Clovis (CLVS) lost 75% of its market cap last week after disclosing a disappointingly low response rate for rociletinib in T790M+ NSCLC patients. Updated response rates were 28%-34%, dramatically lower than the 54-60% response rate reported at ASCO 2015. According to the company, the dramatic difference stems from analyzing the same data set based on more stringent criteria (confirmed response rate).
When a patient achieves a response (30% reduction in tumor burden), the response has to be confirmed by a subsequent scan (~6 weeks later). If the subsequent scan shows progression then the response is not considered a confirmed response. As companies report results of early stage in parallel to recruiting new patients, it is common to see unconfirmed responses as part of the efficacy evaluation.
From a clinical perspective, unconfirmed responses are far less attractive as they represent a transient benefit that is followed by disease progression after less than two months. When a drug leads to an unconfirmed response rate of 50% but the confirmed response rate is 30%, this means that almost of half of responders could not maintain their response for 2 months. This potentially impacts the overall clinical benefit including PFS and overall survival.
As data mature (typically in the context of P2 or P3) there is enough follow up to distinguish between confirmed and unconfirmed responses, making the latter irrelevant for assessing response rate. This is particularly important when companies file for accelerated approval based on response rates. In other words, at the end of the day an unconfirmed response rate is not considered a meaningful (let alone approvable) endpoint. This is why companies and investigators often distinguish between confirmed and unconfirmed responses. Below are two examples from Array Biopharma (ARRY) and Immunogen (IMGN) (I picked them simply because I am going to discuss them below, red arrows are my addition).
I was very surprised to learn that Clovis has been consistently using unconfirmed response rates in its data analysis without ever telling investors a significant portion of these responses were transient. While the company presents the delta between confirmed and unconfirmed responses as an unexpected new finding, looking retrospectively at the results calls into question this claim.
Clovis started presenting data for therapeutic doses of rociletinib’s new formulation at ASCO (June) 2014, starting from a cohort of 40 patients with T790M+ NSCLC and a response rate of 58%. At the time, many patients had limited follow up so technically it was impossible to confirm most responses. The spider plot below clearly demonstrates this.
Five months later at the EORTC meeting (Nov 2014), Clovis presented an update which included 27 efficacy evaluable patients treated at the doses the company decided to take forward (most responses had already been reported at ASCO 2014). Response rate was 67% and although many patients had significant follow up, there was still no breakdown of confirmed and unconfirmed responses. Based on this and the fact that patient recruitment picked up significantly in H2 2014, it is safe to assume that the company started 2015 with a ~60-patient database with sufficient follow up to get a reliable confirmed response rate.
At ASCO 2015, Clovis had a 243-patient efficacy database. Response rate was 53% and again, despite the fact many patients had significant follow up to distinguish between confirmed and unconfirmed responses, Clovis did not disclose anything about this issue.
Before last week’s announcement, Clovis never reported efficacy based on confirmed responses even though it is a common practice and despite the fact it had plenty of opportunities to do so. In addition, AstraZeneca started to report efficacy based on confirmed responses more than one year ago (see link). Clovis presented its data, sometimes back to back with Tagrisso (AZD9291), but still used unconfirmed overall response rate.
Using a confirmed response rate, rociletinib’s response rate of ~30% is inferior to that of Tagrisso, which has a confirmed response rate of 51%-59% in T790M+ NSCLC. Rociletinib appears to have a similar PFS to Tagrisso and the difference may be explained by a different patient population pursued by the two companies but for now Tagrisso looks like the undisputed winner.
As an investor, I find the behavior of Clovis’ management disturbing because I fail to see how the issue of response confirmation could be overlooked. Not only is distinguishing between confirmed and unconfirmed responses a common practice, Clovis knew AstraZeneca used confirmed response rate and did not bother to do so as well, thereby leading investors to assume the drugs are comparable in terms of response rate. To clarify, it is OK when an initial signal is not corroborated by a larger data set. My concern is that Clovis knew that responses are not durable in many of the patients but still covered this up. Clovis did not lie to investors, but it certainly did not provide the full picture.
Array’s recently announced partnership with Pierre Fabre for binimetinib and encorafenib was not well received by the market due to the limited size and scope of the deal. The market (myself included) expected a rich global deal with a leading pharma (I actually thought an acquisition makes more sense). Instead, Array did a limited deal (covering Europe and other markets) and retained full rights in US and Japan.
Financially, the deal itself does not look that bad at second glance as Array keeps US and Japan and retains a 20+% royalty rate in Pierre Fabre’s territories while offloading 40% of global development costs. Still, the fact Array could not land a richer and broader deal is disappointing because it implies there was limited interest from large biopharmas, who could easily acquire Array for under $1B.
Admittedly, MEK inhibitors haven’t lived up to expectations and BRAF+ melanoma is the only setting in which a clear benefit has been observed. Results in other indications were either negative (uveal melanoma) or not exciting enough even as combination with BRAF inhibitors (BRAF+ colon cancer). Nevertheless, there are still plenty of settings in which MEK inhibitors are being evaluated, which could translate to a significant commercial opportunity.
Extrapolating from the melanoma experience, MEK inhibitors may be used to enhance activity of BRAF inhibitors. Beyond melanoma, BRAF inhibitors may be applicable in niche indications with BRAF mutations. Two recent publications (see link #1 and link #2) evaluated Zelboraf in BRAF+ tumors (excluding melanoma) and reported responses across various tumor types including a 42% response rate in BRAF+ NSCLC and a 96-100% response rate in a rare leukemia. Based on experience in melanoma and colon cancer, adding a MEK inhibitor may augment efficacy and prolong treatment duration.
Larger indications such as RAS-mutated tumors may be relevant for other combination regimens with MEK inhibitors. These include KRAS+ NSCLC where AstraZeneca is evaluating Array’s selumetinib with chemotherapy. Companies are also exploring this indication, which represents a multi-billion opportunity, with PD-1/PD-L1 antibodies combined with MEK inhibitors.
There are also small indications in which MEK inhibitors may have single agent activity, with neurofibromatosis as a recent example (see press release). Array will soon report P3 results for binimetinib as single agent in NRAS+ melanoma. The study was initiated by Novartis at the time based on a single arm study that demonstrated modest efficacy (~15% response rate, PFS of 3.6 months) and since the P3 trial compares binimetinib with chemotherapy, I view the readout as a high risk event. Combination with a CDK4/6 inhibitor appears more promising but toxicity is an issue.
I plan to hold Array going into data readouts in 2016. Failure in NRAS+ melanoma could present an attractive entry point ahead of the 2016 readouts (especially Astra’s SELECT-1 in KRAS+ NSCLC).
Earlier this month, ImmunoGen got a step closer to breakthrough therapy designation with updated results for IMGN853 (mirvetuximab soravtansine) in ovarian cancer. Results provided only limited additional information in terms of number of patients and follow up but provided a strong indication regarding the company’s ability to select patients more likely to respond to the drug.
In the overall population, response rate was 50% (10/20 patients) vs. 53% (9/17) reported at ASCO but stratifying patients according to FRα (mirvetuximab’s target) expression demonstrated a striking correlation between FRα expression level and responses. Of the 10 patients with high expression, 7 (70%) had a confirmed response including two ongoing complete responses. Two additional patients had an unconfirmed response, bringing the unconfirmed response rate to 90%. In contrast, patients with medium and low expression had limited and no benefit, respectively (see figure below).
Response duration in high expressors was encouraging as of the 7 responders, 5 stayed on treatment for 6 months or longer, including 4 patients who were still ongoing at the time of presentation. One patient received mirvetuximab for a year.
If corroborated by future results, the high response rate in high expressors is a clear positive for Immunogen because it increases likelihood of accelerated approval using a small P2. Obviously, the addressable market is smaller but still represents a $1B global opportunity (conservatively assuming 15,000 eligible patients). At ASCO 2015, the company will report data for 20 additional patients, which will be crucial for corroborating activity in FRα-high tumors. A confirmed response rate of 50% with a durability of 5-6 months in these heavily-pretreated patients will validate ImmunoGen’s clinical strategy and make mirvetuximab one of the most promising agents for ovarian cancer.
We are selling Clovis following the recent update and regulatory setback as well as Ocata (OCAT) following its acquisition by Astellas.