Adjuvant endocrine therapy uptake, toxicity, quality of life, and prediction of early discontinuation

Abstract

Background

Many patients receiving adjuvant endocrine therapy (ET) for breast cancer experience side effects and reduced quality of life (QoL) and discontinue ET. We sought to describe these issues and develop a prediction model of early discontinuation of ET.

Methods

Among patients with hormone receptor–positive and HER2-negative stage I-III breast cancer of the Cancer Toxicities cohort (NCT01993498) who were prescribed adjuvant ET between 2012 and 2017, upon stratification by menopausal status, we evaluated adjuvant ET patterns including treatment change and patient-reported discontinuation and ET-associated toxicities and impact on QoL. Independent variables included clinical and demographic features, toxicities, and patient-reported outcomes. A machine-learning model to predict time to early discontinuation was trained and evaluated on a held-out validation set.

Results

Patient-reported discontinuation rate of the first prescribed ET at 4 years was 30% and 35% in 4122 postmenopausal and 2087 premenopausal patients, respectively. Switching to a new ET was associated with higher symptom burden, poorer QoL, and higher discontinuation rate. Early discontinuation rate of adjuvant ET before treatment completion was 13% in postmenopausal and 15% in premenopausal patients. The early discontinuation model obtained a C index of 0.62 in the held-out validation set. Many aspects of QoL, most importantly fatigue and insomnia (European Organization for Research and Treatment of Cancer QoL questionnaire 30), were associated with early discontinuation.

Conclusion

Tolerability and adherence to ET remains a challenge for patients who switch to a second ET. An early discontinuation model using patient-reported outcomes identifies patients likely to discontinue their adjuvant ET. Improved management of toxicities and novel more tolerable adjuvant ETs are needed for maintaining patients on treatment.

Breast cancer remains the most common cancer among women (1). The majority (70%) of patients present with hormone receptor–positive and HER2-negative tumors and benefit from adjuvant endocrine therapy (ET), either tamoxifen or aromatase inhibitors (AI) (2). Standard duration of adjuvant ET is 5 years. However, extended adjuvant ET for 7 or 10 years clearly showed more favorable outcomes in patients with high-risk breast cancer (3,4). Conversely, early discontinuation of ET before 3 years as well as nonadherence lead to shorter disease-free survival among all patients (5,6).

Prior studies showed that approximately 20%-30% of patients prematurely discontinue ET before the first 3 years of therapy (7,8). Nonadherence is a complex phenomenon that is likely to also be impacted by the numerous adverse effects associated with sustained estrogen deprivation such as fatigue, musculoskeletal symptoms (eg, joint pain), and gynecological symptoms (eg, hot flashes, night sweats, vaginal dryness) (9,10). Previous analysis on the Cancer Toxicities (CANTO) cohort underlined the persistent deterioration of quality of life (QoL) 2 years after breast cancer diagnosis in patients receiving ET, with a negative impact on multiple domains of daily life (11). These adverse effects may be mitigated by targeted clinical management strategies aimed to improve the QoL of breast cancer survivors and promote adherence to ET (12,13).

However, adverse effects are not the only determinants of nonadherence. Prior studies highlighted psychological determinants such as attitude toward ET and belief in ET efficacy as being consistently associated with adherence (14,15).

In this study, we aimed to describe real-world patterns of ET uptake including patients’ discontinuations of ET, toxicities, and QoL up to 5 years from ET initiation. We also developed a predictive model of early ET discontinuation. To achieve both aims, we used CANTO (NCT01993498), a multicenter, nationwide, prospective cohort study of 10 150 women with stage I-III breast cancer that aims to describe the toxicities of cancer treatment and their impact on QoL over 5 years after the end of primary treatment (16).

Methods

Data source

We used data from a prospective cohort of women diagnosed with stage I-III primary breast cancer (CANTO; NCT01993498). Data were prospectively collected in 26 French investigational sites at 5 time points: at inclusion just after breast cancer diagnosis (T0); after primary treatment (primary surgery, chemotherapy, or radiotherapy) and a few months after ET initiation (T1); 1 year after T1 (T2); 3 years after T1 (T3); and 5 years after T1 (T4) (see Figure 1). The CANTO study design was previously described (16). All patients provided written informed consent. The study was approved by the ethics committee (registration number ID-RCB : 2011-A01095-36,11-039).

Figure 1.

Design of the CANTO study. Median duration and interquartile range of the different time intervals in the 6488 patients eligible for descriptive analyses are given. CANTO = Cancer Toxicity study; ET = endocrine therapy; PRO = patient-report outcomes; T0 = just after breast cancer diagnosis; T1 = few months after ET initiation; T2 = 1 year after T1; T3 = 3 years after T1; T4 = 5 years after T1.

Study cohort

The study included 9630 women enrolled in the CANTO cohort between March 21, 2012, and February 7, 2017, and followed until January 2022. We restricted our analysis to patients with hormone receptor–positive and HER2-negative invasive breast cancer who initiated AI or tamoxifen after 2012 (n = 6488). Among that population, 4122 patients were postmenopausal, 2087 were premenopausal, and 279 had unknown hormonal status. Menopausal status was assessed at inclusion. A total of 162 patients declared premenopausal at diagnosis received AI without ovarian suppression and thus were classified as unknown menopausal status.

Finally, upon exclusion of patients with incomplete data collection at 1 or multiple time points, the population of interest included 5282 women (see Figure 2). This population was randomly split into 2 sets: 4225 participants in the training set and 1057 in the validation set. This split between the training and validation set was stratified on occurrence of early discontinuation events to ensure an adequate number of events in the validation set.

Figure 2.

Flowchart for descriptive analyses and machine-learning model development. CANTO = Cancer Toxicity study; ET = endocrine therapy; LHRH = luteinizing hormone release hormone; QoL = quality of life; T1 = few months after ET initiation.

Variables and outcomes

The outcome of interest was time to early discontinuation defined as the time from first ET initiation to final ET discontinuation with no subsequent therapy recorded in the electronic Case Report Form (eCRF). Patients whose reason for discontinuation was treatment completion were censored at the time of discontinuation. Patients who continued treatment after 5 years were censored at that time. For the machine-learning model development, because most patients’ descriptors were measured at T1 (a few months after ET initiation), time to early discontinuation was defined using T1 as the index date instead of time of first ET initiation. ET discontinuation was collected from study eCRFs that captured patients’ statements on ET uptake at the prespecified time points.

Secondary outcomes were time to discontinuation of the first, second, or third ET. Time to first ET discontinuation was defined as the time between first ET initiation to first ET final discontinuation as recorded in the eCRF. Temporary interruptions were not considered as discontinuations. Similarly, time to second or third discontinuation was defined as the time between second or third ET initiation and second or third ET discontinuation. For all outcomes, participants who were lost to follow-up, progressed, or died were censored at the time of the event. There are a variety of reasons for discontinuation including toxicity, patient decision, and medical decision including switch for change in menopausal status.

Toxicities and patient-reported outcomes (PROs) were collected at baseline, T1, T2 (T1 + 1 year), T3 (T1 + 3 years), and T4 (T1 + 5 years). A set of physical toxicities was collected by a clinical research nurse during a face-to-face examination using the Common Toxicity Criteria Adverse Events Scale (CTCAE), version 4.0 (17). QoL was evaluated using the European Organization for Research and Treatment of Cancer QoL questionnaire (QLQ-C30) and breast cancer module (QLQ-BR23) (18,19). Anxiety and depression were assessed using the Hospital Anxiety and Depression Scale (HADS) (20). We considered these variables separately in postmenopausal and premenopausal participants.

Statistical analysis

Kaplan–Meier (KM) curves stratified by postmenopausal and premenopausal participants were estimated for time to early discontinuation as well as time to first, second, and third ET discontinuation.

Proportion of toxicities and mean PRO scores were estimated with corresponding 95% confidence intervals (CIs) at inclusion (PRO only) at T1, T2, and T3 for postmenopausal and premenopausal participants. We then estimated toxicities and PRO scores according to the number of previous switches from one ET to another (ie, discontinuation of an ET followed by initiation of a new ET). The 3 populations with different treatment patterns considered were 1) participants still on their first ET at T3, 2) participants who discontinued a first ET and initiated a second but not a third ET before T3, and 3) participants who initiated a third ET before T3.

Confidence intervals for proportion are computed with the Clopper–Pearson exact method. The threshold for statistical significance is .05, and all tests are 2-sided. Analyses were performed using SAS 9.4 and R.

Machine-learning model

We identified 189 variables available at T1 to predict time to early discontinuation. Dependent variables included 47 variables measured at inclusion (medical history, medical status), TNM staging, 13 treatment variables (describing surgery, chemotherapy, radiotherapy, first ET drug, and number of ET discontinuations before T1), and 128 variables measured at T1 including 74 toxicities and 25 PRO scores (QLQ-C30, QLQ-BR-23, and HADS) as well as 25 changes in PRO scores between inclusion and T1. Rare toxicities with less than 1% occurrences were excluded.

On the whole dataset (training and validation set combined) and before imputing missing data, we performed a univariate analysis of association between early discontinuation and each individual variable using a Cox regression. We corrected for multiple tests using the Bonferroni correction on the 189 univariate tests. This univariate analysis was not used to guide machine-learning model development. The list of variables and the results of the univariate analysis are available in Supplementary Table 4 (available online).

Missing data was imputed using Missforest in the training set and was imputed to the median value of the training set in the validation set (21).

We considered 2 complementary machine-learning models: L2-penalized Cox regression and gradient boosted trees with a Cox partial likelihood (XGBoost) (22,23). Although linear models are simple and interpretable, tree-based models can capture interactions between variables that can bring higher performance. Variables were standardized for the linear model and for the univariate analysis.

We performed nested cross-validation to evaluate which model performed best on the training set, using the concordance index metric (C index) with 0.5 corresponding to random predictions and 1 to perfect discrimination (24). We then evaluated the resulting model on the validation set. We trained models on the full dataset of 189 variables as well as on 51 variables: the 50 PRO and the number of previous discontinuations at T1.

Finally, we evaluated the models on the validation set. We illustrated the performance of the best compact model by estimating KM curves of time to early discontinuation in the validation set, stratified by tertiles of the hazard ratios output by the model.

Results

Treatment patterns and ET discontinuations

Median follow-up time from the initiation of the first ET was 4.8 years (1st quartile [Q1]-3rd quartile [Q3]: 3.3-5.4 years). In the postmenopausal population, the first ET was a nonsteroidal AI (ie, anastrozole or letrozole) for 87% of women, 49% of patients who switched to a second ET received the steroidal AI exemestane, and tamoxifen was the most prevalent choice of third ET. In the premenopausal population, the first ET was tamoxifen for 98% of the women, an AI was prescribed to 87% of patients who switched to a second ET, and AI and tamoxifen were equally prescribed as the third ET (Table 1, Figure 3).

Table 1.

Population characteristics and treatment landscape separately for postmenopausal and premenopausal participants

Characteristics	Postmenopausal patients	Premenopausal patients	P
	(n = 4122)	(n = 2087)
Age, median (IQR), y	64.0 (58.6-68.9)	46.2 (42.2-49.7)	<10-10
Diagnostic method, No. (%)			<10-10
Clinical examination	968 (24.3)	1022 (50.7)
Organized screening program	2105 (52.9)	478 (23.8)
Individual screening (opportunistic radiological tests)	906 (22.8)	516 (25.5)
Missing values	143	71
Stage, No. (%)			<10-10
Stage I	2235 (54.2)	920 (44.1)
Stage II	1559 (37.8)	901 (43.2)
Stage III	328 (8.0)	266 (12.7)
Initial treatment, No. (%)a
Conservative surgery	3417 (82.9)	1518 (72.7)	<10-10
Mastectomy	896 (21.7)	700 (33.5)	<10-10
Sentinel node dissection	3387 (82.4)	1517 (72.8)	<10-10
Axillary dissection	1314 (32.0)	921 (44.2)	<10-10
Radiotherapy	3750 (91.0)	1908 (91.5)	.59
Chemotherapy	1421 (34.5)	1204 (57.7)	<10-10
First ET, No. (%)			<10-10
Tamoxifen	324 (7.9)	2038 (97.7)
Anastrozole	1507 (36.6)	6 (0.3)
Letrozole	2075 (50.3)	14 (0.7)
Exemestane	216 (5.2)	11 (0.5)
With LHRH agonists	0 (0)	29 (1.4)
Second ET, No.	958	540	<10-10
Tamoxifen, No. (%)	152 (15.9)	36 (6.7)
Anastrozole, No. (%)	159 (16.6)	140 (26.0)
Letrozole, No. (%)	172 (18.0)	248 (45.9)
Exemestane, No. (%)	472 (49.3)	84 (15.6)
With LHRH agonists, No. (%)	1 (0.1)	49 (9.1)
Only LHRH agonists, No. (%)	1 (0.1)	31 (5.7)
Unspecified other ET, No. (%)	0 (0)	1 (0.2)
Third ET, No.	268	150	.028
Tamoxifen, No. (%)	94 (35.1)	67 (44.7)
Anastrozole, No. (%)	54 (20.1)	27 (18.0)
Letrozole, No. (%)	58 (21.6)	16 (10.7)
Exemestane, No. (%)	60 (22.4)	31 (20.7)
With LHRH agonists, No. (%)	0 (0)	10 (6.7)
Only LHRH agonists, No. (%)	0 (0)	8 (5.3)
Fulvestrant, No. (%)	2 (0.7)	0 (0)
Unspecified other ET, No. (%)	0 (0)	1 (0.7)

^aThe different primary and node surgeries sum above 100% because of surgical resumption. Tests for equality of distribution of ET were done on the 4 main ETs. ET = endocrine therapy; LHRH= luteinizing hormone-releasing hormone; IQR = interquartile range.

Figure 3.

Repartition of ET received in postmenopausal patients and premenopausal patients according to first, second, or third ET prescribed. ET = endocrine therapy; LHRH = luteinizing hormone-releasing hormone agonists (goserelin, leuprorelin).

The definitive early discontinuation (ie, patients who discontinue ET before treatment completion) rate at 5 years was 12.7% (95% CI = 13.8% to 11.4%) and 14.7% (95% CI = 12.9% to 16.5%) in postmenopausal and premenopausal patients, respectively. Discontinuation rate of the first adjuvant ET at 4 years was 30.1% (95% CI = 28.6% to 31.6%) in postmenopausal and 34.7% (95% CI = 32.5% to 36.9%) in premenopausal patients. Among patients who started a second ET, the 1-year discontinuation rate on this second prescribed adjuvant ET was 33.2% (95% CI = 30.0% to 36.2%) in postmenopausal and 31.2% (95% CI = 26.9% to 35.2%) in premenopausal patients. Among patients who started a third ET, the 1-year discontinuation rate on the third adjuvant ET was 35.0% (95% CI = 28.7% to 40.7%) in postmenopausal and 30.4% (95% CI = 22.1% to 37.8%) in premenopausal patients (Figure 4).

Figure 4.

Discontinuations of ET. A) Early discontinuation before treatment completion. B) Discontinuation of the first prescribed ET. C) Discontinuation of the second prescribed ET. D) Discontinuation of the third prescribed ET. The y axis for early discontinuation (A) matches that of Figure 7, A. ET = endocrine therapy.

Toxicities and PROs according to menopausal status

Prevalence of toxicities in the descriptive analyses population as well as in the postmenopausal and premenopausal subgroups at T1, T2, T3, and T4 are reported in Supplementary Table 1 (available online). Mean PRO scores in the same populations at study inclusion, T1, T2, T3, and T4 are reported in the same table.

Joint pain, gynecological infections, and hypertension were more frequent for postmenopausal than premenopausal patients at T3. On the other hand, premenopausal women experienced more hot flashes, night sweats, attention disorders, headaches, and leukorrhea compared with counterparts (Figure 5).

Figure 5.

Toxicities by menopausal status. Prevalence (and 95% confidence interval) of selected toxicities (CTCAE) at T3 in the post- and premenopausal populations. CTCAE = Common Toxicity Criteria Adverse Events Scale; T3 = 3 years after T1.

Postmenopausal and premenopausal patients reported similar overall QoL and health and systemic therapy side effects as captured by the PRO questionnaire at T3. Premenopausal patients experienced better physical and sexual functioning as well as sexual enjoyment but more fatigue.

Toxicities and PROs according to previous discontinuations of ET

At T3, patients who have switched ET suffered from worse toxicities in most categories (Figure 6; Supplementary Table 2, available online). They also reported reduced QoL, health, and role functioning and worsened insomnia, fatigue, and pain. This is particularly true in postmenopausal patients; the results are less conclusive for premenopausal patients.

Figure 6.

Prevalence of toxicities (CTCAE) by category and selected PRO (QLQ-C30) scores by ET sequences separately for post- and premenopausal patients. A) Toxicities in postmenopausal patients. B) Toxicities in premenopausal patients. C) PROs in postmenopausal patients. D) PROs in premenopausal patients. CTCAE = Common Toxicity Criteria Adverse Events Scale; ET = endocrine therapy; GI = gastrointestinal; QLQ-C30 = European Organization for Research and Treatment of Cancer QoL questionnaire 30; PRO = patient-reported outcome; QoL = quality of life; T3 = 3 years after T1.

Performance of the early discontinuation predictive models

Early discontinuation was observed for 416 of 4225 patients in the training set and 105 of 1057 in the validation set.

Using the 189 variables on the training set, the penalized Cox regression and the XGBoost model obtained a C index of 0.64 in cross-validation. Using the 51 selected variables (PRO and previous discontinuations) on the training set, the penalized Cox regression and the XGBoost model obtained a C index of 0.62 in cross-validation.

Finally, when applied on the held-out validation set, we obtained a C index of 0.62 (95% CI = 0.56 to 0.68), 0.62 (95% CI = 0.56 to 0.68), 0.61 (95% CI = 0.56 to 0.66), and 0.60 (95% CI = 0.55 to 0.66) with the penalized Cox regression on the full variable set, XGBoost on the full variable set, the penalized Cox regression on the selected variables, and XGBoost on the selected variables, respectively.

The KM curves of time to early discontinuation in the validation set, stratified by tertiles of the predictions of the penalized linear model based on selected variables, illustrate the discriminative ability of the model (Figure 7, A). The definitive early discontinuation rate at 5 years after T1 for patients in the lowest tertile of risk was 8.7% (95% CI = 5.2% to 12.1%) compared with 11.9% (95% CI = 8.2% to 15.5%) for the medium-risk patients, and 16.1% (95% CI = 11.4% to 20.5%) for the high-risk patients.

Figure 7.

Performance of the predictive model and variable importance. A) Early discontinuation in the validation set stratified by tertiles of the penalized linear model based on QoL variables and ET discontinuations before T1. B) Top 10 coefficients of the penalized linear model based on QoL variables and ET discontinuations before T1. All PRO scales are measured at T1. Changes in PRO scales are computed between T1 and inclusion. Variables that are statistically significant at the Bonferroni level in the univariate analysis are written in bold. B23 = European Organization for Research and Treatment of Cancer breast cancer module; C30 = European Organization for Research and Treatment of Cancer QoL questionnaire; coeff = coefficients; ET = endocrine therapy; PRO = patient-reported outcome; QoL = quality of life; T1 = few months after ET initiation.

Determinants of definitive early discontinuation

The list of the 189 variables used to develop the predictive model of early discontinuation and the univariate association results are reported in Supplementary Table 4 (available online). There were 25 variables statistically significantly associated with early discontinuation after the Bonferroni correction. The top association was that patients who have already discontinued an ET before T1 were more likely to discontinue ET definitively. As measured by QLQ-C30, higher QoL, physical functioning, role functioning, social functioning, emotional functioning, and cognitive functioning scores at T1 as well as improvement in social and emotional functioning between inclusion and T1 were associated with lower rates of early ET discontinuation. Receiving mastectomy and radiotherapy was also associated with lower rates of early ET discontinuation. As measured by QLQ-C30 or QLQ-BR23, fatigue, pain, dyspnea, insomnia, systemic therapy side effects, breast symptoms, and arm symptoms at T1 as well as increase in insomnia and systemic therapy side effects between inclusion and T1 were associated with higher rate of early ET discontinuation. Similarly, peak muscular pain, muscular pain at rest, presence of gastrointestinal symptoms, CTCAE grade of headache, and nausea as well as the sum of the CTCAE grade of all reported adverse events as collected and measured by the clinical research nurse were associated with higher likelihood of early ET discontinuation.

When we considered the top 10 coefficients in the penalized linear model based on QoL variables and ET discontinuations before T1 (Figure 7, B), 7 variables were also statistically significant at the Bonferroni level.

Discussion

In this study, we have shown that the majority of patients who switched to another ET continued to experience major toxicities and remain at high risk of discontinuing the subsequent ETs and that PROs (QLQ-C30, -BR23, and HADS) capture well the determinants of early discontinuation.

In clinical practice, when the burden of side effects heavily affect a patient’s QoL, it is common to change the ET in an attempt to mitigate treatment toxicity. However, we showed that ET side effects continued to take their toll on patients who changed ET, and thus, patients were likely to discontinue the second and potentially third ET prescribed by physicians, with a higher likelihood of definitive early ET discontinuation for patients who had multiple ET changes in a fairly short time. Therefore, our findings highlight that ET toxicity management is still an unmet medical need, and the development of more tolerable ET is eagerly awaited.

Two prior studies evaluated the effect of ET switch on treatment toxicity and reported limited persistence on the new treatment. Indeed, in the articular tolerance of letrozole (ATOLL) study, only 71.5% of patients at 6 months from the switch to letrozole from anastrozole were still on treatment with the persistence of mainly musculoskeletal symptoms; likewise in the Exemestane and Letrozole Pharmacogenetics (ELPh) trial, only 38.6% were still on the alternate AI with a median follow-up of 13.7 months (25,26). Therefore, although switching from one ET to another allows the patients to be treated longer, it is not ideal in terms of mitigating ET side effects in the long-term. Patients who have needed to switch ET might be more susceptible to ET toxicities or there might be a cumulative effect of toxicity when trialing different ETs, but in any case, these patients experience impaired QoL.

Furthermore, our analysis expands on prior knowledge of factors associated with ET discontinuation, as it provides a robust predictive model of early permanent discontinuation of ET before treatment completion. This early discontinuation model, although it requires an external validation, has the potential of identifying patients who are likely to discontinue their adjuvant ET and that may benefit from personalized interventions that reduce ET dropout and improve survival outcomes.

We found that PROs (QLQ-C30, -BR23, and HADS), increasingly implemented in routine medical care (27), are adequate tools to capture the determinants of early discontinuation a few months after ET initiation. Thus, they represent a useful descriptor set for machine-learning models aiming to predict important aspects of patient drug uptake behavior. Interestingly, some actionable ET side effects such as headaches and nausea were also associated with discontinuation. These symptoms can be easily assessed in clinical practice and in most cases are manageable with appropriate medical or nonmedical interventions (12,13).

We also described other factors, beyond ET symptoms, such as prior treatments that were also associated with ET uptake. For example, we found that mastectomy followed by radiotherapy was associated with lower risk of ET discontinuation, as it is likely that such patients are more aware of the disease severity and related risk of recurrence.

We also observed that premenopausal and postmenopausal patients experience different ET toxicities although age may also contribute to those differences. Consistently with a previous study comparing tamoxifen side effects with exemestane side effects in premenopausal women (28), premenopausal patients who are mainly receiving tamoxifen experience more hot flashes, night sweats, and leukorrhea, and postmenopausal patients receiving AIs experience more joint pain. Despite those differences, premenopausal and postmenopausal patients reported similar overall QoL.

Some limitations should be acknowledged. First, treatment uptake was not measured by patients’ diaries, pill counts, or validated treatment-adherence questionnaires, and we relied only on patients’ self-declarations collected by clinical research nurses, which can overestimate treatment uptake (29). Second, although CANTO is an observational cohort, it cannot be considered a complete real-world population as patients are aware that they are observed over time. This can influence their behavior and treatment uptake. In addition, CANTO is a French cohort, and findings might not apply to other countries. Finally, we excluded 16% (1016 of 6488) of patients for model development because they missed visit 1 or the associated questionnaires, but reassuringly, these excluded patients had a similar early discontinuation rate than the included patients (Supplementary Methods, Supplementary Table 3, Supplementary Figure 1, available online).

In conclusion, using data from one of the largest prospective cohorts of patients with early breast cancer, we described how ET toxicities impact patients’ QoL in the long term, often leading to switch of ET without successful results on treatment uptake and QoL outcomes. However, we highlighted that predictive models of early discontinuation based on PROs can be used to prioritize patients in need of additional support to remain on ET. Modifiable risk factors of treatment dropout can also be corrected. Predictive models of ET discontinuation could enable the early identification of at-risk patients that require personalized interventional programs for managing treatment toxicities and more tolerable adjuvant ETs to improve their daily QoL and survival outcomes.

Data availability

Access to data from the CANTO cohort is subject to limitations due to Institutional Review Board constraints. Requests for data access should be directed to UNICANCER.

Author contributions

Félix Balazard, PhD (Data curation; Formal analysis; Methodology; Software; Validation; Visualization; Writing - original draft; Writing—review & editing), Ines Vaz-Luis, MD, PhD (Conceptualization; Investigation; Supervision; Writing—review & editing), Alexandra Jacquet, PhD (Project administration; Writing—review & editing), Anne-Laure Martin, PharmD (Conceptualization; Supervision), Catherine Coulouvrat, MD (Conceptualization; Methodology), Megan Rice, PhD, MPH (Conceptualization; Methodology), Medha Sasane, PharmD, PhD (Conceptualization), Keri J.S. Brady, PhD, MPH (Conceptualization; Validation), Mario Campone, MD, PhD (Investigation), Marie-Ange Mouret-Reynier, MD (Investigation), Christelle Lévy, MD (Investigation); Laurence Vanlemmens, MD (Investigation), Olivier Tredan, MD, PhD (Investigation), Christina Herold, MD (Conceptualization; Supervision; Writing—review & editing), Paul-Henri Cottu, MD, PhD (Investigation), Patrick Soulie, MD (Investigation), Christelle Jouannaud, MD (Investigation), Marion Fournier, MD (Investigation), Charles Coutant, MD, PhD (Investigation), Olivier Rigal, MD (Investigation), Asma Dhaini Merimeche, MD (Investigation), Gautier Paux, MSc (Conceptualization; Methodology), Nathalie Briot, MSc (Data curation; Formal analysis; Software), Julie Blanc, MSc (Data curation; Formal analysis; Software), Stéphane Mulard, MSc (Formal analysis; Project administration; Software; Writing—original draft), Elise Bordet, PhD (Conceptualization; Project administration), Aurélie Bertaut, MD, PhD (Data curation; Methodology; Writing—original draft; Writing—review & editing), Florence Lerebours, MD (Investigation), and Barbara Pistilli, MD (Conceptualization; Investigation; Supervision; Writing—review & editing).

Funding

The analysis presented here was funded by Sanofi. Sanofi designed the analysis. However, Sanofi had no role in the design of the CANTO cohort.

The CANTO cohort is supported by the French government under the Investment for the Future Program, which is managed by the National Research Agency, Grant No. ANR-10-COHO0004 and by the Ligue Nationale Contre le Cancer.

Conflicts of interest

FB: Salary; Owkin. AB: None. EB: Salary; Sanofi. SM: Salary; Owkin. JB: None. NB: None. GP: Salary; Sanofi. ADM: None. OR: Consulting Fees (eg, advisory boards); Pfizer, Pierre Fabre. C. Coutant: Consulting Fees (eg, advisory boards); ROCHE, MSD, AstraZeneca, Seagen. MF: Other; Symposium fees for Myriad genetics, May 2021, 1000€, for the SFCO. CJ: Consulting Fees (eg, advisory boards); Daiichi Sankyo, AstraZeneca, Pfizer. Other; travel/accommodation/expense from Novartis. PS: None. FL: Royalty; AstraZeneca, Eisai, Eli Lilly, Novartis, Pierre Fabre. Consulting Fees (eg, advisory boards); Advisory board honoraria: AstraZeneca, Eisai, Genomic Health, Eli Lilly, Pierre Fabre, Roche. Other; Travel/Accomodation/Expenses: AstraZeneca, Eisai, Novartis, Pfizer, Pierre Fabre, Roche. PC: Consulting Fees (eg, advisory boards); Honoraria: Pfizer, Roche, Lilly, Pierre Fabre (as well as Novartis and NanoString Technologies to author’s institution), Consulting or advisory role: Roche/Genentech, Pfizer, Lilly. Fees for Non-Continuing Medical Education (CME) Services Received Directly from Commercial Interest or their Agents (eg, speakers’ bureaus); Novartis, Pfizer. Other; Travel, Accommodations, Expenses: Roche and Pfizer. OT: Royalty; Novartis-Sandoz, Pfizer, Lilly, Astra-Zeneca, Daiichi Sankyo, Eisai, Pierre Fabre, Seagen, Roche, MSD-Merck. Contracted Research; Roche, BMS, MSD-Merck. LV: None. CL: None. MMR: Contracted Research; Pfizer, Roche, MSD, Lilly, Novartis, Astra-Zeneca, Myriad. MC: Consulting Fees (eg, advisory boards); Honoraria: Lilly-Accord-GT1-Pfizer, Speakers’ bureau: Novartis. Other; Consulting or advisory role to the author’s institution: AstraZeneca, Sanofi, Servier, Abbvie, Accord, Eli Lilly, Pfizer. KB: Salary; Sanofi. MS: Salary; Sanofi. MR: Salary; Sanofi. CC: Salary; Sanofi. AM: None. AJ: None. IVL: Consulting Fees (eg, advisory boards); AstraZeneca, Amgen, Pfizer, Edimark to author’s institution. CH: Salary; Sanofi. BP: Royalty; Novartis, AstraZeneca, MSD Oncologie, Pfizer. Consulting Fees (eg, advisory boards); Consulting/Advisor: Puma Biotechnology, Novartis, Myriad Genetics, Pierre Fabre. Contracted Research; Daiichi, Puma Biotechnology, Novartis, Merus, Pfizer, AstraZeneca.