Temporary treatment cessation compared with continuation of tyrosine kinase inhibitors for adults with renal cancer: the STAR non-inferiority RCT

Summary of advances in new agents to treat renal cell carcinoma

Since the start of the STAR trial in 2012, there have been a number of treatment advances in systemic therapy for metastatic renal cancer. These include the routine use of additional TKIs such as axitinib and tivozanib and agents which are combined TKI and c-MET inhibitors (small molecules which inhibit activity of c-MET tyrosine kinase, the receptor for hepatocyte growth factor, HGF/SFs) such as cabozantinib, new combinations such as everolimus and lenvatinib (a newer TKI), all given orally. 33,34

Renal cancer is an immunosensitive disease. Over recent years, there has been great success in a number of cancers from immunotherapeutic agents, for example in melanoma, lung cancer and bladder cancer. Treatment with PD-1 and PD-L1 inhibitors and CTLA4 (cytotoxic T-lymphocyte-associated protein 4) inhibitors has now also become the standard of care in renal cancer. 35 There is however still a role for first-line TKI treatment in a number of patients.

Progressive disease-1 and PD-L1 inhibitors are a group of checkpoint inhibitor anticancer drugs that block the activity of PD-1 and PD-L1 immune checkpoint proteins present on the surface of cells. CTLA-4 is a protein receptor that functions as an immune checkpoint and downregulates immune responses. CTLA-4 is constitutively expressed in regulatory T cells but only upregulated in conventional T cells after activation. This phenomenon is particularly notable in cancers. Initially, agents such as nivolumab were given as monotherapy in second-line metastatic renal cancer treatment. However, more recently, ipilimumab (CTLA-4 inhibitor) and nivolumab (PD-1 inhibitor) in combination has been given as standard of care in the first-line metastatic setting in renal cancer (approved by NICE in 2019). 36

Over the last 2–3 years, immunotherapy agents combined with TKI therapy have also become a standard of care in the first-line setting in mRCC. 35 Such immunotherapies are given intravenously and include agents such as nivolumab, ipilimumab, pembrolizumab and avelumab.

Currently, for some patients, TKIs such as pazopanib and sunitinib remain first-line therapy, either because patients prefer to start with oral therapies or because they are unsuitable for combination immunotherapy, or TKI with immunotherapy, due to contraindications such as pre-existing autoimmune disease or comorbidities making them not suitable for combination therapy. In addition, those patients who are treated with immunotherapy in the first-line setting are likely to receive TKI therapy as second, third or fourth line, depending on their fitness for further therapy. Notably, with immunotherapy alone or in combination with TKIs, questions regarding treatment breaks similar to those addressed in the STAR trial remain very relevant, as these combinations are expensive, have significant toxicity and are currently continued until disease progression.

Although in 2021 first-line treatment of advanced renal cancer remains non-curative, the hope is that with further treatment advances, survival will increase.

Adjuvant therapy in renal cell carcinoma

Patients may initially present with metastatic renal cancer at the time of diagnosis or following potentially curative nephrectomy for localised renal cancer. In the latter group which is particularly common in intermediate or high-risk Leibovitch groups, individuals may develop metastases. In high-risk groups, the risk is up to 50% over a 5-year period. Until recently, most adjuvant trials have not convincingly demonstrated significant advantages of giving TKI therapy in the adjuvant setting, so the standard of care has been regular follow-up and imaging. However, with the increasing success of immunotherapy in treating metastatic renal cancer, encouraging signals are starting to emerge using immunotherapy in the adjuvant setting in intermediate- or high-risk renal cancer where there is a substantial risk of developing metastatic disease in the next 5 years following nephrectomy (as defined by Leibovitch score at nephrectomy). 37–39 In this situation for those patients who progress on adjuvant immunotherapy, TKIs may also remain the standard of care in the first-line metastatic setting.

Prognostic risk score

At the start of the STAR trial, the Motzer score remained the main prognostic index in RCC and was therefore used for stratification in control and treatment arms for this study. The Motzer score is used as a tool to determine survival and is based on data from a study reported in 1999 in which 670 patients were treated at Memorial Sloan Kettering Institute in the USA. The Motzer score looks at five key indices, scoring one for each ‘yes’ category for the following: PS (Karnofsky score) < 80%; time from diagnosis to systemic therapy < 12 months; haemoglobin less than the lower limit of the normal range; corrected calcium levels > 10 mg/dL and lactate dehydrogenase (LDH) > 1.5 × upper limit of normal (ULN). This divides the patient population with metastatic renal cancer into three prognostic groups: ‘favourable’ who have a Motzer score of 0, ‘intermediate’ who have a score of 1 or 2 and ‘poor’ with a score ≥ 3, with calculated survival reducing from favourable to poor groups. This score was however developed in patients treated with cytokines and interferon and not with more modern treatments such as TKIs and immunotherapy.

More recently, the International Metastatic RCC Database Consortium (IMDC) Risk Score for RCC has been introduced and predicts OS in patients treated with systemic therapy. It has become increasingly used as it has a number of advantages in predicting survival in prospective studies over the Motzer score using more modern treatments. This IMDC score was developed by Daniel Heng from Tom Baker Cancer Centre, Calgary, Canada, who developed a new set of variables that included neutrophil count. 40 Upon validation in a kidney cancer trial, Heng’s model came closest to predicting death after a 2-year period compared with a number of other models. IMDC is now one of the standard models currently used for mRCC.

The factors included in the IMDC model are shown in Appendix 1 with the scores that correlate with each factor. The total score is then categorised into favourable (0 risk factors), intermediate (1–2 risk factors) and poor (> 2 risk factors).

Risk categories in advanced renal cancer according to the IMDC score

More recently, a further adaption has also been made to the original Motzer score, which has been extended to include not just the original five factors but also primary radiotherapy and ≥ 2 metastatic sites (the Mekhail Extension41); however, it remains broadly similar.

For the STAR trial, in terms of the statistical prospective analysis, the IMDC score was also calculated (as well as the original Motzer score) for individual patients and included as a subgroup analysis for the primary analyses. This was felt important as, with development of immunotherapy and combined TKI therapy, determination of risk score in individual patients has become more important to guide treatment choice.

Axitinib

Axitinib is an oral multitargeted TKI with antitumour activity. It selectively inhibits VEGF receptors 1, 2 and 3, platelet-derived growth factor receptor, and c-kit, which may inhibit angiogenesis in tumours. Axitinib alone is not currently recommended or NICE-approved for the first-line setting in advanced RCC. However, more recently, it has been considered in combination with immunotherapy first line.

A large trial (AXIS) assessed axitinib for the second-line treatment of patients with advanced RCC. AXIS was a Phase III, international, multicentre, randomised, open-label, active-controlled trial comparing axitinib with sorafenib for treating advanced or mRCC after failure of first-line systemic therapy. 42 This study resulted in axitinib being recommended as an option for treating adults with advanced RCC, after failure of treatment with a first-line TKI or a cytokine, and resulted in it becoming part of standard second-line treatment in RCC. It is given orally, initially starting at a dose of 5 mg twice daily, but can be increased or titrated to blood pressure (BP) readings in the patient to 7 or 10 mg twice daily. Similarly, it can be decreased in those with toxicity to 3 mg or a minimum dose of 2 mg twice daily as needed.

Tivozanib

The main clinical evidence for routine use of tivozanib came from TIVO-1, an open-label randomised controlled trial (RCT) that primarily investigated whether tivozanib (n = 260) prolongs time to disease progression compared with sorafenib (n = 257). 43 Most of these patients were recruited from Eastern Europe rather than the UK. Upon disease progression, patients in the sorafenib group could switch (cross over) to treatment with tivozanib. As a result of this study, tivozanib was approved by NICE for its use as a first-line treatment for adult patients with advanced RCC. 17 It is given for 3 weeks on and 1 week off for a 4-week cycle. The starting dose of 1340 mg with (as in other TKIs) treatment continues until disease progression or unacceptable toxicity. Only one dose reduction is allowed. The cost of treating RCC with tivozanib is likely to be lower than the cost of treating with sunitinib or pazopanib (mainly because of drug costs), but tivozanib is also likely to be less effective on the basis of the trial data. However NICE felt that the estimated cost savings are high enough to compensate for the estimated lower effectiveness (using the tivozanib open access scheme) and it was approved by NICE in 2018. This approval is relevant for future NICE considerations of TKI intermittent treatment strategies in this setting. The toxicity pattern is slightly different to pazopanib and sunitinib, but it has not proven to be better and survival data also appeared to be inferior to S/P. 16

Cabozantinib

Cabozantinib is a combined MET and AXL inhibitor, in addition to inhibiting VEGF receptors. It is given orally with a standard approved dose of 60 mg daily and allows two dose reductions for toxicity to 45 mg and then 20 mg daily. Initial approval from NICE was given in August 2017 as an option for treating advanced RCC in adults, after VEGF-targeted therapy, as a subsequent line of therapy. During this time, axitinib and nivolumab were also available post first line in the advanced RCC setting. Therefore, NICE allowed cabozanitinib to be used after one or two lines of prior therapy in the advanced setting.

This NICE approval44 was largely based on data from the METEOR study comparing cabozantinib with everolimus after failing initial TKI therapy, which showed PFS superiority in the cabozantinib arm compared with everolimus (median 7.4 and 3.9 months, respectively; HR 0.51, 95% CI 0.41 to 0.62; p < 0.0001). OS improved with cabozantinib compared with everolimus (median 21.4 and 16.5 months, respectively; HR 0.66, 95% CI 0.53 to 0.83; p-value 0.00026). This led to the widespread use of cabozantinib in either the second- or third-line setting in advanced RCC.

Subsequently, the CARBOSUN Phase 2 study compared sunitinib with cabozantinib in the first-line advanced RCC setting. Initial results showed improved investigator-assessed PFS, with a median PFS of 8.2 months for cabozantinib compared with 5.6 months for sunitinib (p = 0.012). Updated results from the CABOSUN trial showed cabozantinib significantly prolonged PFS per the independent review committee compared with sunitinib as first-line therapy for advanced RCC of poor or intermediate risk. 45

These data made a major contribution to the NICE approval of cabozantinib in the first-line setting in 2017. However, in practice, many clinicians have preferred to continue to use cabozantinib in the second- or third-line setting rather than first line to increase the number of treatment options available to patients and also as CABOSUN was a Phase II, not a Phase III, trial.

Lenvatinib in combination with everolimus

Lenvatinib targets VEGFR1, 2 and 3, PDGFRα, fibroblast growth factor receptor (FGFR) and the KIT and RET tyrosine kinases and was initially developed for use in differentiated thyroid carcinoma refractory to standard therapy. 46 Evidence for its efficacy in treatment of RCC came from a randomised Phase II trial of 153 patients with metastatic or unresectable, locally advanced, ccRCC who had received prior antiangiogenic therapy. 47

Three arms with 1 : 1 : 1 randomisation were compared: patients received lenvatinib alone (24 mg/day) or everolimus alone (10 mg/day) or lenvatinib (18 mg/day) plus everolimus (5 mg/day) in 28-day cycles until progression or unacceptable toxicity. The lenvatinib/everolimus combination resulted in significant improvement in median PFS (14.6 months) compared with everolimus alone (5.5 months), but not compared with lenvatinib alone (7.4 months). Lenvatinib alone significantly improved PFS versus everolimus alone. However, toxicity events Grade > 2 occurred in more patients in the lenvatinib alone arm (79%) and in the combined arm (71%) than in the everolimus alone arm (50%).

This combination was approved by NICE as a second-line treatment option in 2018. The evidence from a single clinical trial suggests that, on average, people live around 10.1 months longer if they have lenvatinib plus everolimus rather than everolimus alone. 47 In the trial, lenvatinib plus everolimus caused side effects, leading many patients to interrupt or even stop treatment. This is despite the patients enrolled in the trial being relatively fit (i.e. they had an ECOG PS score of 0 or 1), so it is a treatment option for the fittest patients only.

Protocol v4.0 amendment (approved April 2013)

STAR had originally mandated that sunitinib was the only drug permitted for use in the Phase II trial with a pre-planned reconsideration of pazopanib (another TKI drug under consideration by NICE and the subject of a comparative study against sunitinib, COMPARZ) prior to opening the Phase III part of the STAR trial, based on the available data at that time. However, the COMPARZ trial reported data early (October 2012) which showed pazopanib to be non-inferior to sunitinib in terms of its primary end-point PFS and with no significant difference in OS. NICE approved pazopanib for first-line treatment in 2011. Based on these data, it appeared likely that some clinicians would wish to offer pazopanib in standard practice, as an alternative to sunitinib, which would potentially reduce the number of participants taking sunitinib and therefore be eligible to participate in the STAR trial. Therefore, the decision was made [following discussion with the funder (NIHR HTA), TSC, DMEC, key investigators and patient representatives] that the protocol be amended (v4.0) to include the option of using pazopanib, with the type of TKI as a stratification factor.

Protocol v7.0 amendment (approved September 2014)

Following a review of the interim analysis results in July 2014 (end of Phase II), the trial oversight committees concluded that both the Stage A and B end points had been met, including no evidence that a DFIS was inferior to a CCS arm in terms of TSF (Stage B). Therefore, the DMEC advised continuation of the trial to Phase III (with both sunitinib and pazopanib) and the TSC approved trial continuation.

It was also decided in consultation with the clinical advisors to remove the requirement for participants to meet the maximal radiological response threshold prior to commencing a treatment break in the DFIS arm. Experience from the Phase II aspect of the study demonstrated that it would not be feasible for local sites to reliably determine maximal radiological response. In addition, only around 5% of participants had not reached the maximal radiological response threshold at 6 months. This change was approved by the TSC and DMEC.

Protocol v10.0 amendment (approved July 2017)

In 2016, the DMEC and TSC carried out a re-evaluation of some of the key assumptions made within the original sample size calculations. Although not a pre-planned review of the sample size estimates, this was deemed appropriate due to the fact that the trial had accumulated a significant amount of follow-up data since first opening to recruitment and thus had more accurate information for some of the required sample size estimation values, specifically the 2-year survival estimate in the CCS arm, the (extended) period of recruitment and the overall dropout rate. As a result of this, the sample size was reduced from 1000 to 920 participants, where 720 events were required for the analysis to have 80% power.

Protocol v11.0 amendment (approved July 2019)

Recruitment closed on 12 September 2017 following the recruitment of 920 participants. The trial was planned to continue follow-up until the target of 720 OS events had been observed.

A review of the number of events (death from any cause) for the co-primary end point of OS showed that the number of events was lower than expected and it was considered highly unlikely that the target of 720 events would be reached by the planned end of follow-up in September 2019. An analysis based on data estimated to be collected until September 2019 was considered to have between 72% and 73% power (as opposed to the target of 80% power) for OS, meaning that the trial would have insufficient evidence to be able to demonstrate NI.

On discussion with the Chief Investigator and Co-Chief Investigator, their opinion was that the main reason for the lower event rate was due to the increasing availability of second- and third-line treatment options (nivolumab and cabozantinib), which therefore improved OS and reduced the number of events seen within a specified time. To enable the final analysis to be carried out with the appropriate 80% power, the follow-up period was extended up to an additional 15 months (based on event rate modelling by Renfro et al. 54) in order to observe the required 720 events.

Protocol v12.0 (approved January 2021)

Ongoing monitoring of the event rate showed that, due to multifactorial reasons, as discussed previously, the event rate continued to decrease and a repeat of the modelling, carried out in July 2020, shifted the predicted time frame to observe 720 survival events.

As the study was in the tails of any applied distribution, it was not possible to accurately predict when 720 events would be observed; however, it was clear that it would not be until significantly beyond the planned end of follow-up. As such, it was not feasible to extend the trial for a further fixed duration and the TMG, DMEC and TSC agreed that follow-up would be completed on 31 December 2020 as planned, regardless of whether 720 survival events had been observed.

The end of trial definition was amended to be defined as the date of the collection of the last tissue sample or 31 December 2021, whichever came sooner. This change allowed tissue block collection, which had been interrupted by the COVID pandemic, to proceed beyond the end of follow-up until no later than 31 December 2021 (with the possibility that it could end before this if the TMG considered that it was not feasible to collect any further tissue blocks).

The trial schema is shown in Appendix 3.

Inclusion criteria

Patients were permitted to participate if they met all of the following criteria:

1. Male or female aged ≥ 18 years.

2. Histological confirmation of a component of clear cell RCC.

3. Inoperable loco-regional or metastatic disease.

4. No prior systemic therapy for advanced disease (inoperable loco-regional and/or metastatic disease). Previous treatment in the placebo arm of the SORCE study was permitted.

5. ECOG PS 0–1 assessed prior to randomisation and within 16 days of starting treatment with either sunitinib or pazopanib. 55

6. Uni-dimensionally measurable disease as defined by RECIST criteria.

7. Full blood count56 was performed prior to randomisation and within 16 days of starting treatment with either sunitinib or pazopanib.

   • Haemoglobin (Hb) ≥ 9 g/dL57 (blood transfusions were acceptable).

   • Absolute neutrophil count (ANC) ≥ 1 × 10⁹/L.

   • Platelets ≥ 80 × 10⁹/L.

8. Renal biochemistry58 was performed prior to randomisation and within 16 days of starting treatment with either sunitinib or pazopanib. Measured or calculated glomerular filtration rate (GFR) ≥ 30 ml/minute was permitted. (Cockcroft and Gault or Wright formula were used according to local practice.)

9. Hepatobiliary function59 was performed prior to randomisation and within 16 days of starting treatment with either sunitinib or pazopanib.

   • Aspartate transaminase (AST) or ALT ≤ 2.5 × ULN.

   • Bilirubin (BR) ≤ 1.5 × ULN, or in patients with Gilbert syndrome BR ≤ 3 × ULN and direct BR ≤ 35%.

10. Provided written informed consent prior to any trial-specific procedures.

11. Able and willing to comply with the terms of the protocol including:

    • commencement of sunitinib or pazopanib within 5 (actual not working) days of randomisation;

    • temporarily stopping sunitinib or pazopanib if randomised to the DFIS arm;

    • capable of oral self-medication;

    • commencement of sunitinib or pazopanib within 42 days of the baseline CT scan;

    • capable of reporting toxicity and completing QoL and medical resource utilisation (MRU)/Health Economics questionnaires.

12. If female and of child-bearing potential, must:

    • have a negative pregnancy test within 72 hours prior to randomisation, and should not be breastfeeding;

    • agree to use adequate, medically approved, contraceptive precautions (oral or barrier contraceptive) under the supervision of a general practitioner (GP) or Family Planning Clinic during and for 30 days after the last dose of sunitinib or pazopanib.

13. If male with a partner of child-bearing potential, must agree to use adequate, medically approved, contraceptive precautions (oral or barrier contraceptive) under the supervision of a GP or Family Planning Clinic during and for 30 days after the last dose of sunitinib or pazopanib.

14. Requirement to start first-line therapy with either sunitinib or pazopanib and decision already made as to which TKI to be used according to local standard practice.

Allowed situations included:

• primary renal cancer in situ or previous nephrectomy;

• previous brain metastases treated with complete surgical resection. Stereotactic brain radiation therapy (SBRT) or Gamma Knife with no subsequent evidence of progression (patients treated only with whole brain radiotherapy are not eligible);

• previous radiotherapy and/or previous/ongoing bisphosphonates or bone antiresorptive drugs for the treatment of symptomatic bony metastasis. Care should be taken to follow dental guidelines for the antibone resorptive drug.

Exclusion criteria

Patients were excluded from participation if they met any of the following criteria:

1. Pulmonary or mediastinal disease causing obstruction or clinically significant bleeding/haemoptysis.

2. Patients with an estimated life expectancy of < 6 months.

3. Known contraindications to the particular TKI to be used (i.e. sunitinib or pazopanib).

4. Any previous treatment with sunitinib, pazopanib or other TKI (including in the adjuvant setting).

5. Untreated brain metastases60: patients were eligible if previous brain metastases were treated with complete surgical resection, SBRT or Gamma Knife with no subsequent evidence of progression. Patients were not eligible if brain metastases were treated only with whole-brain radiotherapy.

6. Any concurrent or previous other invasive cancer that could confuse diagnosis or end points. Allowed situations included (but not limited to): non-melanomatous skin cancer or superficial bladder cancer.

7. Hypersensitivity to the particular TKI to be used (i.e. sunitinib or pazopanib).

8. Any concomitant medication or substances forming part of local ongoing care known to significantly affect, or have the potential to significantly affect, the activity or pharmacokinetics of the particular TKI to be used (i.e. sunitinib or pazopanib).

9. Poorly controlled hypertension despite maximal medical therapy. 7 It was recommended that subjects should have a systolic BP of either < 150 mmHg, and/or a diastolic BP of < 90 mmHg. Antihypertensive drugs could be used to achieve these values.

10. Any other serious medical or psychiatric condition which in the opinion of the investigator could affect participation in the STAR trial, including gastrointestinal abnormalities limiting the effectiveness of orally administrated drugs, uncontrolled infections, current or recent history of clinically significant cardiovascular disease, significant haemorrhage or gastrointestinal perforation or fistula which, in the opinion of the local investigator, would render the patient unsuitable for standard sunitinib or pazopanib therapy.

Recruitment procedure

Patients were approached during routine oncology appointments and were provided with verbal and written details about the trial. The verbal explanation of the trial and the version of the participant information sheet (PIS) and consent form (CF) appropriate for the TKI recommended for use (sunitinib or pazopanib) were provided by the patient’s clinical team.

An optional DCE-MRI substudy was undertaken at St James University Hospital, Leeds, and participants approached at this site were also provided with an additional PIS and CF regarding the DCE-MRI substudy.

Following information provision, patients were given a minimum of 24 hours to consider trial participation. Assenting patients were then formally assessed for eligibility and invited to provide written informed consent.

Informed consent

Informed, written consent was obtained prior to randomisation into the study, subject to the patient meeting the eligibility criteria. A record of the consent process detailing the date of consent and all those present was kept in the participant’s notes. The original signed and dated CFs were held securely as part of the trial site file; copies were filed in the hospital notes (as per local practice) and copies were returned to the Clinical Trials Research Unit (CTRU).

Registration for DCE-MRI substudy participants

Patients participating in the DCE-MRI substudy were required to undergo a baseline DCE-MRI scan prior to the commencement of sunitinib or pazopanib treatment on the STAR trial. Given the narrow window specified between randomisation and commencement of sunitinib or pazopanib, baseline DCE-MRI substudy scans were scheduled prior to randomisation. Patients agreeing to participate in this substudy were therefore registered with the CTRU prior to their baseline DCE-MRI scan in order to confirm their eligibility for the main trial, consent and participation in the substudy.

Registration was carried out by the CTRU automated 24-hour telephone randomisation service. Participants were allocated a unique trial identification number at registration which was used at randomisation and throughout their study participation.

Randomisation

Randomisation took place as soon as possible after consent and confirmation of eligibility and was no more than 5 days prior to the start date of sunitinib or pazopanib treatment. The decision regarding the TKI (sunitinib or pazopanib) to be used was at the discretion of the treating clinician and was made prior to randomisation.

Randomisation was carried out by the CTRU automated 24-hour telephone randomisation service. Participants were randomised in a 1 : 1 ratio between the CCS and DFIS treatment arms.

A computer-generated minimisation programme that incorporated a random element was used to ensure that treatment groups are well balanced by:

• Motzer/MSKCC (Memorial Sloan-Kettering Cancer Centre) prognostic group. 61

  • Favourable risk (0 factors).

  • Intermediate risk (1–2 factors).

  • Poor risk (≥ 3 factors).

• Trial site.

• Gender.

• Age:

  • < 60 years.

  • ≥ 60 years.

• Disease status at the time of randomisation:

  • Metastatic.

  • Locally advanced.

  • Previous nephrectomy:

  • Yes.

  • No.

• TKI:

  • Sunitinib.

  • Pazopanib.

Participants who had not been registered for the study were allocated a unique trial identification number at randomisation.

Participants, medical staff and clinical trial staff were informed of the allocated treatment arm. The treatment allocation was not blinded as accurate radiological evaluations were fundamental to the Stage B end point, and for this reason all radiological evaluations in the Phase II component of the study (Stages A and B) were performed centrally. Central reporting of radiological evaluations was not carried out in Phase III of the study.

Baseline assessments

Participants were required to have cross-sectional imaging (chest, abdomen and pelvis were strongly recommended) within 42 days before the start of protocol treatment. A contrast CT scan (chest abdomen pelvis) was the preferred modality of cross-sectional imaging. If this was not possible (e.g. in the case of contrast allergy or renal insufficiency), then a non-contrast CT (chest abdomen pelvis) scan was performed, assuming the disease was evaluable by this method. If the disease was not evaluable using a non-contrast CT scan, a MRI scan of the abdomen and pelvis and a non-contrast CT scan of the chest were performed. All subsequent follow-up scans were required to be in the same modality (CT or MRI) and performed using the same technique.

Following informed consent and prior to randomisation (within 16 days prior to commencing trial treatment) patients were assessed with regard to medical history, physical examination (including height, weight, ECOG PS, vital signs, heart rate and BP), full blood count, biochemistry [urea and electrolytes (UE) including urea, creatinine, sodium and potassium], liver function tests (LFTs) including alkaline phosphatase (ALP), ALT/AST, total BR and albumin, LDH, thyroid function tests (TFTs), bone profile (calcium) for calculation of Motzer score and pregnancy test (if woman of child-bearing potential) within 72 hours prior to randomisation.

In addition, if a bone scan was carried out as part of standard local practice, this was performed in accordance with routine time frames but was not mandated by the protocol.

The baseline QoL (booklet A) (FACT-G and FSKI-15, EQ-5D-3L/EQ-VAS and MRU/Health Economics questionnaires) was completed prior to randomisation, as close as possible to commencement of treatment.

Treatment assessments

Irrespective of the allocated treatment arm (DFIS or CCS), participants were assessed clinically for symptoms and toxicity 6-weekly at the start of each treatment cycle.

The following assessments were conducted within 5 days prior to each treatment cycle and/or clinical review (while on planned treatment break for DFIS arm participants): clinical assessment including weight, ECOG PS and vital signs (HR and BP), AE reporting/toxicity assessment (CTCAE v.4.0), full blood count, UE, LFT and TFT (q 12 weeks).

More frequent monitoring of liver function was required for participants receiving pazopanib at timings recommended in the pazopanib Summary of Product Characteristics62 at weeks 3, 5, 7 and 9, then at months 3 and 4, with additional tests as clinically indicated. Periodic testing should then continue after month 4.

Imaging

Computerised tomography scan imaging was carried out prior to commencement of cycle 3 and every 12 weeks thereafter. During Phase II of the trial, central reporting of scans was performed to ensure consistency. During Phase III, scans were reported locally according to RECIST.

The timing of the radiological assessments was the same in both the CCS and DFIS arms. However, if the scan scheduling fell out of sync with the cycles of treatment (e.g. a delay due to toxicity or other medical reasons), the scan could be delayed by up to 4 weeks to allow the scan to coincide with the usual treatment cycles.

In the event of clinical evidence of disease progression at a time other than that when radiological reassessment is due, radiological assessments were performed to confirm progression, unless there was a compelling reason that this was not possible.

All scans performed during the treatment break were compared to the scan immediately before the treatment break started as per the standard RECIST reporting guidelines.

Follow-up

After permanent discontinuation of protocol treatment, participants were followed up until the end of the trial follow-up period. SAEs, serious adverse reactions (SARs) and suspected unexpected serious adverse reactions (SUSARs) collected for 30 days following the end of the trial follow-up period were included in the final analysis. If it was not possible for the participant to attend clinic at this time point, events were collected by telephone call if considered to be appropriate by the treating clinician.

Participants were seen in clinic 6 months after permanently discontinuing protocol treatment and annually thereafter until the end of follow-up on 31 December 2020. Details of the participant’s status and any subsequent treatment received for renal cancer were collected at the follow-up visits along with the EQ-5D-3L/EQ-VAS QoL questionnaire.

All randomised participants were followed up for survival unless consent was withdrawn for further data collection.

Quality of life

Information from all questionnaires (FACT-G, FSKI-15 and EQ-5D-3L/EQ-VAS) was collected at clinic visits at baseline (before the participant was informed of their randomisation allocation) and at day 1 of cycles 2, 3 and 4 during which time participants on both arms received sunitinib or pazopanib as clinically appropriate.

After cycle 4 (24 weeks post randomisation), the EQ-5D-3L/EQ-VAS questionnaires were collected every 2 weeks and were completed by participants at home. This intensive QoL collection continued until 48 weeks post randomisation and covered the 24-week period after participants had taken up their randomised treatment allocation (DFIS or CCS). After this point, questionnaires were once again collected in clinic on a 6-weekly basis.

The FSKI-15 and FACT-G questionnaires were collected every 6 weeks at clinical assessment visits for the duration of trial treatment.

The 2-weekly questionnaire completion was acknowledged to be a significant burden for participants; however, it was considered key to informing the QALY co-primary end point as differences between the treatment strategies were likely to be the greatest immediately following participants taking up their randomised treatment allocation.

In addition, 2-weekly QoL was considered relevant for participants receiving sunitinib as it was given over 28 days followed by a 14-day off-treatment period. In comparison, pazopanib was administered for the full 42 days of the 6-weekly treatment cycles.

In order to capture any differences in QALYs between the arms after treatment strategy failure, EQ-5D-3L/EQ-VAS information was collected for all participants (where possible) until the end of follow-up.

Due to the importance of QoL data in this trial, measures were taken to ensure maximum compliance of questionnaire completion. Participants consented to receive e-mail or text message reminders from the research team at CTRU (this was optional) over the 24-week period where participants were required to complete 2-weekly QoL questionnaires at home. Where a QoL questionnaire was missed at a hospital clinic visit, the local research team posted the questionnaire to the participant’s home (after checking the participant’s status to establish it was appropriate to do so).

Phase II

Phase III

Phase II

Phase III

Analysis populations

The analysis required the following populations: ITT, PP, safety, EQ5D, FKSI and FACT-G. All randomised participants were considered for inclusion in each population. Participants were excluded from the ITT population if they did not have RCC. Participants were excluded from the PP population if they were deemed to be major protocol violators by the TMG or they reached 6 months post randomisation but do not take up their randomisation allocation. Note that participants in Phase II were required to reach maximum radiological response (MRR) prior to taking up their first treatment break. Any participants in the Phase II part of the trial who did not go on a treatment break because they did not achieve MRR will not be excluded from this population. Participants were excluded from the safety population if they did not receive any of their protocol treatment (sunitinib or pazopanib). Participants were excluded from the QoL populations (EQ5D, FKSI, FACT-G) if their baseline questionnaire could not be scored. With the exception of the safety population, all analysis was conducted according to what participants were randomised to receive. For the safety population if a participant in the DFIS arm declined or did not take up a treatment break in error at 6 months post randomisation and it is not rectified by their next scan they were summarised and included in the CCS arm.

Interim analyses

Descriptive analysis

Key baseline characteristics [minimisation factors of the trial (excluding centre), age, gender, ethnicity, PS, haemoglobin, neutrophils, platelets, calcium, LDH bone involvement and time since diagnosis] were compared between randomisation arms in all analysis populations. In addition, the treatment participants received on trial, along with the results of their on-study assessments, and the treatment they received in follow-up were summarised between randomisation arms in the ITT population.

Co-primary end-point analysis

As both primary outcomes were to assess NI, these analyses were conducted in both the PP and ITT populations. If DFIS could show NI in both OS and QALY in both populations, then the analysis would conclude that DFIS was non-inferior to CCS.

Secondary end-point analysis

For time to event secondary end points (TSF, TTF, PFS and SPFI), analysis was conducted on the ITT population. The number and proportion of events at the time of analysis were summarised both by randomisation allocation and by TKI received. Time-to-event curves with median survival and corresponding 95% CIs were plotted using the Kaplan–Meier method, and event estimates (i.e. the proportion of participants event-free) at each year following randomisation were presented for each treatment arm, along with their corresponding 95% CIs.

Cox regression analysis was used to formally compare the end points between the treatment arms, accounting for the minimisation factors, except for randomising centre. The PH assumption was assessed using the supremum test, where, if violated, it was pre-specified that additional analysis methods would be implemented.

The safety analysis summarised the observed AEs, SAEs, SARs and SUSARs for the safety population. Summaries included overall statistics as well as by treatment strategy (DFIS or CCS) and TKI treatment (sunitinib or pazopanib). The confirmed cases of osteonecrosis of the jaw (ONJ) and any notifications of trial participant pregnancies or their partners were also reported.

Each QoL measure (total FKSI, FKSI Disease-Related Subscale, FACT-G total and subscales, EQ-5D-3L index and EQ-VAS) was summarised using the median and IQR within the relevant QoL population at each possible time point. Additional analysis was conducted for all of the measures except the EQ-5D-3L index and the EQ-5D-3L VAS. For the additional analysis, the measure was considered in a multilevel repeated measures model. Time, treatment strategy, baseline QoL and an interaction for treatment strategy by time were included as fixed effects with participant and participant by time included as random effects. The model was also adjusted for the minimisation factors of the trial, excluding the centre. Note that time, measured in 6-weekly intervals, was treated as continuous.

Ancillary analysis

For OS the following ancillary analysis was conducted:

1. A piecewise hazards model was applied to account for participants in both arms being treated the same for the first 6 months of the trial, provided the piecewise hazards model is not used within the analysis of primacy.

2. The analysis described in the section Co-primary end-point analysis was repeated where the Motzer stratification factor was combined into two groups rather than three.

For QALYs the following ancillary analysis was conducted:

1. The analysis described in the section Co-primary end-point analysis was repeated where QALYs were measures from week 24, the point where participants were due to take up their randomisation allocation.

2. The analysis described in the section Co-primary end-point analysis was repeated where QALYs were measured up to:

   1. 12 months post randomisation.

   2. 24 months post randomisation.

   3. 36 months post randomisation.

3. A multivariable linear regression model was applied on the imputed data.

4. The analysis described in the section Co-primary end-point analysis was repeated using only complete case data.

5. The analysis described in the section Co-primary end-point analysis was repeated on data where observations thought to be missing not at random (MNAR), defined to be those with the reason for missing ‘Inappropriate to give to participant/participant too unwell’, were imputed to be the worst health state (−0.59).

For both PFS and TTF, a piecewise hazards model was considered with both two (splitting at week 24) and three intervals (splitting at week 24 and approximately week 42, the end of the average length of a treatment break).

In addition, for TTF, an ancillary analysis, the same as described in the section Secondary end-point analysis, was conducted which derived TTF, excluding the time spent on a treatment break in the DFIS arm. For the CCS arm, this was the same as TTF as defined above. For this DFIS arm, this was calculated as the TTF defined above, minus the sum of the treatment breaks. The duration of a treatment break was defined as the time between the expected end date of the last cycle prior to the break (cycle start date + 42 days) and the start of the next treatment cycle following the treatment break. For this sensitivity analysis, if a participant was still on a treatment break at the time of analysis or had been told to restart trial treatment following a break but the treatment information was outstanding, then their end point was censored at the scan date which told them to start their current treatment break.

For the FKSI-15 and FACT-G questionnaires, the analysis described in the section Secondary end-point analysis was repeated where QoL was measured up to:

1. 12 months post randomisation.

2. 24 months post randomisation.

3. 36 months post randomisation.

Subgroup analysis

Exploratory subgroup analysis was conducted for both co-primary end points within the PP population using the correct stratification factors. The subgroups investigated were:

• Body mass index (BMI) as recorded at baseline (underweight or normal < 25, overweight or obese ≥ 25).

• Comorbidities as recorded at baseline (0, 1, 2 or more).

• Age at randomisation (> 70 or ≤ 70 and > 75 or ≤ 75).

• Bone involvement as recorded at baseline (Yes/No).

• Liver metastases as recorded at baseline (Yes/No).

• IMDC score (Favourable, Intermediate, Poor).

• The minimisation factors of the trial.

For OS the subgroup analysis considered the number of participants who died within each subgroup category as well as the 2-year OS estimate and 95% CI using the Kaplan–Meier method. The analysis was then extended to consider a Cox regression model similar to that described within the section Co-primary end-point analysis with the addition of a term for the subgroup being investigated (except for when the minimisation factor is the subgroup under consideration) and an appropriate interaction term. Heterogeneity was determined by a likelihood ratio test on the inclusion of the interaction term.

For QALYs the mean and 95% CI of QALYs, calculated over the course of the trial, for participants in each subgroup category will be calculated by randomisation using the imputed data. The difference in means QALYs between the randomisation arms was also calculated.

Missing data

Missing data were considered for the EQ-5D-3L utility index due to its involvement in the derivation of QALYs. However, due to the extensive collection of questionnaire data, missing data were only considered to be an issue during follow-up when the questionnaire collection is more infrequent. However, if a baseline questionnaire was missing, it was imputed using mean imputation. 71

The pattern of missingness was investigated along with the key baseline demographics of those with and without questionnaires at each follow-up time point. The distribution of the utility index was plotted during follow-up and imputed using predictive mean matching in a multiple imputation by chained equations framework. 72–74 The imputation model included:

• Randomisation allocation.

• The observed/imputed EQ-5D-3L utility score at the previous follow-up time points.

• The value of the utility score at the participant’s last on-study review (F05/a) at which the EQ-5D-3L questionnaire was completed.

• The minimisation factors of the trial (excluding randomising centre).

The number of imputed data sets was selected to be the maximum percentage of missing questionnaires across the time points which were being imputed, where imputation was conducted in time order and followed the following rules:

• If a participant had withdrawn from QoL completion or was lost to follow-up, then questionnaires were imputed up to the time they were last known to be alive.

• If a participant was not lost to follow-up, missing questionnaires were imputed for the entire follow up period or until death.

• Only participants still alive at the relevant time point were included in each imputation model.

Trace plots were used to investigate how well the imputation converged.

Clinical pathway

The trial recruited participants between 13 January 2012 and 12 September 2017. Follow-up continued until 31 December 2020. The Consolidated Standards of Reporting Trials (CONSORT) diagram showing how participants flowed through the clinical aspect of the study is shown in Figure 4. Overall, 2197 potential participants were screened for trial entry with 920 (461 CCS, 459 DFIS) participants being randomised.

FIGURE 4.

Consolidated Standards of Reporting Trials diagram for the STAR Trial. a, Indicates the number of participants who received each TKI rather than the number who were randomised under each TKI. b, Indicates that the reasons given are none mutually exclusive.

During the course of the trial, 878 (95.4%, n = 920) [CCS: 453 (98.3%, n = 461), DFIS: 425 (92.6%, n = 459)] participants discontinued their trial treatment prior to the end of trial follow-up (31 December 2020). This equates to 42 participants (4.6%%, n = 920) [CCS: 8 (2.7%, n = 461), DFIS: 34 (7.4%, n = 459)] still receiving trial treatment at the end of the trial. Of these, 432 participants discontinued prior to week 24. This was the case for a similar proportion of participants in both arms [CCS: 221 (47.9%, n = 461): DFIS 211 (46.0%, n = 459)]. For the majority of participants, radiological disease progression played a role in them discontinuing trial treatment (see Report Supplementary Material 1, Table 1). This was the case for a similar proportion of participants in both arms prior to week 24 (CCS: 35.2%, DFIS: 37.1%). However, unsurprisingly, post week 24 the CCS arm had a higher proportion of participants stopping treatment due to radiological progression (CCS: 60.7%, DFIS: 40.7%), whereas the DFIS arm had a higher proportion of cessation due to death (CCS: 0.7%, DFIS: 8%) and clinical lead withdrawal (CCS: 3.3%, DFIS: 9.1%). The other reasons for discontinuing trial strategy are listed in Report Supplementary Material 1, Table 2. Of those who continued past week 24, 52.9% (n = 488) were randomised under the Motzer/MSKCC prognostic group of intermediate or poor risk.

Quality of life

Figure 5 displays how participants moved through the QoL study, the number of expected questionnaires and the number returned at a questionnaire booklet level. Recall that Booklet A was due at baseline, Booklet B at weeks 6, 12 and 18 from randomisation, Booklet C at weeks 24, 30, 36 and 42, Booklet D at 2-weekly intervals between weeks 24 and 46, Booklet E at 6-weekly intervals from week 48 while the participant was still on treatment and Booklet F at 6 months following the end of trial treatment and annually thereafter. Time point-specific information can be found in Report Supplementary Material 1, Table 3. The reasons for missing questionnaires by booklet are shown in Table 1. The other reasons for missing questionnaires include completed on an incorrect date, destroyed by site, misplaced by site English-language barriers.

FIGURE 5.

Overview of how participants moved through the QoL aspect of the STAR trial.

Withdrawal

In total, 63 participants withdrew from some aspect of the trial across 64 occurrences (see Report Supplementary Material 1, Table 4). Report Supplementary Material 1, Figure 6 shows how participants could withdraw consent from different aspects of the trial. The majority of instances withdrew from QoL (71.9%), a higher proportion in the CCS arm (84.4%) than the DFIS arm (59.4%). Of the 28 (43.8%) who withdrew from trial follow-up, only 7 of them allowed data to be collected at standard visits. This resulted in 21 (2.82%, n = 920) participants being formally lost to follow-up. An additional one participant in the DFIS arm was lost to follow-up due to the participant moving away but this was not formally recorded. However, participants’ long-term follow-ups were included in all analyses. If recorded, the reasons for withdrawal are given in Report Supplementary Material 1, Table 5.

Protocol violations

In total, 76 protocol violations were observed from 71 participants. The majority of protocol violations (65.8%) were due to a breach in eligibility criteria (see Report Supplementary Material 1, Table 6). The eligibility criteria breached are shown in Report Supplementary Material 1, Table 7.

Treatment cycles

The median (IQR) number of treatment cycles was 4 (2, 10) overall and similar between the two arms [CCS: 5(2, 10), DFIS: 4 (2, 9)]. Prior to week 24, the median (IQR) number of treatment cycles was identical between the two arms 4 (2, 4). Post week 24, the median (IQR) was again similar between the two arms with the medians being identical [CCS: 6 (2, 13), DFIS: 6 (3, 12)]. In terms of short cycles, cycle delays and dose reductions, the percentage of short cycles was similar across the randomisation arms both overall (CCS: 25.7%, n cycles = 3602, DFIS: 26.3%, n cycles = 3285) and pre and post cycle 4 (pre CCS: 32.3%, n cycles = 1439, DFIS: 31.1%, n cycles = 1449, post CCS: 21.3%, n cycles = 2163, DFIS: 22.5%, n cycles = 1836), as were the percentage of delays (overall — CCS: 13.8%, n cycles = 3602, DFIS: 13.4%, n cycles = 3285, pre CCS: 14.1%, n cycles = 1439, DFIS: 13.0%, n cycles = 1449, post CCS: 13.5%, n cycles = 2163, DFIS: 13.8%, n cycles = 1836), and the percentage of dose reductions (overall — CCS: 7.4%, n cycles = 3602, DFIS: 8.3%, n cycles = 3285, pre CCS: 12.6%, n cycles = 1439, DFIS: 13.6%, n cycles = 1449, post CCS: 3.9%, n cycles = 2163, DFIS: 4%, n cycles = 1836). However, the median (95% CI) time to first delay and time to first dose reduction was longer in the DFIS arm compared to the CCS arm [delay — CCS: 6 months (4, 7), DFIS: 10 months (7, 11), reduction – CCS: 6 months (6, 8), DFIS: 10 months (5, 14)] accounting for the fact that when on a treatment break participants cannot have treatment delayed or reduced.

Treatment breaks

In total, 248 participants (56.2%, n = 459) in the DFIS arm continued on trial after 6 months. Of these, 210 (84.7%) started their first treatment break according to the protocol at week 24. A similar proportion of participants on each TKI [sunitinib: 89 (84.0%, n = 106), pazopanib: 121 (85.2%, n = 142)]. Thirty-eight participants took their first treatment break later than protocol advised, where 15 of these (39.5%) did not take up a treatment break despite continuing at week 24. These participants continued from week 24 onwards in error due to the clinician’s decision and then either withdrew from trial treatment or experienced radiological progression on a later scan. Of the 38, 12 (31.6%) took up their first break at 36 weeks, 6 (15.8%) at 48 weeks, 2 (5.3%) at 60 weeks and 3 (7.9%) at other time points. The three other time points at which the treatment break was taken up were week 168, week 30 and week 42. For the majority, the reasons for continuing past week 24 were due to clinical decisions or during Phase II of the trial that maximal radiological response had not been reached.

The median number of treatment breaks was one in both TKI groups. The maximum number of breaks taken by a participant was nine (see Table 3). The median (IQR) length of a treatment break was 87 (84–119) days and similar between the two TKIs [sunitinib: 85.5 (84–112), pazopanib 87.5 (84–137)].

Anticancer treatment post trial

Overall, 61.6% of participants received any systemic anticancer therapy treatment during follow-up: a slightly higher proportion in the CCS arm (68.5%, n = 461) compared to the DFIS arm (54.6%, n = 458), in addition to a higher proportion of those on sunitinib (65.5%, n = 385) compared to pazopanib (58.8%, n = 534). The remaining types of treatment recorded post trial (radiotherapy, surgery, palliative care) were similar across the two randomisation arms (see Table 4) and TKI received (see Report Supplementary Material 1, Table 14). The medication names for the anticancer therapy received, distinct per participant, are given by randomisation allocation in Report Supplementary Material 1, Tables 15 and 16. These were categorised into first-line TKI, TKI, immunotherapy (alone or in combination), Mammalian target of rapamycin (alone or in combination), other cancer treatment or investigational product by a TMG clinician. Note that anticancer therapies considered not to aim to improve survival were not included in this re-categorisation. The number of distinct types of therapy a participant received is shown in Report Supplementary Material 1, Table 17. Note that zero in Report Supplementary Material 1, Table 17 refers to only STAR trial treatment received. The number of participants who received each type of therapy at least once is shown in Table 5. Note that immunotherapy was received by a similar proportion of participants in both arms.

Intention to treat

Of the 920 randomised, 919 were included in the ITT population (CCS: 461, DFIS: 458). One participant in the DFIS arm was excluded as they did not have RCC.

Per-protocol population

Of the 920 randomised, 871 (94.7%) were included in the PP population (CCS: 453, DFIS: 418). A higher proportion of participants randomised to the CCS arm (98.3%, n = 461) compared to those randomised to the DFIS arm (91.1%, n = 459). The reasons for exclusion from the PP population are shown in Table 7.

Safety population

Nine hundred and sixteen (99.6%) of all randomised were included in the safety population (CCS: 485, DFIS: 431). Four participants, three in the DFIS arm and one in the CCS arm, were excluded due to never commencing their TKI treatment. In addition, 24 (5.2%, n = 459) participants randomised to the DFIS arm are included in the CCS safety arm.

Quality-of-life populations

Eight hundred and sixty-nine (94.5%) randomised participants were included in the EQ-5D-3L population (CCS: 438, DFIS: 431). Fifty-one participants were excluded due to having baseline questionnaires which could not be scored. Similarly, 856 (93.0%) randomised participants were included in the FACT-G population (CCS: 425, DFIS: 431). Finally, 882 (95.9%) randomised participants were included in the FKSI population (CCS: 436, DFIS: 446).

Overall survival

Quality-adjusted life-years

Time to strategy failure

Time to strategy failure is defined in Report Supplementary Material 1, Figure 1. In total, 850 (92.5%) participants in the ITT population (n = 919) died, progressed or required another anticancer systemic treatment prior to the end of follow-up on 31 December 2020. Of these, 438 were from the CCS arm, accounting for 95.0% of those in the ITT population randomised to CCS (n = 461). The remaining 412 were from the DFIS arm, accounting for 90.0% of those in the ITT population randomised to DFIS (n = 458). In terms of TKI, 356 TSF events were randomised under sunitinib (91.8%, n = 388) and 494 (93.0%, n = 531) under pazopanib.

Median TSF (95% CI) in the CCS arm was 8 months (8 to 9) and 11 months (9 to 13) in the DFIS arm. At 24 months, there was a higher proportion (95% CI) of participants event-free in the DFIS arm compared to the CCS arm [CCS: 15.6% (12.4% to 19.1%), DFIS: 26.3% (22.3% to 30.4%)]. Figure 8 shows the Kaplan–Meier curve for TSF by randomisation allocation.

FIGURE 8.

Kaplan–Meier figure for TSF by randomisation allocation in the ITT population.

Table 14 shows the Cox PH model results for TSF. The HR for randomisation allocation suggests that the DFIS arm has a risk of strategy failure 0.75 times the risk of strategy failure in the CCS arm. This comparison is statistically significant at the 1% significance level (p < 0.001). The supremum test for PH showed that no variables included in the regression model violated the PH assumption at the 1% significance level.

Time to treatment failure

Time to treatment failure is defined in Report Supplementary Material 1, Figure 2. In total, 877 (95.4%) participants in the ITT population (n = 919) stopped their trial strategy prior to the end of follow-up on 31 December 2020. Of these, 453 were from the CCS arm, accounting for 98.3% of those in the ITT population randomised to CCS (n = 461). The remaining 424 were from the DFIS arm, accounting for 92.6% of those in the ITT population randomised to DFIS (n = 458). In terms of TKI, 371 TTF events were randomised under sunitinib (95.6%, n = 388) and 506 (95.3%, n = 531) under pazopanib.

Median TTF (95% CI) in the CCS arm was 7 months (5 to 8) and 8 (6 to 9) months in the DFIS arm. At 24 months, there was a higher proportion (95% CI) of event-free participants in the DFIS arm compared to the CCS arm [CCS: 12.1% (9.4% to 15.3%), DFIS: 22.5% (18.8% to 26.4%)]. Figure 9 shows the Kaplan–Meier curve for TTF by randomisation allocation. Table 15 shows the Cox PH model results for TSF. The HR for randomisation allocation suggests that the DFIS arm has a risk of treatment failure 0.75 times the risk of treatment failure in the CCS arm. This comparison is statistically significant at the 1% significance level (p < 0.001). The supremum test for PH showed that both randomisation allocation and randomised under TKI violated the PH assumption at the 1% significance level. Therefore, piecewise hazards models were applied both with two intervals (see Report Supplementary Material 1, Table 32), separating at week 24, and three intervals (see Report Supplementary Material 1, Table 33) separating at week 24 and approximately week 42, the end of the average length of a treatment break. In both cases, the randomisation allocation comparison remains statistically significant at the 1% significance level [HR (95% CI); two intervals: 0.71 (0.62 to 0.82), three intervals: 0.72 (0.63 to 0.82)]. However, the goodness-of-fit statistic for both piecewise models was significant at the 1% model suggesting that the piecewise model is not a good fit to the data. As this did not change the conclusions to the Cox regression model, no further investigations were conducted.

FIGURE 9.

Kaplan–Meier figure for TTF by randomisation allocation in the ITT population.

Progression-free survival

Note that this section refers to the first progression recorded on the trial, as defined in Report Supplementary Material 1, Figure 3, irrespective of whether it resulted in cessation or recommencement of trial treatment.

In total, 868 (94.5%) participants in the ITT population (n = 919) progressed or died prior to the end of follow-up on 31 December 2020. Of these, 431 were from the CCS arm, accounting for 93.5% of those in the ITT population randomised to CCS (n = 461). The remaining 437 were from the DFIS arm, accounting for 95.4% of those in the ITT population randomised to DFIS (n = 458). In terms of TKI, 368 PFS events were randomised under sunitinib (94.8%, n = 388) and 500 (94.2%, n = 531) under pazopanib.

Median PFS (95% CI) in the CCS arm was 11 months (9 to 11) and 8 months (8 to 8) in the DFIS arm. At 24 months, there was a higher proportion (95% CI) of participants in the CCS arm event-free compared to the DFIS arm [CCS: 21.4% (17.8% to 25.3%), DFIS: 10.2% (7.6% to 13.2%)]. Figure 10 shows the Kaplan–Meier curve for PFS by randomisation allocation. Table 16 shows the Cox PH model results for PFS. The HR for randomisation allocation suggests that the DFIS arm has a risk of progression or death 1.37 times the risk of progression or death in the CCS arm. This comparison is statistically significant at the 1% significance level (p < 0.001). The supremum test for PH showed that both randomisation allocation and randomised under Motzer score violated the PH assumption at the 1% significance level. Therefore, piecewise hazards models were applied both with two intervals (see Report Supplementary Material 1, Table 34), separating at week 24, and three intervals (see Report Supplementary Material 1, Table 35) separating at week 24 and approximately week 42, the end of the average length of a treatment break. In both cases, the randomisation allocation comparison remains statistically significant at the 1% significance level [HR (95% CI); two intervals: 1.40 (1.22 to 1.60), three intervals: 1.33 (1.16 to 1.52)]. However, the goodness-of-fit statistic for both piecewise models was significant at the 1% level suggesting that the piecewise models are not a good fit to the data. As this did not change the conclusions to the Cox regression model, no further investigations were conducted.

FIGURE 10.

Kaplan–Meier figure for PFS by randomisation allocation in the ITT population.

Summative progression-free interval

Summative progression-free interval is defined in Report Supplementary Material 1, Figures 4 and 5. In total, 783 (85.2%) participants in the ITT population (n = 919) ended their last SPFI with an event prior to the end of follow-up on 31 December 2020. Of these, 416 were from the CCS arm, accounting for 90.2% of those in the ITT population randomised to CCS (n = 461). The remaining 367 were from the DFIS arm, accounting for 80.1% of those in the ITT population randomised to DFIS (n = 458). In terms of TKI, 341 SPFI events were randomised under sunitinib (87.9%, n = 388) and 442 (83.2%, n = 531) under pazopanib.

Median SPFI (95% CI) in the CCS arm was 8 months (8 to 10) and 10 months (9 to 11) in the DFIS arm. At 24 months post randomisation, there was a higher proportion of participants (95% CI) event-free in the DFIS arm compared to the CCS arm [CCS: 14.9% (11.7% to 18.5%), DFIS: 24.6% (20.5% to 28.9%)]. Figure 11 shows the Kaplan–Meier curve for SPFI by randomisation allocation. Table 17 shows the Cox PH model results for SPFI. The HR for randomisation allocation suggests that the DFIS arm has a risk of SPFI ending in an event 0.77 times the risk of SPFI ending in an event in the CCS arm. This comparison is statistically significant at the 1% significance level (p < 0.001). The supremum test for PH showed that no variables included in the regression model violated the PH assumption at the 1% significance level.

FIGURE 11.

Kaplan–Meier Figure for SPFI by randomisation allocation in the ITT population.

Quality of life

Ancillary analysis

Adverse events

There were 913 participants in the safety population who experienced AEs across the trial with only 3 not experiencing any. Out of 485 participants, 484 (99.8%) on the CCS experienced an AE, while 429 (99.5%) out of 431 participants on the DFIS experienced an AE. All 384 participants on sunitinib experienced an AE and 529 (99.4%) of the participants on pazopanib experienced an AE. During cycles 1–4 of the treatment, both strategies were identical. During this time, the number of participants who experienced an AE is identical to the overall summary. Only 488 participants in the safety population continued on to cycle 5 or further. Of these, 484 (99.2%) experienced an AE, 262 (98.9%) on the CCS and 222 (99.6%) on the DFIS out of a total of 265 and 223, respectively. Two hundred and twenty-two participants on sunitinib continued past cycle 5 and 220 (99.1%) experienced an AE at this point. Two hundred and sixty-four (99.2%) out of 266 participants on pazopanib experienced an AE from cycle 5 onwards.

Overall, 671 (73.3%) out of 916 participants in the safety population experienced an AE of grade 3 or above, with 343 (70.7%) out of 485 on the CCS and 378 (76.1%) out of 431 on the DFIS. Two hundred and eighty-two (73.4%) of participants on sunitinib and 389 (73.1%) of participants on pazopanib experienced AEs of CTCAE grade 3 or higher. From cycles 1 to 4, both treatment strategies were identical. During this time, 539 (58.8%) out of 916 participants experienced an AE of CTCAE of grade 3 or higher. 278 (57.3%) out of 485 participants on the CCS and 261 (60.6%) on the DFIS experienced an AE of CTCAE of grade 3 or above. Two hundred and twenty-one (57.6%) and 318 (59.8%) of those on sunitinib and pazopanib, respectively, experienced an AE of grade 3 or higher during this time. Of the 488 participants who continued trial treatment past cycle 4, 275 (56.4%) experienced an AE of grade 3 or above. These were experienced by 134 (50.6%) participants on the CCS and 141 (63.2%) participants on the DFIS. One hundred and eighteen (53.2%) and 157 (59.0%) participants on sunitinib and pazopanib, respectively, experienced AEs of grade 3 or above from cycle 5 onwards.

Of the AEs which were pre-specified to be of interest, noticeably most participants experienced fatigue (88.3%) and hypertension (69.2%) (see Report Supplementary Material 1, Table 73). In the trial, there was little difference in the reporting between randomisation allocations. However, pazopanib had a noticeably higher proportion of participants experiencing hypertension (pazopanib: 73.1%, sunitinib: 63.8%) and hepatotoxicity (pazopanib: 49.2%, sunitinib: 22.4%). Alternatively, sunitinib had a noticeably higher proportion of participants experiencing neutropenia (pazopanib: 11.1%, sunitinib: 31.3%), thrombocytopenia (pazopanib: 16.4%, sunitinib: 29.4%) and mucositis/stomatitis (pazopanib: 44.4%, sunitinib: 66.9%). Considering the maximum grade of each CTCAE (see Report Supplementary Material 1, Table 74), it seems that the differences between TKI received occur in the higher CTCAE grades.

Overall, 451 (93%, n = 485) participants on the CCS and 414 (96.1%, n = 431) on the DFIS reported other AEs. Overall, 865 (94.4%, n = 916) of participants experienced an other AE. Three hundred and seventy-one (96.6%, n = 384) of participants on sunitinib and 494 (92.9%, n = 532) of participants on pazopanib experienced an other AE. In total 13,007 other AEs were reported with 6643 occurrences from participants on the CCS and 6364 occurrences from participants on the DFIS. The majority of other AEs were of grade 1, the lowest grade, 4881 (73.5%) from the CCS and 4567 (71.8%) from the DFIS (see Report Supplementary Material 1, Table 75). Six thousand one hundred and thirteen and 6894 other AEs came from participants on sunitinib and pazopanib, respectively. There was little observed difference between the grades of the AEs from the two TKIs. Other AEs of grade 3 or above were re-categorised to the CTCAE term (see Report Supplementary Material 1, Table 76). The most commonly reported other AE of grade 3 or above was back pain (4.5%), followed by lung infection (4.5%) and abdominal pain (4.0%). Other AEs grade 3 or above which could not be coded to a CTCAE term are listed in Report Supplementary Material 1, Table 77.

Serious adverse events

Osteonecrosis of the jaw

There have been four cases of ONJ reported during trial treatment; all four cases were in participants on sunitinib with two participants on the CCS and two on the DFIS. The full details of each case are given in Report Supplementary Material 1, Table 89.

Pregnancies

One pregnancy was reported on the trial. This was the partner of a trial participant who was receiving pazopanib and was on the DFIS arm. The pregnancy was 40 weeks and resulted in a healthy birth.

Within-trial cost-effectiveness analysis

Decision economic model analysis

A decision-analytic model (DAM) was developed to estimate the outcomes occurring beyond the 2-year follow-up and outwith the trial period (to a lifetime horizon). These were combined with the observed costs and benefits accumulated up to 2 years.

Within-trial analysis

The results from the primary analysis (ITT) and sensitivity analyses (listed in Numbers analysed including PP, and complete case analyses) are included in Table 20 and Appendix 5, Table 46. In all analyses, DFIS is found to be both cost saving and providing additional QALYs over CCS (i.e. it is a dominant strategy). The differences in total costs appear to be largely driven by the additional medicine costs in the CCS arm. There are high levels of missing data on self-reported resource use (ranging from 19% of observations for items listed under primary and community care to 45% for items listed under outpatient and inpatient care), with slightly more missing data in the CCS than in the DFIS arm (Appendix 5, Table 45). However, the general pattern across these analyses is that the greater the level of imputation or extrapolation, the smaller the differences in costs and QALYs become. This indicates that the imputation and extrapolation methods used in our primary analysis have not led us to overestimate the QALY gains or cost savings associated with DFIS.

The DFIS appears to be the optimal strategy, regardless of the sample used in the analysis and remains robust when analysis assumptions are relaxed. Figure 18 shows the cloud of ICERs on the cost-effectiveness plane; the simulated ICERs fall predominantly in the south-east quadrant meaning that DFIS is highly likely to reduce costs and increase HRQoL compared to CCS. The bootstrap analysis found that DFIS would have a 99% chance of being cost-effective. Given these results, the CEAC was not considered informative and is not included.

FIGURE 18.

Cost-effectiveness plane for the primary analysis.

The MNAR sensitivity analyses indicated the sensitivity of the results to the MAR assumption except where changes were applied to both arms (see Appendix 5, Figure 28).

Decision model analysis

The cost-effectiveness results yielded by the lifetime decision model are included in Table 21. In all analyses, DFIS is found to be both cost saving and yielding QALY gains over CCS. This finding is robust to changes in the modelling period, discount rates and survival curve estimation. The PSA indicates that, at a willingness-to-pay threshold of £20,000 per QALY gained, DFIS has a 95% chance of being cost-effective. DFIS has a 96% chance of being cost saving and a 68% chance of yielding QALY gains. The cost-effectiveness plane, CEAC and net benefit distributions generated by the lifetime analysis are included in Appendix 5, Figures 31–33.

The EVPI (£135 per patient; £884,250 per population in year 1) is low and reflects the degree of certainty regarding which strategy is optimal. The partial EVPI bar chart indicates the absence of decision uncertainty across parameters. Only the mean (distribution scale) parameter for the OS curve is determined to have a positive EVPPI (Appendix 5, Figure 34). However, these values are low (£21.66 for the CCS OS mean) and over the relevant population (£141,773). In tandem with the overall EVPI results, given the likely patient population, the cost of decision uncertainty does not appear to warrant additional research.

Design

A qualitative inductive study design using in-depth interviews with a pre-designed topic guide was employed.

Patient and public involvement

Patients with RCC were consulted about adapting information including information sheets and topic guides so they were patient friendly. A patient recruitment video was developed for research nurses to use and to demonstrate the patient perspective.

Topic guides

The topic guides (see Appendix 6) contained questions to explore why patients had declined to take part in the study, their reasons for agreeing to join the trial and their experience of the treatment and extended breaks.

Participants

We aimed to approach 24 patients from three clinical research sites (Leeds, Manchester, Cambridge) for each study. This was increased to four sites to increase recruitment numbers (Hull). For study 2, participants would be selected at variable points in the trial, for example, some prior to their planned treatment break, some during their first or subsequent treatment breaks and all were recruited on a consecutive basis.

Inclusion criteria

For study 2, participants who had been randomised to the modified sunitinib/pazopanib arm of the STAR trial in the DFIS were invited to interview. They had to be well enough to take part in an interview either face to face or over the telephone, able to provide written, informed consent and be willing to talk about their experiences in the trial.

Exclusion criteria

Patients were excluded if they were unable to provide written informed consent, were unwell or declined to take part.

Recruitment method

Patients were approached by their clinical team and given an information sheet and a demographic questionnaire. Verbal consent to pass personal details to researchers was obtained. Patients had a week to decide to take part or decline by completing a reply slip, sending an e-mail or by telephone. The interview was scheduled and signed consent was obtained. This recruitment strategy was preferred because it minimises response bias and potentially increases the methodological rigour of the research. 95

Data management and analysis

Interviews were transcribed verbatim and managed using Nvivo. 96 Data were stored according to standard procedures at the University of Leeds. An analysis plan and coding frame were developed a priori.

A thematic analysis was used,97,98_ENREF_98. Interview data were analysed after each interview to aid additional prompts at the interview. Transcripts were read and coded for themes and subthemes. The analysis was refined by using a constant comparison and contrastive approach and identifying negative cases to examine similarities and differences between participants. 98,99 The analytical process, the themes and subthemes and the interpretations resulting from it were refined and agreed in discussions with a second researcher (JH).

Study 1: acceptability of STAR trial

Study 1 employed qualitative interviews and a self-completion questionnaire to obtain views of those who declined to take part in the trial. Only one participant was referred to be interviewed for study 1. They declined trial participation due to perceived health reasons, not the treatment break. Recruitment to the STAR trial itself was very successful, so the study of decliners – intended to facilitate recruitment – was discontinued in Phase 3.

Study 2

Recruitment to the trial and the substudy

Recruitment and retention figures in the STAR trial were high. The accounts of participants showed that anticipated benefits of QoL became a reality. The recruitment figures speak for themselves: many eligible patients were content with the treatment alternatives represented by STAR. The qualitative work, which must be interpreted in that context, affords insight into some more specific reasons for acceptability. Participants were asked directly why they had taken part, but responses to other questions also threw light on the appeal of participation from their point of view.

The risk of harm continued to be acknowledged and was seldom far from respondents’ minds, but they did not regret their decision to take part in the trial, even if they had personally experienced the return of unpleasant symptoms or clinical progression. They understood that allocation to the DFIS arm of the STAR trial would entail an early and fairly predictable improvement in daily life. They were less clear about how long treatment might be effective for them, as it was tailored to their clinical circumstances. The question of whether the break might prove later to have been enjoyed at the expense of an unpredictable reduction in treatment effectiveness was not at the forefront of their minds. They were happy that two treatment strategies were being assessed and that the health professionals caring for them had their best interests in mind.

The initial context for STAR was favourable: participants trusted their doctors not to suggest anything that was against their interests. As in other studies, ‘wanting to give something back’ and altruism featured in accounts,100 as did having few other options. In STAR itself, if allocated to the DFIS arm, participants knew they would be monitored closely and reinstated on drug treatment if the need arose. They had high levels of resilience (evidenced by strategies for getting on with usual activities) and good support networks.

Also relevant was the nature of the treatment under evaluation: unlike non-crossover intervention trials (common), in the experimental DFIS arm of STAR, at the point of disease progression off treatment, participants were planned to re-start previous treatment and could then continue without treatment breaks as per participant choice. Of note, a number of participants chose to have multiple planned treatment breaks (see Treatment breaks).

Quality of life in treatment breaks

The accounts suggest that the DFIS was not just acceptable, it was an attractive option to many and a definite preference for some. They welcomed the opportunity to feel more able to resume valued activities in their lives, even if only temporarily. It follows that STAR recruitment may have been enhanced because a treatment option preferred by many patients was not available outside of the trial. Preference trials such as ProtecT, comparing very different treatment modalities, in this case, active monitoring, surgery and radiotherapy, would not benefit from this kind of effect as all arms were available outside of the trial and patients would choose their preferred arm rather than being randomised to one of the three options. 101

Study limitations

A highly selected group of participants in the DFIS arm of the STAR trial were interviewed. These participants had maintained resilience, hope and personal social networks of support. Participants on the continuous arm, those who dropped out or participants for whom treatment was no longer working were not interviewed. It is possible that an alternative research design that incorporated the perspectives of those on the continuous arm would enable a more in-depth comparison of long to short breaks. This may have illustrated additional psychosocial issues. A further design issue is sampling patients with different numbers of treatment breaks. Ideally, you would stratify the sample to have equal numbers in each category, so this limited any comparison of those experiences.

We were not able to reach the recruitment target, although we believe we reached sufficient information saturation about the issues being evaluated. The main pressure on recruitment was the limited number of sites for the qualitative substudy and staff changes at the site. A high number of participants who agreed to be contacted declined an interview (6/17). It is possible that the qualitative assessment did not include the less resilient or those living alone or who were deemed not well enough to take part due to the protective actions of the nurses caring for them. The additional burden of taking part in an interview should not be underestimated. Those trial participants who took part in the interviews were required to focus on the adversity of diagnosis and treatment and it was made clear that interview participation was entirely voluntary.

Implications for care

If such a strategy was implemented in practice, some thought should be given to how patients should be supported during the extended break to cope with and alleviate worries. The overall trial results will help with this, as no substantial detrimental effect on OS was shown. It will be important to support patients to maintain resilience and receive adequate information about disease progression and tumour regrowth along with professional and peer support. As part of this, patients on a treatment break require close monitoring and access to rapid clinical assessment if they become unwell.

Implications for trials

For understandable reasons, much of the literature on the added value of qualitative work to the conduct of trials has focused on recruitment,102 and initially, that was also the focus in STAR. When recruitment proved not to be a problem, attention switched to understanding the experience of receiving the novel treatment and how that might relate to the acceptability of DFIS as a distinct departure from usual care.

By collecting data on the experience of being in the intervention arm, STAR extends the methodological literature on the role of qualitative research in trials. The data throw light on the acceptability, even attractiveness, of the intervention and the reasons for that acceptability, which in turn help explain the high recruitment rate. A distinctive feature of STAR is that eligible patients could relate to both arms of the trial: the DFIS form of ‘active monitoring’ to such experienced patients held considerable appeal and contributed to their willingness to take part in the trial.

More generally, the experience of STAR is a reminder that even when a patient’s treatment options entail materially different trade-offs between process and outcome, some people will wish to weigh up of pros and cons – including those relating to uncertainty – before a treatment decision is made. The usual care approach of maximising effectiveness within the limits of tolerability will undoubtedly be the preferred option for some, but it is unlikely to be the preferred option for all.

Design

Imaging assessment with dynamic contrast-enhanced magnetic resonance imaging

All DCE-MRI examinations were performed on a Siemens (Erlangen, Germany) 1.5 T system. The DCE-MRI assessed up to five target lesions (the largest five) within the abdomen and pelvis identified at baseline (before TKI initiation), 4 weeks (post initiation of TKI therapy) (± 4 days) and 10 weeks (± 4 days) after STAR trial randomisation.

A detailed description of the dedicated imaging technique and parameters was published. 103

Post dynamic contrast-enhanced magnetic resonance imaging acquisition image analysis

The DCE-MRI imaging data were anonymised and post-processed using the software Platform for Research in Medical Imaging Version 0.4 (PMI 0.4). The post-DCE-MRI acquisition image analysis to assess the parameters such as perfused tumour volume, K^trans, ECV and MTT were described in detail103 by two experienced observers.

Statistical analysis

Two-tailed paired t-tests were used to analyse the change in the DCE-MRI parameters between the three time points (baseline and 4 weeks, 4 weeks and 10 weeks and baseline and 10 weeks). The differences in DCE-MRI parameters between the participants with PD at 24 weeks and those with no progression were evaluated using an independent samples t-test for normally distributed data and the Mann–Whitney U test was used for non-normally distributed data determined by using a Kolmogorov–Smirnov normality test. For participants with more than one lesion identified on the DCE-MRI, only the largest lesion was selected to analyse the changes to tumour perfusion characteristics over the three time points in relation to the primary end point as these were least affected by partial volume effect. Receiver operating characteristic (ROC) curve analysis and AUC values were calculated for parameters that were associated with PD at 24 weeks. The statistical significance level was set at p < 0.05. All statistical tests were performed using SPSS Statistics software (Version 21.0; IBM Corp., Armonk, New York, USA). Interobserver agreement was assessed using the intraclass correlation coefficient (ICC) with ICC values scored as excellent (> 0.81), good (0.61–0.80), moderate (0.41–0.60), fair (0.21–0.40) and poor agreement (< 0.2).

Interobserver agreement

The interobserver agreement was excellent; perfused tumour volume [ICC (95% CI): 0.928; (0.869 to 0.959)]; K^trans [ICC (95% CI): 0.949; (0.918 to 0.969)] and ECV [ICC (95% CI): 0.910; (0.800 to 0.961)].

Participants and treatment

Participants who were eligible and who consented to the STAR trial125 provided additional (optional) consent to participate in this CT substudy. Participants with suitable CT scans were included in the analysis defined as RECIST measurable disease at one or more sites on baseline pre-treatment imaging and IV contrast-enhanced CT imaging. Participants were not included in the analysis if there was non-measurable disease at baseline; non-contrast CT at the required time points; or incomplete imaging data sets.

Computerised tomography imaging

Computerised tomography imaging was performed pre-treatment and 12 weekly thereafter for the main trial. CT scanning parameters required for inclusion in the substudy population are summarised in Appendix 8.

Image analysis

For this substudy, contrast-enhanced CT scans from baseline, 12 weeks and 24 weeks of therapy were reanalysed centrally. Image analysis was carried out by an experienced radiologist (7 years of body imaging) on a workstation using commercial software (Syngo 2012c, Siemens Healthcare, Erlangen, Germany). Target lesions were selected, as per RECIST 1.1. guidelines (≥ 10 mm long axis in size or ≥ 15 mm short axis if nodal, up to two lesions per organ, up to five per patient), and their locations recorded.

Target lesions were assessed in the arterial or portal phase and for a subset of lesions where both arterial and portal venous phase imaging was acquired through the lesion; assessment was performed in both phases. Assessment of tumour size (maximum target lesion longest dimension or nodal short-axis dimension) and attenuation (in Hounsfield units, HU) were undertaken using a semi-automated process. Whole lesion measurements were obtained by drawing a freehand region of interest around the lesion perimeter which automatically propagated to subsequent slices and could be manually corrected for contouring accuracy.

A region-of-interest (ROI) was also placed within the descending aorta at the level of the diaphragmatic hiatus and the aortic attenuation was recorded for the purpose of signal normalisation. The process was repeated for the same lesions in follow-up studies.

Radiomic analysis was performed using in-house software based on MATLAB (Matlab 2013, The Mathworks, Inc., Natick, MA, USA) that has been validated as part of the International Biomarker Standardisation Initiative. 126 Again, freehand ROIs were drawn around each target lesion and a range of radiomic features including locoregional second- and high-order features were extracted automatically by the software.

Statistical analysis

Participants

In total, 182 participants were enrolled from 27 sites. Fifty-three participants were excluded, the majority for missing data or corrupt imaging data, leaving 129 participants [94 male, 35 female, mean ± SD age 64 ± 9 years (range: 40–85 years)] for final analysis. The participant flowchart is shown in Figure 21.

FIGURE 21.

Participant flowchart.

Participant characteristics are summarised in Table 26. The majority of participants (71%, 92/129) had undergone prior nephrectomy. Of these, 56% (72/129) received sunitinib therapy, the remainder pazopanib. In total, 233 target lesions were evaluated; the majority were nodal in location (29%, 69/233). Of these target lesions, 42% (99/233) were imaged in both phases, 35% (82/233) were imaged in the arterial phase only and 22% (52/233) in the portal phase only.

Imaging response categorisation and outcome prediction

Prediction of 24-week response categorisation using baseline radiomic variables

Exploratory analysis was undertaken for 116 target lesions from 76 participants; 81 baseline radiomic variables were considered. Nine baseline radiomic features were found to be significantly associated with patient response (see Table 35). Associated features were all derived from grey-level dependence matrix (GLDM) and the grey-level co-occurrence matrix (GLCM).

Trial design and analysis

The STAR trial utilised a novel study design, in particular relating to its powering on both, OS and QALYs and the inclusion of a number of end points specific for intermittent treatment strategies.

For a study spanning almost 12 years from concept to completion with a large number of participants in a large number of sites, there was inevitably a need for the study to be able to respond to changing external circumstances, such as new drug approvals. A major strength of the STAR trial was the ability to maintain, against this background, rigorous conduct and reporting according to CONSORT recommendations, with analyses conducted according to a predefined SAP agreed with the TMG and reviewed by the DMEC.

The STAR trial was designed to be pragmatic, aiming to recruit a real-world population of patients receiving TKI therapy as first-line treatment for advanced RCC. The inclusion criteria were therefore kept as broad as possible. The intervention of a DFIS changed only the timing of treatment cycles and hence was not predicted to cause any additional toxicity to participants or any significant logistical issues in delivery at the site. We, therefore, enabled sunitinib (and later sunitinib or pazopanib) to be used in line with local practice to ensure that the majority of patients suitable for treatment with first-line TKI would also be eligible to participate in STAR. This was done to ensure the generalisability of the trial results and the baseline characteristics of the participants confirm the real-world population represented.

The trial was robustly designed to answer the overall Phase III aims. It was however recognised during conception that there were significant challenges to being able to do this, for example, in terms of recruitment of the sample size required to demonstrate NI, the duration of recruitment required, ensuring that the approach of planned treatment breaks was deemed acceptable to clinicians and health professionals and the limited data available to inform the SAP. For these reasons, the three-stage trial was proposed with a Phase II to Phase III seamless design: Phase II including Stage A to address feasibility in terms of recruitment rate, Stage B to address efficacy in terms of TSF in addition to Phase II secondary end points to confirm the accuracy of the assumptions made to inform the Phase III sample size. This was a key strength of the study design as it permitted the Phase II data to provide assurance that the study should continue on to a Phase III study.

Careful consideration was made in selecting all the trial end points and a strength of the study is in the selection of those appropriate to an intermittent treatment strategy. Standard end points such as PFS are not appropriate in intermittent strategy trials as earlier progression is expected with the inclusion of a treatment break, the question relates to whether disease control can be regained on retreatment. Time to strategy failure (also referred to as time to failure of strategy) was proposed as an appropriate end point for intermittent strategies. 64,135 For this reason, TSF was selected to be the primary outcome of Stage B.

The large sample size, and the fact that renal cancer is termed a ‘rare’ cancer, required a relatively long duration of recruitment from a high number of sites. This prolonged recruitment meant that there were changes in the treatment options available for RCC during the time it was open. Although STAR was planned to be a pragmatic study, it was essential to ensure that any agreed amendments would not compromise the trial integrity and/or interpretation of the results. When considering the results now in line with current practice, it is also important to account for the situation and the time of trial design and delivery.

This ability to respond to the changing treatment landscape was another strength of the STAR trial. For example, during initial trial development, the potentially practice-changing first-line trial of sunitinib compared to IFNα5 had been published, and sunitinib in this setting was under consideration by NICE and was expected by the clinical community, but was yet to be formally approved (this occurred March 2009). There were accumulating data that alternative TKIs such as pazopanib may provide similar benefits in the first-line setting potentially while causing less toxicity, but data confirming this were awaited. It was appreciated by the TMG that in order for the trial to be feasible, the CCS arm needed to align with standard UK practice. Horizon scanning anticipated pazopanib approval (occurred February 2011) and consideration of inclusion of pazopanib alongside sunitinib was planned pre-opening the Phase III part of the trial. During recruitment into Phase II, it became apparent that not allowing clinicians to utilise pazopanib as an alternative TKI would limit recruitment after the publication of the COMPARZ trial (October 2012)13 which demonstrated comparability with sunitinib. Waiting to include pazopanib only in Phase III would have therefore jeopardised demonstrating the feasibility of continuing to Phase III and caused premature closure of the trial. After careful consideration of the pros and cons, and with the full support of key investigators, patient representatives, TSC, DMEC and the NIHR HTA, the eligibility criteria were updated to allow this in April 2013. The TKI used was added in as an additional stratification factor.

STAR was one of the first Phase III trials to have a patient-reported outcome measure feed into the co-primary end point (QALY). Overall, the return rate of QoL questionnaires was excellent [13,147 out of 16,726 (78.6%) questionnaire booklets were returned during the trial]. However, the nature of the end point and the trial meant that careful consideration was required during the analysis. Missing data in patient-reported outcome measures typically cannot be chased and therefore plans to address missingness were included in the SAP as summarised in the section Missing data. The results of the primary analysis which imputed missing data during the follow-up period reached the same conclusion as the complete case analysis and alternative imputation methods. Therefore, it is likely that missing questionnaire data were due to chance (MAR) rather than for underlying reasons related to the QoL of patients (MNAR). A way to determine between data that are MAR and MNAR is to collect the reasons for missing questionnaires. A limitation of this study is that this information was not available for the majority of questionnaires. However, given the concordance in the results for the QALY analysis where both the PP and ITT analyses concluded NI, this is less of a concern for this study. We would recommend that any future studies considering patient- reported outcome data, either as a primary or secondary end point, collect the reasons for missing questionnaires from the outset of the trial to aid their analysis.

The decision to impute questionnaires during the follow-up period was supported by the frequent collection of QoL questionnaires during treatment which resulted in an accurate measure of QALYs even when questionnaires were missing. However, this will have increased the burden on the research staff at the site as all questionnaires were completed at a site on paper. Since trial conception, more up-to-date methods of collection have been adopted where participants can complete questionnaires at home on phones, computers or tablets. We would recommend that future trials adopt these more modern methods of QoL collection.

A limitation of the secondary QoL analysis is that multiple models were fitted onto the same data set. However, no adjustments for multiplicity were made. This was deemed appropriate as they were unpowered secondary end points and on consideration of the results, while statistically significant effects were observed, the effect sizes were not clinically meaningful.

As highlighted previously, when the STAR trial was designed there was a very real concern that planned treatment breaks may not be acceptable to patients and/or clinicians. In fact, this was not the case, supported by the qualitative outcomes with few patients withdrawing from the trial stating a wish for continuous treatment as the reason. However, during trial design, these concerns led to the STAR trial mandating only one planned treatment break in the DFIS arm, with subsequent breaks taken at the discretion of the patient and clinician. The median number of planned treatment breaks was one, but a broad range was observed up to nine. A consideration in similar trials in the future would be to test the ongoing strategy to include multiple planned treatment breaks as PP.

Radiological reporting

Another strength of the trial design was that it also permitted a number of changes to be implemented around the time of transition to the Phase III part of the trial. A key change was the removal of the requirement for central radiological reporting and the transition to local radiological reporting. Central radiological reporting was required in the Phase II part to reassure a number of clinicians who, at the time of trial conception, were concerned that participants on the DFIS arm who at 24 weeks had a disease that was continuing to respond (i.e. still shrinking) should not have their TKI stopped until the response stabilised. For participants to be eligible to take up their treatment break in Phase II, they were required to have achieved MRR; this required comparison to the previous scan rather than baseline (the latter is standard for RECIST reporting).

In practice, the implementation of central radiological reporting was very challenging. A key issue was the short turnaround time available between imaging and the clinic appointment for informing patients of the results. At the time, there was no way that images could be sent electronically, so each scan had to be anonymised and downloaded onto a CD at the site and then sent centrally to Leeds. A considerable amount of work was involved in the anonymisation of data and transfer of images and reports between sites; associated with this were logistical difficulties with the collection and delivery of images and timely central reporting, so it was not always possible to have the central report at the time of clinic appointment and meant that clinicians at the site were sometimes required to make treatment decisions without this.

The central reporting had not been planned to continue into Phase III and it was apparent that it would not be feasible to make radiologists at the site report scans in a new way (by comparing to the previous scan). Reassuringly, when the data from the central reporting were reviewed at the end of Phase II, fewer than 5% of participants had not had MRR at 24 weeks. This provided reassurance to clinicians that simplifying the pathway was clinically justifiable and that participants in the DFIS who had not progressed by 24 weeks could take up their planned treatment break at this point (unless a strong clinical rationale to continue). This proposal was reviewed by the TMG and DMEC and ratified by the TSC.

PPI input and patient perceptions

The STAR trial included significant PPI input throughout, from initial focus groups to throughout the study with PPI membership on both the TMG and TSC. As mentioned previously, there was concern regarding whether patients would be willing to stop a treatment known to be working. Consideration was therefore given to the timing of approach to participate and randomisation as well as to how best to present the trial to patients. Based on clinical considerations from other relevant studies and patient discussion, the decision was made to perform randomisation at baseline prior to patients receiving any TKI, with patients taking up their randomisation at 24 weeks assuming no progression. A DVD was also developed for patients assisted with the presentation of the trial. The benefits of this approach were demonstrated by achieving the recruitment rate required in Phase II to enable progression to Phase III and the high take-up rate of allocated treatment arm.

The decision to not attempt to blind the trial to patients, medical staff or clinical trial staff was deemed to be justified in view of the additional cost and logistical issues which outweighed the benefits.

Amendments to sample size and event rate

In the trial design, the required sample size was 1000 participants (see Phase III). This assumed that 2-year survival in the CCS arm was 54% and that a maximum of 10% of participants would be lost to follow-up. As a result of the recruitment rate for the trial slowing down in 2016, these assumptions were re-assessed. It was found that the 2-year survival rate in the CCS arm was 48.5% and only 2% of patients had been lost to follow-up. Therefore, following the DMEC and TSC agreement, the sample size was updated to reflect these estimates requiring 920 participants and 720 events.

As discussed previously, during the lifespan of the trial, the landscape of treatment for advanced renal cancer changed significantly. From a situation at the start with only one TKI approved (sunitinib), there are now multiple treatment options spanning 3+ lines of therapy, and with these changes, there has been a significant improvement in outcomes for patients. This is evidenced through the 2-year survival estimates in the CCS arm of 55.5% in the ITT population and 55.2% in the PP population, a higher proportion than the assumed 48.5%. For the trial, this meant that fewer events (deaths) occurred that were predicted during follow-up, which had a consequent impact on the overall power of the study. Follow-up was extended to include more events, but this was not sufficient, and a very significant further extension would have been required to attain the number of events required to reach 80% power. Therefore, after careful consideration, this was not felt to be justifiable in terms of resources for a minimal increase in potential power. This was a limitation as it meant that the overall power of the trial was slightly reduced. Clearly, this slight reduction in power may have been a reason why the study only fell short of demonstrating NI in OS for the PP population. In addition, the change in the event rate could have motivated a change in the NI margin. However, as it was pre-specified to conclude NI based on the relative difference of 0.812, rather than the absolute difference of 7.5% and relative differences are less affected by changes in event rate,136 0.812 was kept as the NI margin. Note that because a similar proportion of patients in both arms received immunotherapy treatment post trial (see Anticancer treatment post trial), subsequent therapy was not adjusted for in any ancillary analysis.

Outcomes of main study

The STAR trial was one of the largest non-commercial studies in advanced RCC ever carried out in the UK. It has provided extensive evidence demonstrating patient acceptability of treatment breaks. As outlined above, the appetite of patients for a treatment break was surprisingly high such that only 12 patients in the DFIS arm (2.6%, n = 459) withdrew from the study so they could have continuous treatment. In addition, although only one treatment break was mandated, more than 43% of patients opted to take two or more treatment-breaks and the qualitative substudy (see Qualitative assessment) clearly showed patient acceptability for the treatment break strategy, although the knowledge that they were being closely monitored during a break remained crucial. In a trusted and supportive clinical context, patients welcomed the opportunity that a break afforded to resume valued activities in their lives, even if only temporarily.

The trial also demonstrated the successful use of the seamless Phase II/Phase III design, with all patients contributing to the final Phase III outcome data as well as using Phase II data to demonstrate adequate recruitment rate and early indication of efficacy via TSF.

The formal co-primary end point required that both the OS end point and the QALY end point meet the pre-set boundary conditions set out in the grant application in both PP and ITT populations in order to be able to conclude NI. For the OS end point, although the ITT population met this condition, the PP population marginally failed to do so (Cox HRs 0.83 and 0.80, respectively, compared to a requirement of ≥ 0.81 in order to show NI, based on the pre-stated ≤ 7.5% difference in OS, Figure 23). For the QALY end point, this condition was met for both the PP and ITT populations, based on the pre-stated ≤ 10% difference in QALY. Because the PP population did not meet the OS requirement, this means that formally we cannot conclude that the NI condition was met. This difference may be explained through the lack of power for the comparison due to the reduced number of events observed in the PP population potentially explained by the changing treatment landscape throughout the trial and the nature of the PP population.

FIGURE 23.

Comparison of lower CI bounds to NI margin – OS.

During the STAR trial, other systemic therapies were introduced into the RCC treatment landscape, and OS improved for STAR participants by the introduction of immunotherapy and new therapies such as nivolumab and cabozantinib in the second- and third-line setting. While this improvement in OS is very welcome, it has resulted in a lower event rate (death) for STAR patients. With the agreement of the DMEC and TSC, trial follow-up ceased before the required number of survival events (720) were observed for 80% power. In addition, due to the nature of the strategy, 49 (5.3%) of patients were excluded from the PP population, more in the DFIS arm compared to the CCS arm (see Figure 4) resulting in less power for the PP comparison to the ITT comparison (720 events were required, 678 events were observed in the ITT population and 648 events were observed in the PP population).

Nevertheless, it may be seen from Figure 7 that for the ITT population, the Kaplan–Meier OS plots for the two arms are almost superimposable, except for the small numbers of patients with OS longer than about 6 years and there is no clinically meaningful difference in the median survivals (28 months in the CCS arm, compared with 27 months in the DFIS arm). Similarly, for the PP population, again the Kaplan–Meier OS plots for the two arms are almost superimposable (see Figure 6), except for the small numbers of patients with OS longer than about 6 years and there is no clinically meaningful difference in median survivals (28 months in the CCS arm, compared with 27 months in the DFIS arm). Notably also, the respective median OS values are the same in the ITT and PP populations, yet the ITT population met the pre-defined NI margin, but the PP population marginally did not.

The NI margin of ≤ 7.5% was chosen following discussions with the UK and US communities and NCRI Renal Clinical Studies Group (CSG). Earlier designs had proposed a difference of ≤ 10% to be clinically acceptable. However, this was not accepted due to the data available on sunitinib over IFNα (54% compared to 46% at 2 years). During the discussions around the determination of the NI margin, it was agreed that a limited reduction in OS would be clinically acceptable if benefits in terms of reduced drug toxicity and improvements in QoL and health economic parameters were observed. Given the results observed, while NI cannot be concluded, the DFIS is considered to be clinically acceptable based on these parameters.

We also carried out sensitivity analysis for OS using a piecewise model. This analysis accounted for the fact that patients in both arms were treated identically for the first 24 weeks after randomisation. These analyses (see Overall survival – piecewise model) showed similar results to the main primary analysis where NI could not be concluded in the PP population by a small margin (95% CI Lower Bound 0.80 < 0.812).

The cost-effectiveness analysis showed that, at 2 years, the DFIS arm was associated with substantial cost savings (£6408 and £3235 per patient in complete case and imputation analyses, respectively). These savings are driven by overall reduced treatment costs. We also estimated QALY gains for DFIS versus CCS over the same period (0.049 and 0.165 for imputation and complete case analyses, respectively). These findings were mirrored by the decision modelling analyses over a lifetime horizon. Both analyses concluded that DFIS was highly likely to be the most cost-effective strategy. Sensitivity analyses indicate this conclusion is robust to changes in various assumptions and analytical approaches. Cost-effectiveness estimates were increased in favour of DFIS when the costs of subsequent therapies were included in the analyses.

Although not a primary end point for the overall Phase III study, the ‘Time to Strategy Failure’ (defined as progression on sunitinib/pazopanib treatment, need for a change in systemic treatment or death) proved to be valuable as an interim Phase II end point and also a secondary Phase III end point. In essence, this was a measure of the time for which the trial strategy (CCS or DFIS) was working. Figure 8 shows a plot for the ITT population, demonstrating a clear and significant increase in TSF for the DFIS arm (HR 0.75, 95% CI 0.65 to 0.86); p-value < 0.001 compared with the CCS arm (median 11 months vs. 8 months), although this does not ultimately translate into an OS advantage.

It is clear that PFS cannot be used in the conventional way in a trial with treatment breaks as the conventional PFS does not take into account the successful rechallenge with sunitinib/pazopanib following progression on a treatment break. Figure 10 illustrates this point where it is clear that, for the first 6 months, the PFS curves are superimposable for the CCS and DFIS arms, but the DFIS arm shows shorter PFS following the treatment breaks. A better way of comparing the two arms is to use summative PFS (see Figure 11), where it is clear that the proportion of participants progression-free is significantly higher in the DFIS arm than in the CCS arm.

Comparison of toxicity data is also complex because raw AE and SAE data do not take into account that participants in the DFIS arm were recording them for longer due to their greater time on trial treatment. However, on consideration of SAEs (see Serious adverse events) deemed to be related to TKI treatment (SARs) which are recorded in the same period for both arms, a smaller proportion of participants in the DFIS arm experienced an event than in the CCS arm and participants in the DFIS arm accounted for fewer of the overall events compared to the CCS arm, illustrating the potential safety benefits of the DFIS arm.

Substudies

The substudies add valuable detail and context to the overall STAR trial which have been fully discussed above. The key messages were:

• A key conclusion from the qualitative substudy is that the data throw additional light on the acceptability and attractiveness of the intervention and the reasons for that acceptability, which in turn help explain the high recruitment rate. A distinctive feature of STAR is that eligible patients felt able to relate to both arms of the trial: the DFIS form of ‘active monitoring’ to such patients held considerable appeal and contributed to their willingness to take part in the trial.

• The quantitative DCE-MRI feasibility clinical study associated with the STAR trial was the first to use longitudinal serial assessments to detect changes in biomarkers following sunitinib or pazopanib treatment in mRCC. It showed that DCE-MRI-derived biomarkers of tumour perfusion were potential surrogate biomarkers to predict early disease progression following TKI therapy in advanced RCC. The study also demonstrated that these DCE-MRI assessments were reproducible and that further larger prospective clinical studies are justified to test its wider application in the context of routine clinical practice.

• The CT substudy took advantage of the STAR trial to conduct a prospective multicentre study to assess the possibility of predicting response/non-response earlier in treatment, thus allowing for re-evaluation of therapeutic strategy as appropriate. The study confirmed that assessment of enhancement as well as size change alters the categorisation of response. Use of mChoi criteria may allow for earlier detection of PD, and more representative separation of participants with PR versus SD. Notably, while published literature has suggested an association of mChoi criteria with TTP, PFS and OS in terms of prediction of early progression within 24 weeks of treatment initiation, no association was shown in the STAR cohort. Assessment of changes in tumour heterogeneity may complement standard response assessment but current data remain limited and further work is still required to advance this field.

Potential benefits and harms

Because of the nature of the STAR research question, much of the above discussion already addresses the potential benefits and harms of a treatment-break strategy. The qualitative substudy showed perceived patient benefits in the opportunity to take treatment breaks which was also reflected in the low number of patients (12 in total) opting to withdraw at the randomisation point in order to have continuous treatment. In terms of OS, STAR has shown that there is no clinically meaningful difference between the DFIS and CCS arms when considering median OS (27 and 28 months, respectively). Further, ongoing analysis is looking at whether we can predict which patients will most benefit from treatment breaks and multiple treatment breaks. We plan to further analyse the STAR data in this regard. The health economic evaluations of cost-effectiveness indicate that DFIS is the optimal strategy and that, in all base cases and most sensitivity analyses, DFIS was shown to be cost saving and providing QALYs gains over CCS. Overall, the potential benefits of DFIS appear to outweigh any potential harms.

The STAR study, the largest of its kind which includes an OS end point, is expected to make a substantial contribution to the literature on the concept of treatment breaks, especially since it combines extensive assessment of patient preferences with quantitative outcome data. In RCC, while the STAR study was ongoing, a single-centre Phase II study by Rini and co-workers32 also provided evidence for the benefits of treatment breaks.

Other ongoing trials which have taken the STAR design into account include the REFINE study138 which is looking at treatment breaks in a range of cancer types including RCC patients on immunotherapy and studies such as DANTE139 which is exploring treatment breaks in immunotherapy treatment of melanoma.

Relating to the main trial

The main trial demonstrated that treatment breaks are acceptable to patients and clinicians, by virtue of > 40% of patients who continued past week 24 having greater than or equal to two treatment breaks and 27% greater than or equal to three treatment breaks. This suggests that there may be a subset of patients who are more appropriate for and will benefit most from the utilisation of a DFIS. There is work ongoing currently, aiming to define this population. The trial protocol mandated only one treatment break; however, it would be interesting when designing similar trials in the future to test the DFIS fully by including multiple treatment breaks rather than only the initial one. Another population of interest are the subset of patients who experienced exceptionally prolonged durations of disease control during treatment breaks, and it will be useful to further define these participants. Further research is warranted to consider if there are certain biological factors which result in some patients benefiting from a treatment break more than others.

As described at length previously, patient-reported outcomes were key outcomes in this trial. We would recommend that any future studies considering patient-reported outcome data, either as a primary or secondary end point, collect the reasons for missing questionnaires from the outset of the trial to aid their analysis. We would also suggest that in future trials, consideration is given to more contemporaneous methods of data collection enabling participants to complete questionnaires at home on phones, computers or tablets. These more convenient collection methods would hopefully reduce missing data. Thought should also be given to prompts to remind patients to complete questionnaires.

Since the conception of the STAR trial, treatments for advanced RCC have changed, such that it is a minority of patients who are treated with single-agent TKI treatment. TKI remains an important treatment in advanced RCC, and this may increase further first line with the recent approval of adjuvant IO in a subset of patients. Further research should now be considered into the potential benefits of treatment breaks in more contemporary RCC treatment, for example IO monotherapy or TKI/IO combination therapy.

In addition, more broad learning around intermittent treatment strategy trials should be taken from this trial and used to extend this research area to other types of cancer and their treatments more generally. With the number of available systemic treatments in cancer increasing with associated high costs in terms of finance and side effects, research to define appropriate treatment duration is increasingly important. Such trials are frequently large, costly and can be challenging to deliver; hence, ensuring that they are designed in the most appropriate way is essential. The STAR trial has demonstrated patient and health professional interest and support for exploring these alternative approaches.

Relating to health economics

The use in the STAR trial of a QALY-based primary end point remains novel. Its accuracy was reliant in part on PROMSs and hence was potentially impacted due to the assessment schedule, missing data and the breadth and recall of the HRQoL measure that we used. It is possible that the full QoL impact of treatment intervals has not been captured here due to these reasons; thus future research seeking to estimate the value of treatment breaks and dose reductions should adopt a more nuanced approach to QoL capture. This approach of capturing benefits should also explore patients’ preferences for such strategies, including their willingness to trade off the associated risks and benefits. Another area of future research identified while performing the STAR analyses was the need to develop methodological approaches to adjusting outcomes following subsequent treatments in trials.

Relating to qualitative work

The STAR qualitative work is related primarily to recruitment. However, if a DFIS approach is implemented in practice, thought should be given to how patients should be supported during the extended break to cope with and alleviate worries. The overall trial results will help with this, but additional qualitative work with patients who are approaching or on a treatment break would be useful to aid understanding of how patients are feeling during this time and could inform how patients could best be supported.

Relating to computerised tomography substudy

The STAR CT substudy confirmed that assessment of enhancement, as well as well-recognised size change, alters the categorisation of response. Use of mChoi criteria may allow for earlier detection of PD, and more representative separation of participants with PR versus SD. Assessment of changes in tumour heterogeneity may complement standard response assessment but current data remain limited. Further work is required to explore how to incorporate these findings alongside standard size assessments for response and also to explore the clinical relevance of these findings, in terms of the impact of earlier changes in treatment.

Relating to magnetic resonance imaging substudy

This small feasibility substudy has shown DCE-MRI-derived biomarkers of tumour perfusion (perfused tumour volume, K^trans and ECV) as potential surrogate biomarkers to predict early disease progression following treatment with TKI therapy in advanced RCC, and in addition, demonstrated reproducibility of these DCE-MRI biomarkers. Going forward, larger prospective clinical studies are required to confirm these biomarkers as predictive markers of early progression, and importantly lead on from this and test their wider application and clinical relevance.

Participants

We thank our 920 participants for their willingness to take part in research in order to inform future generations. Through ongoing clinical trials, we continue to strive for better outcomes for patients with resected RCC.

Trial Steering Committee

Professor Barry Hancock (chairperson), Dr Bernard Escudier, Dr Wedi Qian, Jackie Low (PPI).

Data Monitoring and Ethics Committee

Mr James Paul (chairperson), Dr Richard Jackson, Dr Peter Hall, Dr Uschi Hoffmann.

Trial Management Group

Professor Janet Brown (chairperson), Dr Fiona Collinson, Professor Julia Brown, Dr Christy Ralph, Professor Peter Selby, Professor Jenny Hewison, Dr David Meads, Dr Janine Bestall, Dr Adam Martin, Dr Tze Min Wah, Dr Pat Hanlon, Kara-Louise Royle, Jayne Swain, Chris Linsley.

Previous members: Dr Cat Handforth, Dr Kevin Chan, Dr Abdulazeez Salawu, Dr Sebastian Trainor, Professor Walter Gregory, Dr Luis Daverede, Dr Sandy Tubeuf, Silviya Nikolova, Dr Helen Howard, Lucy McParland, Emma Best, Emma Batman, Laura Allen, Vicky Hiley, Katie Neville, Heather Cook, Jamie Oughton, Cait Kielty-Adey, Alex Smith.

We would like to acknowledge the significant contribution of Pat Hanlon, our TMG patient and public involvement representative, who sadly passed away in January 2020. Pat had a huge commitment to kidney cancer research. He played an important role on the TMG and we will miss him very much.

We would also like to acknowledge the contribution of Kate Hayward who provided invaluable guidance during trial design.

MRI substudy group

Dr Tze Min Wah, Professor David Buckley, Dr Jim Zhong.

The MRI substudy was funded by the Leeds Hospitals Charity at St James’s Hospital, Leeds.

CT substudy group

Professor Vicky Goh, Dr Christian Kelly-Morland.

Investigator sites

We would like to thank all the trial teams members involved in the study at the participating sites, and it would not have been possible to carry out this work without their support.

The following sites and Principal Investigators were involved in the recruiting and treating trial participants:

Nottingham University Hospital (Nottingham University Hospitals Trust); Professor Poulam Patel.

Weston Park Hospital (Sheffield Teaching Hospitals NHS Foundation Trust); Dr Omar Din.

The Clatterbridge Cancer Centre NHS Foundation Trust; Dr Judith Carser (former Principal Investigator), Dr Richard Griffiths.

The Royal Marsden NHS Foundation Trust; Professor James Larkin.

St James’ University Hospitals NHS Foundation Trust; Professor Janet Brown (former Principal Investigator), Dr Christy Ralph.

Castle Hill Hospital (Hull University Teaching Hospitals NHS Trust); Professor Anthony Maraveyas.

The Christie NHS Foundation Trust; Professor Fiona Thistlewaite (former Principal Investigator), Dr Tom Waddell.

Mount Vernon Hospital (East and North Hertfordshire NHS Foundation Trust); Dr Paul Nathan.

Addenbrooke’s Hospital (Cambridge University Hospitals NHS Foundation Trust); Professor Tim Eisen.

The Beatson West of Scotland Cancer Centre (NHS Greater Glasgow and Clyde); Professor Rob Jones.

Southampton General Hospital (University Hospital Southampton NHS Foundation Trust); Dr Matthew Wheater.

Northern Centre for Cancer Care (The Newcastle upon Tyne Hospitals NHS Foundation Trust); Dr Rhona McMenemin (former Principal Investigator), Dr Ashraf Azzabi.

St Bartholomew’s Hospital (Barts Health NHS Trust); Professor Thomas Powles.

Scarborough General Hospital (York and Scarborough Teaching Hospitals NHS Foundation Trust); Dr Mohan Hingorani (former Principal Investigator), Dr Khaliq Rehman (former Principal Investigator), Dr Mohammad Khan.

Belfast City Hospital (Belfast Health and Social Care Trust); Dr James McAleer (former Principal Investigator), Dr Alison Clayton.

Yeovil District Hospital NHS Foundation Trust; Dr Geoffrey Sparrow (former Principal Investigator), Dr Urmila Barthakur (former Principal Investigator), Dr Emma Gray (former Principal Investigator), Dr Erica Beaumont (former Principal Investigator).

Royal Cornwall Hospitals NHS Trust; Dr Alastair Thomson.

Royal United Hospital (Royal United Hospitals Bath NHS Foundation Trust); Professor Mark Beresford.

Queen Alexandra Hospital (Portsmouth Hospitals University NHS Trust); Dr Joanna Gale.

Royal Bournemouth Hospital (University Hospitals Dorset NHS Foundation Trust); Dr Thomas Geldart.

Essex County Hospital (East Suffolk and North Essex NHS Foundation Trust); Dr Dakshinamoorthy Muthukumar.

Royal Free London NHS Foundation Trust; Professor Thomas Powles (former Principal Investigator), Dr Ekaterini Boleti.

Royal Shrewsbury Hospital (Shrewsbury and Telford Hospitals NHS Trust); Dr Narayanan Srihari.

Royal Devon and Exeter NHS Foundation Trust; Dr Denise Sheehan (former Principal Investigator), Dr Rajaguru Srinivasan.

Great Western Hospital (Great Western Hospitals NHS Foundation Trust); Dr Omar Khan.

Birmingham Heartlands Hospital (University Hospitals Birmingham NHS Foundation Trust); Dr Anjali Zarkar.

Norfolk and Norwich University Hospitals NHS Foundation Trust; Dr Gaurav Kapur.

Kent and Canterbury Hospital (East Kent Hospitals University NHS Foundation Trust); Dr Carys Thomas.

Maidstone Hospital (Maidstone and Tunbridge Wells NHS Trust); Dr Sharon Beesley (former Principal Investigator), Dr Kathryn Lees.

Medway Maritime Hospital (Medway NHS Foundation Trust); Dr Henry Taylor (former Principal Investigator), Dr Christos Mikropoulos (former Principal Investigator), Professor Stergios Boussios.

Cheltenham General Hospital (Gloucestershire Hospitals NHS Foundation Trust); Dr David Farrugia (former Principal Investigator), Dr Marios Decatris.

Cumberland Infirmary (North Cumbria Integrated Care NHS Foundation Trust); Dr Anil Kumar.

Blackpool Victoria Hospital (Blackpool Teaching Hospitals NHS Foundation Trust); Dr Falalu Danwata.

King’s Mill Hospital (Sherwood Forest Hospitals NHS Foundation Trust); Dr Santhanam Sundar.

Royal Preston Hospital (Lancashire Teaching Hospitals NHS Foundation Trust); Dr Natalie Charnley.

Musgrove Park Hospital (Somerset NHS Foundation Trust); Dr Mohini Varughese (former Principal Investigator), Dr Emma Gray.

Broomfield Hospital (Mid and South Essex NHS Foundation Trust); Dr Gopalakrishnan Srinivasan (former Principal Investigator), Dr Abdel Hamid.

Charing Cross Hospital (Imperial College Healthcare NHS Trust); Dr Naveed Sarwar.

University Hospital Coventry (University Hospitals Coventry and Warwickshire NHS Trust); Dr Andrew Stockdale (former Principal Investigator), Dr Jane Worlding (former Principal Investigator), Professor Stergios Boussios.

Eastbourne District General Hospital (East Sussex Healthcare NHS Trust); Dr Caroline Manetta.

Conquest Hospital (East Sussex Healthcare NHS Trust); Dr Caroline Manetta.

Ninewells Hospital (NHS Tayside); Dr Angela Scott (former Principal Investigator), Dr Mark Baxter.

Western General Hospital (NHS Lothian); Professor Duncan McLaren (former Principal Investigator), Dr Aravindhan Sundaramurthy.

Dorset County Hospital (Dorset County Hospital NHS Foundation Trust); Dr Richard Osborne (former Principal Investigator), Dr Renata Dega (former Principal Investigator), Dr Melanie Harvey.

Darent Valley Hospital (Dartford and Gravesham NHS Trust); Dr Roy Vergis (former Principal Investigator), Professor Seshadri Sriprasad (former Principal Investigator), Dr Patryk Brulinski (former Principal Investigator), Dr Amanda Clarke.

Leicester Royal Infirmary (University Hospitals of Leicester NHS Trust); Dr Guy Faust.

Glan Clwyd Hospital (Betsi Cadwaladr University Health Board); Professor Nicholas Stuart (former Principal Investigator), Dr Carey MacDonald-Smith.

Ysbyty Gwynedd (Betsi Cadwaladr University Health Board); Professor Nicholas Stuart (former Principal Investigator), Dr Carey MacDonald-Smith (former Principal Investigator), Dr Anna Mullard (former Principal Investigator), Dr Pasquale Innominato.

James Cook University Hospital (South Tees Hospitals NHS Foundation Trust); Dr Janine Graham.

Velindre Cancer Centre (Velindre University NHS Trust); Dr Jason Lester (former Principal Investigator), Dr Nachi Palaniappan.

Derriford Hospital (University Hospitals Plymouth NHS Trust); Dr Martin Highley.

Royal Derby Hospital (University Hospitals of Derby and Burton NHS Foundation Trust); Dr Prabir Chakraborti (former Principal Investigator), Dr Prantik Das.

Royal Surrey County Hospital (Royal Surrey NHS Foundation Trust); Dr Agnieszka Michael.

Torbay Hospital (Torbay and South Devon NHS Foundation Trust); Dr Anna Lydon.

Guy’s Hospital (Guy’s and St Thomas’ NHS Foundation Trust); Dr Sarah Rudman.

Preparation of the participant DVD

Dame Leslie Fallowfield and Professor Valerie Jenkins the team at Sussex Health Outcomes Research and Education in Cancer (SHORE-C) at the University of Sussex.