Tislelizumab, a novel anti-PD1 in hepatocellular carcinoma? Quality-of-Life, Non-Inferiority, and Duration of Treatment in RATIONALE-301.

Quality-of-life data should be interpreted with extra caution when censoring rates are high.

In a recent work co-authored with Alyson Haslam and Vinay Prasad, we question the validity of QoL results in trials with high censoring rates. When patients drop out of a clinical trial, one should consider whether their quality of life (QoL) metrics differ from those who remain. If the QoL analysis only includes patients who stay in the trial, it may bias the reported QoL results. This work was prompted by the purported improvement in QoL in the immunotherapy arm (anti-PD1 tislelizumab) of the RATIONALE-301 trial.

RATIONALE-301 was a non-inferiority, open-label, global, phase 3 randomized controlled trial. Patients with untreated advanced hepatocellular carcinoma were randomized between tislelizumab – an anti-PD1 antibody – or sorafenib. Overall survival, with a non-inferiority margin at 1.08, was the primary endpoint. The trial was considered positive (non-inferiority was met), with a median survival of 15.9 months in the tislelizumab group and 14.1 months in the sorafenib group (HR = 0.85, 95.003 % CI, 0.71–1.02).

QoL data with high rates of PFS/OS censoring.

In a follow-up publication, it was concluded that the group receiving the new anti-PD-1 therapy – tislelizumab – had better QoL.

However, we estimated (see below) that significant censoring occurred early in the trial for PFS, with 4.7% of patients censored at the first time point in the tislelizumab group and 19.3% in the sorafenib group. Lower censoring rates were observed for OS, though a similar discrepancy existed.

In other words, patients disproportionately dropped out from the control arm. It is likely that these patients were not randomly selected but rather patients with more treatment options and a willingness to seek alternatives, either off-trial or on-trial.

These patients could potentially have had better QoL data, but their exclusion from the analysis could have resulted in an apparent improvement in the experimental arm.

Non-Inferiority Against Sorafenib?

The idea that patients who left the control arm early may have been in better condition is supported by the fact that sorafenib is a poor control arm. One could argue that sorafenib showed superiority over placebo in the SHARP trial, which is correct.

However, sorafenib's real-world benefit was questioned when Medicare data showed that both treated and untreated patients had a median survival of only 2–3 months. Median survival in real life was half that of patients receiving a placebo in the trial. This is a clear example of the “efficacy-effectiveness gap” where positive results seen in clinical trials with carefully selected patients may erode in broader, real-world populations.

“Consequently, aiming to show non-inferiority as compared to sorafenib, which was comparable to no treatment in real-life, is questionable.”

In an open-label trial like RATIONALE-301, patients were aware of their treatment assignments. Informed patients in good condition may have sought other treatment options than sorafenib, either in or off trials.

Shouldn’t Duration of Sorafenib Treatment be Increasing Over Time?

One issue in RATIONALE-301 was the duration of sorafenib therapy. One would expect that physicians' ability to administer the drug would improve over time. Better toxicity management and increased familiarity with the drug should naturally extend the duration of treatment, or at the very least, not shorten it.

“Here, in the RATIONALE-301, the 2.7 months median duration of sorafenib administration is the shortest across all phase 3 trials in this setting, to our knowledge (Table 1). Investigators, for reasons that are partly understandable, are not incentivized to push drugs when given as control therapy, as well as they are when this is a new therapy. Yet, such imbalance in managing the control drug is likely to influence the final results, by weakening the control group.”

Conclusions

In trials with early, unbalanced censoring affecting PFS or OS, QoL data only reflects those still in the study. QoL results should be interpreted with extra caution, especially in non-inferiority trials where improved QoL might serve as justification for changing practice.

Careful examination of the comparator arm – in this case, sorafenib – is needed, including the duration of treatment. This allows for a better understanding of whether investigators are pushing the drug or, conversely, stopping a weak comparator early.