Critical flaws in: “Dupilumab and lymphoma risk among patients with asthma: a population-based cohort study”

Extract

We read with interest the article by Ma et al. [1], suggesting an association between dupilumab treatment and lymphoma risk in patients with asthma. With expanding prescribing of biologics, including dupilumab, for asthma, such findings are of consequence, necessitating in-depth assessment of data, methods and conclusions. We identified three major areas of concern regarding this article – crucial methodological flaws, biological implausibility, and incident risk of disease focus – that question the interpretation of the results.

Shareable abstract

There are three major areas of concern identified in the article on dupilumab treatment and lymphoma risk in patients with asthma by Ma and co-workers – crucial methodological flaws, biological implausibility, and incident risk of disease focus https://bit.ly/4a3nT2R

To the Editor:

We read with interest the article by Ma et al. [1], suggesting an association between dupilumab treatment and lymphoma risk in patients with asthma. With expanding prescribing of biologics, including dupilumab, for asthma, such findings are of consequence, necessitating in-depth assessment of data, methods and conclusions. We identified three major areas of concern regarding this article – crucial methodological flaws, biological implausibility, and incident risk of disease focus – that question the interpretation of the results. These shortcomings could inadvertently cause detrimental impacts on prescriber practices and patient outcomes.

Propensity score matching was used; however, the model and covariates used to derive scores are not well-described, nor were additional sensitivity analyses performed (i.e. alternative propensity score matching algorithms and inverse probability weighting) to validate results and justify inferences [2, 3]. Generally, propensity score matching cannot fully match every characteristic one-to-one, especially for unmeasurable/unmeasured confounders, and does not consider, for example, immunosuppression status of a patient in one group versus the other [2, 3]. Treating propensity score-matched samples as randomised samples introduces bias without further adjustments for unmeasured confounding factors [2, 3] and, despite having enough apparent observations to match >1:1, which could substantially reduce bias [4], 1:1 matching was used. Missing data, common in observational studies, limit assumptions that can be drawn [5]. The authors do not state how absence of certain factors was handled. Consequently, uncertainty in potential associations between dupilumab exposure and lymphoma risk is not adequately quantified. Furthermore, comparing inhaled corticosteroid/long-acting beta-agonist (ICS/LABA) recipients with dupilumab recipients is suboptimal, given the likely differences in disease courses, healthcare utilisation patterns, and potential differences in baseline risk of lymphoma development. A more appropriate comparator would have been the inclusion of patients of similar disease severity treated with another asthma biologic.

Kaplan–Meier curves and proportional hazards assessments were absent, preventing the reader from determining the validity of reported hazard ratio estimates. A hazard analysis is a time-to-event analysis, so the origin/index date definition is important. It is unclear which evaluation periods the authors used to draw conclusions. Categorical sample sizes, such as the approximated ≤10 sample size, lead to information loss and can introduce modelling instability; this cut-off should be defined as a whole number instead of a range (0–10). Retrospective studies with no prespecified end-points and no correction for multiple testing have no type 1 error control, increasing the likelihood of spurious associations. Therefore, inflated alpha/type 1 errors (false-positives) exist: many nominal p-values <0.05 reported in these analyses would support the presence of false-positives and the probability of increased risk of lymphoma occurring by chance/random variation. Considering the large sample size (n=749 062), non-robust p-values further raise concerns around stability of the findings. A more robust comparator group is needed to draw inferences.

We question the clinical relevance of the findings. In the authors’ assessment of lymphoma risk by treatment duration, they report a persistent increased risk of T/NK-cell lymphomas with dupilumab exposure of >16 weeks; however, to demonstrate this, a dose- and duration-dependent risk increase should be shown. The biological plausibility of developing lymphoma within 16 weeks of starting therapy is dubious, especially for slow, indolent malignancies, such as lymphoma. There is a high possibility that the malignancies may have been present in patients before dupilumab initiation. A 3-month look-back period to exclude atopic dermatitis or assess risk of slow-growing lymphoproliferative disorders is problematic due to the short timeframe and frequent misdiagnosis of cutaneous T-cell lymphoma as atopic dermatitis, meriting further examination through biopsies.

The global age-standardised incidence rate (ASIR) for non-Hodgkin lymphoma between 2019 and 2021 was 6.7–7.1 per 100 000 and, in 2019, for Hodgkin lymphoma was 1.1 per 100 000 [6, 7]. Assuming these rates remained unchanged over 5.7 years (the timespan assessed), we anticipate a combined lymphoma ASIR of approximately 44–47 per 100 000, equating to 7 cases per 14 936 patients treated with dupilumab and 323–345 cases per 734 126 treated with ICS/LABA, a ratio of 1:46–49. Although the authors’ reported risk of new-onset lymphoma across treatment arms is higher than we would expect from crudely extrapolating from global ASIR rates, the ratio of cases across treatment arms is the same (1:47).

LIBERTY ASTHMA TRAVERSE, a phase 3, single-arm, open-label extension study demonstrated long-term efficacy and safety of dupilumab (up to 3 years) in patients with moderate-to-severe asthma [8]. Of 2062 patients enrolled in TRAVERSE from the phase 2b/3 LIBERTY ASTHMA QUEST studies, only one dupilumab recipient developed Hodgkin lymphoma (deemed unrelated to treatment) during the 96-week treatment period – an observational period equivalent to 3374 patient-years. Based on the authors’ findings, a much higher lymphoma rate should have been observed during TRAVERSE, further supporting our concern that overdiagnoses/misdiagnoses may have impacted their findings. Meaningful differences in baseline clinical characteristics between trial populations and real-world cohorts may account for the observed discrepancy in lymphoma incidence, regardless of dupilumab exposure.

It should be highlighted that Ma et al. [1] found that patients treated with dupilumab have lower all-cause mortality risk. If a signal did exist for increased lymphoma risk, which we feel is based on analyses with considerable limitations and flaws, it is dwarfed by the substantial mortality benefit. This suggests that there may be residual or unmeasured confounding factors related to overall health or healthcare access unaccounted for in the study analyses.