Lessons from Hong Kong NPC Study Group Trials

Case Study 4Lessons from the Hong Kong Nasopharyngeal Carcinoma Study Group TrialsPrincipal Investigator:Dr. Anne Lee and the International Nasopharyngeal Carcinoma Study Group

Materials NPC-9901 publications: Lee, et al. JNCI 102 (2010), pp. 1-11 and Lee, et al. To appear in Cancer (2018). NPC-0501 publication: Lee, et al. Cancer, Publication online on 12/19/2014 Protocols for NPC-9901, NPC-9902, and NPC-0501. Key Words: Endpoints, Randomization, Stratification, Imbalanced Randomization, Patient Population. Rick Chappell, Ph.D. Professor, Department of Biostatistics and Medical Informatics University of Wisconsin Medical School Stat 542 – Spring 2018

The International Nasopharyngeal Carcinoma Study Group • (1998 - present), a Hong Kong based cancer group conducting randomized clinical trials on NPC originally organized by Dr. Anne Lee, currently Clinical Professor and Head, Department of Clinical Oncology, the University of Hong Kong. • Organized three trials • NPC-9901: Radiotherapy (RT) vs. chemoradiotherapy (CRT) • NPC-9902: 2 x 2 factorial of RT vs. CRT and accelerated vs. conventional fractionation • NPC-0501: trial of chemotherapy timing, concurrent-adjuvant vs. induction-concurrent

Outline • Medical Background and Justifying the Patient Population – NPC-9901 • Choice of Endpoint and Interpretation – NPC-9901 • Preserving the Randomization – NPC-0501

NPC-9901 final publication (Cancer, 2017 – initial publication in JNCI, 2010): A multicentre, phase 3, randomized trial on concurrent chemoradiotherapy plus adjuvant chemotherapy versus radiotherapy alone in patients with regionally advanced nasopharyngeal carcinoma: 10- year outcome on efficacy and safety. Anne W.M. Lee, Stewart Y. Tung, W.T. Ng, Victor Lee, Roger K.C. Ngan, Horace C.W. Choi, Lucy L.K. Chan, Lillian L. Siu, Alice W.Y. Ng, T.W. Leung, Harry H.Y. Yiu, Brian O’Sullivan, and Rick Chappell.

NPC-9901 final publication (Cancer, 2017 – initial publication in JNCI, 2010): From the abstract: • Purpose ... compared the results achieved by concurrent chemoradiotherapy (CRT) versus radiotherapy (RT) alone for nasopharyngeal carcinoma (NPC) with advanced nodal disease. • Patients and Methods Patients with nonkeratinizing/undifferentiated NPC staged T1-4N2-3M0 were randomized to CRT or RT. ...

I. Medical Background What was the evidence in 1990 for chemotherapy in the treatment of NPC, and what was the motivation for new studies in China?

Undifferentiated Nasopharyngeal Carcinoma: the most common cancer in Guangzhou – 80 times more common here than in Beijing. Also quite common in other parts of Southeast Asia

This study needed to be done in Southeast Asia • Although Laos, Vietnam, and Cambodia could greatly benefit from such a trial (chemotherapy was prohibitively expensive there), they didn’t have the resources to undertake it. • Also without expertly designed RT, the results might be doubted (as in IG-0099) • Singapore could and did have the resources to do a small trial. • Only China had - and has! - the resources to do a large study.

II. Choice of Endpoint –NPC-9901 • Should we use progression-free survival (PFS; time to local, regional, or metastatic recurrence or death due to any cause) or overall survival as the primary endpoint? • PFS is quicker but OS is more relevant. • We could use both but then we would need a correction for multiple comparisons, such as Bonferroni (multiplying the p-value by 2).

We decided using both outcomes in sequence. • In the original analysis done at the study’s conclusion we used PFS as the primary outcome. • In the 5-year and later followups we used OS. • Because no two outcomes were considered primary at the same time, we didn’t need to correct for multiple comparisons. • Do OS and PFS tell us the same thing?

Is progression a surrogate outcome for overall survival? That is, does the effect of CRT on PFS reflect an effect of CRT on OS? ? CRT vs. RT OS PFS

Is progression a surrogate outcome for overall survival? That is, does the effect of CRT on PFS reflect an effect of CRT on OS? ? CRT vs. RT OS PFS Medical translation: Are progressions in the control arm (mostly) salvageable with CT, resulting in OS rates about the same as newly diagnosed patients?

Overall survival curves with 10 year followup (2017 paper):

Is progression a surrogate outcome for overall survival? That is, does the effect of CRT on PFS reflect an effect of CRT on OS? ? CRT vs. RT PFS OS Medical translation: Are progressions (mostly) salvageable with CT, resulting in OS rates about the same as newly diagnosed patients? Answer: Mostly, but apparently not completely. There may be a residual benefit of CT on OS.

From the abstract of the 2010 JNCI paper: Results “... 348 eligible patients were randomly assigned; the median follow-up was 2.3 years. The two arms were well-balanced in all prognostic factors and RT parameters. The CRT arm achieved significantly higher failure-free survival (72% v 62% at 3-year, P .027), mostly as a result of an improvement in locoregional control (92% v 82%, P .005). However, distant control did not improve significantly (76% v 73%, P .47), and the overall survival rates were almost identical (78% v 78%, P .97). In addition, the CRT arm had significantly more acute toxicities (84% v 53%, P .001) and late toxicities (28% v 13% at 3-year, P .024).”

From the abstract of the 2010 JNCI paper: Conclusion Preliminary results confirmed that CRT could significantly improve tumor control, particularly at locoregional sites. However, there was significant increase in the risk of toxicities and no early gain in overall survival. Longer follow-up is needed to confirm the ultimate therapeutic ratio.

From the abstract of the 2018 Cancer paper: Results “The early findings of significant improvement in tumor control were maintained: the CRT group achieved significantly higher 10-year overall failure-free rate (62% vs 50%, P=0.01) and progression-free survival (56% vs 42%, P=0.006) due to superior locoregional-control (87% vs 74%, P=0.003), whilst the impact on distant-control remained insignificant (68% vs 65%, P=0.24). The initial differences in toxicity profile subsequently narrowed to become insignificant ... The OS rate in the CRT group consequently improved to reach significant superiority at 10-year (62% vs 49%, P=0.047).”

From the abstract of the 2018 Cancer paper: Conclusion“Long-term results confirmed that adding concurrent cisplatin plus adjuvant cisplatin-fluorouracil to conventional-fractionated RT could significantly improve OS without excessive late toxicities for patients with regionally advanced NPC. However, more potent therapy is needed for improving distant control, especially for patients with stage IVA-B disease.” [Continuing confusion on why chemotherapy should fail to prevent metastases but help local control, despite radiation being the {in the usual theory} local therapy.]

From the Abstract: “BACKGROUND: A current recommendation for locoregionally advanced NPC is conventional fractionated radiotherapy with concurrent cisplatin plus adjuvant cisplatin and fluorouracil (PF). In this randomized trial, the authors evaluated the potential therapeutic benefit from changing to an induction-concurrent chemotherapy sequence, replacing fluorouracil with oral capecitabine, ... RESULTS: In total, 803 patients were accrued, ...”

Preserving the randomization • In designing NPC-0501, a trial of chemotherapy timing, concurrent-adjuvant vs. induction-concurrent, there was a problem with logistics. • One of the drugs, capecitabine, was oral and could be administered on an outpatient basis. • The other drug, 5-FU, required overnight stay • One of the largest participating centers, Queen Elizabeth Hospital, was facing a bed shortage if they participated. • “Could the patients at QE be randomized to capecitabine, and the other patients be randomized to 5-FU?”

No.

No. This is not randomization. In Hong Kong and around the world, patients at different hospitals are different. • However, we could randomize 2:1 in favor of capecitabine at QE and against it elsewhere. • As long as we stratify the analysis based on hospital, this would not be biased. • However, the imbalance would cause a loss in efficiency and inflate the sample size. • QE ended up finding the beds for the trial so randomization was balanced.

Above, I wrote: • “However, we could randomize 2:1 in favor of capecitabine at QE and against it elsewhere. • As long as we stratify the analysis based on hospital, this would not be biased.” Why is stratified (adjusted) analysis important with imbalanced but not with balanced randomization?

Above, I wrote: • “However, we could randomize 2:1 in favor of capecitabine at QE and against it elsewhere. • As long as we stratify the analysis based on hospital, this would not be biased.” Why is stratified (adjusted) analysis important with imbalanced but not with balanced randomization? Because the imbalance can confound the relation between treatment and outcome. An artificial example:

Suppose in a two-arm (“Treatment vs. Control”) comparison, • QE Hospital patients do 20% better than Other patients. • The T –C Treatment effect is 10% for both hospitals. Then a possible scenario is Other Control Patients: 50% Other Treatment Patients: 60% QE Control Patients: 70% QE Treatment Patients: 80%.

Also suppose that there are 90 patients randomized 45:45 in QE and 90 patients 45:45 in Other (balanced randomization). Then the effect for Other is 60% - 50% = 10%. the effect for QE is 80% - 70% = 10%. The stratified analysis combines these to get an effect of 10%. The grouped (unstratified) effect is the same, (80% + 60%)/2 – (70% + 50%)/2 = 10%.

Also suppose that there are 90 patients randomized 45:45 in QE and 90 patients 45:45 in Other (balanced randomization). Then the effect for Other is 60% - 50% = 10%. the effect for QE is 80% - 70% = 10%. The stratified analysis combines these to get an effect of 10%. The grouped (unstratified) effect is the same, (80% + 60%)/2 – (70% + 50%)/2 = 10%. Wow.

But now suppose we have an imbalanced randomization: 90 patients randomized to T:C as 30:60 in QE and 60:30 in Other. Then the effect for Other is 60% - 50% = 10%. the effect for QE is 80% - 70% = 10%. Just as before, the stratified analysis combines these to get an effect of 10%. The grouped (unstratified) effect is now different because of confounding of treatment with hospital: (80% x 1/3 + 60% x 2/3) – (70% x 2/3 + 50% x 1/3) = 3.3%.

Lessons from Hong Kong NPC Study Group Trials