Atopic dermatitis (AD) is one of the most common chronic inflammatory skin diseases affecting approximately 129 million individuals worldwide (1). For Germany, a prevalence of 4.21% has recently been published (2). A first milestone has been reached with the approval of dupilumab for adults with moderate to severe AD in 2017.
This fully humanized immunoglobulin G4 monoclonal antibody blocks the shared IL-4alpha subunit of the IL-4 receptor and inhibits signalling of the key T helper (h) 2- cytokines IL-4 and IL-13 (3). As it was the first innovative systemic drug to be approved for AD, information on its use in routine care is available primarily for dupilumab and from registries so far (4, 5). From phase III studies as well as registry and real-world evidence data it can be concluded that a distinct proportion of dupilumab treated adult patients with AD does not achieve sufficient improvement or reach the primary efficacy endpoint of an eczema area severity index (EASI)-75 or investigator global assessment (IGA) 0/1 after 16 weeks, respectively (3, 5, 6).
In literature, the proportion of dupilumab “non- or poor responders” varies depending on the definition of the target-to-treat and the use of concomitant topical anti-inflammatory treatment. A recently published review on currently available data on the effectiveness of biologics in AD indicates a percentage of 14.8% to 57.3% of adults who do not reach an improvement in EASI-75 after 16 weeks of treatment while after 52 weeks this proportion ranges from 0.8% to 58.1% (7). The present study aimed to identify the percentage and characteristics of patients with insufficient treatment response to dupilumab based on data from dermatological routine care.
For this purpose, patient data from the TREATgermany registry population were analysed.
METHODS
Primary and secondary study objectives
The primary objective of this study was to assess the proportion of patients with an insufficient treatment response to dupilumab within the first year of treatment (“poor responders”). The secondary objective was a more-in-depth patient characterization of dupilumab poor responders to identify potential risk factors. Both study objectives were pursued using data from the TREATgermany registry.
The TREATgermany registry
TREATgermany received approval from the ethics committee of the Medical Faculty of the Technical University Dresden, Germany (No. EK 118032016) as well as all responsible ethics committees at participating study sites and is registered at ClinicalTrials.gov (NCT03057860) and the ENCePP Resource Database (EMA). All participants gave written informed consent and patient anonymity (pseudonymization) was preserved using methods approved by the ethics committee.
Only patients with moderate to severe AD (i.e. objective scoring atopic dermatitis index (8) (oSCORAD)>20 points or current systemic treatment for AD or systemic treatment within the last 24 months) are eligible for inclusion. Study data were collected and managed using REDCap electronic data capture tools hosted at the TU Dresden (Koordinierungszentrum für klinische Studien). Further details on the registry’s inclusion criteria and assessments have been published previously (9–12).
Definition of the dupilumab response groups
The data analysis was conducted using the TREATgermany dataset of n=2,171 AD patients included between 06/2016 and 12/2024 who initiated dupilumab therapy (“DupiStart”). For patients enrolled in the registry after initiation of dupilumab therapy, start of dupilumab treatment was no more than 14 days prior to enrollment. Only patients with moderate to severe AD at DupiStart,i.e.
EASI≥7 and oSCORAD≥20, and 2 consecutive follow-up visits (FUV) both with EASI assessments under continuous dupilumab treatment were considered for further data analysis: The 1st FUV was performed 8 to 20 weeks after DupiStart. The time interval for the 2nd FUV is immediately adjacent and extends up to 52 weeks after DupiStart (Fig. 1).
The selected time intervals for FUV in this study were established through comprehensive preliminary analyses, which carefully balanced clinically relevant periods for the drug’s effectiveness, while yielding a sufficiently large sample size to conduct robust statistical analyses. The patients were classified into dupilumab response groups (Fig. 1).
Data extraction from the TREATgermany questionnaires
Demographic data, patient reported outcomes (PROMS) on AD, information on social life and lifestyle risk factors [marital status, occupational activity, smoking, body mass index (BMI, calculated with height and weight)], and the use of topical treatment for AD were captured directly by the patient. The patient’s age (calculated based on birth year at DupiStart visit), biological sex, comorbidities, objective AD severity (EASI, oSCORAD) were assessed by the physician. We limited our focus to the biological sex as defined by the treating physicians, since there were only very few deviations from the patient’s self-identified gender.
Statistics
Descriptive analyses were conducted for all groups. Statistical testing was restricted to a comparison between dupilumab stable responders and poor responders. For nominal variables, the χ2 or Fisher exact test were applied.
For continuous variables the independent samples t-test or Wilcoxon rank-sum test were conducted, depending on the data’s distribution. Paired-samples t-test or Wilcoxon signed-rank were used to assess differences in disease severity between DupiStart and the 1st and 2nd FUV. For these statistical tests, the significance level was set to α=0.05.
To identify potential influencing factors, logistic or linear regression models were used, depending on the nature of the outcome variable. For these, the α was set to 20%.
Owing to item-specific missing values, sample sizes may vary across analyses. Statistical analyses were conducted using available data per item; missing values were reported. Due to the exploratory nature of the analyses, no alpha adjustment was performed.
SAS software (Version 9.4 for Windows) was used for statistical analyses and paneled bar charts. GraphPad Prism software (Version 8.0 for Windows) was used for boxplots and bar charts, and Microsoft Power Point 365 Version 2510 was used for pie charts and classification schemes.
RESULTS
Number of patients in the dupilumab response groups
Between 06/2016 and 12/2024, dupilumab treatment was initiated in 476 TREATgermany patients (EASI≥7 and oSCORAD≥20) with 2 consecutive FUV during continuous dupilumab therapy (1st FUV between week 8 and 20 and 2nd FUV until week 52). 319 patients were included in university and non-university hospitals, and 157 patients in dermatological practices. The rate of poor responders was 4% (n=19/476) while the percentage of dupilumab stable responders was 44% (n=209), (Fig.
2). Fifteen patients (3%) could be classified as late responders, 102 patients (22%) as intermediate responders, and 131 patients (27.5%) as undulating responders. An overview of the dupilumab response groups as well as the distribution of the 1st FUV across the groups is provided in Figs S1 and S2.
Preliminary descriptive and inductive data analysis indicates the patient’s sex and distinct lifestyle risk factors as risk factors for being a dupilumab poor responder
Descriptive analyses revealed a trend suggesting differences between the dupilumab stable responder and poor responder groups with respect to the patient’s sex, smoking status, arterial hypertension and BMI (Table SI). Descendingly grouping the patients according to their response to treatment into dupilumab “stable responders”, “late responders”, “intermediate responders”, “undulating responders” and “poor responders” illustrated, the higher the proportion of smokers, the lower the response to treatment (Table SI).
Subsequently, inductive data analysis restricted to exploratory statistical comparison between poor responders and stable responders revealed (a trend toward) a significantly higher proportion of men in the dupilumab poor responder than in the stable responder group (78.9% vs 50.2%, p=0.016 (χ2 test)), (Table SII). A higher proportion of smokers was found in dupilumab poor responders than in stable responders (36.8% vs 18.5%, p=0.072 (Fisher exact test)). Poor responders also showed a higher prevalence of arterial hypertension (47.4% vs 16.1%, p=0.003 (Fisher exact test)) and a significantly higher BMI (median [IQR] 27.0 [25.4–33.5] vs 24.4 [22.3–27.5], p=0.003, Wilcoxon rank sum test).
Proportion and risk profile of patients excluded from the predefined analysis population due to missing mandatory 2nd FUV items or unmet study eligibility criteria
In total, 85 patients were excluded from data analysis due to missing 2nd FUV (n=36) or 2nd FUV outside this study’s predefined visit window (n=19), discontinuation of dupilumab in 27 patients, including 2 cases with lack of effectiveness, or missing EASI at 2nd FUV (n=3).
Thirteen out of these 85 patients achieved an EASI response ≤40 at 1st FUV vs 39 with reduction in EASI of ≥75%. Both groups differed regarding sex, BMI and arterial hypertension but not smoking, with a clear trend towards the above-mentioned risk profile of poor responders in patients with an EASI response below 40% improvement at 1st FUV (Table SIII).
Sensitivity analysis according to the severity of AD at DupiStart
Preliminary induction data analysis restricted to the stable responder and poor responder group revealed a statistically significant difference in severity of AD measured by EASI and oSCORAD at DupiStart (EASI range 7.3 to 70.8 stable responders vs 7.8 to 24.7 poor responders, Table SIV).
To ensure better group comparability regarding disease severity, the stable responder, late responder, intermediate responder and undulating responder groups were selected approximating the poor responder’s EASI range of 7 to 28. Cases with EASI scores outside this range were excluded. After this, the total number of patients with moderate to severe AD at DupiStart decreased from n=476 to n=368 patients (n=252 in university and non-university hospitals, n=116 in dermatological practices).
149 patients were in the stable responder group while the number of patients in the poor responder group remained unchanged (n=19), (Fig. 3).
Mean reduction in EASI and oSCORAD under dupilumab exceeds the minimal clinically important difference in the stable responder, but not in the poor responder group
Fig. 4 depicts the absolute EASI values for all dupilumab response groups at DupiStart, 1st and 2nd FUV in the sensitivity analysis. Statistically significant reductions in mean EASI at 1st FUV could be observed for the stable, late-, intermediate, undulating responder groups.
In the stable responder group as well as in the intermediate responder and undulating responder group, but not in the late responder group, the reduction in mean (mean difference, MD) EASI exceeded the minimal clinically relevant difference (MCID) of 6.6 points (13). At 2nd FUV, there was even a greater reduction in mean EASI compared to DupiStart in the stable responder group as well as late-, intermediate- and undulating responder group. As expected, patients of the poor responder group did not reach a significant reduction in absolute EASI neither at 1st nor 2nd FUV and the MD remained below the MCID.
Regarding the percentage of patients who fell below an absolute EASI of 7 points, this was 100% at the 1st and 2nd FUV in the stable responder group. In the late responder group, 14.3% of patients showed an EASI<7 at the 1st FUV and 100% at the 2nd FUV (data not shown). Although the main outcome of this study was the EASI, the change in oSCORAD at 1st and 2nd FUV was calculated as well.
The MD in oSCORAD in the stable responder group clearly exceeded the MCID (13) of 8.2 points at both FUV. The same observation was made for the late-, intermediate responder and undulating responder group (Table I). Patients of the poor responder group statistically significantly decreased in oSCORAD at both FUV as well, but the reduction remained below the MCID.
Chance for being classified as a dupilumab responder for patients with an EASI≥75 response at 1st FUV
We investigated whether the chance for an EASI≥75 response at 2nd FUV was increased for patients with an EASI≥75 response at 1st FUV. Sensitivity analyses revealed an Odds ratio of 5.06 (95% CI: [3.19; 8.03]), p<0.001 (χ2 test) indicating a clear association of a good dupilumab response after 8 to 20 weeks and a sustained and robust long-term treatment up to 52 weeks.
Mean reduction in POEM and DLQI under dupilumab is considerably stronger in the stable responder than poor responder group
In the dupilumab stable responder group, a clear reduction in POEM and DLQI exceeding the MCID of 3.4 and 4 points (POEM) (13, 14), respectively, and four points (DLQI) (15) was evident at both FUV. Similar results were obtained for the late-, intermediate responder and undulating responder group.
In the poor responder group, a substantial reduction in POEM and DLQI could only be observed at 2nd FUV, and a reduction exceeding the corresponding MCID could only be observed for the POEM.
Sensitivity analysis after EASI matching confirms previous results on the patient’s sex as well as smoking behaviour, arterial hypertension and an increased BMI as risk factors for being a dupilumab poor responder
After the abovementioned sensitivity analysis by matching the EASI ranges, previously observed (trends towards) statistically significant differences related to the patient’s sex, smoking behaviour, arterial hypertension and BMI between dupilumab stable responders and poor responders could be confirmed (Table SV, Fig. 5). There were still significantly more men in the group of poor responders than stable responders (78.9% vs 48.3%; OR 4.01 (95% CI: [1.27; 12.65]), p=0.012 (χ2 test)).
The proportion of active smokers remained higher among poor responders than among stable responders (36.8% vs 16.3%; p=0.031 (χ2 test)). But due to insufficient patient numbers, the result of the more conservative exact test according to Fisher should also be considered, which was just above the threshold for statistical significance at a significance level of α=0.05 (p=0.054). However, with an OR of 3.00 (95% CI: [1.07; 8.37]), a clear trend for active smoking as a risk factor for being a poor responder is noticeable.
Logistic regression revealed that the risk for smokers being a poor responder was independent from the patient’s age (OR 0.97 (95% CI [0.90; 1.04], p>0.20)).
For a significantly higher proportion of the dupilumab poor responder group than stable responder group arterial hypertension (47.4% vs 15.9%; OR 4.77 (95% CI: [1.75; 13.04]), p=0.003, Fisher exact test) and an increased BMI (median [IQR] 27.0 [25.4–33.5] vs 24.8 [22.3–27.8], p=0.012, Wilcoxon rank sum test) was observed. However, within the group of poor responders there was no difference between men and women regarding the BMI (p>0.05, Wilcoxon rank sum test).
Two multivariate regression models were also performed to determine whether these were independent risk factors. Smoking, male sex, and BMI were each significantly associated with an increased risk of being a poor responder after adjustment for the other variables included in the model. However, after the addition of arterial hypertension to the model, only male sex and smoking remained statistically significant (Table SVIa and VIb).
A difference in the use of concomitant topical treatment between stable responders and poor responders could not be determined (see Appendix S1).
DISCUSSION
This is the first study attempting a more-in-depth clinical characterization of patients with AD and insufficient treatment response to dupilumab based on data from a prospective, multicentre non-interventional AD registry. The proportion of poor responders was only 4% in this study. A relatively small proportion of cases were not considered for data analysis due to the predefined inclusion and exclusion criteria of this study, particularly with regard to complete data on continuous dupilumab therapy and key data items, as well as appropriate visit windows for the FUVs.
The EASI-40 cut-off may seem unconventional at first glance, but was introduced for the purpose of clear differentiation from EASI-50 because this has been defined as one of the treat-to-targets for systemic therapy of AD after 3 months of treatment (5, 14, 16). 44% of patients were classified as stable responders. An additional group of 30% of all patients achieved a treatment response exceeding an EASI-75 at one of the 2 follow-up visits.
The decrease in mean EASI exceeded the MCID at both FUV in the group of stable responders, but not poor responders (13). A substantial decrease in objective severity of AD in stable responders was confirmed by oSCORAD measurements. Furthermore, stable responders, but not poor responders, showed an improvement of POEM and DLQI exceeding the MCID at both FUV.
The rate of dupilumab poor responders in our study is much lower than in previous publications. This is very probably due to a heterogeneity mainly concerning the studies” measuring instruments and observational periods. In a single-centre retrospective-cohort study based on IGA measurement, 492 patients (90%) achieved a sufficient treatment response while 52 patients were “non-responders” (17).
However, a treat-to-target after a distinct treatment period was not part of the study concept. In another retrospective cohort study, the rate of dupilumab non-responders (<SCORAD-50 after 4 months) was 42.5% (37/96) (18). Data from prospective studies whose design included a definition of “dupilumab non- or poor responders” are not yet available.
A more-in-depth patient characterization revealed a significantly higher proportion of males and a higher BMI, frequency of arterial hypertension and smokers in the group of poor responders in our study. Group comparisons within the patient cohort excluded from the predefined analysis population due to missing mandatory 2nd FUV items or unmet study eligibility criteria reinforce this risk profile in patients with an EASI response below 40% improvement at 1st FUV.
Sharif et al. stated that “although some RCTs have reported outcomes stratified according to sex, they often neglect to consider a sex-based perspective in their discussions of findings, which may result in potential sex bias.” Interestingly, this research group succeeded in discovering sex-specific differences in the safety profile of dual interleukin (IL)-4/IL-13 blockade by “big data- and machine learning-based analysis of a global pharmacovigilance database” (VigiBase) (19). An impaired treatment response in males could be observed in an observational, retrospective and multicentre study including 198 adult patients diagnosed with moderate-to-severe AD. Women were 3.6 times more likely to achieve an EASI-75 response under dupilumab than men (20).
A male sex was demonstrated to be associated with lower drug survival rates in an Italian study (21). In contrast, Eichenfield et al. did not find a difference between the sexes in an analysis of an electronic medical records dataset of 435 patients to assess the real-world effectiveness of dupilumab in AD (22). In summary, sex-related data published on the use of dupilumab for AD so far come from conceptually different studies, but seem to be more consistent with our observation.
Another key finding was a comparatively higher proportion of smokers in the poor responder than stable responder group. This fits well with a previous TREATgermany data analysis that demonstrated a higher lesion’s intensity, pruritus and less weeks with well-controlled AD in smoking compared to non-smoking AD patients (23). Linder et al. also observed a higher rate of smokers in patients with a lower treatment response (18).
Other clinical studies rather focused on the prevalence or incidence of AD and explored whether smoking is per se a risk factor, but did not consider potential effects on the response to systemic treatment (24–26). Regarding data from basic research, there remains a lack of groundbreaking data elucidating how “smoking” and “skin inflammation” in AD might be connected (27–31).
We were able to confirm that smoking and arterial hypertension coincide in patients who show a low response to dupilumab. Indeed, an “AD blood signature” supporting a dysregulation of inflammatory and cardiovascular risk markers in AD patients has been published (32) and synergistic inflammatory pathways leading to persisting disease activity are currently discussed (33). In a systematic review and meta-analysis a higher odds ratio for arterial hypertension in patients with AD, particularly in case of more severe disease, compared to healthy control was demonstrated albeit this was most consistently observed in studies from the U.S. than Europe and Asia and cyclosporine is discussed having contributed to this outcome (34).
The impact of obesity is also intensively discussed in the field of AD. Although in a previous study from TREATgermany the prevalence of obesity was lower compared to data from other countries, there was a significant correlation between the BMI and severity of AD (35). We found a significantly higher BMI in dupilumab poor responders, but we did not identify a cut-off for the BMI that may serve as a distinguishing feature.
Nicolosi et al. also discuss a role of an increased BMI for the response to dupilumab measured by EASI although dividing the participants into groups using a BMI cut-off of 25 revealed no statistically significant differences (36) while in another retrospective study, stratification by a BMI below vs equal or more than 25 revealed a lower short-term effectiveness of dupilumab in patients with a higher BMI (37). Melé-Ninot et al. also identified both a male sex and an increased BMI as predictors for an impaired response to dupilumab treatment (measured by EASI and IGA at week 52) in a retrospective multicentre study (20). A possible connection between an increased BMI and a reduced treatment response to dupilumab is further supported by findings from integrated metabolomics and lipidomics (38).
The strengths of this study are its prospective design and the use of patient data from routine care. It is the first study predefining an EASI cut-off for dupilumab poor responders in order to assess the proportion and characteristics of this subpopulation. On the other hand, the “real-world” character of this study represents a limitation.
In contrast to controlled studies, the use of routine data commonly requires distinct approaches to ensure evaluations, such as matching the EASI ranges for the groups at DupiStart in this study. A corresponding data analysis did not indicate that excluding patients not fulfilling the predefined inclusion and exclusion criteria of this study resulted in an underreporting of individuals with insufficient treatment response to dupilumab at 1st FUV.
In conclusion, this TREATgermany study demonstrates a proportion of 4% dupilumab poor responders within the first year of treatment. Group comparisons between dupilumab poor- and stable responders further revealed male sex, smoking, arterial hypertension and increased BMI as potential risk factors for being a dupilumab poor responder. Although these observations are generally supported by previously published findings, the data available here is still incomplete overall.
Further translational research is needed to confirm these clinical findings, but also to provide deeper insight into underlying pathomechanisms in order better assess possible clinical implications of the patient’s sex and lifestyle (risk) factors for the response to dupilumab treatment.