Strong preclinical, supporting, and early clinical data make INmune Bio's Phase 2 readout of XPro1595 in Alzheimer's disease a high risk/reward opportunity.
Very good overview of the situation. There are a few details you left out that are important. First, the very clearly communicated reason for the ex-US P2 trial is due to the FDA hold that went much longer than expected. The company was eager to include US sites but for the FDA’s heel-dragging. Second, in case of a successful P2 outcome, the company’s cash position shouldn’t be any concern, as $30 million in warrants will be immediately exercised to provide close to a year of capital. And excellent results will likely lead to non-dilutive licensing deals months later. I do expect them to use their shelf to raise $100 million+, but that shouldn’t affect the stock price. Finally, given the qualifications of the neuroscientists handling the XPro data and trial design, the only reason to be skeptical of the P1b results not being fully disclosed would be that management is hiding something. BUT, they own a very large percentage of the company and have bought more shares at several recent financings, so they are competent enough to design a trial that they truly believe will succeed and have invested the money to support that. Unless they’re delusional, we have little reason to question the “missing” data. So these seemingly small details you left out provide just more evidence for why the stock is heavily shorted and the bet is so asymmetric.
Yes the warrants are definitely an important point about cash position that I failed to include, someone else alerted me to this as well. Going to make a quick edit to include it.
Regarding the US vs. ex-US trial sites: yes, the clinical hold was the reason for them having to go exclusively ex-US. My point is that regardless of why it is the case, the lack of US sites might be an incrementally negative thing in the eyes of some investors.
And similarly on the fact that INmune never disclosed any of the numerous cognitive biomarkers that were measured in the Phase 1b: I agree if you take a step back it is likely not a big deal, though just pointing out that selective disclosure from a company that otherwise has been very transparent and forthcoming is a bit odd.
Fair enough. Given that the ex-US sites are UK, Canada, Australia, and EU (and not developing nations that are less credible), I don't see this as an issue, especially if results are clearly good. Remember also that this is a P2, so it's understood that the standards aren't as high as the P3 would be (which would be global). And the selective disclosure may seem odd, but again, if the company were deliberately hiding something, they wouldn't have been buying more stock. So you could argue that the lack of transparency there suggests added risk going into the P2 readout, but once the data comes out, the P1b and any transparency issues become irrelevant to the stock price.
There is some recency bias going on here. Three decades old AChEIs (donepezil etc) had effect size in cohen's d 0.2 to 0.25 against placebo in mild AD with ADAS-cog that has close to same properties as CDR-SB, cohen's d are interchangeable between them. That is ~ 2x larger than lecanemab against placebo in 6 months.
What is the possibility that XPro1595 yields equivalent or better results against placebo than donepezil? XPro1595 has shown large objective benefits from the baseline in dMRI (FW, RD, AFD), NfL (-84%), p-tau217 (-45%) which are three best markers longitudinally within-subjects to correlate with cognition decline (correlations are in range of 0.3 to 0.6 with cognition). One can find same type of correlations from other diseases, but has to look for CRP in rheumatoid arthritis or ProBNP in heart failure. They are that tight correlations.
EEG change in alpha waves (+0.5Hz) and power (+18%) was 2-3x larger in 4 weeks than what is seen with AChEIs at 6-12 months vs placebo. Such big alpha wave increase is typically seen only after spontaneous recovery of mild traumatic brain injury in few months (9.0->10Hz).
Diffusion MRI changes were a lot larger than for any MS drug approved for RRMS even though in MS they follow different locations than in AD and results in dMRI are not truly interchangeable between any two diseases. With XPro1595 the dMRI free water change -45% at 12 months imply that moderate AD patient has gone from avg mod AD --> healthy aging level of FW (there is naturally little FW along the white matter tracts) and this has occurred with approx the same magnitude in all subjects as the error bars are so narrow for N of 3 to 5.
I encourage you to find out how free water increase in white matter tracts leads to decrease in electrical conduction speed, how this leads to glial cells sensing that "you dont need these synapses anymore" and then they phagocytose them away from you due to activity-dependent modulation of synapses by microglia, astrocyte and neuronal pathways.
Your statistical test calculation is off the mark as they will NOT analyze the longitudinal data with yours truly Student's t-test. Every amyloid antibody phase 3 trial has used some version of mixed-effects model for repeated measurements and covariate adjustment and I think one can be sure that they have used a biostatistician for power calc and the statistical analysis plan. MMRM can adjust for the fact that there is large within-subject correlation of the cognition assessment values during 0,3,6 months of measurement timepoints and some baseline variables like APOE4+ and high CRP should predict faster than avg decline and can be used as covariates reducing the overall variation between-subjects. Depending on how high within-subject correlation (r ~0.75-0.85) one wants to assume, the smaller effect size is needed. I anticipate cohen's d = 0.25-0.30 versus the placebo group decline (cohen's d ~ 0.5 as in lecanemab trial) to be enough for p <0.05 even when bypassing the musings on lower variance inflammed subjects have had in some small ADNI cohort.
In CDR-SB and such cognition scales two easily eliminated sources of variation are the change in rater (neuropsychologist or similar rater) and changes in informant (e.g. spouse or child) as explained in the EMACC webinar INMB has published. With the help of J. Jaeger, they have likely tried to minimize all the noise in cognition measurement scales to enable detect as small differences as possible with the chosen N.
Very much appreciate the insight on the statistical powering side, where my analysis is definitely more high-level and thus leaves a lot of room for refinement. Thanks for the explanation. Your expectation of 0.25-0.30 cohen's d being enough for statistical significance lines up nicely with my analysis, where I basically said XPro1595 needs to do about 0.30 points better than placebo on CDR-SB, which with a 1.0 standard deviation assumption is right in line. (Unless I am missing something again haha).
My estimate of cohen's d of 0.25 to 0.30 is in the conservative side: small amount drop-outs, no decrease in variation compared to other trial data, no increase in within-subject correlations to other available data. No meaningful decrease in variation from prognostic factor adjustment. Little more aggressive estimates and it is cohen's d 0.20-0.25 range.
P2 trial baseline data for CDR-SB in poster showed SD of 1.5. CDR-SB values increases in progression and often SD increases a bit with higher absolute values, let assume between-subject SD 1.8 at 6 months. Then if placebo group avg increases 0.5 cohen's d (SD of baseline most often used): 0.5 * 1.5 = +0.75 points. In aggressive estimates 0.20 cohen's d would be 0.2 * 1.5 = 0.3 points better than placebo (0.75-0.30) = +0.45 points at 6 months.
Very good overview of the situation. There are a few details you left out that are important. First, the very clearly communicated reason for the ex-US P2 trial is due to the FDA hold that went much longer than expected. The company was eager to include US sites but for the FDA’s heel-dragging. Second, in case of a successful P2 outcome, the company’s cash position shouldn’t be any concern, as $30 million in warrants will be immediately exercised to provide close to a year of capital. And excellent results will likely lead to non-dilutive licensing deals months later. I do expect them to use their shelf to raise $100 million+, but that shouldn’t affect the stock price. Finally, given the qualifications of the neuroscientists handling the XPro data and trial design, the only reason to be skeptical of the P1b results not being fully disclosed would be that management is hiding something. BUT, they own a very large percentage of the company and have bought more shares at several recent financings, so they are competent enough to design a trial that they truly believe will succeed and have invested the money to support that. Unless they’re delusional, we have little reason to question the “missing” data. So these seemingly small details you left out provide just more evidence for why the stock is heavily shorted and the bet is so asymmetric.
Yes the warrants are definitely an important point about cash position that I failed to include, someone else alerted me to this as well. Going to make a quick edit to include it.
Regarding the US vs. ex-US trial sites: yes, the clinical hold was the reason for them having to go exclusively ex-US. My point is that regardless of why it is the case, the lack of US sites might be an incrementally negative thing in the eyes of some investors.
And similarly on the fact that INmune never disclosed any of the numerous cognitive biomarkers that were measured in the Phase 1b: I agree if you take a step back it is likely not a big deal, though just pointing out that selective disclosure from a company that otherwise has been very transparent and forthcoming is a bit odd.
Fair enough. Given that the ex-US sites are UK, Canada, Australia, and EU (and not developing nations that are less credible), I don't see this as an issue, especially if results are clearly good. Remember also that this is a P2, so it's understood that the standards aren't as high as the P3 would be (which would be global). And the selective disclosure may seem odd, but again, if the company were deliberately hiding something, they wouldn't have been buying more stock. So you could argue that the lack of transparency there suggests added risk going into the P2 readout, but once the data comes out, the P1b and any transparency issues become irrelevant to the stock price.
Yes, agree with all of that.
There is some recency bias going on here. Three decades old AChEIs (donepezil etc) had effect size in cohen's d 0.2 to 0.25 against placebo in mild AD with ADAS-cog that has close to same properties as CDR-SB, cohen's d are interchangeable between them. That is ~ 2x larger than lecanemab against placebo in 6 months.
What is the possibility that XPro1595 yields equivalent or better results against placebo than donepezil? XPro1595 has shown large objective benefits from the baseline in dMRI (FW, RD, AFD), NfL (-84%), p-tau217 (-45%) which are three best markers longitudinally within-subjects to correlate with cognition decline (correlations are in range of 0.3 to 0.6 with cognition). One can find same type of correlations from other diseases, but has to look for CRP in rheumatoid arthritis or ProBNP in heart failure. They are that tight correlations.
EEG change in alpha waves (+0.5Hz) and power (+18%) was 2-3x larger in 4 weeks than what is seen with AChEIs at 6-12 months vs placebo. Such big alpha wave increase is typically seen only after spontaneous recovery of mild traumatic brain injury in few months (9.0->10Hz).
Diffusion MRI changes were a lot larger than for any MS drug approved for RRMS even though in MS they follow different locations than in AD and results in dMRI are not truly interchangeable between any two diseases. With XPro1595 the dMRI free water change -45% at 12 months imply that moderate AD patient has gone from avg mod AD --> healthy aging level of FW (there is naturally little FW along the white matter tracts) and this has occurred with approx the same magnitude in all subjects as the error bars are so narrow for N of 3 to 5.
I encourage you to find out how free water increase in white matter tracts leads to decrease in electrical conduction speed, how this leads to glial cells sensing that "you dont need these synapses anymore" and then they phagocytose them away from you due to activity-dependent modulation of synapses by microglia, astrocyte and neuronal pathways.
Your statistical test calculation is off the mark as they will NOT analyze the longitudinal data with yours truly Student's t-test. Every amyloid antibody phase 3 trial has used some version of mixed-effects model for repeated measurements and covariate adjustment and I think one can be sure that they have used a biostatistician for power calc and the statistical analysis plan. MMRM can adjust for the fact that there is large within-subject correlation of the cognition assessment values during 0,3,6 months of measurement timepoints and some baseline variables like APOE4+ and high CRP should predict faster than avg decline and can be used as covariates reducing the overall variation between-subjects. Depending on how high within-subject correlation (r ~0.75-0.85) one wants to assume, the smaller effect size is needed. I anticipate cohen's d = 0.25-0.30 versus the placebo group decline (cohen's d ~ 0.5 as in lecanemab trial) to be enough for p <0.05 even when bypassing the musings on lower variance inflammed subjects have had in some small ADNI cohort.
In CDR-SB and such cognition scales two easily eliminated sources of variation are the change in rater (neuropsychologist or similar rater) and changes in informant (e.g. spouse or child) as explained in the EMACC webinar INMB has published. With the help of J. Jaeger, they have likely tried to minimize all the noise in cognition measurement scales to enable detect as small differences as possible with the chosen N.
Very much appreciate the insight on the statistical powering side, where my analysis is definitely more high-level and thus leaves a lot of room for refinement. Thanks for the explanation. Your expectation of 0.25-0.30 cohen's d being enough for statistical significance lines up nicely with my analysis, where I basically said XPro1595 needs to do about 0.30 points better than placebo on CDR-SB, which with a 1.0 standard deviation assumption is right in line. (Unless I am missing something again haha).
My estimate of cohen's d of 0.25 to 0.30 is in the conservative side: small amount drop-outs, no decrease in variation compared to other trial data, no increase in within-subject correlations to other available data. No meaningful decrease in variation from prognostic factor adjustment. Little more aggressive estimates and it is cohen's d 0.20-0.25 range.
P2 trial baseline data for CDR-SB in poster showed SD of 1.5. CDR-SB values increases in progression and often SD increases a bit with higher absolute values, let assume between-subject SD 1.8 at 6 months. Then if placebo group avg increases 0.5 cohen's d (SD of baseline most often used): 0.5 * 1.5 = +0.75 points. In aggressive estimates 0.20 cohen's d would be 0.2 * 1.5 = 0.3 points better than placebo (0.75-0.30) = +0.45 points at 6 months.