Voriconazole and the Risk of Keratinocyte Carcinomas Among Lung Transplant Recipients in the United States

This population-based cohort study examines associations of voriconazole and other antifungal medications with risk of keratinocyte carcinomas in recipients of lung transplants.

S olid organ transplant provides potentially curative treatment for patients with end-stage organ disease. The number of transplants has grown over time, with 34 770 solid organ transplants occurring in the United States in 2017, of which 2478 were lung transplants. 1,2 Solid organ transplant recipients have elevated risk for infections as well as many types of cancer, particularly virus-related cancers, which largely results from the immunosuppression caused by medications used to prevent graft rejection. 3,4 For unclear reasons, transplant recipients have a strongly elevated risk for keratinocyte carcinomas (KCs), which comprise cutaneous squamous cell carcinoma (SCC) and basal cell carcinoma (BCC). 5 Notably, cutaneous SCC is the most common cancer among transplant recipients and causes substantial morbidity, with rates especially elevated among recipients of lung transplants (hereinafter referred to as lung recipients). 6 Keratinocyte carcinoma risk factors among transplant recipients include white race (particularly for individuals with Fitzpatrick skin types I-III 7 ), residence in regions with high ambient UV radiation (UVR), older age, and history of skin cancer. 5,6,8 Although immunosuppression likely plays a role, SCC is not known to be caused by a virus, and risk is increased to a much smaller degree in immunosuppressed individuals with HIV infection. 9 Severe fungal infections are a major cause of mortality among lung recipients. 10 Voriconazole is a broad-spectrum, oral triazole antifungal medication that was first approved in the United States in May 2002. It is commonly given to lung and allogeneic hematopoietic stem cell recipients to prevent and treat invasive aspergillosis. Voriconazole prophylaxis is often administered immediately after transplant for a period ranging from several months to more than 1 year, and voriconazole may also be used preemptively after detection of aspergillus colonization. [11][12][13] Voriconazole is currently the first-line agent for prevention and treatment of aspergillus infections in lung recipients based on high-quality evidence. 13 Voriconazole is associated with adverse events, including skin phototoxicity reactions, [14][15][16][17][18] and case reports have described aggressive cutaneous SCCs occurring in solid organ and bone marrow transplant recipients during voriconazole treatment. [14][15][16] These reports led to a few modest-sized studies of solid organ transplant recipients (range, 68-543 recipients, which included a range of 17-86 SCC cases), 14 previous studies 14,[19][20][21][22][23][24] were small and performed at a single transplant center. Generalizing results of studies performed at a single site or a small number of sites is challenging, because transplant centers use diverse posttransplant treatment protocols. Moreover, only 1 study 27 examined the association between voriconazole and cutaneous BCC in lung recipients. That study reported no association. Finally, previous studies were underpowered to examine the association between newer antifungal agents, including posaconazole, 20 and KC risk.
The goal of the present study was to examine the association between voriconazole use and both SCC and BCC in a large population-based cohort of lung recipients. An ancillary aim was to examine the association between clinical alternatives to voriconazole, such as posaconazole, and KC risk.

Methods
This cohort study was approved by the human subjects committee at the National Cancer Institute, which waived the need for informed consent for use of registry data collected for surveillance and research purposes. We used data from the Scientific Registry of Transplant Recipients (SRTR), which captures all US solid organ transplants, linked with outpatient pharmacy claims for 2007 to 2016 from Symphony Health, a nationwide aggregator of health insurance claims. The SRTR provided information on incident SCC and BCC collected from transplant centers as part of required reporting of posttransplant malignant neoplasms. Because this ascertainment has low sensitivity (estimated, 14%-41%), 6,28 we missed a large proportion of KC cases. However, SRTR KC reporting has high positive predictive value (71%-88%) and specificity (99% for SCC). We obtained Symphony Health claims for immunosuppressant medications (include tacrolimus, cyclosporine, mycophenolate mofetil, azathioprine, sirolimus, everolimus, and corticosteroids) and antifungals (voriconazole, itraconazole, posaconazole, and other antifungals). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Our study included non-Hispanic white individuals who underwent lung transplantation from January 1, 2007, to

Key Points
Question What is the association between voriconazole, an antifungal used to treat aspergillosis infections, and the risk of keratinocyte carcinomas among recipients of lung transplants?
Findings In a population-based cohort study of 9599 non-Hispanic white recipients of 9793 lung transplants in the United States (2007-2016) with linkage to pharmacy claims, increasing cumulative voriconazole exposure was associated with an increased risk of cutaneous squamous cell carcinoma.
Meaning These findings suggest that physicians caring for lung transplant recipients at high risk for aggressive keratinocyte carcinomas should limit voriconazole exposure when possible and encourage skin protection behaviors and more frequent cancer screenings.
December 31, 2016, because our preliminary analyses indicated that claims for immunosuppressant medications were available for approximately 80% of recipients from 2007 onward. Other racial/ethnic groups experienced fewer than 5 outcomes and were thus excluded from the study. For included recipients, follow-up began at transplant and ended at the first cancer diagnosis (SCC or BCC, depending on the analysis), transplant failure or retransplant, death, loss to follow-up, or end of study (December 31, 2016). For each specific analysis, the other outcome (SCC or BCC) was ignored rather than censored. If a person had multiple successive transplants, each transplant was treated as a separate period of follow-up.
Follow-up was partitioned into 30-day intervals beginning at transplant. We assigned each medication claim a start date equal to the dispensed date and an end date equal to the start date plus the days' supply plus a 15-day grace period to account for small gaps in use or pill splitting. Exposure status for each medication was ascertained at the beginning of each 30-day interval. Because all recipients must use maintenance immunosuppressant medications to prevent rejection, we assumed that lack of immunosuppressant claims implied incomplete pharmacy claims information for an interval. We then excluded transplant recipients who had documented immunosuppressant coverage for less than 25% of their 30-day follow-up intervals.
Because many included recipients lacked immunosuppressant claims for some 30-day intervals, our classification of antifungal medication exposure included an unknown category. Specifically, a person was considered exposed to an antifungal medication if a claim indicated use of that medication, and unexposed if no claim for the antifungal was observed but a claim for at least 1 immunosuppressant medication was observed. We then assigned unknown antifungal exposure status for an interval if there was no claim for the antifungal medication and no claim for an immunosuppressant.
On the basis of the assignments for each 30-day interval, we created time-dependent variables that captured use of each antifungal medication at the start of each interval. Everexposed status was assigned if the current interval or any prior interval indicated exposure to that medication. If no current or prior interval had an exposed status and more than 75% of all current or prior intervals had unexposed status, we assigned never-exposed status. This approach was necessary because a large proportion of the population had unknown exposure status during their first interval (likely resulting from initial inpatient receipt of medications, which is not captured by the Symphony Health pharmacy claims). Thus, a person could transition from unknown to unexposed status once a preponderance of claims indicated lack of exposure. All other intervals were classified as having unknown exposure status. For descriptive purposes, we present estimates of antifungal medication use according to time since transplant and calendar year of transplant. Specifically, the proportion of the cohort exposed to each antifungal medication in these categories was calculated as the proportion of exposed person-time divided by the known person-time (ie, exposed or unexposed).
Data were analyzed from March 1, 2018, to February 13, 2019. From the SRTR, we obtained data on recipient demo-graphic and transplant characteristics. Mean annual ambient exposure to UVR was derived by mapping recipient residence as previously described. 29 Univariate associations were examined between these variables and SCC or BCC in separate Cox proportional hazards regression models. We used time since transplant as the timescale and stratified the baseline hazard by calendar year of transplant (in 2-year increments). Variables that were statistically significantly associated with the outcome were then included in a multivariable model, and those that remained associated with the outcome after mutual adjustment were retained for a final base model (eMethods in the Supplement). We included ambient UVR exposure in the final models regardless of statistical significance because of voriconazole's documented photosensitizing effects. 14 We then used Cox proportional hazards regression models to assess associations between antifungal medications (treated as time-dependent variables) and the risk of SCC or BCC. We evaluated current or prior antifungal exposure using variables that captured ever, never, or unknown and cumulative exposure based on the total number of exposed intervals (unknown intervals did not contribute to this calculation). We also considered different intervals to lag the exposure (0, 180, and 360 days) to assess delayed effects of the medications. When cumulative exposure of an antifungal medication was associated with the outcome, we report results for both the continuous measure (per 30-day interval) and based on percentile categories of use (approximately <33rd, 33rd-66th, 66th-90th, and >90th percentiles).

Results
A total of 18 144 first or subsequent lung transplants were performed in the United States from 2007 to 2016, of which 16 180 were performed in non-Hispanic white individuals. After excluding 6371 transplants with immunosuppressant coverage for less than 25% of 30-day intervals and 16 transplants lacking smoking information or mappable UVR exposure, we included 9793 lung transplants among 9599 unique individuals in the present study ( Table 1) Table 2). There were 347 BCCs observed during follow-up (incidence, 101 per 10 000 person-years). Our model-fitting procedure identified ever use of itraconazole and ever use of posaconazole as risk factors for BCC, but there was no association with cumulative exposure (eTable 3 in the Supplement); the model with no lag fit the data well. In the final model (

Discussion
In this population-based study of lung recipients, voriconazole exposure was strongly associated with SCC risk, such that individuals with more than 15 months of exposure experienced 3-fold increased risk of SCC compared with unexposed individuals. Although itraconazole was also associated with SCC, the association was modest in magnitude and of borderline significance, and no association with duration of use was found. In addition, both itraconazole and posaconazole were associated with BCC risk, but neither was associated with BCC in a cumulative fashion.
Our results showing an association between cumulative voriconazole exposure and SCC risk are similar to those reported by Hamandi et al 25 for lung recipients and in a recent meta-analysis of studies of lung and hematopoietic stem cell transplant recipients. 26 Our study of 9793 lung transplants is much larger than prior studies and included 1031 SCC cases compared with 55 in Hamandi et al 2 5 and 185 in the meta-analysis. 26 This large size allowed us to model the association with increasing duration of voriconazole and adequately examine associations with clinical alternatives to voriconazole. Similar to previous reports, 21,22,25,27,30 we found that increasing age and male sex were risk factors for SCC and BCC. Smoking history, which was not assessed in prior studies, was also positively associated with SCC and BCC in the present study. Exposure to UVR was included as a potential confounder, although it was not significantly associated with skin cancer in the multivariate models. Although voriconazole can cause severe phototoxic reactions, [15][16][17][18] carcinogenic mechanisms are poorly understood. Voriconazole exposure modifies keratinocyte gene expression, specifically affecting genes related to the cell cycle, including the FOXM1 pathway-which is associated with SCCand terminal differentiation pathways. 30 Voriconazole's primary metabolite (voriconazole n-oxide) and UV-A radiation can act together to damage DNA and inhibit apoptosis. 31 Moreover, voriconazole can induce upregulation of cyclooxygenase 2, which is critical to SCC progression. 31  In contrast to our findings for SCC, we found no evidence of an association between voriconazole and BCC. Our BCC results are also comparable to those of the recent meta-analysis, 26 but the meta-analysis included only 2 studies with a total of 41 BCC cases. Both itraconazole and posaconazole were associated with increased BCC risk, although we did not find a cumulative effect. The positive associations are puzzling because no other evidence links either medication to BCC risk, and limited evidence suggests that both may actually limit BCC progression. [32][33][34] We speculate that our BCC results may be due to confounding by physician prescribing preferences. After publication of reports documenting a possible association between voriconazole and aggressive skin cancer first appearing in 2007 15 and 2010, 16 physicians may have channeled highrisk patients (eg, those with fair skin or a history of skin cancer) away from voriconazole and toward alternative medications. This pattern of prescribing, if present, may also explain the small increased risk for SCC associated with itraconazole. Moreover, it might also have led us to underestimate the voriconazole-SCC association if patients less susceptible to SCC were preferentially prescribed voriconazole.
Antifungal medication use was highest in the first few years after transplant, corresponding to when lung recipients are at highest risk of aspergillosis and other fungal infections. Interestingly, voriconazole was the most commonly used antifungal immediately after transplant, but use declined precipitously compared with other antifungals after the first year after transplant, which may reflect physicians' perception, based on earlier studies, that longer voriconazole use increases SCC risk. 22,25 Voriconazole use decreased during the study period, and there appeared to be an increase over time in use of posaconazole, a clinical alternative to voriconazole that received US Food and Drug Administration approval in late 2006.

Strengths and Limitations
Our study has several notable strengths. It is population based and the largest study to date, including 61% of all lung transplants among non-Hispanic white recipients in the United States during 2007 to 2016. The study's large size enabled us to statistically adjust for all measured risk factors and assess alternative antifungal medications to voriconazole. By linking pharmacy claims, we had precise medication exposure information. Incorporation of a grace period allowed for late medication pickup and pill splitting, and our requirement for immunosuppressant coverage enhanced confidence in the classification of antifungal use. In addition, smoking history, an SCC risk factor 35 not assessed in previous studies, was largely complete in our data.
Nonetheless, some limitations should be considered. We lacked data on skin cancers before transplant. As previously mentioned, physicians may have chosen antifungal medications based on this history, which could have biased associations, most likely in a conservative direction with respect to voriconazole. We assessed mean ambient UVR exposure based on recipients' residence at the time of transplant. 29 However, this measure does not capture previous residences, time spent outdoors, or sun protection behaviors, which may explain why we did not see strong associations between this measure and risk of KC.
Our linkage to pharmacy information was incomplete for some individuals and was commonly missing immediately after lung transplant, when they were likely in the hospital or using medications that they brought home with them. We attempted to mitigate this limitation by creating a category for unknown antifungal exposure status when immunosuppressant claims were not observed for a given interval. We also assigned individuals who began with unknown antifungal status to never-exposed status after they accumulated sufficient unexposed intervals (>75% of their previous intervals), which increased the likelihood that these individuals were truly unexposed.
Finally, SCC and BCC were ascertained from SRTR based on transplant center reports. Although this ascertainment has low sensitivity, the positive predictive value was high, which supports the validity of the captured cases. As a result of this misclassification, the AHRs may be slightly attenuated and our associations somewhat conservative. In addition, we lacked information on identified cancer features and thus cannot determine whether voriconazole is associated with deadly or lowrisk SCC.

Conclusions
In this study, voriconazole was associated with increased SCC risk among lung recipients, especially after prolonged exposure. Voriconazole is a highly effective treatment for aspergillosis-a major source of mortality among lung recipientsand is thus the recommended first-line treatment against suspected or confirmed infections. 13 Although the efficacy of antifungal medications is the key consideration for the treatment of established infections, the balance of benefits and risks likely differs for prophylaxis, and broadly accepted guidelines are lacking. 36 Given the high morbidity associated with SCC in the transplant setting, it is reasonable that physicians should consider the risk-benefit ratio of shorter durations of voriconazole prophylaxis or using alternatives to voriconazole in patients with high KC risk, encourage screening, and advise them about sun protective measures. Future research should focus on identifying the minimal time necessary for effective prophylaxis in lung recipients and quantifying the risks and benefits of prophylaxis with various medications in specific populations, including further assessment of posaconazole and itraconazole as alternatives with potentially lower SCC risk.