Metastatic melanoma patient outcomes since introduction of immune checkpoint inhibitors in England between 2014 and 2018

Abstract Immune checkpoint inhibitors (CPIs) have radically changed outcomes for patients diagnosed with metastatic melanoma globally in the last 10 years, based on evidence of overall survival (OS) benefits generated from international randomised controlled trials (RCTs). Since RCTs do not always reflect real‐world prescribing, we interrogated established national databases to track prescribing of CPIs approved for first line treatment of metastatic melanoma patients in England since 2014 and determined patient outcomes associated with OS, as well as treatment‐related toxicity. Between April 2014 and March 2018, 5465 melanoma patients were diagnosed and treated with systemic anticancer therapy (SACT), 2322 of which received first‐line CPIs. There was good 3‐year OS concordance with RCT outcomes for ipilimumab (32%), ipinivo (56%) and nivolumab (51%), but OS was lower than expected for pembrolizumab (40%). Comparing patients prescribed ipinivo with those prescribed pembrolizumab, ipinivo‐treated patients were younger (88% vs 49% patients <70 years, P < .001) and fitter (60% vs 38% patients with Eastern Cooperative Oncology Group [ECOG] performance status 0, P < .0001). Emergency hospital admission rates from the earliest and last treatment dates were higher for patients prescribed ipinivo (37% and 55%) compared to those prescribed pembrolizumab (17% and 29%). The 30‐day mortality rates favoured ipinivo patients (3.8% ipinivo, 9.1% pembrolizumab, P < .0001) and likely reflected marked differences in median treatment durations: 63 (range 7‐440) days for ipinivo and 192 (range 5‐943) days for pembrolizumab. The dominant treatment‐related condition linked to hospital admission was colitis, recorded for 25% of patients prescribed ipinivo compared to 4% of patients prescribed pembrolizumab. Our population data has demonstrated that RCT outcomes can be achieved in routine care settings with careful patient selection.


| INTRODUCTION
Immune checkpoint inhibitors (CPIs) have markedly changed the treatment landscape for patients with metastatic melanoma over the last 10 years, due to proven impact on overall survival (OS) reported in international randomised controlled trials (RCTs). [1][2][3][4][5] However, realworld populations do not always reflect RCT populations controlled by strict trial eligibility criteria and poorer outcomes for patients treated outside clinical trial settings have been reported. 6,7 CPIs are high-cost drugs with complex side effects that vary from very mild and manageable to life-changing, as well as life-threatening. 8 Since the first reported OS benefits reported in patients diagnosed with metastatic melanoma, this new class of immunotherapy drugs is being used increasingly to treat a wide range of cancers and the resource implications for healthcare systems worldwide is concerning. 9 The translation from trials to clinical practice in metastatic melanoma patient management provides a unique window through which to observe how CPIs were adopted across a national health service and have influenced population outcomes.
Since 2011, a series of CPI antibodies have been approved for the treatment of metastatic melanoma (encompassing unresectable locally advanced and distant metastatic spread). The CTLA-4 antibody, ipilimumab, was the first systemic therapy to demonstrate OS benefit in this patient population in a RCT. 1 Subsequently, the anti-PD-1 antibodies, nivolumab and pembrolizumab were shown to be more effective compared to either ipilimumab, or chemotherapy, respectively, 2,3 while also generating fewer severe, or life-threatening side-effects. The combination of ipilimumab and nivolumab (ipinivo) evaluated in the CheckMate 067 trial has the highest objective response rate and OS of all current first-line metastatic melanoma immunotherapy options. 4,5 However, the OS gain associated with ipinivo compared to anti-PD-1 antibody alone was reported to be small (52% vs 44% at 5 years), while toxicity was far more severe (58% vs 23% severe or life-threatening adverse events).
There is currently no published national or international consensus guidance on selecting between anti-PD1 monotherapy or ipinivo as first line treatment for patients with metastatic melanoma. However, the implications of treatment choice for both patients and healthcare systems can be profound, given that these antibodies are high cost drugs with complex and unpredictable side-effects, administered for variable durations lasting months to several years in some instances.

| MATERIALS AND METHODS
This work uses data that has been provided by patients and collected by the NHS as part of their care and support. The data is collated, maintained and quality assured by the National Cancer Registration and Analysis Service, which is part of Public Health England (PHE).
All individuals diagnosed with a melanoma skin cancer (ICD10 C43) in England between April 1, 2010 and December 31, 2017, were extracted from the National Cancer Registration Dataset. 11 Cases were linked to the national SACT dataset 12 using unique patient NHS numbers to identify those patients with confirmed unresectable locally advanced stage III or IV metastatic melanoma receiving first-line CPI therapy with ipilimumab, pembrolizumab, nivolumab or ipinivo. First-line therapy was confirmed by ensuring there was no record of any systemic therapy being prescribed for advanced disease prior to being prescribed a CPI.
The patient data was linked to the Hospital Episodes Statistics (HES) 13 Inpatient and Accident and Emergency (A&E) datasets to establish how many patients either attended or were admitted within 30 days of their first and last CPI treatment record in the SACT database. 14  unplanned care. There are several routes through which a patient can be admitted for emergency care, including managed routes such as via a General Practitioner, as well as being admitted following an A&E attendance. Thirty-day emergency hospital attendance or admission rates measured from first CPI treatment date were assessed, aiming to capture events more likely to be due to treatment-related toxicity; 30-day emergency hospital attendance or admission rates measured from the last CPI treatment date were expected to capture events more likely to be associated with treatment cessation, which could include disease progression, as well as treatment-related toxicity.
The conditions associated with emergency hospital admissions were identified by ICD-10 codes recorded locally. These were collated and explored to determine whether and to what extent treatmentrelated toxicities-in particular, immune-related adverse events (IrAEs)-contributed to emergency hospital admissions. To examine for IrAEs of special interest, specific ICD-10 codes (refer Table S1 for code names) were grouped to identify episodes of colitis (K521, K529 and A099), endocrinopathy (E059, E222, E230, E231, E236 and E871) and hepatitis (K711, K712, K716, K718, K754 and K759).
In addition, we looked for any other frequently occurring ICD-10 codes which accounted for >10% total recorded codes in any one treatment group.

| Statistical analysis
OS was calculated using Kaplan-Meier methodology from a patient's earliest CPI treatment date recorded in the SACT database and the patient's date of death, or the date the patient was last traced for whether they had died (vital status). All patients were traced for their vital status using the Personal Demographics Service (PDS) 15 on October 2, 2019; this date was used as the censor date if patients were still alive. Patients were followed for a minimum of 18 months. To establish significant differences in OS, we compared 95% confidence interval estimates. Thirty-day mortality rate was calculated from a patient's last treatment date recorded in the SACT database to their date of death. 14 Thirty-day emergency hospital attendance or admission rates were calculated from a patient's first and last treatment date recorded in the SACT database to any emergency attendance or admission dates recorded in the HES database. We used a two-samples test of proportions to compare clinical characteristics and outcomes associated with patients prescribed ipinivo and patients prescribed pembrolizumab, or ipilimumab, and used the more conservative significance value of α = 0.01 to correct for these multiple statistical tests.   with ipinivo ( Figure 3).

| Toxicity outcomes
Thirty-day emergency hospital attendance or admission rates measured from first CPI treatment date were assessed, aiming to capture events more likely to be due to treatment-related toxicity. From patients' first CPI treatment (Table 2), 30-day rates of emergency admissions were higher in ipinivo (37%) compared to ipilimumab (23%) (although this difference was not significant following the multiple comparisons correction [P = .013]) and pembrolizumab (17%, P < .0001). Thirty-day rates of A&E attendances from a patient's earliest CPI treatment were also significantly higher in association with ipinivo (16%) compared to ipilimumab (9%, P < .0001) or pembrolizumab (8%, P < .0001).
Thirty-day A&E attendance or emergency admission rates measured from the last CPI treatment date were expected to capture events more likely to be associated with treatment cessation, which could include disease progression, as well as treatment-related toxicity. Table 2 demonstrates emergency admissions were significantly higher with ipinivo (55%) compared to ipilimumab (40%, P < .0001), or with pembrolizumab (29%, P < .0001). Rates of 30-day A&E attendances following a patient's last recorded CPI treatment were not significantly different in the three groups (19% with ipinivo, 14% with pembrolizumab and 15% with Ipilimumab, P = .33).
After excluding for ICD-10 codes associated with melanoma diagnosis or co-morbidities, a total of 922 ICD-10 codes were linked to 889 unique patients admitted as an emergency. Only a small minority of these ICD-10 codes could be strongly attributed to IrAEs (Figure 4)  reflecting rapid uptake of the combination CPI regimen reported to offer the highest OS compared to single-agent CPI regimens in RCTs. 4,5 There was remarkably strong concordance between 3-year OS outcomes in national clinical practice for patients prescribed ipilimumab, nivolumab and ipinivo compared to the respective RCTs.
The exception was for those patients prescribed pembrolizumab, who had a lower than expected 3-year OS of 40% compared to 51% reported in the KEYNOTE 006 trial. 17 Access to prescribing most highcost drugs in England is limited to populations reflecting the registration trial entry criteria. However, while access to ipinivo is limited to patients with ECOG PS 0 or 1 with no active brain metastases (reflecting better prognostic groups), when pembrolizumab was first made available in England, access was not strictly limited in this way.
The preponderance of older, less fit patients prescribed pembrolizumab may account for their poorer OS outcomes: entry into the KEYNOTE 006 trial 16 was restricted to ECOG PS 0-1 patients, while 9% of our real-world pembrolizumab population had ECOG PS 2 or higher.
Selecting initial treatment with a single agent anti-PD-1 antibody or ipinivo is particularly challenging for melanoma specialists, since good predictive biomarkers are not available to help patients and clinicians weigh up the relatively modest gain in survival chances associated with ipinivo vs significantly increased risk of severe, potentially life-threatening sideeffects. 5 We identified 2-fold higher rates of emergency hospital attendances and admissions within 30 days of starting ipinivo therapy compared to those occurring after starting pembrolizumab. This is despite the younger and fitter patient cohort prescribed ipinivo in comparison to pembrolizumab. The hospital admission rate after last treatment was also doubled in patients receiving ipinivo compared to pembrolizumab.
Assessment of events occurring within 30 days of first treatment was anticipated to identify causes more likely to be treatment-related, while events occurring within 30 days of last treatment was anticipated to identify causes more likely to be disease-related.
The ipinivo regimen approved for treatment of metastatic melanoma is high dose ipilimumab (3 mg/kg) plus nivolumab (1 mg/kg) given on the same day once every 3 weeks for a total of 4 cycles, followed by maintenance nivolumab monotherapy; pembrolizumab is given intravenously on a 3-weekly cycle; both regimens can be continued for as long as there is clinical benefit. The highest risk of toxicity events associated with ipinivo is during the first 12 weeks of combination therapy; median number of treatment cycles reported in the CheckMate 067 trial was 4, with 39% of patients discontinuing due to treatment-related adverse events. 4 In our population, the median duration of ipinivo treatment was 9 weeks, equivalent to receiving 3 out of 4 planned combination cycles. Therefore, it is reasonable to assume that events occurring after both first and last As this was an observational rather than a randomised study, we cannot conclude rates of hospital admissions are a direct result of their CPI treatment or their metastatic melanoma rather than due to other different pre-existing characteristics of the patient groups.
However, the higher rates of colitis (25% vs 4%), hepatitis (6% vs 2%) and endocrinopathies (5% vs 1%) in patients receiving ipinivo compared to pembrolizumab mirror the higher rates of these IrAEs observed with the combination CPI regimen compared to single-agent anti-PD-1 antibodies in clinical trials. 4,5 The incidence of these specific IrAEs recorded for our patient populations are in fact lower than those reported in the randomised trials. We have reported events associated with hospital admissions, while a proportion of patients experiencing these toxicities will be successfully managed in the clinic.
Furthermore, reporting toxicities in real life populations is challenging.
Review of the huge volume of hospital admission ICD-10 codes identified that there were no "one size fits all" codes directly reflecting