The earth was flat for a very long time. Then all of a sudden it wasn't. Narratives (like the flat earth) are only true for a finite amount of time, and then become false when they are outdated. Biotech investing, which isn't for the faint of heart, is a fertile field for narratives. The combination of nascent technology and complex science leads to general statements about the prospects of a company and its technology. Detail seemingly isn't important, because at 30,000 feet the narrative holds true for a short amount of time anyway. And in these modern times, a tweet from a self-styled biotech expert on twitter and the narrative can be immediately propagated globally in milliseconds. These narratives can lead to wrong predictions all the time, even from the so-called experts. For those who think this isn't possible, in 2007 a well-known biotech commentator said of ipilimumab (Yervoy), owned by Medarex at the time "I don't believe the current phase 3 studies are going to yield positive data strong enough to warrant the drug's approval". But the drug did get approved, it became a blockbuster, Medarex got bought by Bristol-Myers Squibb (NYSE:BMY), and the rest is history.
The IMPRESS Trial and the Bear Thesis
Enter NewLink Genetics and the bear thesis surrounding its cancer vaccine - Algenpantucel-l. The IMPRESS trial is a randomized phase 3 study of standard adjuvant therapy alone or in combination with Algenpantucel-l in patients who have undergone surgical resection for pancreatic cancer. Pancreatic cancer has limited treatment options and is projected to be the second deadliest malignancy in the U.S. by 2030. An estimated 15,000 patients are eligible for resection and an additional 10,000 patients have locally advanced disease. NLNK is also targeting this latter patient subset in another phase 3 trial (now fully enrolled) for Algenpantucel-l in patients with borderline resectable or locally advanced unresectable pancreatic cancer (PILLAR).
According to the bears, the Algenpantucel-l vaccine, like all other cancer vaccines and all other pancreatic cancer-targeted therapeutics, won't work. In addition, NLNK management's control arm overall survival assumptions in IMPRESS are way off. And even if by some miracle Algenpantucel-l does get approved, no one will use it anyway. Add to this a hard-to-interpret set of phase 2 results and continued insider selling then you have all the makings of a bearish narrative. Some skeptics have even suggested that because the IMPRESS trial has not been stopped at either of the pre-specified interim analysis points, the final analysis is also doomed to fail. Such an argument shows that you don't need to have a basic understanding of statistics to be a skeptic. If that wasn't enough, NewLink Genetics' Iowa location is also apparently a bearish sign, but Harry Stine can tell you that it is actually possible to live in Iowa and be successful at the same time.
The question is, after almost 6 years since enrollment of the first patients in the IMPRESS trial, is the bear case still valid, or is an alternative bullish narrative emerging? Does the detail matter at all or is a simple rehash of an ingrained bearish narrative sufficient to predict the IMPRESS result? The purpose of this article is to critically dissect each part of the IMPRESS trial bear thesis.
Bear Argument # 1 - Cancer vaccines don't work.
The bear case argues that single agent cancer vaccines are not sufficient to cause immune activation, and this view is supported by the failure of previous cancer vaccine trials. However, bears seemed to have failed to recognize that in the IMPRESS trial Algenpantucel-l is being administered concurrently with immune-modulating chemotherapy that is known to take the brakes off the immune system, in a similar way that checkpoint inhibitors do. As we will show below, immune modulation by chemotherapy to bolster the effects of cancer vaccines is well known. Bears seem to have either ignored or underappreciated the significance of this combination in the IMPRESS trial.
In the IMPRESS trial, Standard of Care (SOC) chemotherapy consists of gemcitabine with or without 5-Fluorouracil (5-FU). Both of these chemotherapy agents are known to deplete Myeloid-Derived Suppressor Cells (MDSCs). MDSCs put brakes on the immune system by inhibiting CD8+ and CD4+ T cells and increasing Treg populations. Recent studies have showed that both gemcitabine and 5-FU suppress tumor-associated MDSCs and increase immunological visibility of tumors by increasing expression of tumor associated antigens on their surface. In both preclinical and clinical studies vaccine and chemotherapy combinations have proven to be more effective than the single agents alone (Figure 1). This could go some way to explaining the unusually high (96% vs. 70% SOC) 1-year survival rate observed in the phase 2 trial.
The ability of gemcitabine and 5-FU to transform so-called "cold" or non-immunogenic pancreatic tumors into "hot" or immunogeneic tumors was discussed very recently at the 2016 ASCO GI conference by Dr Vonderheide. This chemotherapy/immunotherapy combination approach has also been validated by big pharma, with Ira Mellman, Genentech's vice president of cancer immunology, suggesting that "chemotherapy may in fact be, to some extent, immunotherapy."
Figure 1: Examples of the immune-modulating effect of gemcitabine/5-FU to bolster the effectiveness of cancer vaccines.
Folfirinox, which is likely to be commonly administered in the salvage setting, also contains 5-FU, meaning immune tolerance can be potentially reduced in the IMPRESS trial even when the patient has recurred. In the phase 2 trial, complete responses were observed in the salvage setting, which is a very noteworthy observation, since relapsed pancreatic cancer is invariably insensitive to chemotherapy. Specifically, in the phase 2 trial three patients that had received Algenpantucel-l therapy subsequently received either 1. FOLFIRINOX, 2. Gemcitabine + Capecitabine + Erlotinib or 3. Gemcitabine + Taxotere. All of these patients experienced durable (6-36 months) complete responses. A common feature was that all three of these salvage regimes contained immune-modulating chemotherapy that reduces immune tolerance by depleting MDSCs.
In addition, a unique characteristic of NLNKs Hyperacute therapy is elevation of eosiniphil levels (70% of the patients in the phase 2 trial showed eosinophilia, with 30% lasting over a year). A recent high-profile article in Nature has shown that eosinophils orchestrate cancer rejection by normalizing tumor vessels and enhancing infiltration of CD8(+) T cells.
Notably, the phase 2 study of the Algenpantucel-L vaccine in combination with chemotherapy and radiation for patients with resected pancreatic cancer yielded encouraging results, particularly for patients who mounted humoral immunological responses against several tumor-associated epitopes.Thus, the multi-faceted immune response from Hyperacute vaccines of antibodies, T cells and eosinophils combined with the immune-modulating effect of chemotherapy cannot be compared to previous different single-agent vaccines and appears to provide an ideal combination for tumor destruction.
Conclusion: Bears seem to have ignored that the immune-modulating chemotherapy used in this trial can bolster the efficacy of the Algenpantucel-l vaccine, including in the salvage setting.
Bear Argument # 2 - NLNK management's control arm overall survival assumptions in IMPRESS are way off.
Bears contend that the delayed rate of events in this trial is a predictor of a negative outcome. Furthermore, despite NewLink Genetics stating that the MOS of the combined arms in the trial is 30 months, bears argue this can be explained by control patients living longer than expected.
In reality, relying too heavily on a single MOS value to build a bullish or bearish case for the outcome of the IMPRESS trial probably isn't a very wise idea. This is because it is the shape of the two curves that matters, rather than just a MOS value. For example, theoretically the two arms could have an identical MOS value but the long-term survival in the trial arm could be double that of the control arm, leading to a successful trial result. This type of scenario would not be unexpected from an immunotherapy trial, where typically the survival benefit is apparent at the tail of the Kaplan-Meier curve.
For this article we used our statistical model to highlight two scenarios based on different curve shapes. The first scenario assumes no survival benefit to the trial arm and is aligned with the bear thesis. The second scenario assumes only a small survival benefit to the trial arm, but only at the tail of the Kaplan-Meier curve. Both cases assume a 30 month control MOS, which is highly conservative and is used to stress-test the bear-case. A more reasonable control MOS is likely to be in the mid-twenties, which in itself would be a notable improvement from the 20 month MOS observed in the RTOG-9704 trial. However, we wanted to ask if the trial can still succeed even if the control MOS (30 months) exceeded everyone's expectations.
Below are some details of how the modeling and analysis was performed;
- Statistical powering assumptions and patient enrollment data points were sourced from company reports and prior management disclosures and guidance.
- The model developed for this study has a single entry for each patient (n=722), who is randomized to either the control or trial arm. The survival of each patient is randomly generated using MOS and standard deviation inputs. Additional levers are added to modify curve shape.
- At each simulated analysis point (222, 333 or 442 events) each individual patient is either recorded as an event or censored. These data are then inputted into statistical modeling software to simulate Kaplan-Meier (KP) curves, log-rank analysis and reduction of death method.
- Assume the 442nd event occurs mid-March of this year.
Scenario 1: (trial arm = control arm = 30 months, 5-year survival = 20%).
Scenario 1 assumes no survival benefit between the trial and control arm and obviously would result in the failure of the trial. However, this scenario did not fit the known and expected rate of events in the trial, with events occurring meaningfully faster than what is actually being observed. For example, under this scenario, 265 events would have occurred at the first interim analysis, rather than 222 events, 395 events would have occurred at the 2nd interim analysis, rather than 333 events and 520 events would have occurred by around March of this year, rather than 442 events. Since the current rate of events, although highly variable, is likely only 5-10 per month, this difference is noteworthy and is far beyond the range of modeling uncertainty. In fact, to fit the actual rate of events, long-term survival must be adjusted significantly higher (3 year survival = 50%, 5 year survival = 35%), meaning the shape of the control curve must now resemble that of an immunotherapy agent if no benefit to the trial arm exists. It is for these reasons that scenario 1 doesn't appear valid. It does however show that the continued delay in reaching 442 events is not due to a longer control MOS as bears would suggest, it is because of higher than expected long-term survival.
Scenario 2: Trial arm benefit with only very moderate curve separation until about 15 months (control arm = 30 months, 5-year survival = 20%).
Scenario 2 assumes almost no curve separation before 15 months, followed by curve flattening in the trial arm relative to the control arm (Figure 2). This very conservative assumption is consistent with the relatively late curve separation observed with successful immunotherapy trials. However, it assumes very little initial benefit for the vaccine in the first 12 months while it is being administered, which was not observed in the phase 2 trial and thus is another built-in conservative assumption. As shown in Table 1, under this scenario the trial continues to the final interim analysis using log-rank analysis and succeeds at the final analysis with either the log-rank or reduction of death risk analysis methods. NLNK clearly made a mistake in the design of the trial with a successful stop at the first interim analysis point of 222 events being virtually impossible.
In this scenario, the significant benefit attributable to the trial arm is only seen at the final analysis due to the length of time patients have been in the trial. Assuming the 442nd event occurs in mid-March of this year, the last enrolled patient would have been in trial for just over 30 months (first censored patient) the middle (361st) patient would have been in trial almost 50 months, while the first enrolled patient would have been in trial some 70 months.
Figure 2: Simulated Kaplan-Meier curve of Scenario 2 observed in the middle of March, 2016 (442 events). Vertical tick marks starting at month 32 are patients in the trial that remain alive at the time of analysis (censored). Analysis includes the on average 1.5 month recovery period between surgery and treatment.
1st Interim: Log-Rank
2nd Interim: Log-Rank
Reduction of Death*
Table 1: Summary of trial over control benefit in scenario 2 using log rank and reduction of death risk method at the 1st, 2nd and final IMPRESS analysis points.*Analysis made at month 40.
A notable feature of Table 1 is the significant benefit to the trial arm when the reduction of death method is applied at the final analysis of 442 events. The FDA has recently made this the analysis method of choice for immunotherapy trials because it is a way of revealing a long-term survival benefit in situations where initial delayed curve separation occurs. This analysis relies on counting the number of long-term survivors at a given point in time in both arms, and as Figure 2 shows, significantly more patients are alive at 442 events in the trial arm than in the control arm. Based on the evidence presented here, a doubling of long-term survival in the IMPRESS trial arm relative to the control arm is not out of the question.
Conclusion: The bear argument that the slower than expected rate of events in the IMPRESS trial is solely attributable to a higher than expected control arm MOS appears invalid. Even under highly conservative control MOS assumptions, long-term survival in the trial arm could be double than what is observed in the control arm, suggesting the possibility of a highly successful outcome for patients in this trial. Bears need a significant improvement in BOTH MOS and long-term survival in the control arm for the trial to fail. Arguing a case for a sudden doubling of long-term survival in surgically resected pancreatic cancer relative to historical controls is very difficult.
Bear Argument # 3 - Even if Algenpantucel-l does get approved, no one will use it anyway.
One analyst has stated that the final readout of the IMPRESS trial to the final analysis point means that the final results will come as a number of other high-profile Phase 3 pancreatic cancer programs read out in 2016, leaving NewLink Genetics in an increasingly crowded market. This author failed to recognize that not one of the competing trials mentioned are actually for surgically resected pancreatic cancer. If positive, Algenpantucel-l is likely to become standard-of-care adjuvant therapy in post-resection pancreatic cancer and achieve peak sales of >$1B.
Conclusion: Bears have this completely wrong. There currently is no FDA approved drug for patients with resected pancreatic cancer. Algenpantucel-l would become the first.
Does this create an opportunity for investors?
As the discussion above highlights, the narrative surrounding the failure of NewLink Genetics' IMPRESS trial appears to be dated and probably false, creating an opportunity for risk-tolerant investors. Currently there is little or no value priced in at these levels for phase 3 IMPRESS data, or anything else vaccine-related. This may appear surprising given the future model of combination immunotherapy which calls for immune stimulation - vaccines/CAR-T, combined with agents to block tolerance and escape - checkpoint antibodies and IDO inhibitors (Figure 3).
NewLink Genetics' core HyperAcute platform is one of the more innovative technologies in the oncology space and their pipeline of earlier-stage vaccine candidates provides additional multiple opportunities on the collaborative front in the future. Meanwhile, NLNK remains eligible to receive significant milestones (>$1B) from its collaboration with Roche (OTCQX:RHHBY) on IDO - a significant portion of which are triggered by early-stage development milestones. In addition, investors haven't placed any value on any of their other products, such as their wholly-owned IDO inhibitor, Indoximod (despite recently reporting impressive results) and their Merck (NYSE:MRK) -partnered Ebola vaccine.
Valuing small/mid-cap biotech is always a perilous exercise, but it is worth noting that analysts INCY)+at+Outperform,+$110+PT/11231067.html" target="_blank">assume that Incyte's (INCY) IDO inhibitor, Epacadostat, will generate adjusted sales of $2.3B in 2023. This suggests that NLNKs current valuation has more to do with being below the radar screen, rather than the strength and breadth of its immunotherapy platform.
Given the low expectations, if the IMPRESS trial is a success, the stock would undergo a massive revaluation to the upside and perceptions would change overnight. For example, the PILLAR trial would be viewed as having a very high chance of success, while Hyperacute lung, melanoma, prostate and renal which, are all in the clinic, would be ripe for lucrative partnerships and combination trials, including with their own IDO inhibitors. Under such a scenario, the company would not sell out to big pharma for less than $10 billion, which is $300+ per share. Although the stock is heavily shorted (5.6 m shares short or about 26% of the float), many funds and institutions are likely buying the stock for the IDO/TDO platform alone. In addition, NLNK has a relatively healthy cash cushion (ca. $180 m) well into 2016, meaning that there is likely no need to issue significant dilutive equity before significant milestone payments occur. While a positive read out in the IMPRESS trial would be a transformative event (see above), a negative readout, which in our opinion would be surprising, could send the stock 20-40 % lower than current trading levels.
Figure 3: View of immuno-chemotherapy in the future with IDO inhibitors, vaccines/CARTs, checkpoint antibodies and radio/chemotherapy in combination therapy.
This article questions the long-standing bearish narrative surrounding NLNKs IMPRESS trial. The bear argument, which hasn't changed since the company went public in late 2011, relies on assumptions which appear to be outdated and incorrect. The earth may now be round.
Additional disclosure: I wrote this article myself, and it expresses my own opinions. I have no business relationship with any company whose stock is mentioned in this article. All model development in this report come from public knowledge through company transcripts, editorials, and peer reviewed journal articles. Although every effort was put into simulating the IMPRESS trial as closely as possible, there are no guarantees that the results presented here are accurate; therefore the information contained is for informational purposes only. It is recommended that readers do their own due diligence before making any investment decisions.
Disclosure: I am/we are long NLNK.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. I have no business relationship with any company whose stock is mentioned in this article.
Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.