Prospective evaluation of genomics-guided off-label treatment Authors: K. Verkerk, A. C. Spiekman, S. F. Haj Mohammad, F. A. J. Verbeek, H. Timmer, M. A. van Maren, L. J. Zeverijn, B. S. Geurts, V. van der Noort, P. Roepman, A. M. L. Jansen, W. W. J. de Leng, H. Gelderblom, H. M. W. Verheul, E. E. Voest, H. M. W. Verheul, H. M. W. Verheul, E. E. Voest, H. H. Nienhuis, J. M. van Dodewaard-de Jong, M. Labots, D. J. A. de Groot, C. van Herpen, A. Hoeben, H. Gelderblom, L. V. Beerepoot, E. D. Kerver, H. M. Westgeest, A. D. Bins, A. P. Hamberg, E. Boon, G. Vreugdenhil, G. J. de Klerk, T. van Voorthuizen, A. Vulink, F. van den Berkmortel, M. van Rooijen, D. Houtsma, A. L. T. Imholz, H. P. van den Berg Published: 2026 ABSTRACT Anticancer drugs are frequently used off-label for tumours that are genetically similar to the approved indication. However, outcomes are rarely captured systematically, limiting evidence-based decision-making and risking repeated futile treatment. The Drug Rediscovery Protocol (DRUP; ClinicalTrials.gov ID: NCT02925234) prospectively evaluates such off-label use in patients in the Netherlands with advanced solid tumours who lack standard treatment options and harbour actionable genomic alterations1. Here we present results of 1,610 patients who began treatment with 37 different off-label drugs between July 2016 and May 2024 in the DRUP trial. Of these patients, 1,363 were response-evaluable, including 533 (39.1%) with rare cancers. The clinical benefit rate (confirmed response or stable disease for at least 16 weeks) was 34.9% (95% confidence interval, 32.2–37.6) and the objective response rate was 15.7% (95% confidence interval, 13.7–17.9). Median progression-free and overall survival were 3.4 months (95% confidence interval, 2.8–3.5) and 8.2 months (95% confidence interval, 7.6–8.8), respectively. Grade 3 or higher treatment-related adverse events occurred in 28.4% of patients. Notably, evidence generated in DRUP was used for reimbursement decisions by the regulatory bodies in the Netherlands2. Although activity across all tumour–drug combinations was modest, defined molecular subgroups and exceptional responders (7.0%) achieved meaningful benefit. To maximize patient benefit, we recommend that off-label precision medicines should be used only within frameworks that systematically evaluate efficacy and toxicity, support biomarker refinement and enable stepwise assessment toward potential future label expansion. These frameworks should prioritize high-confidence targets, early intervention, regulatory-aligned end-points and international collaboration. Main Patients with advanced or metastatic cancer who have exhausted all approved treatments often seek additional therapeutic options. The rise of social media, patient advocacy groups and other information sources have made patients and physicians increasingly well-informed and proactive3, including regarding preliminary studies and case reports suggesting that drugs approved for one cancer type may be effective against another. This is especially relevant for patients with rare cancers, who face limited treatment options and poorer outcomes owing to reduced trial opportunities and commercial incentive for drug development in small populations4. The expanding role of targeted therapies in oncology further reinforces this trend, as many act on molecular alterations that occur across tumour types, albeit at different frequencies5. These molecular similarities provide a scientific rationale for physicians to consider using approved therapies outside their registered indications as a last-resort option, favouring their established safety profiles and dosing over experimental phase I agents6. Consequently, off-label use, defined as prescribing drugs beyond their US Food and Drug Administration or European Medicines Agency approval, is widespread: a recent analysis of 165,912 US patients with 14 common cancer types found that 18.6% received at least one off-label therapy7. Given that the development of successful anticancer drugs is an expensive and lengthy process, with estimated attrition rates exceeding 95% (ref. 8), repurposing approved therapies beyond their original indications offers a pragmatic strategy to accelerate access to treatments and improve the overall efficiency of drug development. This potential is exemplified by imatinib, initially developed for chronic myeloid leukaemia, which later demonstrated substantial efficacy in patients with KIT-mutated gastrointestinal stromal tumours9. Another notable example is the HER2-targeted therapy trastuzumab, first approved for breast cancer in 1998 and subsequently for HER2-positive gastric or gastro-oesophageal junction cancer in 2010 (refs. 10,11). Yet, despite the strong biological rationale, a 12-year regulatory gap separated the two approvals, underscoring missed opportunities to extend the benefits of approved, effective therapies to additional patient groups in a timely and evidence-based manner. Conversely, not all targeted therapies yield comparable results across tumour types. For example, whereas BRAF inhibitors achieve high response rates in melanoma, they show minimal efficacy when used as monotherapy in patients with colorectal cancer carrying the same mutation12,13. The SHIVA trial further emphasized these limitations by failing to show improved outcomes from a limited range of off-label targeted therapies compared with standard treatment in a tumour-agnostic, randomized setting, along with reporting higher toxicity in the experimental group14. Even though the trial faced methodological limitations, including targets with low predictive value outside their canonical tumour types or poorly matched to therapies, these findings still underscore the importance of carefully evaluating each new indication to ensure patient benefit and avoid unnecessary harm. However, whereas clinical trials are carefully controlled, there is no formal framework to systematically capture data on the outcomes of off-label use15. This lack of infrastructure contributes to limited and uneven access and hampers the ability of the medical community to learn from off-label treatment efforts15,16. This increases the risk of repeatedly administering ineffective therapies and exposes vulnerable patients to unnecessary toxicity. Taken together, it is essential to find a balance between the search for additional clinical benefit from repurposing approved drugs and protecting patients from harm due to off-label treatments. To address these challenges, DRUP (NCT02925234) was launched in 2016. DRUP aims to provide patients with advanced solid tumours who have exhausted all standard-of-care treatment options with access to approved targeted and immunotherapies outside their label indications, while systematically capturing outcome data to contribute to evidence-based treatment1. Each parallel cohort in this investigator-initiated, multicentre, platform trial is defined by a molecular target, a corresponding study drug and either a specific tumour type or a tumour-agnostic approach (Fig. 1a). The primary end-points are clinical benefit (defined as confirmed objective response or stable disease sustained for at least 16 weeks) and safety. DRUP follows a stepwise design: in stage 1, eight patients are enrolled. If at least one patient experiences clinical benefit, the cohort expands to stage 2 with an additional 16 patients. A cohort is considered successful if 5 or more of the 24 patients in stage 2 show clinical benefit. In such cases, a stage 3 expansion cohort may be initiated to confirm the clinical benefit rate (CBR) in a larger population and to generate evidence that can support label extension or reimbursement17.Fig. 1Trial design and characteristics of included patients.a, Schematic overview of the study and cohort design. For each study drug, a theoretically unlimited number of cohorts can be opened in parallel, depending on the tumour types and targets of submitted patient cases and drug availability. Patients are enrolled in a two-stage design within each cohort. Clinical benefit is defined as either a confirmed complete or partial response, or stable disease lasting at least 16 weeks. After successful completion of stage 2, a stage 3 confirmatory cohort may be opened to validate findings in a larger subgroup, potentially supporting label extension or reimbursement. The sample size for stage 3 depends on the observed CBR in stage 2. Created in BioRender; Voest, E. https://BioRender.com/0lqb5il (2026). b, Bottom, tile plot displaying the distribution of targets and tumour types among evaluable patients following the above study design. Tumour types occurring in less than 1% of the population and targets occurring in less than 0.3% of the population are grouped under ‘Other’. Notably, 130 patients with MSI-H tumours were included in a stage 3 cohort; all other patients are in stage 1 or 2. Top, bar plot highlighting the distinction between different alterations within each target, which often define separate cohorts. Additionally, for several targets, different drugs were used. HRD, homologous recombination deficiency. RAS/RAFwt, RAS or RAF wild type. Trial design and characteristics of included patients. a, Schematic overview of the study and cohort design. For each study drug, a theoretically unlimited number of cohorts can be opened in parallel, depending on the tumour types and targets of submitted patient cases and drug availability. Patients are enrolled in a two-stage design within each cohort. Clinical benefit is defined as either a confirmed complete or partial response, or stable disease lasting at least 16 weeks. After successful completion of stage 2, a stage 3 confirmatory cohort may be opened to validate findings in a larger subgroup, potentially supporting label extension or reimbursement. The sample size for stage 3 depends on the observed CBR in stage 2. Created in BioRender; Voest, E. https://BioRender.com/0lqb5il (2026). b, Bottom, tile plot displaying the distribution of targets and tumour types among evaluable patients following the above study design. Tumour types occurring in less than 1% of the population and targets occurring in less than 0.3% of the population are grouped under ‘Other’. Notably, 130 patients with MSI-H tumours were included in a stage 3 cohort; all other patients are in stage 1 or 2. Top, bar plot highlighting the distinction between different alterations within each target, which often define separate cohorts. Additionally, for several targets, different drugs were used. HRD, homologous recombination deficiency. RAS/RAFwt, RAS or RAF wild type. Here we present results of 1,610 patients treated with 37 different off-label drugs, highlighting their implications for patients, clinicians and researchers. We also identify critical challenges and offer insights to support the development of a sustainable framework for the responsible use of off-label anticancer therapies. Patient population Between 22 July 2016 and 29 May 2024, a total of 2,934 patient cases were submitted to the central study team for review. This count included only cases that were formally considered; preliminary inquiries on cases deemed clearly ineligible were not documented. Of these, 1,621 (55.2%) patients met the eligibility criteria and initiated study treatment. All patients included in the efficacy analysis had at least 4 months of follow-up at data cut-off (1 November 2024); 11 patients who started treatment after 1 June 2024 were excluded. Additionally, 247 (15.2%) of the included patients were non-evaluable, mostly owing to insufficient treatment cycles. A full overview of the accrual and follow-up is presented in Extended Data Fig. 1. Baseline characteristics are summarized in Table 1. Of the 1,363 evaluable patients, 533 (39.1%) had rare cancers (incidence of fewer than 6 cases per 100,000 persons per year18). A total of 103 different tumour types were included, carrying 75 different treatment targets. Microsatellite instability (MSI-H; n = 180, 13.2%), high tumour mutational load (TML-H; n = 138, 10.1%), ERBB2 amplification (n = 93, 6.8%) and BRAF mutations (n = 83, 6.1%) were the most frequently observed. Patients were treated with 37 approved drugs (Extended Data Fig. 2): 347 (25.5%) with immune checkpoint blockade, 162 (11.9%) with poly (ADP-ribose) polymerase (PARP) inhibitors, and 854 (62.6%) with tyrosine kinase inhibitors or targeted monoclonal antibodies.Table 1Baseline characteristicsEvaluable(n = 1,363)Non-evaluable(n = 247)P valueSex, n (%)0.420 Male720 (52.8)123 (49.8) Female643 (47.2)124 (50.2)Age, median [IQR]62 [53–70]62 [54–70]0.468ECOG performance statusa, n (%)<0.001 0467 (34.9)47 (19.5) 1797 (59.5)151 (62.7) 275 (5.6)43 (17.8)Previous treatment lines, n (%)0.958 0164 (12.0)28 (11.3) 1-3977 (71.7)176 (71.3) 4-6183 (13.4)36 (14.6) > 639 (2.9)7 (2.8)Baseline LDHb, median [IQR]221.5 [179.0–297.0]253.0 [188.0–403.0]<0.001Baseline albuminc, median [IQR]39.0 [35.0–42.0]37.0 [31.5–41.0]<0.001Sum of target lesionsd, median [IQR]68.0 [40.0–114.0]81.0 [44.0–132.0]0.006Tumour type, n (%)0.190 Colorectal cancer199 (14.6)31 (12.6) Lung cancer181 (13.3)44 (17.8) Primary brain tumour148 (10.9)28 (11.3) Prostate cancer100 (7.3)8 (3.2) Sarcoma81 (5.9)15 (6.1) Breast cancer77 (5.6)16 (6.5) Bile duct cancer55 (4.0)16 (6.5) Ovarian cancer47 (3.4)12 (4.9) Endometrial cancer45 (3.3)11 (4.5) Head and neck cancer44 (3.2)6 (2.4) Pancreatic cancer34 (2.5)10 (4.0) Skin and adnexal cancer39 (2.9)4 (1.6) Carcinoma of unknown primary39 (2.9)3 (1.2) Oesophageal cancer33 (2.4)7 (2.8) Thyroid cancer32 (2.3)4 (1.6) Cervical cancer25 (1.8)7 (2.8) Bladder cancer27 (2.0)4 (1.6) Gastric cancer21 (1.5)2 (0.8) Neuroendocrine tumour19 (1.4)4 (1.6) Small intestine cancer16 (1.2)4 (1.6) Other101 (7.4)11 (4.5)Rare cancer, n (%)533 (39.1)91 (36.8)0.524Baseline characteristics of all patients that started treatment and all patients that are evaluable for response according to the protocol. Data missing for part of the cohort: a1.9%, b17.7%, c20.9%, d2.1%. For patients evaluated using RANO (response assessment in neuro-oncology) criteria, tumour size is calculated as the product of the two largest perpendicular diameters of each lesion. The square root of the sum of these products across all lesions is used to determine the sum of target lesions to make it comparable with Response Evaluation Criteria for Solid Tumors (RECIST). IQR, Interquartile range; LDH, lactate dehydrogenase. Table 1Baseline characteristicsEvaluable(n = 1,363)Non-evaluable(n = 247)P valueSex, n (%)0.420 Male720 (52.8)123 (49.8) Female643 (47.2)124 (50.2)Age, median [IQR]62 [53–70]62 [54–70]0.468ECOG performance statusa, n (%)<0.001 0467 (34.9)47 (19.5) 1797 (59.5)151 (62.7) 275 (5.6)43 (17.8)Previous treatment lines, n (%)0.958 0164 (12.0)28 (11.3) 1-3977 (71.7)176 (71.3) 4-6183 (13.4)36 (14.6) > 639 (2.9)7 (2.8)Baseline LDHb, median [IQR]221.5 [179.0–297.0]253.0 [188.0–403.0]<0.001Baseline albuminc, median [IQR]39.0 [35.0–42.0]37.0 [31.5–41.0]<0.001Sum of target lesionsd, median [IQR]68.0 [40.0–114.0]81.0 [44.0–132.0]0.006Tumour type, n (%)0.190 Colorectal cancer199 (14.6)31 (12.6) Lung cancer181 (13.3)44 (17.8) Primary brain tumour148 (10.9)28 (11.3) Prostate cancer100 (7.3)8 (3.2) Sarcoma81 (5.9)15 (6.1) Breast cancer77 (5.6)16 (6.5) Bile duct cancer55 (4.0)16 (6.5) Ovarian cancer47 (3.4)12 (4.9) Endometrial cancer45 (3.3)11 (4.5) Head and neck cancer44 (3.2)6 (2.4) Pancreatic cancer34 (2.5)10 (4.0) Skin and adnexal cancer39 (2.9)4 (1.6) Carcinoma of unknown primary39 (2.9)3 (1.2) Oesophageal cancer33 (2.4)7 (2.8) Thyroid cancer32 (2.3)4 (1.6) Cervical cancer25 (1.8)7 (2.8) Bladder cancer27 (2.0)4 (1.6) Gastric cancer21 (1.5)2 (0.8) Neuroendocrine tumour19 (1.4)4 (1.6) Small intestine cancer16 (1.2)4 (1.6) Other101 (7.4)11 (4.5)Rare cancer, n (%)533 (39.1)91 (36.8)0.524Baseline characteristics of all patients that started treatment and all patients that are evaluable for response according to the protocol. Data missing for part of the cohort: a1.9%, b17.7%, c20.9%, d2.1%. For patients evaluated using RANO (response assessment in neuro-oncology) criteria, tumour size is calculated as the product of the two largest perpendicular diameters of each lesion. The square root of the sum of these products across all lesions is used to determine the sum of target lesions to make it comparable with Response Evaluation Criteria for Solid Tumors (RECIST). IQR, Interquartile range; LDH, lactate dehydrogenase. Baseline characteristics Baseline characteristics of all patients that started treatment and all patients that are evaluable for response according to the protocol. Data missing for part of the cohort: a1.9%, b17.7%, c20.9%, d2.1%. For patients evaluated using RANO (response assessment in neuro-oncology) criteria, tumour size is calculated as the product of the two largest perpendicular diameters of each lesion. The square root of the sum of these products across all lesions is used to determine the sum of target lesions to make it comparable with Response Evaluation Criteria for Solid Tumors (RECIST). IQR, Interquartile range; LDH, lactate dehydrogenase. Cohort composition A total of 313 cohorts was initially established in DRUP. At the time of analysis, 209 cohorts remained, as per protocol, several were merged following consultation with and approval of the Independent Data Monitoring Committee (for example, all CDK4/6 alteration cohorts treated with palbociclib and ribociclib were analysed jointly19). These cohorts encompassed 512 unique drug–target–alteration–tumour combinations, reflecting the complexity and substantial heterogeneity inherent to precision oncology. An overview of the target–tumour combinations is shown in Fig. 1b, and outcomes of completed cohorts to date are summarized in Extended Data Table 1. Of the 15 cohorts that had completed stage 2 at the time of analysis, 14 met the protocol-defined success criteria, with at least 5 patients achieving clinical benefit. To date, one stage 3 cohort has been initiated: 130 patients with MSI-H tumours treated with nivolumab in a tumour-agnostic setting, which subsequently supported tumour-agnostic reimbursement for this indication in the Netherlands20. An additional stage 3 cohort (olaparib for BRCA-mutated tumours) is currently in preparation in our European network21. As stage 3 cohorts serve a confirmatory purpose—rather than exploratory as in the other DRUP cohorts—only stage 1/2 outcomes will be included in the subsequent sections to prevent overestimating the overall results. Clinical outcomes Among 1,233 evaluable patients treated in stage 1/2 DRUP cohorts, 17 (1.4%) achieved a confirmed complete response and 177 (14.3%) a confirmed partial response, resulting in an objective response rate (ORR) of 15.7% (95% confidence interval, 13.7–17.9). An additional 236 patients (19.1%) had stable disease for at least 16 weeks, contributing to the CBR—the primary outcome—of 34.9% (95% confidence interval, 32.2–37.6; Fig. 2a). Outcomes and patient characteristics on a per-patient level are included in Supplementary Table 2. Cohort-agnostic differences in baseline characteristics between patients with and without clinical benefit and objective response are shown in Extended Data Tables 2 and 3, respectively. Notably, the extent of prior treatment was significantly associated with a decreased likelihood of both clinical benefit and objective response (P < 0.001), as well as a higher baseline LDH (P < 0.01) and larger tumour burden (calculated as the baseline sum of target lesions, P < 0.01). A higher Eastern Cooperative Oncology Group (ECOG) performance status was associated with decreased chance of clinical benefit, but not objective response (P = 0.019 versus P = 0.513). Overall duration of response (DOR), progression-free survival (PFS) and overall survival were 9.4 months (95% confidence interval, 7.8–13.4), 3.4 months (95% confidence interval, 2.8–3.5) and 8.2 months (95% confidence interval, 7.6–8.8), respectively (Fig. 2b–d).Fig. 2Clinical outcomes across the entire study population and by treatment class.a, Bar plot showing the best overall response in all evaluable patients treated in stage 1/2. CR, complete response; PR, partial response; SD, stable disease. b, Kaplan–Meier curve displaying the DOR in all responders, including the 194 confirmed responders from A and 43 unconfirmed responders, of whom 16 sustained stable disease for at least 16 weeks and 27 did not. ICB, immune checkpoint blockade; TT, targeted therapy. c, Kaplan–Meier curve of PFS. d, Kaplan–Meier curve of overall survival. All plots show outcomes of all patients combined as well as per drug type. Clinical outcomes across the entire study population and by treatment class. a, Bar plot showing the best overall response in all evaluable patients treated in stage 1/2. CR, complete response; PR, partial response; SD, stable disease. b, Kaplan–Meier curve displaying the DOR in all responders, including the 194 confirmed responders from A and 43 unconfirmed responders, of whom 16 sustained stable disease for at least 16 weeks and 27 did not. ICB, immune checkpoint blockade; TT, targeted therapy. c, Kaplan–Meier curve of PFS. d, Kaplan–Meier curve of overall survival. All plots show outcomes of all patients combined as well as per drug type. Although per protocol, non-evaluable patients are excluded from the efficacy analyses in DRUP, a supplementary intention-to-treat analysis of all 1,468 patients that started treatment in stage 1/2 DRUP cohorts yielded an ORR of 13.2% (95% confidence interval, 11.5–15.1), CBR of 29.3% (95% confidence interval, 27–31.7), median PFS of 2.1 months (95% confidence interval, 2.0–2.6) and median overall survival of 6.8 months (95% confidence interval, 6.4–7.4). Figure 3a shows the CBR, ORR and median PFS for the largest tumour type and target subgroups. After multivariable adjustment for tumour type, mutations in MET, BRAF p.V600E mutations, MSI-H, high tumour mutational burden (TMB-H) and BRCA1/2 inactivation were significantly associated with improved CBR and PFS compared with the rest of the cohort, whereas RAS/RAF-wild type, CDK4 amplification, KIT mutation and CDKN2A inactivation were linked to poorer outcomes in both measures. Similarly, after adjusting tumour types for target distribution, patients with breast and pancreatic cancer had significantly worse outcomes compared with other tumour types. Although thyroid cancer and cancer of unknown primary showed superior outcomes, this was not significant after adjusting for targets, suggesting that the effect was primarily driven by favourable targets within those subgroups (Fig. 1b). In thyroid cancer, for example, the CBR was 86.7% in BRAF p.V600E-mutated patients (n = 15) versus 25.0% in others (n = 16; P = 0.001). Additionally, comparison of tumour subgroups should be interpreted cautiously, as differences in CBR and PFS can reflect intrinsic variations in disease biology rather than differences in treatment efficacy.Fig. 3Clinical outcomes across tumour type and molecular target subgroups.a, Clinical outcomes by target (left) and tumour type (right) comprising at least 1% of the population or labelled as ‘Other’. Bottom, CBR (light green) and ORR (dark green); horizontal grey lines indicate CBR (top) and ORR (bottom) across the whole cohort. Top, forest plots display median PFS with 95% confidence intervals (whiskers); the dotted grey line and shaded area indicate the median PFS across the whole cohort and its 95% confidence interval. Asterisks denote subgroups with significantly better or worse outcomes based on multivariable logistic regression (CBR) or Cox models (PFS), adjusted for tumour type and target (reference ‘Other’ for both). Drugs used per included target are summarized in Supplementary Table 3. b, Bar plots showing CBR by tumour type within each drug–target subgroup that included at least one tumour-specific subgroup and a comparator group of at least eight patients. The colour of the bars reflects the CBR values. P values reflect global significance from likelihood ratio tests comparing models with (clinical benefit ~ tumour type) versus without (clinical benefit ~ 1) tumour type. Black dashed lines indicate the overall CBR in the drug–target subgroup. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Clinical outcomes across tumour type and molecular target subgroups. a, Clinical outcomes by target (left) and tumour type (right) comprising at least 1% of the population or labelled as ‘Other’. Bottom, CBR (light green) and ORR (dark green); horizontal grey lines indicate CBR (top) and ORR (bottom) across the whole cohort. Top, forest plots display median PFS with 95% confidence intervals (whiskers); the dotted grey line and shaded area indicate the median PFS across the whole cohort and its 95% confidence interval. Asterisks denote subgroups with significantly better or worse outcomes based on multivariable logistic regression (CBR) or Cox models (PFS), adjusted for tumour type and target (reference ‘Other’ for both). Drugs used per included target are summarized in Supplementary Table 3. b, Bar plots showing CBR by tumour type within each drug–target subgroup that included at least one tumour-specific subgroup and a comparator group of at least eight patients. The colour of the bars reflects the CBR values. P values reflect global significance from likelihood ratio tests comparing models with (clinical benefit ~ tumour type) versus without (clinical benefit ~ 1) tumour type. Black dashed lines indicate the overall CBR in the drug–target subgroup. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Given the lack of consensus on what constitutes durable clinical benefit in end-stage disease, a post hoc analysis was included using a threshold of stable disease for at least 24 weeks (a cut-off also frequently applied in precision oncology trials22,23). Out of 236 patients with stable disease for at least 16 weeks as best overall response, only 123 (52.1%) sustained clinical benefit beyond 24 weeks. Together with the responders, this yielded an adjusted CBR of 25.7% (95% confidence interval, 23.3–28.2). Tumour-agnostic activity DRUP was designed with the understanding that tissue context influences treatment response1. However, factoring in tumour type increased trial heterogeneity from 113 drug–target–alteration subgroups in stages 1/2 to 501 when tumour type was included. To assess its actual effect, we compared outcomes across tumour types within sufficiently sized drug–target subgroups (Fig. 3b). Tumour type significantly affected CBR in 4 out of 17 subgroups (23.5%). For example, in patients with MET amplifications treated with crizotinib, tumour type was a significant predictor of clinical benefit (P = 0.031), with lung cancer showing superior outcomes. Similar tumour type-dependent differences were observed in two PARP inhibitor-treated subgroups (both showing improved outcomes in prostate cancer), and for pembrolizumab in TML-H (worse outcomes in colorectal cancer compared with other tumour types). Exceptional responders To identify which patients showed most benefit from off-label treatment, we defined exceptional responders as those who achieved a confirmed complete response or remained progression-free for at least two years. In the 958 evaluable stage 1/2 patients who started treatment before 1 November 2022 and thus had a minimum follow-up of 2 years on 1 November 2024, 67 (7.0%) were defined as exceptional responders. The targets of the exceptional responders included MSI-H in 21 cases (31.3%), TMB-H or TML-H in 15 cases (22.4%), 15 BRAF p.V600E mutations (22.4%), 4 MET alterations (6.0%; including two alterations leading to MET exon 14 skipping, 1 MET amplification and 1 MET tyrosine kinase domain mutation (p.H1094Y)), 3 ALK fusions (4.5%), 2 BRAC1/2 inactivations (3.0%), 2 NRAS mutations (3.0%; p.G12D and p.Q61R), and 1 (1.5%) FGFR2 fusion, ROS1 fusion and PIK3CA p.H10647R mutation. A swimmer plot of the exceptional responders is shown in Fig. 4.Fig. 4Treatment duration and outcomes of exceptional responders.Swimmer plot showing the time on retreatment and PFS without treatment for all exceptional responders. Stars indicate the first measurements of complete response, diamonds indicate the first measurements of partial response, and circles indicate the first measurement of progressive disease (all per treatment course). Arrows indicate continued treatment after data cut-off and ticks the start of a treatment pause. The plot is annotated with the tumour types and targets. Amp, amplification; loss, biallelic inactivation. Treatment duration and outcomes of exceptional responders. Swimmer plot showing the time on retreatment and PFS without treatment for all exceptional responders. Stars indicate the first measurements of complete response, diamonds indicate the first measurements of partial response, and circles indicate the first measurement of progressive disease (all per treatment course). Arrows indicate continued treatment after data cut-off and ticks the start of a treatment pause. The plot is annotated with the tumour types and targets. Amp, amplification; loss, biallelic inactivation. Safety A total of 699 adverse events with a grade of at least 3 that were deemed at least possibly treatment-related occurred in 458 out of 1,610 patients (28.4%) who started treatment. These included 630 grade 3 events in 436 patients, 58 grade 4 events in 45 patients and 11 grade 5 events in 10 patients (Extended Data Table 4). The prevalence of treatment-related adverse events with a grade of at least 3 for each treatment, shown alongside the corresponding CBR, is presented in Extended Data Fig. 3. In total, 218 out of the 699 treatment-related adverse events (TRAEs) were also serious adverse events (SAEs). Treatment was discontinued owing to adverse events in 151 patients (9.4%). Discussion Over the past nine years, DRUP provided opportunities for off-label treatment to more than 1,600 patients with advanced cancers with no remaining standard therapeutic options. This included a substantial proportion (39.1%) of patients with rare cancers—patients who typically face limited histology-based options and who benefited equally from the precision strategy24. Across patients treated in stage 1/2 cohorts, 34.9% experienced clinical benefit, 15.7% achieved an objective response and 7.0% were exceptional responders. The CBR and ORR of DRUP have remained stable for the duration of the study1,24 and are in line with those reported in other prospective precision oncology basket trials, such as NCI-MATCH (Molecular Analysis for Therapy Choice; ORR 10.3%), MyPathway (ORR 23%) and in meta-analysis (ORR 14%)25–27. Off-label drugs are frequently prescribed outside trial settings and without equivalent oversight, which can result in significant harm. Patients, who are often in fragile conditions after exhausting standard treatments, risk serious toxicity—as seen in DRUP, where 28.4% experienced grade 3 or higher TRAEs—and face uneven access15. For healthcare systems, high drug prices, inconsistent reimbursement and the cost of managing toxicities contribute to financial burden15,16,28,29. These concerns argue for stricter regulation of off-label use. Structured initiatives such as DRUP, TAPUR30 (Targeting Agent and Profiling Utilization Registry), CAPTUR31 (Canadian Profiling and Targeted Agent Utilization Trial) and European DRUP-like clinical trials united under the PCM4EU (Personalised Cancer Medicine for all EU Citizens) and PRIME-ROSE (Precision Cancer Medicine Repurposing System Using Pragmatic Clinical Trials) consortia21 help reduce these risks by granting controlled access to treatment while generating evidence. They offer options for patients with rare cancers, monitor outcomes and identify early signals of activity or, importantly, lack thereof. To illustrate, high submission and inclusion rates for DRUP cohorts targeting alterations in the CDK4/6 pathway through single-agent inhibitors reflected strong interest from clinicians, but efficacy proved limited19. Capturing and publishing such negative results is essential to protect future patients from ineffective strategies. Including negative cohorts in clinical guidelines can further help prevent unnecessary costs, false expectations and avoidable toxicity. Despite the overall modest activity across many cohorts, DRUP also demonstrated that substantial benefit is achievable. Several exceptional responders emerged across different molecular subgroups, illustrating the potential of precision oncology to deliver profound and durable benefit, even in rare, heavily pretreated or difficult-to-treat cancers (for example, eight exceptional responders with BRAF p.V600E-mutated brain tumours). Moreover, one DRUP expansion cohort generated sufficient evidence to achieve full national reimbursement for its off-label indication20, underscoring the idea that structured evaluation can meaningfully expand patient access when efficacy is clear. These positive signals motivated a deeper examination of the variables in our study design that influence patient benefit, including relevance of molecular targets, number of prior treatments, tissue context and cohort expansion. First, the highest-performing targets in DRUP shared a common denominator: a strong biological rationale and prior clinical evidence of activity. This is reflected in classification systems such as the ESMO Scale for Clinical Actionability of molecular Targets (ESCAT)32 or OncoKB33, which currently rank these alterations among the most actionable. For example, BRAF p.V600E, MSI-H and TMB-H are now marked as ESCAT Tier I-C and OncoKB level 1 tumour-agnostically34. Similarly, MET mutations in NSCLC and BRCA1/2 inactivation in prostate, breast and pancreatic cancers are classified as tier I-B/C and level 1 within their respective tumour types34. These findings, supported by other studies6,35, emphasize the value of evidence-based frameworks for target prioritization to optimize benefit from genome-guided therapy. Second, improved outcomes were consistently observed in patients with fewer prior treatment lines, a finding echoed in other precision oncology trials22,36. This highlights the potential benefit of administering precision therapies earlier in the disease course, but should be considered in comparison with the benefit of standard treatment strategies available. Third, we addressed the relevance of tissue context. Following the contrasting outcomes of BRAF inhibition in colorectal cancer and melanoma12,13, DRUP was designed on the premise that tissue context affects treatment response. However, this resulted in many small, tumour-specific, often slow-accruing cohorts, complicating outcome evaluation. As tumour-agnostic inclusion would have simplified the study, we analysed whether tumour type actually influenced outcomes. This was true for 4 out of 17 tested drug–target subgroups (23.5%), which aligns with findings from NCI-MATCH, where tumour-agnostic responses were for instance seen in the BRAF cohort treated with dabrafenib and trametinib, but not for AKT-mutated cancers treated with capivasertib or ipatasertib25. Similarly, real-world data show considerable variation in outcomes for tumour-agnostically approved drugs across tumour types37. These observations suggest that ignoring histology altogether may be premature. Finally, to facilitate seamless translation of positive findings to clinical practice, DRUP includes a third-stage expansion option with a personalized reimbursement model17. After more than 8 years, 14 cohorts formally met the predefined success criteria, but only one expansion cohort has been completed. A critical appraisal of this limited progression from stage 2 to stage 3 identified two main factors. First, some cohorts met the predefined success criteria, but lacked sufficient clinical impact to justify expansion. In hindsight, the stage 2 success threshold in DRUP did not align with the current regulatory standards. The criterion, clinical benefit in 5 out of 24 patients (21.8%), including stable disease for at least 16 weeks, may overestimate efficacy, particularly in indolent tumours. Our post hoc analysis revealed that only half of the patients with stable disease for at least 16 weeks sustained stable disease beyond 24 weeks. Regulatory frameworks such as the PASKWIL-criteria (used by Dutch Health Technology Assessment bodies)38 and the ESMO-MCBS (European Society for Medical Oncology–Magnitude of Clinical Benefit Scale)39 do not consider short-term stable disease a strong efficacy signal. Instead, end-points such as PFS of at least 6 months and durable, high ORR better reflect clinical relevance and reimbursement criteria. To avoid extending cohorts that are unlikely to impact clinical practice and to ensure promising findings receive adequate follow-up, future precision oncology trials may align primary end-points more closely with regulatory standards. A second barrier to expanding stage 2 cohorts is patent life. Often, by the time a cohort concluded, the investigated drug faced patent expiry or market competition, reducing pharmaceutical interest in further trials, even under personalized reimbursement models. As a result, patients who could benefit from these effective off-label therapies are left without access. Besides the aforementioned tumour-specific design, other factors that slowed trial progress included limited adoption of molecular diagnostics in clinical practice and the rarity of certain indications. Unlike NCI-MATCH25, DRUP did not provide molecular testing. As access to comprehensive molecular diagnostics remains uneven, many eligible patients were most likely missed40. Nonetheless, even with full identification, some indications were simply too rare to complete cohorts in a country the size of the Netherlands. Faster progress in such settings requires stronger international collaboration. Since its launch in 2016, DRUP has served as a blueprint for national precision oncology trials across Europe, including in Denmark (ProTarget), Sweden (MEGALiT (Molecularly Guided Anti-Cancer Drug Off-Label Trial)), Norway (IMPRESS-Norway) and Finland (FINPROVE)21. The PCM4EU and PRIME-ROSE projects, supported by European grants, now promote shared protocols, aligned end-points and data sharing, helping to accelerate enrollment and validate promising findings within the commercial lifespan of drugs21. Several limitations of our study should be considered when interpreting these findings. First, many of the drugs evaluated in DRUP were developed before the recent wave of next-generation targeted therapies and immunotherapies, and therefore may not fully reflect the potential of contemporary precision oncology. Similarly, several molecular alterations included in the study are now recognized as weak or non-actionable biomarkers, which helps explain the modest activity observed in some cohorts. Second, although all patients were required to have exhausted standard-of-care treatment options, the level of prior treatment remained heterogeneous. Many patients had received multiple additional lines of therapy, including experimental agents from other trials. This real-world variability reflects the context in which off-label use is typically considered, but may have reduced our ability to detect clinical activity in some settings. As the field evolves, improved biological understanding, more selective and potent drugs, and more sophisticated clinical decision-support tools may largely enhance patient selection and treatment benefit. Our results should therefore not be interpreted as evidence that off-label precision medicines are inherently ineffective, particularly given the presence of multiple exceptional responders across molecularly defined subgroups. Rather, our experiences from 2016 to 2024 strongly argue for confining off-label use to structured, data-generating frameworks until new advancements prove otherwise to allow for a better informed and safer choice of treatment. In conclusion, oncologists often treat patients with off-label anticancer treatment on the basis of a scientific rationale and hope for a positive outcome. However, in the absence of a framework to systematically capture outcomes, this practice carries significant risks of toxicity, costs and inequitable access. Our analysis of more than 1,600 patients treated in DRUP highlights both the challenges and the opportunities for off-label use. Although overall activity across all indications was limited, we observed meaningful clinical benefit in defined subgroups and several exceptional responders. These positive signals underscore that effective off-label use is possible, but that it requires robust evidence, careful biomarker selection and prospective evaluation to ensure stepwise drug evaluation towards potential label expansion. Although better drugs and improved decision-support systems may widen the therapeutic window, our experience indicates that for now, off-label precision medicines should only be used within clinical trial settings that ensure robust data collection and systematic outcome monitoring. Methods Study design DRUP is an ongoing investigator-initiated, multicentre, nonrandomized platform trial designed to offer patients with advanced solid tumours access to off-label targeted or immunotherapies matched to their molecular profile. The adaptive design of the trial allows for the creation of an unlimited number of parallel cohorts, each defined by a molecular target, a corresponding study drug, and either a specific tumour type or a tumour-agnostic approach. DRUP has been approved by the Medical Ethical Committee of the Netherlands Cancer Institute in Amsterdam and the institutional review boards of all 35 participating hospitals in the Netherlands (listed in Supplementary Table 1), making it a national study. The study is conducted in compliance with Good Clinical Practice guidelines and the ethical principles outlined in the Declaration of Helsinki. Patient population Eligible patients are adults diagnosed with advanced or metastatic solid tumours who have exhausted all standard-of-care treatment options according to Dutch Society for Medical Oncology guidelines, have no such options available, or have well-justified reasons for opting out of available standard therapies after consensus agreement by the three principal investigators. In addition to the guideline-defined minimum, patients were allowed to have received any number of additional prior therapies, including experimental agents administered in other clinical trials. Prior molecular diagnostics of any type, performed outside the trial as part of routine clinical care or another study, must have identified a potentially actionable molecular target. In cases of uncertainty, the actionability of the target must be confirmed by consensus between two experienced clinical scientists in molecular pathology. The identified target must be associated with a US Food and Drug Administration or European Medicines Agency-approved therapy that is not approved for the patient’s tumour type, but is available within DRUP. For each treatment included in the trial, predefined matching rules were available (Supplementary Table 4). General inclusion criteria include acceptable organ function, an ECOG performance status of ≤2, disease progression on the last line of systemic treatment, and measurable disease according to RECIST v1.141, or other applicable criteria1. Additional drug-specific inclusion and exclusion criteria apply for each treatment. Study procedures Upon enrollment, patients provide written informed consent. Following confirmation of eligibility, a mandatory fresh-frozen tumour biopsy and a 10-ml blood sample for whole-genome sequencing (performed by Hartwig Medical Foundation as previously described5) is collected, except for patients with exclusively intracranial tumour deposits, when the procedure is considered unsafe, or when whole-genome sequencing was previously obtained without systemic treatment since. Treatment then commences according to the specific protocol for the assigned drug and continues until the occurrence of disease progression, death, unacceptable toxicity or voluntary withdrawal from the trial. Tumour assessments are conducted at baseline and subsequently every two months, transitioning to every three months for patients who remain on study beyond six months. Annual survival follow-ups are conducted after treatment cessation. Study end-points The primary end-points of DRUP are clinical benefit and safety. Clinical benefit is defined as a confirmed complete response or partial response, or stable stable disease sustained for at least 16 weeks. Safety is assessed by monitoring serious and TRAEs of grade ≥3, based on the Common Terminology Criteria for Adverse Events (CTCAE) v4.03, from the time of registration until 30 days after the last treatment administration. The Independent Data Monitoring Committee, blinded to response rates during accrual, oversees safety within the trial. Secondary end-points include objective response, defined as complete response or partial response, DOR, PFS and overall survival. Exploratory end-points focus on extensive post hoc biomarker analyses. Per protocol, all patients who initiate treatment are included in the safety analysis. Patients who fail to complete at least one full treatment cycle (or fewer than two administrations for intravenous medications with cycle lengths below four weeks) are considered non-evaluable and replaced in the efficacy and biomarker analyses. Cohort design Within DRUP, a Simon-like two-stage monitoring plan is employed to identify cohorts demonstrating clinical activity42. Initially, 8 patients are enrolled in stage 1. If at least one patient shows clinical benefit, an additional 16 patients are recruited, forming a 24-patient stage 2 cohort. In stage 2, five or more patients with clinical benefit classify the cohort as successful. The null and alternative hypotheses are defined as a CBR of 10% versus ≥30%. This monitoring rule has 85% power, with a one-sided alpha error rate of 7.8%. Following a successful stage 2 cohort, a stage 3 expansion cohort may be considered to confirm the CBR in a larger patient population and support potential label extension or reimbursement. This stage can include a personalized reimbursement model, in which the manufacturer covers treatment costs for the first 16 weeks. If clinical benefit persists beyond this period, financial responsibility shifts to the healthcare insurer17. Statistical analysis Patient characteristics, tumour responses, and TRAEs were summarized using descriptive statistics. Exact 95% confidence intervals for CBR and ORR were calculated using the Clopper–Pearson method. Time on treatment, DOR, PFS and overall survival were estimated using the Kaplan–Meier method. Univariate associations between clinical outcomes and baseline characteristics were assessed using Fisher’s exact test (categorical variables), the Wilcoxon rank-sum test (continuous variables), and the linear-by-linear association test (ordinal variables). To evaluate independent associations with clinical outcomes, multivariable logistic regression (for CBR) and Cox proportional hazards models (for PFS) were performed, including tumour type and molecular target as covariates. This approach accounted for potential confounding, as some either favourable or unfavourable targets were overrepresented in certain tumour types. Additionally, within drug–target subgroups containing at least one tumour-specific and one comparator group with at least eight patients each, logistic regression was used to evaluate the effect of tumour type on outcomes. CBR was modelled as a function of tumour type, and global P values were obtained using likelihood ratio tests comparing the full model (clinical benefit ~ tumour type) with the null model (clinical benefit ~ 1). All P values < 0.05 were considered statistically significant. Analyses were conducted using R v4.4.3. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10405-x. Supplementary information Supplementary InformationThis file contains Supplementary Tables 1 and 3. Supplementary Table 1 provides a complete list of all participating study sites and the corresponding principal investigators. Supplementary Table 3 complements Fig. 3a by providing a complete overview of the drugs used for each target shown in the figure, together with the exact number of patients treated with each drug. Reporting Summary Supplementary Table 2Per patient characteristics of the included patients. This table provides a comprehensive overview of patient-level data for all patients who initiated treatment in the trial. Patients are identifiable through their unique SubjectKey, which can also be used to request additional data for a specific patient through a data access request, as described in ‘Data availability’. For each patient, the table includes sex, age at consent, WHO performance status, number of previous lines of systemic treatment, tumour type, treatment target, drug administered, cohort assignment, treatment start date, best overall response, presence of clinical benefit and objective response, reason for treatment discontinuation, and progression-free and overall survival times and events. Supplementary Table 4Available drugs and matching rules. This table provides an overview of all drugs that were available at any point during the study. To be eligible for a given treatment, patients were required to meet at least one of the listed inclusion criteria and none of the exclusion criteria. For all drugs, the eligibility of individual mutations was determined in consultation with the molecular expert board, consisting of two experienced clinical scientists in molecular pathology. In addition, patients who were eligible for on-label treatment with any of the listed drugs were always excluded from the trial. Peer Review File This file contains Supplementary Tables 1 and 3. Supplementary Table 1 provides a complete list of all participating study sites and the corresponding principal investigators. Supplementary Table 3 complements Fig. 3a by providing a complete overview of the drugs used for each target shown in the figure, together with the exact number of patients treated with each drug. Reporting Summary Per patient characteristics of the included patients. This table provides a comprehensive overview of patient-level data for all patients who initiated treatment in the trial. Patients are identifiable through their unique SubjectKey, which can also be used to request additional data for a specific patient through a data access request, as described in ‘Data availability’. For each patient, the table includes sex, age at consent, WHO performance status, number of previous lines of systemic treatment, tumour type, treatment target, drug administered, cohort assignment, treatment start date, best overall response, presence of clinical benefit and objective response, reason for treatment discontinuation, and progression-free and overall survival times and events. Available drugs and matching rules. This table provides an overview of all drugs that were available at any point during the study. To be eligible for a given treatment, patients were required to meet at least one of the listed inclusion criteria and none of the exclusion criteria. For all drugs, the eligibility of individual mutations was determined in consultation with the molecular expert board, consisting of two experienced clinical scientists in molecular pathology. In addition, patients who were eligible for on-label treatment with any of the listed drugs were always excluded from the trial. Peer Review File