2.17		1.51	1.07; 2.14
3 years	1.41	1.12; 1.78	1.45	1.13; 1.87		1.56	1.17; 2.07
5 years	1.05	0.73; 1.51	1.34	0.91; 1.97		1.60	1.05; 2.44

HR = Hazard Ratio, CI = Confidence Interval; P-value obtained using Likelihood ratio test.

Bold P-values indicate statistically significant differences between groups.

^aSignificant time-dependent coefficient for variable Adjuvant chemotherapy.

^bSignificant time-dependent coefficients for variables Age, Size and Grade of tumor.

^cNo significant time-dependent coefficient.

In sensitivity analyses performed using instrumental variable, AC was also associated with an inferior survival (Supplementary Table S3). Sensitivity analyses accounting for database linkage quality (Supplementary Table S4) and missing data (Supplementary Table S5) yielded consistent results.

Outcome of patients in different subgroups was analysed (Fig. 3 and Supplementary Table S6). Overall, administration of AC was either associated with an inferior or not significantly different OS in all subgroups (Fig. 3). In 1906 patients for whom the SARCULATOR score could be calculated, 675 had a score equal or over 120, the 5-year OS was significantly worse compared to those with a SARCULATOR score under 120 (49% [45–54%] vs. 81.0% [78–84%], P < 0.001), and AC was not significantly associated with OS (Fig. 3A and Supplementary Table S6).

Fig. 3.

Forest plot of hazard ratios for overall survival according to demographic characteristics (A), tumour characteristics (B), surgery quality indicators (C), and tumour histotype (D), Deepsarc study, France, 2012–2017. Hazard ratios were estimated using Cox proportional hazards models. Results presented correspond to propensity score-based analyses using inverse probability of treatment weighting (IPTW) to adjust for confounding. The propensity score included age, sex, histological subtype, tumour grade, tumour size, tumour depth, tumour site, comorbidities, and social deprivation index. Point estimates (central markers) represent hazard ratios, and horizontal lines represent 95% confidence intervals (CI). Values > 1 indicate an increased risk of death associated with adjuvant chemotherapy. HR = hazard ratio; CI = confidence interval; 1 y = 1 year; 3 y = 3 years; 5 y = 5 years; MPNST = malignant peripheral nerve sheath tumour; UPS = undifferentiated pleomorphic sarcoma.

In the 1366 patients with a tumour >5 cm and grade 3, the 5-year OS was significantly worse compared to those with other tumours (49% [46–52%] vs. 72% [70–74%], P < 0.001), and AC was consistently associated with an inferior survival (Fig. 3A and Supplementary Table S6).

Regarding sites, AC was significantly associated with inferior survival at the head and neck (N = 369, 5-year OS 56% [50–62%]), internal trunk (N = 1146, 5-year OS 60% [56–63%]), and trunk wall (N = 823, 5-year OS 62% [58–66%]) sites, whereas other sites showed no significant association (Fig. 3A and Supplementary Table S6).

Regarding quality of surgery (Fig. 3C and Supplementary Table S6): in patients with R0 margins (N = 2,170, 5-year OS: 70% [68–72%]), AC was not significantly associated with survival. In the R1 group (N = 987, 5-year OS: 61% [58–65%]), AC was associated with inferior survival at 3 and 5 years. No significant association was observed in patients with R2 margins (N = 208, 5-year OS: 47% [40–55%]). In patients with re-excision (N = 944, 5-year OS: 74% [71–77%]), AC was not associated with survival whereas in the subgroup without re-excision (N = 2,024, 5-year OS: 62% [60–65%]), AC was consistently associated with inferior survival (5-year OS HR 2.5 [1.7–3.8], P < 0.001).

In all models for both women and men (5-year OS: 65% [63–68%] vs. 63% [61–66%], P = 0.05), and in patients aged under 40 years and over and equal to 40 years (5-year OS: 84% [80–87%] vs. 62% [60–64%], P < 0.001), AC was associated with worst OS (Fig. 3A and Supplementary Table S6). For patients treated between 2012 and 2014 and between 2015 and 2017, AC was associated with an inferior survival with all models. After adjustment, AC was associated with an inferior survival in LPS (N = 1,298, 5-year OS: 71% [68–74%]), in undifferentiated sarcomas (N = 974, 5-year OS: 53% [49–57%]), including UPS (N = 663, 5-year OS: 53% [49–57%]), in malignant peripheral nerve sheath tumours (MPNST) (N = 110, 5-year OS: 51% [42–62%]), and in synovial sarcoma (N = 130, 5-year OS: 78% [70–86%]) (Fig. 3D and Supplementary Table S6). Conversely, administration of AC was not associated with survival in LMS (N = 729, 5-year OS: 64% [60–68%]), angiosarcoma (N = 127, 5-year OS: 33% [26–44%]) and myxofibrosarcoma (N = 414, 5-year OS: 70% [64–75%]) (Fig. 3D and Supplementary Table S6).

Discussion

This nationwide population-based study is the largest to date exploring the outcome of patients with resected localized sarcomas treated or not with AC, allowing for a detailed characterization of both patient- and treatment-related factors in real-world setting. The aim was to tackle the 50-year-old debate of whether AC improves OS in operated patients with localized sarcoma, and/or to identify subgroups in which AC provides improved OS.

The results show that AC is not associated with improved survival. All methods used yielded similar observations of an inferior, or at best similar, survival in the group receiving AC. This pattern was observed in high-risk groups: grade 3 and tumours over 5 cm, those defined by the Sarculator nomogram (<60% survival at 10 years),³⁷ and those with R1 resections. Although Sarculator was developed patients treated in 1994–2007 and may no longer be applicable, prior studies suggested potential value: in a post-hoc re-evaluation of the EORTC 62931 randomized trial,⁹ patients with Sarculator nomogram <60% were observed to have an OS improvement with AC. This was not confirmed in the present study.

Median follow-up (5.8 years) was similar or longer to most RCTs of AC. It remains, however, insufficient to evaluate 10-year outcomes. Prior studies reported trends with survival advantage becoming apparent after extended follow-up periods.^6,7 These long-term effects are not detectable in the current dataset and may require future analyses.

This study has limitations. Only 24,539 patients could be paired to the SNDS, introducing potential biases of selection of unknown nature.^21,22 Linkage might have been more successful for patients who were managed in expert centres and/or those who died, potentially skewing the results.

Among 8331 patients, only 4962 (60%) had exhaustive documentation of prognostic parameters. It is possible that this documentation is better for patients treated in reference centres and/or those who received AC. However, univariate comparisons between included and non-included patients did not yield significant differences (not shown, results cannot be published since they concern individuals outside the scope of the study’s approval). Also, treatment profiles remain consistent with those previously reported: 47% of patients in our cohort received perioperative radiotherapy, comparable to rates reported in other large series (49% in the Sarcoma Meta-analysis Collaboration Meta-analysis³⁸ and 41% in Mathoulin-Pélissier et al.³⁹). Nonetheless, some residual selection bias due to data completeness, linkage quality, or probability of being linked depending on care pathways, may have contributed to imbalance between included or excluded.²¹ Nevertheless, the nationwide population-based design, in a country without specificities in sarcoma histotype distribution or care pathways compared to other high-income countries, supports broad representativeness and generalisability of the results; however, heterogeneity of STS limits their interpretation in routine clinical care.

Although the cohort was large, only 389 patients received AC in the complete-case analysis. All untreated patients were retained through weighting, but the limited number of treated patients reduced statistical power. This limitation is important when considering subgroup analyses, particularly for histologies potentially ‘sensitive’ to AC who received it (88 UPS, 21 synovial sarcoma, and 28 rhabdomyosarcoma).

The study period (2012–2017) should be acknowledged: AC strategies and histological classifications remained stable (even though AC use declined from 12% to 7% during this time), reducing bias due to evolving clinical practice.

As this was not a randomized study, prognostic factors may be unbalanced between groups. As expected, patients receiving AC had worse prognostic features (larger tumours, higher grade, more common R2 surgery) which often favour AC decision in Multidisciplinary tumours (MDT) boards.²⁰ One can also hypothesize that unmeasured prognostic factors, e.g., pain, “speed of clinical progression”, or the requirement for re-excision after unplanned surgery, influence informally clinical decision in routine. These parameters are often discussed in multidisciplinary settings but are not captured by any of the classical prognostic factors and may lead the MDT to propose AC, suspecting higher aggressiveness of the sarcoma. Supporting this hypothesis, delay between diagnosis and surgery was significantly shorter in the group receiving AC (Table 1).

Consequently, the present study does not imply that AC is deleterious; the negative association may primarily result from the unfavourable profiles of patients who received it. The study failed to demonstrate an improved survival with AC in this real-world setting, which was our initial hypothesis.

Despite these limitations, the study offers a complementary perspective to RCTs, which typically address the question “AC vs. no AC” in highly selected populations. AC remains a rarely used treatment strategy in real-world practice—likely due to accumulating scepticism about its risk–benefit balance—and, where prescribed, does not appear to bridge the survival gap between patients with aggressive disease and those with more favourable features.

The consistent association of AC and inferior patient survival deserves consideration regarding the present recommendations in most guidelines, where it is often left at the physician’s discretion. The results challenge both 1) the value of AC for STS patients, and/or 2) the quality and exhaustiveness of prognostic factors used in previous studies and in routine.

The heterogeneity of STS may explain the lack of benefit of AC when evaluated across broad categories. Recent developments in precision oncology suggest that genomics-driven stratification could identify subgroups more likely to benefit from targeted therapies and immunotherapy.⁴⁰ By analogy, such approaches might help refine selection of patients in whom AC could be effective. In addition, the SNDS does not capture chemotherapy agents, dose intensity, or regimen changes within hospitalizations, preventing evaluation of specific adjuvant protocols, which may further contribute to the absence of an observed survival benefit. A shift toward precision medicine, leveraging histo-molecular subtyping and targeted approaches, may redefine the role of AC in future treatment paradigms. Additional prospective work, including RCTs, focusing on more homogenous histotypes and molecular subtypes of sarcomas are needed to explore these questions. They will require international, worldwide efforts, which are particularly challenging in rare cancers.^19,20

In conclusion, AC was not associated with a superior OS, but rather with an inferior OS, in localized STS in this nationwide real-world study. This association persisted across all of the clinical, histological and prognostic subgroups tested, with AC either linked to poorer survival or showing no significant survival benefit. However, these results must be interpreted with caution due to the inherent limitations of an observational study design. These data add to the complexity of previously reported data on adjuvant treatment of STS,1, 2, 3, 4, 5, 6, 7 and should not be interpreted as the end of systemic treatment in high-risk patients, but as evidence of the limitations of pan-histology strategies using conventional regimens across heterogeneous populations, particularly in the context of an observational study design. Designing an emulated trial, using real-world data to simulate a randomized comparison, may help to refine patient selection and better define which, if any, subgroups could derive a benefit from AC.

Contributors

AH, EO, FD, HC, JYB, LP, NP and SMP contributed to conceptualisation and funding acquisition. EO, FD, JYB and SMP contributed to supervision or project administration. AH, CCA and ED contributed to data curation. AH, CCA, ED, and NH contributed to the investigation. AGM, ALC, CN, EB, FG, FLL, GDP, JYB, MK, MT, NF, NM, NP and SB contributed resources. ED, LP, NH and VT contributed to software. JYB, NP, SMP and VT contributed to methodology. NH performed the formal analysis. AGM, ALC, CN, EB, FG, FLL, GDP, JYB, MK, MT, NF, NM, NP, SB and VT contributed to validation. JYB, NP, SMP and VT wrote the original draft. All authors reviewed and edited the manuscript. All authors read and approved the final version of the manuscript.

NH and VT accessed and verified the underlying data.

The French Sarcoma Group/NETSARC+ consortium contributed to the study through the implementation of the NETSARC+ registry, nationwide patient inclusion, prospective clinical and pathological data collection, and data quality control. Consortium members also contributed to data management, coordination, and validation of clinical data, and approved the study concept.

Data sharing statement

The dataset is available through the French Health Data Hub (HDH) catalogue since February 7, 2025.

Access to the data by an independent expert mandated by a scientific publisher shall be carried out under the conditions set out in Deliberation No. 2018–155 of May 3, 2018, approving the French reference methodology for the processing of personal data implemented in the context of research not involving human participants, studies, and evaluations in the field of health (MR-004).

Code sharing is accessible through the same process but cannot be transferred to a git at the time of the manuscript as other studies are in process at the time of this manuscript within the closed gitlab HDH environment. Code is accessible on demand and will be shared at the end of all Deepsarc projects.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT and DeepL IA in order to translate some of the content from French to English. The authors reviewed and edited the content as needed. No AI tools were used to generate ideas, conduct analyses, or influence the scientific content of the manuscript.

Declaration of interests

ALC reports personal honoraria from Deciphera for lectures or educational events. FG reports consulting fees from Deciphera (paid to the institution), participation on an advisory board for Amgen (personal), support for attending meetings from Implantcast France, leadership roles within the French Sarcoma Network (NETSARC+) and the French Sarcoma Group, and stock ownership in Atlanta SAS. MT reports consulting fees from SpringWorks, honoraria for lectures or educational events from Deciphera, Servier and SpringWorks, and support for attending meetings from PharmaMar and Deciphera. NP reports institutional honoraria from Astellas Pharma and MSD, personal support for attending meetings from Deciphera and AstraZeneca, and other interests with PharmaMar (personal) and Bayer (institutional). All other authors (AGM, AH, CCA, CN, EB, ED, EO, FD, FLL, GDP, HC, JYB, LP, MK, NF, NH, NM, SB, SMP, VT) declare no competing interests.