Results of ADAGIO trial for rasagiline suggest possible disease-modifying effect

The reported results from the ADAGIO trial are the culmination of a long effort to determine possible disease-modifying (or to use the less clinically favored term: ‘neuroprotective’) effects of rasagiline in people with PD using the delayed-start study design. We've know the basic message that rasagiline might be disease-modifying for a number of months now, ever since TEVA made public its initial findings earlier this year. But seeing the actual trial results published in the open literature is a good step toward verifying the accuracy of this claim.

The results show that 1mg rasagiline met all three endpoints to be considered 'disease-modifying', while 2mg rasagiline met only two of the three endpoints. So do these data prove that rasagiline is the answer to our dreams? Probably not for at least a couple of reasons. First, the authors indicate that the odd difference in effect of 1mg vs. 2mg rasagiline is difficult to explain. Although the post hoc analysis suggests that the 2mg finding may have simply been a false negative result, it is also possible that the 1mg efficacy finding is a false positive one. Second, the ‘delayed start’ design is still, at the end of the day, a trial based on clinical measures of benefit. Whether rasagiline is truly slowing loss of dopamine neurons or just helping the brain to compensate for the loss (something the authors raise as a possibility) is uncertain. Until the day we have biomarkers of disease process and neurodegeneration, answering this ultimate question will simply have to wait.

No doubt the debate on whether rasagiline is disease modifying or not (and what this means for people with PD) will continue for some time. But from a research and scientific perspective, the trial also raises some broader issues worth discussing and I invite the whole PD Online Research community to weigh in:

1. Is this trial a solid indication that the ‘delayed start’ design is the right way to go when testing drugs for possible disease-modifying effects? How should industry view these results as confirmation of a regulatory pathway for getting ‘neuroprotective’ drugs on the market?
2. An apparent assumption in this current version of the delayed start trial design is that progression of disease is linear over time. Does this fit with what is known about natural PD progression? What do the disease and trial modelers out there think? Are there ways to improve the delayed start design statistically to better measure disease progression? (We've funded a bit of work from Dr. Nick Holford in New Zealand and Jay Nutt in Oregon to play around with these ideas using mathematical modeling.)
3. The idea that a drug may act by activating compensatory processes rather than having a direct protective effect is an intriguing one. Are there markers or other relatively straightforward measures that we should consider testing in the context of drug development to help get at this question? From an industry perspective, would it even matter so long as benefit were seen?
4. Although whether a person with PD should go on rasagiline remains a decision that must be made by doctor and patient, the simple fact that many people may start to go on potentially ‘disease-modifying’ therapies soon after diagnosis raises a problematic challenge for trial design. What has been considered to date the ‘ideal’ trial subject—someone soon after diagnosis who is not yet on antiparkinsonian treatment—may soon no longer exist. How should trial designs deal with this possibility?

This is the second reported delayed start trial with an inconclusive result. The first trial, TEMPO, was with the same drug (Parkinson Study Group, 2004a ) and since then some lessons have been  learned about the design and analysis. We have demonstrated that the statistical properties of this kind of delayed start design compared with a withdrawal design are inferior even when the delayed start design has been optimized and efficient parametric models are used to establish performance (Ploeger et al, 2009).

The interpretation of the ADAGIO results is complicated by using a linear approximation to the non-linear time course of the symptomatic effects which approach an asymptote as seen in Figure 3 of the ADAGIO report (Olanow et al., 2009). Linear rates of change were estimated under the assumption that the full symptomatic effect of rasagiline was reached by 12 weeks. Yet inspection of the mean time course for both doses indicates symptomatic effects were continuing to increase up to 24 weeks. The analysis also does not distinguish explicitly between symptomatic and disease modifying effects. This is important since a disease modifying effect could in fact be a symptomatic effect with a slow onset (or vice versa). Or the effects could be combination of a slow onset symptomatic effect (over years) with a disease modifying effect as we have reported for levodopa (Holford et al., 2006).

FDA scientists have used simulation methods to investigate the delayed start design (Bhattaram et al., 2009). In order to have a power greater than 80%, the FDA study required the effect size of the disease modifying effect to be 40% or greater slowing of progression with 600 subjects per treatment arm.  The observed size of the effect on the first primary endpoint in the 1mg rasagiline treatment arm in ADAGIO (37% slowing) was seen with around 300 subjects per arm. Thus ADAGIO was underpowered to detect the apparent disease modifying effect using this analysis approach.

Our mathematical modeling has described the progression of the DATATOP cohort over 8 years (Holford et al., 2006)  and successfully predicted in quantitative terms a priori the outcome of the ELLDOPA study (Chan et al., 2007). ELLDOPA was another trial with an inconclusive result when originally reported (Parkinson Study Group, 2004b). In this case the design was inadequate for the endpoint comparison because it assumed that all symptomatic effect would have disappeared after 2 weeks withdrawal of treatment – a tenuous assumption (Hauser et al., 2002). Our modeling suggested that levodopa was disease modifying in the DATATOP study (Holford et al., 2006) and this has now been confirmed by our modeling of the ELLDOPA data (reported to MJFF at the end of September 2009). Our analysis of this withdrawal design trial allowed the quantification of symptomatic and disease modifying effects without making assumptions that complete washout of symptomatic effects had occurred.

We would like to explore the possibility of collaboration with the ADAGIO  and TEMPO investigators to analyze the data using a model based approach. The models have been pre-specified (Ploeger et al., 2009) which should temper some of the criticism of an ad hoc exploratory analysis. A model based description may offer further insight into the apparent differences between 1 and 2 mg/day in ADAGIO. We offered insight into the interpretation of ELLDOPA using a model based approach (Chan et al., 2007) and may be able to help understand more clearly what the properties of rasagiline really are.

Turning to other questions raised by Dr Fiske, we would prefer to think that neither the industry nor the regulatory perspective on mechanism is the issue. Surely it is the patient perspective that counts? A patient presumably does not really care whether a biomarker change matched the functional improvement. Indeed, if the biomarker does not match the improvement then one must question the biomarker. The ELLDOPA study illustrated this when the biomarker used to measure dopamine uptake showed apparent worsening in patients who were functionally improved. The most likely explanation for this was the effects of levodopa administered as part of the trial on the activity of dopamine transporters. Would differentiation between slowing of neuronal degeneration and induction of compensatory processes be important if it led to improvement in the patient’s condition? Of course, there is an interest from a scientific perspective (not the primary goal of either industry or regulators) to understand the mechanism of any functional improvement. However, lack of understanding of mechanism should not hold up encouraging early use of disease modifying treatments. The mechanism of the analgesic action of acetaminophen has only begun to be understood in recent years yet it has been used as a (largely) safe and effective medicine for decades.

The problem of add-on therapy for standard of care is well understood in other areas of medical practice. It is considered unethical to deny patients effective treatment unless there is some question that would eventually benefit the patient (or society) by deliberately denying standard of care. Rasagiline alone is a weak treatment – the symptomatic difference from placebo is only 2-3 UPDRS units (similar to our description of deprenyl in the DATATOP study (Holford et al., 2006). This magnitude of symptomatic effects may not justify treatment. The disease-modifying effects of rasagiline remain uncertain with the published reports and the physician and patient must make their judgment on incomplete information.  In contrast levodopa 300 mg/d produced an 8 unit benefit in DATATOP (Parkingson Study Group, 2004b). Given what we now know, we would say it is unethical to withhold levodopa in patients who are likely to benefit both in the short and long term.

Future tests of disease modifying treatments should examine the effects of putative disease-modifying agents in addition to standard of care, such as levodopa, for instance. We have shown (DATATOP and ELLDOPA) that levodopa has disease modifying effects alone and even bigger disease modifying effects when combined with deprenyl (DATATOP) (Holford et al., 2006). Future trials of disease modifying agents will need to be done in patients treated with standard of care – preferably with a withdrawal of the experimental disease-modifying drug which our trial simulation suggested would be a more powerful design than delayed start designs (Ploeger et al., 2009). The continued use of standard of care after withdrawal of the experimental treatment will also minimize one of the criticisms of withdrawal designs – that patients will be denied effective treatment at a time when they have progressed to a more serious state.

As a researcher for some 25 years in Parkinson's disease, and now with oversight of clinical trials in wide-range of disciplines, some distance is helpful. First, whether or not the evidence points to a disease modifying effect for rasagiline, it is impressive that Teva would take the risk with supporting this cliical trial. The risk of a definitive negative finding could have been a barrier to the future conduct of large-scale clinical trials by Pharma for the disease-modifying effects of antiparkinsonian drugs.  Second, in points raised by Brian Fiske and Drs. Holford, Nutt and Ploeger, as well as that  of Dr. Schwarzschild, that the clinical benefit of rasagiline is minimal as compared to the standard treatment of levodopa; what would the patient (and the payer) choose? The cost to benefit ratio would seem very low for rasagiline as compared to standard treatments available. If implemented at very early stage Parkinson's would the clinical benefit be worthwhile? Third, why should "we" endeavor to discover disease-modifying drugs for Parkinson's, given the relatively long life most patients have after diagnosis and treatment with standard antiparkinsonian drugs? I don't mean this in an inhumane way, but as a practical question given our healthcare delivery and payment model.  What would the benefit of any disease-modifying indication have to be in order for the routine approval of reimbursement for a prescription charge that is high as compared to a low-cost clinically effective drug? This question is obviously not limited to Parkinson's, treatment for HIV/Aids has a similar issue with modest effects of a vaccine regimen (NIAID, 2009, ThaiVax Study) when antivirals are clinically effective for symptom treatment. What is the investment for identifying a disease-modifying (or preventing) drug going to be? If found, what will it take to get it paid for by insurers? For the patient, caregiver and treating physician the answer seems almost unethical but there is a mismatch here between hope and reality. The debate over what kind of medial care insurability will exist in the next decades is clear on one point, there will be increasing pressure to lower medical coverage costs by reducing reimbursement for drugs. From a purely practical point of view, there will have to a dramatic impact of a drug on disease progression to allow for the high cost of reimbursement. Determining that cost to benefit ratio should not be left solely to CMS, insurers, and pharmaceutical industry. That decision process, unfortunately, impacts treatment decision far more than any research.

The methodology behind the identification of disease modifying effects in the ADAGIO study is good but the implementation is questionable.  Because of intrinsic characteristics of the metrics in the study, the delayed start protocol will show an effect of delayed start in two situations: (1) during the delay, the disease progresses more in one population than the other, and (2) during the delay, subjects in the placebo group suffer lasting consequences of their untreated symptoms.  Looking at, for example, asymmetries in the falls data, I am not sure that the authors explained how they eliminated the latter in their study.

The neuroprotective effect of rasagiline found in the ADAGIO is consistent with our research on 3,4-dihydroxyphenylacetaldehyde (DOPAL), the toxic monoamine oxidase (MAO) metabolite of dopamine (DA).

MAO-B or-A converts dopamine (DA) to 3,4- dihydroxyphenylacetaldehyde (DOPAL). We showed that DOPAL is present in human substantia nigra (SN) in physiologically relevant amounts (Kristal et al., 2001).  DOPAL is highly toxic to DA neurons in vitro (Kristal et al., 2001) and to DA SN neurons in vivo (Burke et al.,. 2008) DOPAL triggers á-synuclein aggregation to oligomers and to large Lewy-body like aggregates in vitro and in vivo (Burke et al., 2008) In addition, DOPAL activates the mitochondrial permeability transition pore (mPTP) leading to apoptotic cell death (Kristal et al., 2001)

Rasagiline, an MAO inhibitor with neuroprotective properties in Parkinson disease (PD) (Parkinson Study Group, 2002), inhibits both the formation of DOPAL and the activation of mPTP. [5] Rasagiline’s inhibition of DOPAL formation and its mechanism of toxicity (Kristal et al., 2001; Youdim et al., 2005) may explain the neuroprotective properties of this pharmaceutical in PD (Parkinson Study Group, 2002) These experimental results suggest that DOPAL is the endogenous toxin causing PD. Therefore, it is a target for neuroprotective therapy in PD.

In addition the suggestion that low dose L-DOPA may be neuroprotective is also consistent with our results. DA, a metabolite of L-DOPA, forms  a Schiff-base adduct with DOPAL called tetrahydropapverline (THP). THP and all other metabolites of DA are non toxic to DA neurons in vitro and in vivo compared to DOPAL (Kristal et al., 2001; Burke et al., 2008).

I am not in a position to discuss the clinical merits of ADAGIO delayed study since I was involved in the discovery and co-development of rasagiline with TEVA. However, the discrepancy between the end points achieved by 1 versus 2 mg rasagiline in the ADAGIO trial may be related to the three pharmacology actions of the drug, namely, being a potent irreversible MAO-B inhibitor, neuroprotective-antiapoptotic and neutorestorative. In our earlier studies with rasagiline and is optical S-isomer, TVP-1022 we noted that these drugs had neurorescue activity in a neuronal culture deficient of neutotrophic factor support (serum free) (Youdim et al, 2005). This activity was concentration dependent and was achieved at doses much lower than those required to inhibit MAO-B, while effective MAO inhibitory concentrations exerted partial or no neurorescue. Both rasagiline and TVP-1022 exhibited a similar potency, despite the fact that TP-1022 is more than a 1,000 fold less active than rasagiline as MAO-B inhibitor (Youdim et al. 2001a and b), indicating that MAO-B inhibition is not a prerequisite for neuroprotection and neurorscue. This property may stem from the propargylamine moiety (nature) of the drugs, since we lately showed that propargyalmine itself, which is a very poor MAO-B inhibitor, has similar dose dependent activity to rasagiline and TVP1022, and may even be more potent as a neuroprotective agent (Weinreb et al 2005 and 2007; Bar Am et al. 2005). All the above drugs induced neurotrophic factors GDNF and BDNF and pro-survival mitochondrial gene/proteins in a dose dependent and bell shaped manner, indicating that higher concentration are not necessarily better, in fact less potent. The possibility that these results can be translated in animals was studied in the MPTP mice model. Animals were treated for 4 days with 24mg/kg MPTP and 4 days later they started receiving rasagiline at a low dose of 0.05mg/kg orally, once a day for the next 14 days. We clearly showed restoration of dopamine neurons in the substantia nigra pars compacta by a mechanism implicating the survival factors Ras, phosphatidylinositol 3 kinase and Akt (also known as protein kinase B), linked to the neurotrophic factor tyrosine kinase receptor (Sagi et al. 2007). The dose of rasagiline was deliberately chosen (0.05mg/kg) not to inhibit brain MAO-B, while doses between 0.25-1 mg/kg, that inhibit the enzyme, did not produce neurorestoration in this model. We are not aware of any other anti Parkinson drug presently in the clinic that has shown nigrostriatal dopaminergic neurorestorative activity in two animal models of Parkinson’s disease. Whether this is translated in the clinical studies can not be determined, since presently in the clinic that has shown nigrostriatal dopaminergic neurorestorative activity in two animal models of Parkinson’s disease. Whether this is translated in the clinical studies can not be determined, since presently there is no possibility of determining neuroprotection directly in human brain. Rasagiline is 10-20 times more active as MAO-B inhibitor in vivo as compared to selegiline (Youdim et al. 2001b), as well activating the transcription of NGD, BDNF and GDNF (Weinreb et al. 2009). Thus, it may have a narrow window of action as a MAO-B inhibitor that over laps its neuroprotective activity by masking the latter action. Indeed more recently we have almost exactly a similar type of response with our propargylamine derived multifunctional anti-Parkinson drug, M30 (Gal et al. 2009). Thus, the ability of rasagiline to restore and activate endogenous neurotrophins and mediators of classical survival pathways in the affected brain areas, may lead to an incrementing clinical improvement, and possible disease modifying activity. Obviously confirmation of ADAGIO is necessary, but if this is to be initiated I suggest that the besides the 1 and 2 mg dosage, 0.5 mg should seriously be considered. Finally there are a number of examples in the literature where high doses of a drug are not necessarily more effective or better and, in certain cases, less effective. For example I refer you to the recent paper by Raj et al. (2009) where they show low dose of propranolol is more effective than higher dosage in tachycardia.

Moussa B.H. Youdim has received research support from Teva Pharmacertical Co (Israel) and has financial interest in rasagiline, which he co-developed with Teva.