As the number of 1 L and 2 L treatment options for mRCC increases, and the role of effective sequencing of agents evolves, a thorough understanding of the treatment patterns and AE profiles of each drug class in the real-world setting will be critical to providing medical benchmarks, assessing adherence to guidelines and informing benefit:risk decisions in selecting the appropriate mRCC treatment. This retrospective, claims-based analysis evaluated treatment patterns and AEs among a large, real-world, US population of patients with mRCC. A cross-sectional analysis was performed to better understand the patterns of 1 L treatment initiation over a 10-year period, and a longitudinal analysis provided information of treatment choice, baseline correlates and associated AEs in patients treated in routine medical practice.
The cross-sectional trend analysis demonstrated that 1 L treatment initiation patterns for mRCC generally reflected the US Food and Drug Administration approvals and NCCN treatment guidelines, with a few exceptions. The rapid uptake of the TK-targeting agents sunitinib (approved in 2006) and pazopanib (approved in 2009) was noted, establishing TK/VEGF-directed treatment as the most widely prescribed drug class initiated during the study period. Throughout the study period, oral treatments were also more commonly used than IV treatments. These patterns of TK/VEGF inhibitors and oral agents were also recapitulated in the longitudinal analysis. These findings are consistent with those from studies detailing the widespread use of TK inhibitors [26], and the less frequent use of sorafenib and axitinib in the 1 L setting is also consistent with NCCN recommendations for these TK inhibitors [8]. Of note was the finding that the use of bevacizumab as monotherapy was more prevalent than as combination therapy with IFN-α. These findings suggest that increased costs and IFN-related toxicities may render bevacizumab more attractive as monotherapy than as combination therapy with IFN [27]. Moreover, Phase II data support the use of bevacizumab as 1 L treatment and salvage therapy [28, 29]. Additionally, a considerable number of patients received everolimus in the 1 L setting despite the lack of indication in this setting, which reflects the wide variability seen in provider preference and clinical experience.
Provider preferences, patient history and known toxicities associated with drug classes may drive 1 L treatment choice, although it is difficult to ascertain from claims data which characteristics influenced the specific choices of mRCC treatment. In this study, CHF and a DCCI score ≥ 4 were independent predictors of IV treatment choice, while patients with lung metastases were less likely to receive IV treatment than were those without lung metastases. Patients with diabetes were less likely to receive mTOR-directed therapy, while patients with comorbid CHF, liver metastases or bone metastases were more likely to receive mTOR-directed therapy. These data are in agreement with the known cardiotoxicity associated with both sunitinib and pazopanib [23, 25], such as cardiac dysfunction (sunitinib, 11% frequency; pazopanib, 13%) and myocardial infarction/ischemia (sunitinib, 4% frequency; pazopanib, 2%) [11]. Similarly, a preference for TK/VEGF inhibitors in those with diabetes can be explained by the fact that hyperglycemia is a well-known AE associated with mTOR inhibitors (observed in 26% of patients treated with temsirolimus [30] and 50% treated with everolimus [31]) .
Duration of treatment also varies by agent and may influence treatment choice. Median duration of TK/VEGF-directed treatment was much longer than that of mTOR-directed treatment (6.3 and 3.9 months); similarly, oral treatment duration was nearly twice that of IV treatment (6.6 and 3.4 months). The median durations of sunitinib and pazopanib treatment were similar to each other (6.5 and 7.0 months, respectively), but were longer than those of sorafenib (4.7 months), everolimus (4.0 months), and temsirolimus (3.9 months). Reported results of real-world studies are mixed. For example, the observed sunitinib and sorafenib treatment durations were similar to those reported by Feinberg et al. (5.9 and 5.5 months, respectively) [32] and only slightly longer than those reported by Miller et al. (5.6 and 5.3 months, respectively), who also reported durations of 5.3 months for pazopanib and 4.5 months for everolimus [14]. However, these durations are slightly longer than those reported by Hess et al. (3.2 and 4.0 months, respectively; 2.6 months for temsirolimus) [33] and Vogelzang et al. (sunitinib, 4.1 months; pazopanib 4.8 months) [15]. However, it should be noted that direct cross-study comparisons are not possible based on different analysis methods and populations. Treatment durations reported in the randomized clinical trial setting are also sometimes mixed and are not dissimilar to the results from this study. For instance, Phase III data have shown 1 L median treatment durations of 7.6 and 8.0 months, respectively, for TK inhibitors sunitinib and pazopanib [11] and 3.9 months for the mTOR inhibitor temsirolimus [30]. Collectively, these data suggest that there are potential differences in patient selection, outcomes measurement and patient preference between real-world data studies and randomized clinical trials that could have implications for future clinical trial design. These differences may reflect the difficulty in (1) maintaining adequate dosing with these relatively toxic agents and (2) reproducing clinical trial results in the real-world setting for mRCC.
Potential differences in toxicities between the 1 L treatments in this study vs those previously seen were also evaluated, and it was noted that the most common AEs associated with each drug class were generally consistent with previously reported results [34,35,36,37] and with the product labels [19,20,21,22,23,24,25]. Notably, however, substantial latency of onset was observed for several potentially treatment-related toxicities in patients treated with both TK/VEGF- and mTOR-inhibitor classes, which was much different in clinical practice (i.e., onset of fatigue, hypertension and hepatic dysfunction generally occurs quickly). AE latency in the database may be due to capture of only the toxicities that generate a medical claim and may suggest that the toxicities that are observed early on may not receive medical care in clinical practice. There were also AEs potentially associated with checkpoint inhibitors. The latency of these AEs was similar to that observed with checkpoint inhibitors [38, 39]. These results suggest that attributing toxicities to TK/VEGF-directed therapies vs checkpoint inhibitors, when used in combination, may be challenging due to overlapping toxicities.
In contrast to the 1 L treatment data, 2 L treatment data were less defined. In this study, nearly half of all patients did not have evidence of receiving 2 L treatment (47% [n = 940]); however, there is no reliable way to determine the reasons for this from a claims database. Patients who did not receive 2 L therapy in this study may either still be receiving 1 L targeted therapy (and were not captured in the study due to the follow-up period ending) or have attained sustained remission, died before receiving 2 L therapy, refused 2 L treatment or did not receive 2 L therapy for unknown reasons. Consistent with other studies, most patients treated with a 1 L oral TK inhibitor who received 2 L therapy were switched to 2 L mTOR-directed treatment, primarily everolimus [26]; 2 L treatment choice warrants further investigation.
This study had several strengths and limitations. Strengths include the large number of 1 L–treated patients with mRCC over time, including those treated with TK/VEGF- or mTOR-directed therapy. The use of the MarketScan databases also provided strength to this study due to the inclusion of the full continuum of care in all inpatient and outpatient settings, as well as retail and specialty pharmacies, and the longitudinal tracking of patient information. However, the MarketScan databases also have limitations, as they are inherently restricted to insured patients and provide limited social background and demographic data. Secondary metastasis codes were not required for identification of patients with mRCC since the 1 L treatments are approved in the metastatic setting only and to avoid exclusion of eligible patients due to potential underreporting of secondary metastasis codes. However, the study could have included patients who received the treatments off-label in the adjuvant/neoadjuvant setting. Further, administrative claims-based data are prone to coding and data entry errors. Potential underreporting of secondary metastasis and baseline comorbidities may have affected the point estimates in the multivariate models, and the distribution of these variables may not be representative of the RCC population in clinical practice. Moreover, only AEs of sufficient severity to prompt medical attention and generate a claim could be identified, which may have led to underreported IRs. It was also not possible to discern from the claims data whether an AE was reported due to a drug reaction, RCC progression or other cause, nor was it possible to know the severity/grade of the AE. Further, it is unknown from the claims data whether 2 L treatment was received due to toxicity or disease progression. In this study, the treatment duration end date was calculated by adding the days’ supply of oral treatment and cycle length for IV treatment. Since most patients receive their oral drug supply 1 to 3 months in advance, the treatment duration for oral medications may have been overestimated. Additionally, the proportion of patients who switched treatment may have been underreported due to the limited follow-up in this study (median of ≈ 16 months after the index date). However, this median follow-up may be of limited concern considering the low survival rates in this patient population [40, 41]. Lastly, recent approvals (e.g., cabozantinib and nivolumab ± ipilimumab) and treatment recommendations were not captured given that these agents were not approved during the period covered by the analysis [42,43,44].