How effective are antipsychotics at preventing relapse in the real world compared to randomised controlled trials?

For people who have been diagnosed with schizophrenia (and the related diagnosis of schizoaffective disorder) we know that they are at a high risk of having repeated episodes of psychosis, known as a relapse. Psychosis is often thought of as the presence of unusual experiences such as delusions or hallucinations but also includes disturbances of mood, reduced drive & expressiveness of emotion and problems with cognitive functioning. Medications called antipsychotics are often used to treat an episode of psychosis and can be effective in this role with evidence from randomised controlled trials (RCTs) – the gold standard of evidence-based medicine – confirming this even taking into account unpublished trials (Turner et al,2012). If an antipsychotic has been effective in reducing the psychotic symptoms of unusual experiences, then they are often continued afterwards as evidence from RCTs shows they can delay the return of psychotic symptoms (Leucht et al, 2012).

It would be useful to confirm from “real-world” data that this effectiveness in preventing relapse from research is also found in patients in routine clinical practice. It would also be helpful to know which antipsychotics are more effective at this compared to other antipsychotics. Psychiatrists who prescribe these medications will then be better placed to advise patients on the pros and cons of different antipsychotics. Effectiveness at achieving a desired clinical outcome (such as preventing relapse) must be weighed up with the acceptability of side effects.

One issue that affects medication effectiveness is concordance i.e. whether the patient takes the medication or not. Often the patient does not take the medication for a variety of factors – commonly burdensome side effects but also, they may not realise they have a psychotic illness or hope it has gone away. Sometimes misinformation from a variety of sources (which is why it’s good for patients to have access to good quality information). Some antipsychotics come in a long-acting injection form (called LAIs) lasting from 1 week to months depending on the particular version. If a patient has had this injection, then we know they have taken the medication, whereas if they are on a tablet they might say they have taken it but they haven’t. Therefore, it will be interesting to see if in the real world long-acting injections are more effective at preventing relapse than tablets (called ‘oral’ in this research) as they are from research studies (Leucht et al, 2012).

The focus of this blog is an interesting study that looks at how several newer antipsychotics and clozapine are compared to haloperidol (an older antipsychotic) in terms of preventing relapse of schizophrenia and schizoaffective disorder (Efthimiou et al, 2024). In the paper, they use the term second-generation antipsychotic (or SGA) for the newer antipsychotics, but SGA as a term has little coherence in pharmacological science such as describing different activities at neurotransmitter receptors in the brain (Zhou et al., 2022), so I prefer to use ‘newer antipsychotic’.

The authors were also interested in trying to rank antipsychotics as to which was the most effective compared to least effective, but they also looked at how effective antipsychotics were in RCTs (called efficacy) compared to how effective they are in the real world from the national registry data (called effectiveness). It’s important to know that effectiveness in research trials does translate into effectiveness in clinical practice in the real world. It’s also important to see how big the gap is between efficacy in research and effectiveness in the real world (efficacy-effectiveness gap). One potential explanation is that research includes patients who are more likely to respond to treatment without any complicating factors like substance misuse (many illegal drugs can cause a relapse of psychotic symptoms). One other important question is whether the advantage of LAIs is found in routine clinical practice not just research trials.

How big a gap is there between the efficacy of antipsychotics in preventing relapse in research studies and effectiveness in clinical practice?

Methods

The research team used two types of evidence. One was real-world data from national patient data registries from Sweden and Finland. The other type of evidence was combining data from randomised controlled trials (RCTs for short) comparing antipsychotics with placebo and/ or other antipsychotics. Meta-analysis is the statistical technique used to combine data from different RCTs of the same intervention (such as an antipsychotic) which gives you an average estimate of the intervention’s effectiveness across all the studies. When comparisons are made between different interventions against a common comparator (often haloperidol or placebo in this case) this is called a network meta-analysis. For the national registry data, the comparator was the period when the patient was not on any medication. Some claim that stopping antipsychotics causes a withdrawal psychosis, but the evidence is that the withdrawal symptoms are mostly anxiety or physical symptoms, not psychosis (Brandt et al, 2022). The exception is clozapine which is known to cause a withdrawal psychosis. I have also seen a withdrawal psychosis from the short half-life antipsychotic quetiapine.

Amongst the authors are researchers experienced in using meta-analysis, network meta-analysis and the use of Scandinavian registry data to analyse real-world outcomes. They pre-registered what procedures they were going to carry out to do their research and only added two more analyses at the request of people who reviewed the paper before it was accepted for publication.

The real-world part of the study identified people from Swedish and Finnish national registry data with an ICD-10 diagnosis of schizophrenia or schizoaffective and either on haloperidol or a newer antipsychotic and had been stable for at least 12 weeks. This amounted to 90,000 in combined registries – the vast majority were Finnish.

The network meta-analysis identified relapse prevention RCTs of newer antipsychotics compared to each other, haloperidol, or placebo from relapse prevention studies from a database up to 2022. The 10,091 participants in these studies were diagnosed with schizophrenia or schizoaffective disorder and had been stabilised on an antipsychotic before entering the RCT and being randomised to either continue their antipsychotic or receive a placebo.

For both real-world studies and the RCTs, the follow-up period was 12 months. In the real-world study the follow-up may end earlier if the participant died, was hospitalised for something other than psychosis or changed medication.

For both real-world and RCTs the main outcome of interest was whether the participant relapsed by 6 months or 12 months. For real-world studies, relapse was defined as hospitalisation for psychosis. For RCTs, relapse was as defined in each study. The authors quoted evidence for saying hospitalisation was a good indicator of relapse.

Understanding Hazard Ratios

The authors use the statistical Hazard Ratio (HR) measure to show the relapse risk compared to either placebo or no drug. They give the average HR from their analysis and what’s called the 95% confidence interval range (CI range). If an antipsychotic’s CI range is wholly less than 1 then the antipsychotic is likely associated with a lower relapse risk than the comparator. If the CI range is wholly greater than 1 then that antipsychotic is likely associated with a higher relapse risk than the comparator. Often there is an overlap indicating some uncertainty as to whether the antipsychotic’s effectiveness compared to the comparator. You can also compare the average HR for antipsychotics with other antipsychotics to rank them in a league table, but there is often overlap in the CIs indicating some uncertainty in these comparisons.

Various other analyses were carried out. One important additional analysis was looking at how effective antipsychotics were for individuals in the real-world data who could have been eligible to take part in RCTs. These RCT-eligible individuals would be those without the following qualities: be under 18 or over 65 years old, pregnant/breastfeeding, have serious somatic disease, be suicidal, using other psychotropic medication such as antidepressants or mood stabilisers, not respond well to antipsychotic treatments, have intellectual difficulties, or suffer from past or current substance misuse. RCTs tend to recruit participants who are likely to respond well to medications – one factor not mentioned above is that they also tend to exclude individuals who have a patchy record of taking medication.

Results

Only what I think are the important results will be discussed below as these look at the main clinical points of interest: how effective are antipsychotics in research and real-world clinical practice at preventing relapse? These medications form the backbone of relapse prevention for psychosis, so this is an important question. I’ve excluded discussion of the various analyses they did for other purposes such as how consistent the study results were with each other. I’ve also excluded the combined real-world data and RCT data as I think the two types of data are from 2 fundamentally different situations so should not be combined. The outcome data that was presented was for a 6-month follow-up.

• Antipsychotics in RCTs were 2.58 times (or 258%) more effective in RCTs than real-world data: the efficacy-effectiveness gap.
• Antipsychotics were 40% more effective in RCT-eligible patients in the real-world data than those who would not be eligible.
• Long-acting injections (LAIs or depots) were 12% more effective in real-world data than in RCTs.
• LAIs are 27% more effective than oral meds in national registries.
• There was a large overlap in effectiveness between many of the antipsychotics. The 95% confidence intervals (CIs) were often very wide, especially for RCTs (due to smaller numbers than those in each intervention group compared to the national registries).

Many of the antipsychotics overlap in effectiveness with each other and with haloperidol. Rather than give a big, long list where many of the medications overlap it’s simpler to look at the best 3 medications for preventing relapse and the worst 3 interventions including placebo in RCTs and no antipsychotic in real-world data compared to haloperidol.

Top 3 real-world vs Haloperidol (not all drugs included)

1. Clozapine HR 0.64 (CI 0.56 to 0.72)
2. Olanzapine LAI HR 0.73 (CI 0.58 to 0.98)
3. Aripiprazole LAI HR 0.76 (0.45 to 1.29)

Bottom 3 real-world vs Haloperidol

1. Risperidone oral HR 1.04 (CI 0.91 to 1.18)
2. Quetiapine oral HR 1.21 (CI 1.05 to 1.38)
3. No antipsychotic (compared to not being on an antipsychotic to when the patient was taking their prescribed antipsychotic ) HR 1.33 (CI 1.17 to 1.50)

Top 3 RCT vs Haloperidol (no clozapine in RCT analysis)

1. Olanzapine oral HR 0.25 (CI 0.05 to 1.11)
2. Olanzapine LAI HR 0.37 (CI 0.07 to 1.92)
3. Zotepine oral HR 0.48 (CI 0.08 to 2.74)

Bottom 3 RCT vs Haloperidol

1. Ziprasidone oral HR 1.57 (CI 0.38 to 6.46)
2. Lurasidone oral HR 1.74 (CI 0.61 to 4.99)
3. Placebo HR 3.01 (CI 1.08 to 8.34)

The authors also did an analysis combining national registry data and RCT data but for me, this lacked validity as the two samples were just too different, so I’m not mentioning it further.

Antipsychotics had differing effectiveness in preventing relapse compared to haloperidol and were more effective in research trials than in the real world.

Conclusions

Antipsychotics appear effective in preventing relapse but clozapine, olanzapine (LAI and oral), aripiprazole LAI and zotepine oral may be the most effective in preventing relapse.

The gap between efficacy in RCTs and effectiveness in the real world is large. This is partly due to RCTs selecting patients who are more likely to respond well to treatment as well as those who are most likely to take the treatment regularly. There was still a superiority in RCT outcomes for RCT-eligible patients compared to real-world data, but this may still be explained to some extent by RCT patients.

The authors disagree with my conclusion. In their conclusion, they stated:

Though efficacy versus placebo in RCTs might not be directly portable to real-world effectiveness versus no antipsychotic use, we find no evidence of an efficacy–effectiveness gap in head-to-head antipsychotic drug comparisons (p.8).

I think that they mean that RCTs give us a reasonable guide as to which antipsychotics will be most effective in routine practice at preventing relapse. What I’m talking about is how effective they are in routine clinical practice compared to RCTs.

The gap between the efficacy of antipsychotic medication in research and its effectiveness in the real-world setting is large.

Strengths and limitations

• The authors stuck to a pre-registered protocol apart from 2 additional post-hoc analyses of excluding schizoaffective and those on clozapine at baseline. This reduces the risk of researcher shenanigans to get the result they want.
• This study includes large numbers especially for real-world data, and an interesting analysis for real-world RCT-eligible participants provides useful evidence on the efficacy-effectiveness gap.
• Real-world data from Scandinavia may not accurately represent what will happen in other societies. Most of it is from Finland, which might apply to the rest of Scandinavia, but is very different from the US, let alone the Global South.
• The follow-up period was for up to 12 months, but schizophrenia is a condition that can last for many years.
• Not all drugs appeared in both real-world and RCT analysis – many missing from real-world data and clozapine from the RCT analysis. Missing from both RCTs and real-world data were the older antipsychotics especially the older LAIs such as flupentixol LAI, which are still commonly used in clinical practice.

Scandinavian data may not accurately represent the efficacy-effectiveness gap in other countries.

Implications for practice

Antipsychotics in general are effective at preventing relapse in psychosis but some are more effective than others. Long-acting injection antipsychotics have the benefit of making it clear whether the patient is taking the medication or not and have advantages in preventing relapse in the real world.

How do we close the efficacy-effectiveness gap?

• Look at modifiable factors associated with increased relapse.
• Improve concordance with medication by building better relationships with patients, and being more responsive to side effect issues such as taking action to reduce side effects or change antipsychotics.
• Consider the use of long-acting injection antipsychotics.
• Address factors that destabilise patient’s lives and increase the risk of relapse as well as disengagement from care or non-concordance. These issues include stable adequate housing or poverty (so help with access to benefits and employment if willing and capable).
• Alcohol or substance misuse problems should be identified, and support offered such as specialist help and address issues such as previous trauma that may have led to excessive alcohol or substance misuse.

Participants in RCTs often get a good standard of care and are seen frequently, which may explain some of the better outcomes. In the real world, a lower standard of care for many patients may explain some of the efficacy-effectiveness gap. With skilled mental health staff being diverted to early intervention teams and primary mental health teams, they are less available to care for people with chronic psychosis who thus get a worse service or get discharged being told they have ‘recovered’. They may not be seen in outpatient appointments for up to 12 months. They may not have a care coordinator such as a nurse or even if they do, they may hardly see them for weeks due to large caseloads. Social supports such as drop-in centres and other community activities have been cut leading to social isolation and increasing relapse risk. This is the slip twixt cup and lip that we can address by focusing care back on these forgotten patients.

People with chronic psychosis need better care than the current threadbare system of infrequent psychiatry appointments and community home visits.

Statement of interests

Samei Huda has written a book defending the use of the medical model in mental health.

Links

Primary paper 

Efthimiou, O., Taipale, H., Radua, J., Schneider-Thoma, J., Pinzón-Espinosa, J., Ortuño, M., Vinkers, C.H., Mittendorfer-Rutz, E., Cardoner, N., Tanskanen, A. and Fusar-Poli, P. (2024) Efficacy and effectiveness of antipsychotics in schizophrenia: network meta-analyses combining evidence from randomised controlled trials and real-world data. The Lancet. Psychiatry, 2024 Feb;11(2):102-111. doi: 10.1016/S2215-0366(23)00366-8. Epub 2024 Jan 9.

Other references

Brandt, L., Schneider-Thoma, J., Siafis, S., Efthimiou, O., Bermpohl, F., Loncar, L., Neumann, K., Hasan, A., Heinz, A., Leucht, S. and Gutwinski, S., 2022. Adverse events after antipsychotic discontinuation: an individual participant data meta-analysis. The Lancet Psychiatry, 9(3), pp.232-242.

Leucht, S., Tardy, M., Komossa, K., Heres, S., Kissling, W., Salanti, G. and Davis, J.M., 2012. Antipsychotic drugs versus placebo for relapse prevention in schizophrenia: a systematic review and meta-analysis. The Lancet, 379(9831), pp.2063-2071.

Turner, E.H., Knoepflmacher, D. and Shapley, L., 2012. Publication bias in antipsychotic trials: an analysis of efficacy comparing the published literature to the US Food and Drug Administration database. PLoS medicine, 9(3), p.e1001189.

Zhou, C., Nutt, D.J. and Davies, S.J., 2022. Visualizing classification of drugs used in psychotic disorders: A ‘subway map’representing mechanisms, established classes and informal categories. Journal of Psychopharmacology, 36(9), pp.1007-1015.