Commentary on “Towards Enhancing Drug Development Methodology to Treat Cognitive Impairment Associated With Schizophrenia (CIAS) and Other Neuropsychiatric Conditions: Insights from Two Decades of Clinical Trials”

The paper by Horan et al¹ in this issue of Schizophrenia Bulletin considers industry experience in conducting clinical trials targeting cognitive impairment associated with schizophrenia (CIAS) over the past 2 decades and also discusses various scientific developments occurring since the Measurement and Treatment Research to Improve Cognition in Schizophrenia (MATRICS) initiative that are relevant to the study of CIAS. The MATRICS initiative resulted in a proposed set of guidelines in the design and conduct of CIAS trials,² and Horan et al have proposed modifications to these guidelines. Their goal is to enhance drug development in CIAS in order to increase the prospects for approvals of drugs for this aspect of schizophrenia, something that has not yet been achieved, despite 2 decades of CIAS studies.

Since I was the Food and Drug Administration (FDA) representative in the MATRICS initiative in the early 2000s, the editor has asked me to comment on this paper and its proposals for changes to the guidelines. I’m happy to do that but I need to be clear that I left the FDA in 2012 so, of course, I do not represent the FDA, and my comments are entirely my own. Even the original paper resulting from the MATRICS initiative acknowledged that those original guidelines were a “starting point,” and there was a clear expectation that optimal approaches would evolve over time. Some of the proposed Horan et al modifications are quite straightforward and reasonable, in my view, and therefore easy to comment on. Others are more complicated and require a bit of discussion.

I have one other general comment before I discuss the proposed modifications to the guidelines. Sometimes companies complain that they can’t make changes to the way they design and conduct trials because the FDA is too rigid and will never accept their proposed changes, and sometimes that is true. Often, however, there is another reason. Some companies seem afraid to ask the FDA to consider a change from the usual approach, even though the FDA might be quite willing to accept a well-argued and supported alternative approach. Sometimes the conservatism resides within the company and not within the FDA. If, however, the FDA is rigid on a particular issue without good reason, companies can seek recourse at a higher level. Review divisions do have the authority to create policy, however, they are not independent and they do have to justify their positions to higher levels of management. It is possible to challenge the FDA and to prevail.

Optimal Approach to Assessing Cognitive Impairment in Schizophrenia

One of the accomplishments of the MATRICS initiative was to develop a battery of cognitive tests to assess cognitive impairment in schizophrenia, that is, the MATRICS Consensus Cognitive Battery (MCCB). Horan et al nicely summarize the practical difficulties with using the MCCB, and they also note that a single cognitive impairment factor accounts for most of the variance for the MCCB, as for most cognitive batteries. I agree that alternative approaches to assessing cognitive impairment in schizophrenia should be considered, and it is my impression that the FDA, including the clinical outcomes assessment group, has been and would be receptive to such suggestions, assuming they come in the context of good arguments and supportive data. But I also agree that it would be helpful to have more guidance from the FDA on exactly what is needed to support such alternatives.

Optimal Duration for CIAS Trials

Horan et al suggest that the MATRICS initiative recommendation for a trial duration of at least 6 months was derived from a model for targeting cognitive impairment in dementia, where the goal is to prevent or slow decline in cognition in a deteriorating condition, and as such might be expected to take many months. They also suggest that the recommendation for a long trial was partly related to the fact that a cholinergic alpha-7 agonist in consideration at the time of the MATRICS effort was associated with tachyphylaxis. Buchanan et al,² based this recommendation on an expressed need to show an enduring effect on cognition. Horan et al do eventually acknowledge that durability of effect was a concern for the MATRICS guideline and then suggest that durability could be looked at with long-term real-world evidence strategies. Presumably, they are talking about uncontrolled evidence, and I’m not sure regulators would consider such evidence adequate.

Nevertheless, it is not unreasonable to consider separately the effect of a drug on improving cognitive impairment and the durability of that effect. As they point out, many of the drugs being considered for cognitive enhancement would be expected to have a more rapid effect, that is, days and weeks rather than months. It is in fact standard practice for an induction effect and a maintenance effect for a psychiatric drug to be considered separately, with studies of very different designs. But I don’t think a demonstration of durability for a cognitive benefit can be relegated to an uncontrolled, observational design. The randomized withdrawal design is a well-established design for addressing the question of the durability of effect, and there is no reason why it should not be utilized in this situation, but perhaps post-initial approval.

Need For a Co-primary Endpoint

The original MATRICS paper was clear that the FDA expected CIAS trials to have a co-primary endpoint that in some way captured functional improvement associated with cognitive improvement (or at least improved functional capacity, ie, a proxy measure), and that this would also be shown to be superior to placebo. Horan et al suggest that this requirement derived from a similar requirement in dementia drug development programs, and they challenge the application of this model in CIAS programs. I agree with them in part, since many schizophrenic patients never had some of the functional skills being assessed in the first place. However, part of the reason for requiring a co-primary measure in dementia programs was a concern that small changes in scores on cognitive tests have unknown clinical meaning, and the same applies to CIAS. In addition, one might argue that the goal of treatment for CIAS is not so much a desire to see an improvement in scores on cognitive tasks, but rather an improvement in functioning (or at least functional capacity). So, simply dropping this requirement altogether does not seem the right answer to me, but it does beg the question of how best to address this issue in CIAS trials. Horan et al do eventually suggest that, rather than dropping the requirement for a co-primary endpoint, the need for an assessment of ultimate functional impact be accomplished post-approval. This is similar to their suggestion for durability of effect, that is, a separate demonstration of longer-term functional impact in contrast to a more acute demonstration of benefit on cognitive tasks. It is unclear, however, exactly what they are suggesting to demonstrate functional impact. They refer somewhat vaguely to “follow up studies,” and perhaps the use of “digital” technologies. To me, this again sounds like uncontrolled follow-up of patients, and it is unclear to me that the digital technologies implied have any greater clinical interpretability than cognitive tasks in the MCCB. So, this approach is not likely to satisfy regulators. More thought and work are clearly needed to come up with a better way to address functional impact, and it seems unlikely that regulators would consider it sufficient to push it entirely to post-approval.

Optimal Study Design

The MATRICS initiative clearly favored study designs for agents that target cognitive impairment in a so-called residual phase of the schizophrenic illness. This is an adjunctive design, where the new agent or placebo is added to the continued use of an antipsychotic agent in patients who have improved regarding mostly positive symptoms and are stable, but who still have substantial cognitive impairment. In fact, it was viewed as critically important that these be patients who were relatively stable with regard to positive symptoms. The concern was that, if the new agent also had a benefit on positive symptoms, this in itself might secondarily result in improved scores on cognitive tasks. One of the most predictable findings in acute schizophrenia trials is an improvement in negative symptom scores occurring along with an improvement in positive symptom scores. This improvement in negative symptoms is not viewed as a primary effect on negative symptoms, but rather as secondary to the improvement in positive symptoms. Although one can observe improvements in negative symptom scores in these trials, patients are typically still substantially impaired regarding negative symptoms, despite some measurable improvement in negative symptom scores. There is less data on similar improvements in scores on cognitive tasks in acute studies since such tasks are rarely included in these acute schizophrenia studies, but it is suspected that this would be observed if these cognitive tasks were included. Some longitudinal studies³ have shown an association between change in positive symptoms and cognitive impairment, that is, a worsening of cognitive ability associated with especially large increases in positive symptoms. Some have suggested that this is simply an adverse impact on test-taking ability.

In any case, the adjunctive design is not applicable for so-called broad-spectrum agents (BSAs), that is, drugs that improve both positive symptoms and cognitive impairment, since it would not make sense to add a BSA to an antipsychotic agent. Horan et al argue that, while there were no such agents at the time of the original MATRICS initiative, there may be such agents now and it is, therefore, important to develop designs to study such agents (BSAs). Drawing on the recent experience with vortioxetine and cognitive impairment in depression, the authors consider a similar acute design for a BSA, that is, 3 arms (the BSA, an atypical antipsychotic, and placebo), presumably in acutely psychotic patients. Horan et al also, however, acknowledge that this design likely has 2 problems for the FDA: (1) it falls far short of the duration requirement, and (2) there is still the problem of possible confounding by improvement in positive symptoms that could secondarily result in some improvement in scores on cognitive tasks. I agree that the FDA would not likely accept this design. Horan et al also consider an alternative similar monotherapy study in stable schizophrenic patients. However, they quickly reject this design as well because of the likelihood of substantial relapses in patients getting a placebo. This design would likely be problematic for many schizophrenic patients, depending on the duration. However, if it were a shorter trial in patients who had well-established stability, it might be worth giving it further consideration.

Finally, the authors suggest a “switching” design, which is similar to what I had proposed many years ago. Horan et al suggest starting with stable patients, presumably in a residual phase of the illness, who have prominent cognitive impairment. They would then be randomized to monotherapy on either the new BSA or a different atypical antipsychotic than the one they were on. The goal would be to show the superiority of the BSA over a different atypical antipsychotic agent on cognition. I had proposed a similar design in a 2013 paper,⁴ however, in my proposed design, the same stable patients would be randomized to either continue on their current antipsychotic or switch to a new BSA. The goal would be to show the superiority of the BSA over the atypical agent on cognition. Since all patients are getting an active drug, a switching design study could be of longer duration. In my view, either switching design would work, however, Horan et al fail to note the problem in interpreting the data that I allude to in my 2013 paper. Assuming patients assigned to the BSA improve on cognition and the patients remaining on an atypical agent are unchanged in cognition, how should this difference be interpreted? One possibility is that the BSA does improve cognition, while the atypical agent does not. Another possible interpretation is that the BSA does not have an effect on cognition, but the previously used atypical antipsychotic actually worsened cognition, and removing that agent simply returned patients to the level of cognitive impairment that was part of their illness. A third possible interpretation would be that both drugs actually impair cognition, but the BSA has a lesser effect, and so looks better than the atypical agent by virtue of stopping that agent. The point is that, without a placebo arm, interpretation is difficult. If the difference in cognition is meaningful, it could still be celebrated as an advantage for the BSA, but would not serve as an adequate basis for a claim of a primary benefit on cognition. Of course, if the BSA actually returned patients to normal cognitive status, that would be a remarkable, but highly unlikely outcome.

In summary, there are two possibilities for pursuing BSAs. They are both, in essence, switching studies. One would be a 3-arm trial in stable schizophrenic patients. It would not have to be a long-term trial, and, therefore, a placebo arm should be acceptable. However, more thought is needed in thinking through how to interpret the different possible outcomes for such a design, for example, the stable patients switched to a placebo might improve cognition simply based on stopping a drug that has a cognitive impairing effect. Some version of a switching design without a placebo arm is also a possibility, but interpretation of a finding of superiority of the BSA would still be a challenge to work out for labeling.

Enrichment strategies

The original guidance coming from the MATRICS initiative recommended accepting schizophrenic patients within a wide range of cognitive ability, that is, essentially all comers. Horan et al present arguments and data to support thresholds for severity of cognitive impairment, similar to the use of thresholds for severity in virtually all other psychiatric trials. They also suggest establishing standards for the stability of cognitive impairment as another criterion for accepting patients into CIAS trials, again based on arguments and data. Finally, they suggest establishing standards for treatment adherence as an additional criterion for accepting patients into these trials. All of these seem reasonable to me and I think FDA would view such suggestions positively.

However, since we are considering which schizophrenic patients are most appropriate for CIAS trials, I would suggest thinking even more broadly about what patients might benefit most from agents targeting cognitive impairment. It seems notable to me that after 2 decades of studies in CIAS using agents with quite a wide range of pharmacological actions that none of these programs has come even close to reaching an NDA. It is also notable that all of the trials in question used the adjunctive model where a new agent is added to continued treatment with an antipsychotic. The assumption underlying all of these programs was that essentially all schizophrenic patients need to be treated for life with an antipsychotic agent that narrowly targets positive symptoms, and the right approach to best addressing other symptoms such as cognitive impairment and negative symptoms is to add another agent to the antipsychotic (unless, of course, there is an alternative BSA that can address both positive symptoms and cognitive impairment). Maybe the universal failure of these programs is explained in part by the inability of almost any adjunctive agent to overcome the inherently adverse cognitive impairment burden of the underlying antipsychotic agent. In fairness, there is no clear evidence that antipsychotics have a cognitive impairing effect in schizophrenic patients. Most trials of antipsychotics in schizophrenic patients are acute treatment trials in patients having a positive symptom relapse, and possible confounding by virtue of secondary improvement of cognition associated with improvement in positive symptoms could mask a direct adverse effect on cognition. A clear demonstration of this cognitive burden, albeit in a non-schizophrenic population, is seen in the CATIE-AD trial.⁵ The notion that all schizophrenic patients need a lifetime of continuous antipsychotic treatment is increasingly being challenged.⁶ Maybe it is time to consider more refined distinctions among the different subtypes of schizophrenic patients and consider such distinctions in selecting patients for trials, including CIAS trials. There are patients well along in the course of this devastating disease who seem to be troubled more by the functional impact of cognitive impairment and negative symptoms than they are by positive symptoms. Maybe these are patients who could most benefit from cognitive enhancers. While it is undoubtedly true that many schizophrenic patients will need lifetime antipsychotic treatment, it also seems true that some of these patients may not need continuous antipsychotic treatment, and a substantial percentage of these patients are not taking antipsychotics anyway, even if they are being prescribed. These patients might be good candidates for a monotherapy trial that is perhaps not handicapped by a cognitive impairing effect of an underlying antipsychotic agent.

References