Article Text

Original research
Dysthyroidism during immune checkpoint inhibitors is associated with improved overall survival in adult cancers: data mining of 1385 electronic patient records
  1. Mathilde Beaufils1,
  2. Vincent Amodru2,
  3. Manuel Tejeda3,
  4. Jean Marie Boher4,
  5. Christophe Zemmour4,
  6. Brice Chanez1,
  7. Anne Sophie Chrétien5,
  8. Laurent Gorvel5,
  9. Gwenaelle Gravis6,
  10. Damien Bruyat1,
  11. Roxane Mari1,
  12. Anne Madroszyk1,
  13. Thomas Cuny2,
  14. Anthony Gonçalves1,
  15. Aaron E Lisberg7,
  16. Daniel Olive5,
  17. Louis Tassy1,
  18. Frederic Castinetti2 and
  19. Philippe Rochigneux1,5
  1. 1Department of Medical Oncology, Institut Paoli-Calmettes, Marseille, France
  2. 2Department of Endocrinology, Assistance Publique - Hôpitaux de Marseille (AP-HM), Marseille, France
  3. 3Department of Informatics, Institut Paoli-Calmettes, Marseille, France
  4. 4Department of Biostatistics, Institut Paoli-Calmettes, Marseille, France
  5. 5Tumor Immunology, CRCM Marseille, Inserm 1068 - CNRS 7258 - Institut Paoli Calmettes - Aix Marseille University, Marseille, France
  6. 6Medical Oncology, Institut Paoli-Calmettes, Marseille, France
  7. 7Department of Medicine, Division of Hematology/Oncology, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, USA
  1. Correspondence to Dr Philippe Rochigneux; rochigneuxp{at}


Background Dysthyroidism (DT) is a common toxicity of immune checkpoint inhibitors (ICIs) and prior work suggests that dysthyroidism (DT) might be associated with ICI efficacy.

Patients and methods ConSoRe, a new generation data mining solution, was used in this retrospective study, to extract data from electronic patient records of adult cancer patients treated with ICI at Institut Paoli-Calmettes (Marseille, France). Every DT was verified and only ICI-induced DT was retained. Survival analyses were performed by Kaplan-Meier method (log-rank test) and Cox model. To account for immortal time bias, a conditional landmark analysis was performed (2 months and 6 months), together with a time-varying Cox model.

Results Data extraction identified 1385 patients treated with ICI between 2011 and 2021. DT was associated with improved overall survival (OS) (HR 0.46, (95% CI 0.33 to 0.65), p<0.001), with a median OS of 35.3 months in DT group vs 15.4 months in non-DT group (NDT). Survival impact of DT was consistent using a 6-month landmark analysis with a median OS of 36.7 months (95% CI 29.4 to not reported) in the DT group vs 25.5 months (95% CI 22.8 to 27.8) in the NDT group. In multivariate analysis, DT was independently associated with improved OS (HR 0.49, 95% CI 0.35 to 0.69, p=0.001). After adjustment in time-varying Cox model, this association remained significant (adjusted HR 0.64, 95% CI 0.45 to 0.90, p=0.010). Moreover, patients with DT and additional immune-related adverse event had increased OS compared with patients with isolated DT, with median OS of 38.8 months vs 21.4 months, respectively.

Conclusion Data mining identified a large number of patients with ICI-induced DT, which was associated with improved OS accounting for immortal time bias.

  • Immune Checkpoint Inhibitors
  • Immunotherapy

Data availability statement

Data are available on reasonable request. (anonymized database).

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Small series suggests that dysthyroidism might be associated with immune checkpoint inhibitors (ICIs) efficacy, but in limited populations and without accounting for immortal time bias.


  • In this large cohort of 1385 patients, we confirmed that ICI-induced dysthyroidism is associated with improved overall survival (OS), accounting for immortal time bias by landmark analyses and time-varying Cox model.


  • As we demonstrated an improved OS around 40%, the onset of dysthyroidism can help oncologists detecting patients more likely to benefit from ICI.


Medical treatment of solid tumors has irreversibly changed since the development of immune checkpoint inhibitors (ICIs).1–4 In various metastatic adult cancers, ICIs have demonstrated a benefit in long-term overall survival (OS)5 6 and nowadays, ICIs are being developed in neoadjuvant and adjuvant setting.7 8 However, ICIs have a specific toxicity profile, and immune-related adverse events (irAEs) are challenging the historical oncologists’ practices.9 Some AEs are transient whereas others become lifetime condition.10 Among these side effects, hypothyroidism is the most frequent ICI-induced endocrine AE, followed by thyrotoxicosis. Thyrotoxicosis and hypothyroidism occur in 3% and 6% of patient treated with anti-PD1/PDL1, respectively.11 Ipilimumab/nivolumab combination results in more dysthyroidism (DT) (20%)12 than anti-PD1/PDL1 agent or ipilimumab alone (5%).13

In clinical trials, hypothyroidism and thyrotoxicosis are often classified as separate AEs although they may be part of the same disease process called thyroiditis. In some series, thyroiditis account for approximately 7% of patients treated with ICI.14 Thyroiditis is characterized by an inconstant thyrotoxicosis period, followed by hypothyroidism or euthyroidism. The median time to onset of thyrotoxicosis is about 6 weeks, whereas hypothyroidism occurs about 10 weeks after the first ICI dose.14 In most cases, thyrotoxicosis is transient and asymptomatic and does not require specific treatment, whereas hypothyroidism results in a requirement for lifelong supplementation. In both cases, events are mainly grade 1/2 and ICI is continued.15 16

Emerging evidence suggests a link between irAE occurrence and ICI specific survival.17–20 For example, vitiligo has been associated with survival in patients with melanoma treated with ICI.21 Furthermore, multiple irAEs have been shown to correlate with improved survival.22 Small retrospective cohort studies reported that DT might be associated with better prognosis.23 24 However, small retrospective studies are often limited by missing data, patients lost to follow-up or a small number of patients available for analyses.25 Moreover, to set such association is challenging as time-dependent variables might be confounding.26 This leads to immortal time bias that should be managed by specific statistical methods.27 Thus, more evidence is needed to confirm the prognostic role of DT in patients treated with ICI, accounting for immortal time bias.

ConSoRe (Continuum Soin Recherche) is a new data mining tool enabling search and data extraction from electronic patient records (EPRs) developed by UNICANCER, a network of 18 French Comprehensive Cancer Centers. This data mining tool enables the constitution of large cohorts of patients with higher performance than manual curation.28 29 The objective of this study was to explore the association between ICI-induced DT and OS in a large cohort of solid tumor patients using ConSoRe data mining.

Patients and methods

Study design

This retrospective cohort study enrolled patients treated with ICI at Institut Paoli-Calmettes, a Comprehensive Cancer Center in Marseille, France, between 2011 and 2021. The cohort was built with the following inclusion criteria: (1) locally advanced or metastatic adult cancers regardless of tumor localization and (2) treatment with an ICI.

In this cohort (ICI cohort), we identified patients who underwent DT group), including hypothyroidism and thyrotoxicosis. Hypothyroidism was defined as serum thyrotropin (often referred as thyroid stimulating-hormone (TSH)) above normal ranges. Subclinical hypothyroidism was defined as TSH levels (in mU/L) above normal ranges with normal levels of free thyroxin (fT4, in pmol/L). Overt hypothyroidism was TSH levels above normal ranges and fT4 below normal ranges. Thyrotoxicosis was defined as TSH below normal ranges. Subclinical thyrotoxicosis was defined as TSH levels below normal ranges with normal fT4. Overt thyrotoxicosis was defined as serum thyrotropin levels below normal ranges and fT4 above normal ranges.

In the DT group, patients with the following criteria were excluded: (1) medical history of DT before ICI; (2) medical history of hypophysitis and (3) PET scan diagnosis of thyroiditis without biological modification. All patients in the DT group had a normal thyroid function before ICI initiation. Hence, the onset of a DT after the initiation of ICI therapy was defined as ICI-induced DT.

Extracting EPRs with the ConSoRe tool

ConSoRe is an academic data analytics solution developed by UNICANCER (a network of 18 French Comprehensive Cancer Centers) in collaboration with Intel and Sword Group. ConSore performs data mining by aggregating structured and unstructured data from EPRs using natural language processing (NLP).

Using ConSore, we built a cohort of patients (ICI cohort) that received immunotherapy in our center, by screening for the comprehensive list of all ICIs routinely available in France in our time frame: ‘durvalumab’, ‘ipilimumab’, ‘nivolumab’, ‘pembrolizumab’, ‘atezolizumab’, ‘avelumab’. We then extracted the following data: age sex, cancer localization, type of immunotherapy, dates of immunotherapy treatment, date of last news and date of death. Among this ICI cohort, we identified patients with a DT group using the following terms: ‘hypothyroidism’, ‘hyperthyroidism’, ‘thyrotoxicosis’, ‘DT’, and ‘thyroiditis’ terms. Data were manually verified for quality control in 100% of the DT group, and notably for the diagnosis of DT. In the whole cohort, when the error rate was >5% in a variable, all data of the variable were manually checked. In the whole cohort, 28.8% of data were verified.

In an exploratory analysis, we studied cutaneous immune-related AEs (cirAEs), and based our methodology on the largest cohort published in this field (n=7008 patients) described by Tang et al.30 In this study, the cutaneous diagnoses associated with improved survival in patients treated with anti PD-(L)-1 were the following: ‘rash and other non-specific eruption’ (HR=0.70), ‘pruritus’ (HR=0.69), ‘xerosis’ (HR=0.62), ‘psoriasis’ (HR=0.70). Consequently, using our data mining solution (ConSoRe), we screened in our cohort patients who experienced cirAEs during ICI, using the following terms: ‘rash’, ‘erythema’ (accounting for ‘other non-specific eruption’), ‘pruritus’, ‘xerosis’ and ‘psoriasis’.


OS was defined as the duration from the first dose of ICI to death or last follow-up, with no restriction on the cause of death.31 Death status and death date were checked in the publicly available French death register.32 In the survival analysis setting, landmark analysis refers to the practice of designating a time point occurring during the follow-up period (known as the landmark time) and analyzing only those subjects who have survived until the landmark time.33 To avoid immortal time bias in patients exposed to DT, we explored two landmark points. A 6-month landmark was used for analyses, based on previous studies.34 A 2-month landmark was an exploratory analysis, as median time to onset DT was 2 months in our cohort. Follow-up was defined as time from first ICI dose to censoring.

Statistical analysis

Population characteristics were reported using both absolute values and percentages. Univariate and multivariate Cox proportional-hazards models were used to assess the prognostic value of DT, age, gender, smoking status, cancer localization, ICI line and ICI type.35 All variables associated with OS in univariate analysis were included in multivariate Cox model. As DT is a covariate that changes over time during the follow-up period, we took into account this time-varying effects using a time-varying Cox model.36 In this model, DT was treated like a time-varying covariate and adjusted HRs (aHR) were calculated.37 38 A time independent Cox conventional model was also performed to highlight the time-varying effects on analyses.

Survival analyses were performed by Kaplan-Meier method (survminer R package) and survival between the DT group and the non-DT (NDT) group was compared with the log-rank test.39 Landmark analyses were performed to account for immortal time bias.27 Patients who died or were lost to follow-up before landmark point were excluded. As recommended by the literature, patients experiencing a DT within the landmark time were included in the DT group, whereas patients experiencing DT after the landmark stayed in the NDT group.33 40 Median follow-up was assessed using reverse Kaplan-Meier method. All comparisons were two sided, with p<0.05 being considered as significant.

When an interaction between DT and another variable was suspected, we tested the statistical significance of the interaction effect. An interaction test was performed in the Cox model between dysthyroidism and ICI group (defined as monotherapy vs dual therapy) and between DT and cutaneous toxicity. Interaction was considered significant if p<0.05. All analyses were performed by R software (V.4.1.2).


Clinical characteristics

Data from 1385 patients who received checkpoint inhibitors between 2011 and 2021 were included (figure 1). As presented in table 1, patients were in majority male (64%), had a median age of 65 years and presented preferentially lung, bladder, kidney cancers and melanoma. Among these 1385 patients (ICI cohort), 90 patients (7%) had ICI-induced DT group and 1295 patients did not experience DT (NDT group). Most baseline characteristics between the two groups were comparable (age, smoking status, tumor type, type of treatment, treatment line) except that patients in the DT group were more likely to be women (48% vs 36%).

Figure 1

Study flow chart. DT, dysthyroidism; NDT, non-dysthyroidism.

Table 1

Clinical characteristics of the population of 1385 patients treated with immune checkpoint inhibitors (ICI) with or without dysthyroidism

Among the DT group (n=90), 22 patients (24%) had thyrotoxicosis alone, 36 patients (40%) had hypothyroidism alone and 32 patients (36%) had thyrotoxicosis, followed by hypothyroidism. Among patients with thyrotoxicosis alone or followed by hypothyroidism (n=54), 20 (37%) had overt thyrotoxicosis, whereas in hypothyroidism group (n=68), 49 (72%) patients had overt hypothyroidism (online supplemental table 1). In the nivolumab/ipilimumab regimen, the frequency of DT was particularly high (17/90 patients (19%)) compared with anti PD-(L)1 alone (81/1283 patients (6.3%)). No DT occurred in patients treated with ipilimumab monotherapy. The median time to onset of DT was 56 days (IQR=52.5 days). Mean TSH and T4 value for thyrotoxicosis and hypothyroidism are shown in online supplemental table 1.

Supplemental material

Concerning the 6-month landmark analysis population, by definition some patients were excluded because they died or were lost to follow-up before 6 months (503 in the NDT group and 14 in the DT group). Additionally, two patients experienced DT after 6 months and were included in the NDT group. In total, the 6-month landmark population included 868 patients, with 794 in the NDT group and 74 in the DT group. Baseline characteristics of both groups were similar to those of the ICI cohort (online supplemental table 2).

DT during ICI is associated with improved OS

In the ICI cohort, the median follow-up time was 23 months (IQR 10–44 months), 28 months (IQR 15–45 months) for the DT group and 22 months (IQR 10–44 months) for the NDT group. The median OS of the ICI cohort was 16.1 months (95% CI 14.7 to 18.7). As presented in figure 2A, survival was significantly improved in the DT group with a median OS of 35.3 months (95% CI 27.4 to NA), compared with 15.4 months (95% CI 13.6 to 17.5) in the NDT group. The HR for OS was 0.46 (95% CI 0.33 to 0.65, p<0.001) in favor of the DT group.

Figure 2

Overall survival (OS) according to dysthyroidism. Kaplan-Meier curves for (A) OS according to dysthyroidism in ICI cohort (B) OS according to dysthyroidism in 6- month landmark cohort (C) OS according to dysthyroidism in 2-month landmark cohort. P value assessed with log-rank test. DT, dysthyroidism; ICI, immune checkpoint inhibitor; NDT, non-dysthyroidism.

In time-independent univariate Cox model, age >65 years, ICI line ≥1, were statistically associated with poor survival. By contrast, tumor type other than lung, female gender and ipilimumab/nivolumab treatment were positively associated with OS. DT had a positive association with OS with HR 0.46 (95% CI 0.33 to 0.65, p<0.001). In the time-independent multivariate Cox model (table 2), age>65 years was no longer associated with OS in multivariate analysis. On the contrary, female gender, ipilimumab/nivolumab treatment and tumor type other than lung remained associated with improved OS. Importantly, DT remained positively associated with OS with an HR 0.49 (95% CI 0.35 to 0.69, p<0.001).

Table 2

Multivariate Cox analyses for overall survival using time-independent and time-varying models

DT benefit on OS persists when accounting for immortal time bias

When a 6-month landmark was applied, median follow-up time (35 months; IQR: 15–47 months) was longer than in the ICI cohort. In the 6-month landmark cohort, the median follow-up was similar between groups with 27 months (IQR 15–48 months) and 30 months (IQR 18–47 months) for the NDT and the DT groups, respectively. The 6-month landmark analysis confirmed the OS benefit of experiencing a DT during ICI, with an HR 0.67 (95% CI 0.47 to 0.97), p=0.033. In the DT group, median OS was 36.7 months (95% CI 29.4 to not reported (NR)) vs 25.5 months (95% CI 22.8 to 27.8) in the NDT group (figure 2B).

Using a 2-month landmark revealed the same time effect on OS with an HR 0.64 (95% CI 0.44 to 0.94, p<0.024) (figure 2C). Median OS was improved in the DT group (29.4 months (95% CI 21.7 to NR)) compared with the NDT group (19.5 months (95% CI 17.9 to 21.6)).

Then, we performed a time-varying Cox model to account for immortal time bias (see the Methods section). After adjusting for immortal time bias in univariate analysis, DT was still associated with an improved OS with an aHR of 0.59 (95% CI 0.42 to 0.83; p=0002). In the time-varying multivariate model, DT was still associated with an improved OS with an aHR of 0.64 (95% CI 0.45 to 0.90, p=0.010) (table 2).

Finally, we questioned if the OS benefit of DT was driven by the IO-combination (ipilimumab/nivolumab), compared with monotherapy by anti PD-(L)1. As presented in online supplemental figure 2, the positive impact of DT on OS in patients treated with ICIs was consistent among treatment groups, either dual therapy (ipilimumab/nivolumab), or monotherapy (anti PD-1 or anti PD-L1). This finding was replicated in our three analyses (total cohort, 2-month landmark and 6-month landmark). Moreover, the interaction test between ICI type (monotherapy or dual therapy) and DT was not significant (p=0.6), suggesting the absence of interaction between DT and the type of ICI (monotherapy vs dual therapy).

DT benefit on OS is additive with other irAEs

After the demonstration that ICI-induced DT was robustly associated with improved OS, we investigated if this benefit was additive with irAEs in additional organs. Indeed, in the DT group (n=90), 60% of patients experienced another immune-mediated side effect (n=54). Specifically, the most common irAEs associated with DT were cutaneous toxicity (33%), followed by digestive (18%) and rheumatological toxicity (12%) (figure 3A).

Figure 3

Additional immune-related adverse events (IRAEs) in the dysthyroidism group. (A) Pie chart distribution of additional irAE type in percentages. (B) Kaplan-Meier curves for OS according to additional irAE. P value assessed with log-rank test. DT, dysthyroidism; OS, overall survival.

Then, we studied the survival of patients with an isolated DT versus patients with DT and another irAE (multiorgan irAEs). As presented in figure 3B, patients with DT and another irAE had an improved OS compared with patients with an isolated DT, with median OS of 38.8 months (95% CI 31.4 to NR) vs 21.4 months (95% CI 12.1 to NR), respectively. Moreover, in univariate Cox model analysis, multiorgan irAEs was associated with improved OS compared with isolated DT with an HR 0.38 (95% CI 0.19 to 0.73, p=0004). Altogether, these data show that ICI-induced DT is robustly associated with improved OS, and that this benefit on OS is additive with other irAE.

Among additional irAEs, we then decided to focus on cirAEs, another irAEs type which is often not treated with steroids. First, we analyzed the impact of cirAEs in OS in solid tumor patients treated with ICIs (figure 4). Median OS was improved in the cirAEs group with a survival of 65 months vs 13 months in the group without cirAEs (figure 4A). In the cirAEs group, the HR was 0.33 (95% CI 0.26 to 0.42, p=0.001). Then, we performed a 6-month landmark analyses (as previously done for DT) and we found consistent results (figure 4B). To decipher the relative impact of these two variables on OS, we studied the survival of the four groups of patients (with or without DT/with or without cirAEs). As presented in figure 4C, patients who experienced both cirAE and DT had the best OS (median OS: 65 months (95% CI 35 to NA)), followed by patients with cirAE only (median OS: 43 months (95% CI 31 to NA)), followed by patients with DT only (median OS: 27 months (95% CI 21 to NA)) and finally by patients without cirAE nor DT (median OS: 12 months (95% CI 11 to 14)). The 6-month landmark analyses in the same four groups of patients reported consistent results (figure 4D). Taking into account cirAEs in the multivariate Cox proportional-hazards model (online supplemental table 4), DT remained significantly associated with improved OS. However, cirAEs were more strongly associated with OS ((HR 0.35 95% CI (0.28 to 0.46)) than DT ((0.56 95% CI (0.40 to 0.79)). Finally, we evaluated the interaction between DT and cutaneous side effects using an interaction test and found no interaction (p=0.4) between these two variables.

Figure 4

Impact of cutaneous immune-related adverse events (cirAEs) in overall survival (OS) in solid tumor patients treated with immune checkpoint inhibitors. (A) OS according to cirAEs in the all cohort ; (B) OS according to cirAEs in the 6-month landmark cohort; (C) OS according to DT and cirAES in the all cohort ; (D) OS according to DT and cirAES in the 6-month landmark cohort. DT, dysthyroidism; NDT, non-dysthyroidism.


In this retrospective study of 1385 patients treated with ICI, we demonstrated that the occurrence of DT during ICI treatment was associated with improved survival. Indeed, this positive association between DT and improved OS was independent of other prognostic factors and persisted after adjustment for immortal bias. Furthermore, in patients with ICI-induced DT, additional irAEs were associated with improved OS.

One of the strengths of this study is the size of the cohort. To gather this cohort of 1385 patients, we used ConSore, a data mining tool developed in French Comprehensive Cancer Centers. Indeed, data collection in retrospective studies is often made on registries or database with fastidious human intervention. These methods are time-consuming and error-prone. ConSore enables the constitution of large retrospective cohort in an automated manner and helps improving follow-up data using multiple search sources in EPR. Altogether, ConSore helps improving the quantity and the quality of data collected in real-life cohorts.

Additionally, the follow-up of patients in this cohort is longer than any previous studies in the field.34 In this type of real-life data, having a follow-up long enough to detect the maximum of events is essential. Indeed, DT may occur at any time during ICI treatment and can even first appear after ICI discontinuation.41 In our study, median follow-up was about 30 months, whereas it reached only 9 months in Street et al study.34 No follow-up bias was observed since median follow-up was equal in both groups.

Moreover, we carefully addressed possible statistical bias. Notably, taking immortality bias into account is crucial in studies correlating irAEs with clinical outcomes. Immortality bias happens when the occurrence of the event of interest is time-dependent, which is the case for DT, which may occur several months after ICI initiation. By definition, patients experiencing a DT during ICI are alive until this event. They are considered as ‘immortal’ from ICI initiation to ICI-induced DT, which artificially emphasizes the effect of DT on OS. However, many studies on irAEs, and especially in dysthyroidism, do not take this immortality bias into account.17 24 In our study, we used two methods to limit this bias: landmark analysis and multivariate Cox model with DT considered as a time-varying variable. These two validated time-adjustment methods reported consistent results. Hence, we could confirm the association between dysthyroidism and OS, independently of immortal time bias.

Our study presents some intrinsic limitations. First, the incidence of DT (7%) was lower than in other studies and more overt DT were seen.11 12 However, this incidence is strongly dependent on DT definition. In studies considering thyroiditis, incidence is about 7%.14 ConSoRe relies on NLP and uses terms found in EPR in order to search and collect data. Thus, only EPR with written diagnosis (‘hypothyroidism’, ‘hyperthyroidism’, ‘thyrotoxicosis’, ‘DT’, ‘thyroiditis’) were selected. Biological variations of TSH without written diagnosis were not detected by ConSoRe research, which artificially decreased the incidence. Second, Eastern Cooperative Oncology Group score (ECOG) and Charlson Comorbidity Index could not be extracted by our data mining tool. However, in a previous study,34 ECOG score and Charlson Comorbidity Index were associated with OS, but did not change the association between DT and OS in the multivariate analysis. Further developments of ConSore are expected to give access to ECOG and Charlson index. Third, the proportion of tumor types was specific of our center, favoring lung cancer (58%) and urological cancer (22%), with few melanoma patients (3%). Fourth, the patients analyzed in this study are from routine clinics, without biological collection, precluding the possibility of mechanistic investigation. Nevertheless, the advantage of this setting is that ICI-induced DT is studied in real-life patients, and not in overselected patients from clinical trials. Finally, we did not report radiographic response in this study, because our data mining solution, ConSore, is currently not able to extract this kind of data. However, in the context of immuno-oncology, where pseudoprogression and prolonged stable disease are frequent, radiographic tumor evaluation might be considered as a poor surrogate end points for OS.42 Consequently, we decided to focus on OS, which is the ultimate endpoint in oncology.

From a biological perspective, precise mechanisms underlying immune-related DT remain elusive. For skin toxicity, specific T-cell clones recognize shared antigens in lung cancer and in the skin mediating both cutaneous irAE and tumor response.43 The pathogenesis of ICI-induced DT probably involves an interplay among cellular autoimmunity (self-reactive T cells), humoral immunity (antibody-mediated cytotoxicity) and genetics (HLA-DR allele, known to increase the risk of endocrine irAEs).44 Self-reactive T-lymphocytes due to the loss of T-cell tolerance are considered key players of most irAEs, acting either directly or indirectly via stimulating the production of autoantibodies by B-lymphocytes.45 PD-L1 protein is expressed by thyroid follicular cells in areas that also contain abundant PD-1 positive T cells,46 which might explain cross-reactivity. Natural killer (NK) cells have also been implicated, as their frequency increases in ICI-induced thyroiditis.47 Moreover, the presence of preexisting autoantibodies was significantly associated with the occurrence of irAEs in patients treated with ICIs.48 Concerning the association between biological mechanisms linking DT and improved survival, we might hypothesize that ICI-induced DT is a consequence of reactive T-lymphocytes (loss of T-cell tolerance) reinvigorated by ICI (eg, increase of PD-1+ Ki67+ CD8+ T cells during treatment).49 However, more evidence is needed to elucidate possible biological mechanisms linking DT and survival.

From a clinical perspective, consistent with prior studies,50 women were more likely to develop DT. In a recent meta-analysis focusing on sex impact on ICI efficacy, both sex benefit from ICI but the magnitude of the effect was higher in men.51 Further study should evaluate carefully benefit/risk of ICI in women regarding the potential for increased irAEs. Interestingly, DT was more frequent with ipilimumab/nivolumab combination, whereas no DT was observed with ipilimumab monotherapy. These findings are consistent with previous evidence of specific patterns of response and toxicity between ICI classes.52

Compared with the literature, in our cohort DT had a stronger effect on OS with an aHR of 0.59 compared with previous studies and notably the largest cohort from Street et al (aHR 0.80, 95% CI 0.71 to 0.89, p<0.001).34 In this retrospective study, lung cancer accounts for only 20%, whereas it accounts for 58% in our cohort. In Street et al, the link between DT and OS was the strongest in lung cancer versus other tumor types. This result could be explained by the shared embryonic development of thyroid and lung.53 In our cohort, median OS when patient experienced ICI-induced DT increases from 25.4 months to 36.7 months in the 6-month landmark analysis. To our knowledge, no such improvement in median OS according to DT has previously been described in a large cohort of solid tumors.

We previously described the positive impact of ICI-induced DT, but this irAE is often accompanied by irAEs from other organs. Indeed, 60% of patients with DT developed a second irAE, notably cutaneous irAE. Interestingly, Tang et al recently described the positive impact of cutaneous irAE on survival.30 In our exploratory analyses, patients with DT associated with an additional irAE had an increased median OS compared with DT alone. Our results confirm the findings from Shankar et al22 describing the impact of multisystem irAE on OS in an ICI retrospective cohort (aHR=0.86 for one irAE vs aHR=0.57 for multisystem irAE). The study of cutaneous toxicities in solid tumors patients demonstrated that cutaneous irAES were also strongly associated with improved OS. Our results are consistent with the previously published literature on the relationship between cirAEs and OS, notably by Tang et al.30 In our study, when compared with DT, cirAEs appeared to have a stronger association with OS. However, it is hard to formally conclude which of cirAEs or DT is most associated with survival. Indeed, our study was mainly focused on DT and we did not have the date of cirAEs enabling to perform time-dependent Cox model as we did for DT. However, we report that cutaneous and thyroid irAEs are cumulative and are both associated with a clinically meaningful improved OS in patients with solid tumors treated with ICI.

Finally, in daily clinical routine, a close collaboration with endocrinologists is mandatory to manage DT. In the acute phase, it is nowadays recommended to pursue ICI and to avoid antithyroid treatments.15 Moreover, long-term survivors after ICI are becoming more frequent and the endocrinologist follow-up might decrease with time. Endocrinologist follow-up must be pursued in a long-term perspective, as thyroid dysfunction is often neglected post ICI treatment.41

In conclusion, DT is associated with improved OS in patients treated with ICIs regardless of tumor localization and treatment line. The onset of DT might help oncologists refining patient’s prognosis or identifying patients suitable for a rechallenge strategy. Future studies on irAEs must take into account immortal bias with accurate statistical methods.

Supplemental material

Data availability statement

Data are available on reasonable request. (anonymized database).

Ethics statements

Patient consent for publication

Ethics approval

This study involves human participants and was approved by Institut Paoli-Calmettes institutional review board (IRB number: T-CHECK-IPC 2019-053, approval date: January 31, 2020). Participants gave informed consent to participate in the study before taking part.


We thank the bioinformatics team and the clinical research team for their precious support during the conception of the study.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Twitter @MyeLau_Gorvel, @prochigneux

  • MB and VA contributed equally.

  • Contributors Conception of the study: MB, VA, FC and PR. Data extraction: MT and LT. Biostatistical support: JMB and CZ. Patients care: AM, GG and AG. Manuscript writing: all authors. PR is responsible for the overall content as guarantor.

  • Funding This study was funded by Institut Paoli-Calmettes, Marseille, France.

  • Competing interests Pr Olive is co-founder and shareholder of Imcheck Therapeutics, Alderaan Biotechnology and Emergence Therapeutics and has research funds from Imcheck Therapeutics, Alderaan Biotechnology, Cellectis and Emergence Therapeutics. PR reports funds to his institution from Novartis, BMS, and Consultant/Advisory Board from AstraZeneca and GSK. AEL reports the following: Company: Boston Scientific, Immediate family member (wife); Stock (<5% equity), Company: Boston Scientific, Immediate family member (wife); Commercial Research Grants: Daiichi Sankyo, Calithera, Biosciences, AstraZeneca, DracenPharmaceuticals, WindMIL, eFFECTOR Therapeutics; Compensated Consultant/Advisory Board : AstraZeneca, Bristol-Myers Squibb, Leica Biosystems, Jazz Pharmaceuticals, Novocure, Pfizer, MorphoSys, Eli-Lilly, Oncocyte, Novartis; Regeneron, Janssen oncology, Sanofi group of companies, G1 Therapeutics, Molecular Axiom, Amgen, IQVIA.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.