Long-Term Risk of Skin Cancer and Lymphoma in Users of Topical Tacrolimus and Pimecrolimus: Final Results from the Extension of the Cohort Study Protopic Joint European Longitudinal Lymphoma and Skin Cancer Evaluation (JOELLE)

Introduction

According to approved labeling, topical tacrolimus is indicated for the treatment of moderate to severe atopic dermatitis (AD), and topical pimecrolimus, for the treatment of mild to moderate AD. It is known that they are used off-label, too, but the extent of its use has not been investigated.¹ Safety data from systemic use of immunosuppressants in patients with organ transplants, from animal studies, and from case reports raised initial concerns about a potential increase in the risk of lymphoma and skin cancer associated with the use of topical calcineurin inhibitors (TCIs), especially in children. The epidemiologic literature presents insufficient evidence to infer whether TCIs cause malignancy.^2–5 Differentiating the effects of the medication itself from the risks associated with AD or severe AD is inherently difficult.^6,7 Furthermore, reverse causation is a concern because patients in early stages of certain skin malignancies, particularly cutaneous T-cell lymphoma (CTCL), may present clinical manifestations resembling AD and therefore may be treated with the study medications.

Given the long potential latency for cancer development, postmarketing studies with long follow-up are necessary to determine whether there is an association between TCIs and cancer. The European Medicines Agency requested a European study with sufficient follow-up to evaluate long-term risk of cancer. The objective of the JOELLE extension study was to estimate the long-term risks of melanoma, nonmelanoma skin cancer (NMSC), CTCL, non-Hodgkin lymphoma (NHL) (excluding CTCL), and Hodgkin lymphoma (HL) in adults and children.

Patients and Methods

This cohort study included data from the Clinical Practice Research Datalink (United Kingdom) (UK-CPRD) (2002–2017), the PHARMO Database Network (the Netherlands) (NL-PHARMO) (2002–2017), the Danish health databases (Denmark) (2002–2016), and the Swedish health databases (Sweden) (2006–2015).

Within each data source, eligible patients were required to have ≥12 months of continuous database enrollment before the first recorded prescription (new users) within the study period (except for children 0–12 months of age, who were eligible for inclusion with no required period of prior continuous enrollment). Patients with documented history of skin cancer or lymphoma before cohort entry were excluded. New users of topical tacrolimus and new users of topical pimecrolimus were classified into four groups according to age (children <18 years and adults ≥18 years) and TCI exposure (tacrolimus or pimecrolimus). A comparative cohort of users of moderate- to high-potency corticosteroids was matched to each of the four TCI-exposed cohorts. The corticosteroid comparison cohorts included (i) patients with AD diagnosis with a prescription for topical corticosteroids of moderate to high potency (hereafter, “topical corticosteroids”) after the eligibility date and (ii) patients without AD diagnosis with a prescription for topical corticosteroids during the study enrollment period and at least one other prescription within the prior 12 months. The indication for the individual prescriptions was not recorded. A cohort of individuals not treated with any of the study medications (“untreated cohort”), with or without recorded AD diagnosis, was also identified for contextualization.

To control confounding, when selecting the corticosteroid cohort to match to each TCI cohort, we calculated exposure propensity scores (PSs) representing the probability of initiating TCI treatment rather than receiving topical corticosteroids, given a set of baseline covariates. After creating the PSs, trimming was performed to remove non-overlapping and extreme values within the PS distributions. All individuals (both TCI and corticosteroid) above the upper 99th percentile of the corticosteroid PS score distribution were trimmed, as were all individuals below the lowest 1st percentile of the TCI PS score distribution. After trimming, all remaining TCI users were retained as the TCI cohort and PS twentiles within this TCI cohort were identified. Then up to four times as many users of topical corticosteroids, depending on availability, were randomly selected, from all users of topical corticosteroids within each twentile-based stratum, to form the matched corticosteroid cohort.

For contextualization, each user of corticosteroids from the comparator cohort for tacrolimus was matched to four nonusers of any study medication on age, sex, geographic region, and calendar year of start date (untreated cohort).

Follow-up continued from the start date to the earliest of death, database disenrollment, end of the study, or occurrence of one of the study outcomes.

Exposure

Based on records of prescriptions (UK-CPRD) or dispensing (NL-PHARMO, Denmark, and Sweden), exposure to tacrolimus and pimecrolimus was defined as single use (any topical tacrolimus or topical pimecrolimus, but not both) for each of these medications.

Cumulative dose was the total quantity of active substance that a patient received during follow-up (grams) calculated from the strength of the formulation and the package size.

Outcomes

In Denmark and Sweden, outcomes were identified in national cancer registries. For NL-PHARMO, the pathology registry was used, and in the UK-CPRD, outcomes were identified via information from general practices, hospital admissions, and the cancer registry. In NL-PHARMO and UK-CPRD, case validation was performed for all pediatric cases, all CTCL cases, and a random sample of other adult cases. Validation was done in NL-PHARMO by an independent pathologist reviewing pathology excerpts and UK-CPRD by clinical review of electronic medical records.

Covariates

Covariates included immunosuppressive disease and use of immunosuppressive agents; chronic disease; severe skin diseases; AD diagnosis, if available; and measures of healthcare resource utilization. Age, sex, year of start date, and type of prescriber were forced into the PS models. Variables associated with the outcome were also included in the models regardless of whether they were associated with the exposure.⁸

Because information on AD severity was limited or missing in all the data sources, we evaluated the effect of type of prescriber of the first prescription as a proxy for severity of the cutaneous condition. The underlying assumption was that patients with more severe AD would have been seen and treated first by a dermatologist and patients with less severe AD would have been seen and treated first by a GP. The variable was available in Denmark, NL-PHARMO, and Sweden, but not in UK-CPRD.

Statistical Analysis

We evaluated the overall effect as well as the effect of cumulative dose of topical tacrolimus and topical pimecrolimus compared with use of topical corticosteroids. In each data source, cancer events and person-years were stratified between decile boundaries of PSs.⁹ We then used Mantel-Haenszel methods¹⁰ to estimate overall adjusted incidence rate ratios (IRRs) and incidence rate differences, stratifying on study database, deciles of PSs, and sex; and, in NL-PHARMO, Denmark, and Sweden, by type of prescriber (dermatologist, non-dermatologist) of the first prescription. For the main analysis, time at risk started after a lag time of 6 months. Additional analyses with 0-month, 12-month, 24-month, and 48-month lag times were conducted to explore reverse causation and surveillance bias. No hypothesis testing was performed, but 95% confidence intervals (CIs) were calculated around IRRs and incidence rate differences to allow interpretation of the strength of observed effects. No imputation was performed for missing values given the observational nature of the data analyzed in this study. If no prescription for a particular medication existed in the patient’s record, it was assumed the patient was not taking that medication, and if a medical event was not observed in the patient’s medical record, it was assumed they did not have that event.

To further assess whether treatment with the study medications was initiated for signs and symptoms that were compatible with early manifestations of CTCL, we obtained additional information from questionnaires sent to general practitioners in UK-CPRD and reviewed the medical records in Sweden for CTCL cases.¹¹ We analyzed the occurrence of basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) separately to test the role of immunosuppression in the results.

The study was based on medical records. The individual Institutional Review Boards or the Data Protection and Research committees waived the requirement of informed consent due to appropriate handling of patient data and maintenance of patient data confidentiality. We obtained ethical and scientific reviews from the RTI International institutional review board, the Regional Ethical Review Board of Stockholm, the CPRD Independent Scientific Advisory Committee, and the UK National Cancer Intelligence Network. Ethical approval was not required in PHARMO and Denmark. In PHARMO, the study fulfilled the requirements of the PHARMO Compliance Commission, and permission for the use of data from the Dutch National Pathology Registry was obtained. In Denmark, the study was approved by the Danish Data Protection Agency via Statistics Denmark. The study has received the European Network of Centres for Pharmacoepidemiology and Pharmacovigilance (ENCePP) Study Seal. The protocol for the JOELLE study extension phase, protocol version 5.0, dated Jun 30, 2017, is posted in the EU PAS Register, EUPAS21769 #21769.

Results

The study included 126,908 adults and 32,605 children initiating treatment with topical tacrolimus. These patients were compared with 452,996 adults and 117,592 children treated with topical corticosteroids. Similarly, 61,841 adults and 27,961 children initiating treatment with topical pimecrolimus were compared with 244,572 adults and 111,024 children treated with topical corticosteroids. The untreated cohort comprised 1,291,042 adults and 361,584 children. Table 2 shows the distribution of patients by study cohort and population. The distributions by age and sex were similar in all study cohorts. The baseline prevalence of comorbidities and use of medications were similar overall between the topical tacrolimus and topical pimecrolimus cohorts but were different across data sources.

Table 1 STROBE Statement for Cohort Studies with the RECORD Statement Extension—Checklist of Items That Should Be Included in Reports of Observational Studies Using Routinely Collected Health Data

Table 2 Distribution of Users by Study Cohort and Population

Overall follow-up was ≥10 years for 19% of adults and for 32% of children. For users of topical tacrolimus, the median follow-up was 5.7 years in children and 5.0 years in adults. For users of topical pimecrolimus, the median follow-up was 8.9 years in children and 6.5 years in adults.

The median number of prescriptions was 1 for both tacrolimus and pimecrolimus. The median dose of active substance was 0.03 grams for tacrolimus (equivalent to a single 30-gram tube of 0.1% tacrolimus) and 0.3 grams for pimecrolimus (equivalent to a single 30-gram tube of 1% pimecrolimus). Other population characteristics are available in eTable 1 and eTables 4–9 of the Supplementary Materials.

Use of Topical Tacrolimus versus Topical Corticosteroids

In adults, users of topical tacrolimus had an IRR for NMSC of 1.04 (95% CI, 1.00–1.09). The IRR point estimates for melanoma, NHL (excluding CTCL), and HL were <1 (Table 3, Figure 1). The IRR for CTCL with use of topical tacrolimus was 1.80 (95% CI, 1.25–2.58), corresponding to an excess risk of 3 cases per 100,000 person-years (95% CI, 1–6). IRRs for CTCL were 0.81 (95% CI, 0.45–1.47) for a cumulative dose of ≤ 0.05 grams, 2.11 (95% CI, 1.13–3.95) for a cumulative dose from 0.05 to 0.10 grams, and 5.25 (95% CI, 3.21–8.56) for a cumulative dose > 0.10 grams.

Table 3 Pooled Adjusted Incidence Rate Ratios in Users of Topical Tacrolimus and Topical Pimecrolimus Compared with Users of Topical Corticosteroids—Adults

Figure 1 Summary results for tacrolimus and pimecrolimus in children and adults combined. Abbreviations: CI, confidence interval; CTCL, cutaneous T-cell lymphoma; HL, Hodgkin lymphoma; M, Melanoma; NHL, non-Hodgkin lymphoma; NMSC, nonmelanoma skin cancer.

For adult users of topical tacrolimus in whom the time since first exposure to treatment was ≥5 years, the IRR for CTCL was 0.25 (95% CI, 0.03–1.87) (Table 4). In the sensitivity analysis to address reverse causation conducted in UK-CPRD and Sweden, there was little change in the estimated effect on CTCL associated with topical tacrolimus when cases with manifestations of a previous skin condition in the same location as the subsequently diagnosed cutaneous lymphoma were omitted (see eTable 2 and eTable 3 in the Supplementary Materials).

Table 4 Sensitivity Analysis by Time Since Start of Exposure, by Each Type of Malignancy: Adjusted Incidence Rate Ratios in Users of Topical Tacrolimus Compared with Users of Topical Corticosteroids—Adults

In children (Figure 1), the IRR comparing use of topical pimecrolimus with topical corticosteroids was 0.69 (95% CI, 0.20–2.31) for melanoma and 0.63 (95% CI, 0.13–3.13) for NMSC. The IRR comparing use of topical tacrolimus with topical corticosteroids was 2.19 (95% CI, 0.81–5.97) for NHL (excluding CTCL), 2.37 (95% CI, 0.99–5.68) for HL, and 7.77 (95% CI, 0.50–121.45) for CTCL. The IRR for each type of lymphoma was based on few events. For NHL, the IRR was elevated for low cumulative doses, but not for medium or high cumulative doses. For HL, the IRR was elevated for low and high doses, but not for medium doses.

Use of Topical Pimecrolimus versus Topical Corticosteroids

In adults (Table 3, Figure 1), the IRR for melanoma for use of topical pimecrolimus was 1.21 (95% CI, 1.03–1.41), which corresponds to an excess risk of 10 cases per 100,000 person-years (95% CI, 1–18). The adjusted IRR for NMSC with topical pimecrolimus was 1.28 (95% CI, 1.20–1.35), an excess risk of 91 cases per 100,000 person-years (95% CI, 68–114). In adults, the IRR point estimates for NHL (excluding CTCL), HL, and CTCL for users of topical pimecrolimus compared with users of topical corticosteroids were all <1.

In children, the IRR point estimates for each study outcome for topical pimecrolimus compared with topical corticosteroids were based on few events and all were <1 (Figure 1).

Users of Topical Corticosteroids versus Untreated Population

In adults, the IRRs for all outcomes except melanoma were elevated in the cohort of users of topical corticosteroids compared with the untreated population (Figure 2), especially for CTCL (IRR, 5.42; 95% CI, 3.77–7.79). In children, the number of cases was too small for estimating the IRR for the individual outcomes.

Figure 2 Summary results: untreated adults. Abbreviations: CI, confidence interval; CTCL, cutaneous T-cell lymphoma; HL, Hodgkin lymphoma; M, Melanoma; NHL, non-Hodgkin lymphoma; NMSC, nonmelanoma skin cancer.

Discussion

Adult users of topical tacrolimus had IRRs <1 for melanoma, NHL (excluding CTCL), and HL, which indicates that no increased risk of these outcomes was found in topical tacrolimus users compared with topical corticosteroid users. For topical pimecrolimus users, the IRRs for any type of lymphoma were likewise <1. In our study, the results were not homogeneous, and topical tacrolimus and pimecrolimus were not found to produce effects on the same outcomes in contrast with findings from a systematic review that suggested an association between TCI use and risk of lymphoma and no other cancers.⁴

For topical tacrolimus users, the IRR for CTCL was elevated, and the incidence of CTCL increased with increasing cumulative dose. This finding may reflect a causal effect or be the result of confounding by indication. AD is associated with increased risk of malignancies, and the strength of the association is related to the severity of the AD.^6,7 According to the European Medicines Agency–approved labeling, TCIs are indicated only for AD, while topical corticosteroids are also indicated for the treatment of other skin diseases, so a higher proportion of patients without AD are likely included in the topical corticosteroid than in the TCI cohorts. In databases capturing diagnosis from hospital data only, AD diagnoses are often missing, which could hamper the control of confounding and result in overestimation of the risks associated with use of TCIs. Type of prescriber of the first prescription was used as a proxy measure for severity of the underlying cutaneous condition, except in UK-CPRD, where this information was not available. Still, residual confounding is possible and would result in an overestimation of the effect of the study medications, especially for topical tacrolimus, which is indicated for more severe forms of AD.

The IRRs for all outcomes except melanoma were elevated in the analysis of topical corticosteroid users compared with nonusers of any study medication, indicating that either the topical corticosteroids or their indication confers an increased risk of the outcomes studied.

Reverse causation is a concern because patients in early stages of certain skin malignancies, particularly CTCL, may present clinical manifestations resembling AD and therefore may be treated with the study medications. Our efforts to address this produced conflicting results. Although the cumulative dose-response analysis shows the highest risk with the highest cumulative dose, in one sensitivity analysis, the elevated relative risk of CTCL associated with topical tacrolimus was confined to the first years after starting the medication, a pattern that seems more consistent with reverse causation than with a causal effect of topical tacrolimus. However, the other sensitivity analysis for reverse causation indicated little change in the estimated effect of topical tacrolimus on the risk of CTCL when cases with manifestations of a previous skin condition in the same location as the subsequently diagnosed cutaneous lymphoma were omitted.¹¹

In adults, IRRs for melanoma and NMSC in users of topical pimecrolimus were elevated. This may reflect surveillance bias, although that does not explain the greater IRR with greater cumulative doses. In a sensitivity analysis, among NMSC, the overall BCC-to-SCC ratio was >3 in the pimecrolimus and the corticosteroids cohorts. Because SCCs predominate over BCCs in clinical settings where strong immunosuppression is associated with an increased risk of NMSC, these results suggest that systemic immunosuppression is unlikely to be an important cause of the observed associations between exposure to the study drugs and the risk of NMSC in this study.^12–16

Results from this study are consistent with those from other published studies^7,17–19 summarized in a 2011 briefing document to the US Food and Drug Administration:

causality is difficult to determine in light of the potential study biases. (eg, misclassification of lymphoma, protopathic bias, and confounding by indication)²⁰

Moreover, this study shows that any excess risk would be low and confined to a few years after first exposure.

The pediatric population in the JOELLE study was larger than all previous study populations in this area; however, few events occurred among children treated with TCIs. The IRR for each type of lymphoma was elevated for topical tacrolimus at low cumulative doses. Associations that are strongest among those with a low cumulative dose typically are the result of reverse causation or surveillance bias rather than a causal effect because low cumulative dose is correlated with short duration of use and an outcome may already have been present, although clinically undetected, when exposure began. It is also notable that no lymphomas occurred in a recently presented cohort study of 7954 children treated with topical tacrolimus in 2005–2012 (mean follow-up, 6.4 years; 15% were followed for ≥10 years).²¹

This study is the largest to date to evaluate the association of TCIs with skin cancer and lymphomas and includes the longest duration of follow-up of patients in population-based data sources. When analyzing the risk of malignancies associated with long-term follow-up, the data were not consistent with an increased risk of skin cancer or lymphoma as the duration of follow-up increased. However, despite being the largest study of these agents with the longest follow-up, the rarity of the outcomes in children limits the conclusions.

Conclusion

We found little evidence of an association between the use of either topical tacrolimus or topical pimecrolimus and the occurrence of skin cancer and lymphoma. The elevated IRR for CTCL among adult users of topical tacrolimus and the elevated IRRs for melanoma and NMSC among adult users of topical pimecrolimus could represent causal effects or might result from the underlying disease. The IRRs for skin cancer or lymphoma in adults in the fifth and subsequent years since first exposure to the study medications were not increased as might be expected if these were causal effects. Even if causal, the public health impact of these excess risks would be low. Also, the prognosis of these outcomes might be improved through early diagnosis by doctors being alert to the potential development or unmasking of a cutaneous malignancy in patients with a dermatological condition treated with topical immunomodulators.