Long-Acting Injectable vs Oral Antipsychotics for Relapse Prevention in Schizophrenia: A Meta-Analysis of Randomized Trials

Introduction

Because psychopathology and social functioning can worsen with repeated psychotic episodes in patients with schizophrenia,^1,2 relapse prevention is critical. There is strong evidence of antipsychotic efficacy for relapse prevention in chronic and first-episode patients,^3,4 in that the risk of relapse is 2–6 times higher without medication.^3–6 However, because nonadherence rates as high as 50% can limit the efficacy of pharmacotherapy,^7,8 the use of long-acting injectable antipsychotics (LAIs) is an important option.⁹ In practice, patients and clinicians are sometimes reluctant to use LAIs because of stigma, needle pain, time constraints, side effect concerns, and cost.¹⁰ Given these drawbacks to the use of LAIs, convincing data showing the superiority of LAI over oral antipsychotics (OAPs) is needed to support the use of LAIs. The first LAI was introduced in the 1960s. Since then, at least 5 first-generation antipsychotics (FGAs)-LAIs and 3 second-generation antipsychotics (SGA)-LAIs have become available. Our previous meta-analysis found that LAIs were associated with significantly lower relapse rates than OAPs.¹¹ However, new, large, controlled trials showed no benefit of LAIs.^12–15 We performed a meta-analysis that incorporated the new randomized controlled trials (RCTs) and also applied broader inclusion criteria. The new analyses are based on twice as many studies and 3 times as many patients as compared with the earlier ones.

Method

The meta-analysis was performed following PRISMA guidelines.¹⁶

Results

Discussion

To our knowledge, this is the largest meta-analysis comparing LAI vs OAP efficacy for relapse prevention in schizophrenia. Compared with our prior meta-analysis,¹¹ we broadened the inclusion criteria by incorporating studies that enrolled inpatients and outpatients, and also by including studies lasting 6 months to <1 year. However, at the same time, we also examined the result within the studies meeting our previous inclusion criteria (outpatient study, lasting ≥1 year).

We found that pooled LAIs were not superior to OAPs in all of the examined relapse-related outcomes. The only exception was fluphenazine-LAI, which showed significant superiority over OAPs in several relapse-related outcomes. This lack of superiority of LAIs is in contrast to our previous meta-analysis,¹¹ which showed significant superiority of LAIs over OAPs in schizophrenia with a NNT = 10. However, this difference is not due to the use of broadened inclusion criteria for analyzed studies, as results were similar when the analyses were limited to the outpatient studies lasting ≥1 year. Rather, the nonsuperiority of LAIs vs OAPs in the extended analyses of our study was due to a shift in the RR toward the null hypothesis rather than due to increased imprecision indicated by a widening of the CI. In fact, the CI of our primary result was slightly narrower than in the prior meta-analysis (RR = 0.93 [95% CI: 0.80–1.08] vs RR= 0 .70 (95% CI: 0.57–0.87]), suggesting that the point estimate was, if anything, somewhat more precise.

On the other hand, we found in subgroup analyses that FGA-LAIs studies or older RCTs showed significant superiority of LAIs over OAPs but that this was not the case in SGA-LAIs studies or newer RCTs. (In a byline, we note that the distinction between FGAs and SGAs is a questionable one.)²⁸ However these findings require cautionary interpretation. Eight out of 10 FGA-LAI studies were published by 1991, and all 8 RCTs published by 1991 employed exclusively fluphenazine-LAI. Hence, these distinctions were nearly equal, and it is impossible to disentangle the potential reasons for different effect sizes. Moreover, the subgroup analysis comparing the effect sizes of FGA-LAIs and SGA-LAIs relative to OAPs did not yield a significant difference, suggesting that in this indirect comparison, there was no indication that FGA-LAIs are superior to SGA-LAIs. That being said, we consider the following explanations for the differential findings for FGA-LAIs (or older RCTs). One possibility is publication bias. Because the registration system of clinical trials has been introduced, companies are obliged to either publish all studies or make the data publically available, while in earlier years, studies showing less advantages for LAIs may not have been published. Another possibility is changing definitions of relapse. In order to mitigate potential adverse consequences of participating in clinical trials, recent studies tended to utilize lower thresholds for relapse, which can increase the rate of false positives. From the definitions encountered in these analyses, this tendency was not obvious (see table 1). However, it is hard to rule out this possibility, as some older studies used vague definitions rather than quantitative measures, such as predefined changes in PANSS score. In addition to various definitions, thresholds used to determine relapse can also have a big impact. For example, hospitalization as threshold is susceptible to health insurance system variations, social supports, or clinician’s judgment and can differ by geographical regions or over time. Furthermore, the OAPs used in FGA-LAI studies published until 1991 were FGAs, while OAPs used in SGA-LAI studies published since 2005 were generally SGAs. One can argue that patients allocated to FGA-OAPs might have a greater chance of relapse, either due to poorer adherence or differences in neuropharmacologic properties compared with SGA-OAPs. This could be consistent with our recent meta-analysis comparing oral SGAs and FGAs, which showed a small but significant superiority of SGAs in preventing relapse, but due to limited information, we could not determine whether nonadherence was a factor.²⁹ We examined this possibility in the current analyses as well, but we did not find significant group differences. However, adherence is rarely assessed directly, and the evaluations involving these outcomes are very crude, which is a major limitation of the available RCTs. LAIs are thought to be better via improved adherence, not via intrinsically better efficacy. Therefore, it is unclear whether LAIs were not superior because compliance with OAPs was good enough in the context of RCTs.

Regarding CPZ equivalent doses, there were some significant or trend-level differences within and across studies. CPZ equivalents trended toward being higher in OAP arms compared with LAI arms within studies using different antipsychotics, which may have disadvantaged the LAI arms. However, in the studies that used the same medications in the LAI and OAP arm and used also similar CPZ equivalent doses in both arms, there was also no difference in terms of preventing relapse. In SGA-LAI studies, or newer studies, CPZ equivalent doses in the OAP arms also trended to be higher than in the LAI arms. In addition, CPZ equivalent dosages used in FGA-LAI studies and in older studies were at a trend level and significantly higher than in SGA-LAI studies and in newer studies, respectively. These findings are consistent with a trend toward using lower antipsychotic doses since the more widespread use of SGAs. Whether or not this difference is relevant for the superiority of FGA–LAIs and older studies that used higher doses in both the LAI and OAP arms than SGA-LAI and newer studies that used lower doses in both arms is unclear. However, it is possible that higher LAI doses were more effective in showing superiority over OAPs, which may have been dosed higher than necessary leading to dropout or nonadherence due to extrapyramidal side effects. However, the interpretability of the results of potential dose differences is limited by missing information on mean doses in the LAI and OAP arms, especially in old studies, and by substantial differences in the suggested conversion factors from antipsychotic doses to CPZ equivalents, both of which applied mostly to older studies using FGA-LAIs.

In contrast to our results, recent naturalistic studies support the advantages of LAIs over OAPs in relapse prevention^30–32 as do mirror image studies.^33,34 Tiihonen et al.³¹ reported in a nationwide cohort that the risk of rehospitalization with LAIs was one-third that of OAPs. Most LAIs showed significant superiority compared with each OAP counterpart regarding all-cause discontinuation. These results are especially important given the potentially conservative bias in that the patients who receive doctor’s choice LAIs are more likely to be nonadherent and more severely ill compared with patients receiving doctor’s choice OAPs. Moreover, patients consenting to clinical trials of LAIs may not be representative of those prescribed LAIs in real-world settings.³⁵ Participants in clinical trials might overrepresent patients with better engagement with health care providers, better adherence to the treatment, lower illness severity, and better cognitive capabilities to understand complex issues. This difference in procedures can potentially lead to a cohort bias in that less severely ill patients could be enrolled in newer trials, particularly with increasing stringency of consent processes. This may explain the different results of older vs newer studies.

It is also important to recognize that participation in a controlled trial alters the ecology of treatment delivery and experience. Patients in clinical trials are likely to receive more and different types of attention than those in routine care, from measures of adherence to reminders to attend clinical/research assessment sessions, or to the provision of free medication.³⁶ In addition, more frequent monitoring during a trial enables psychiatrists to change dosages according to the symptoms and provide supportive psychotherapy. It is difficult to determine what role these factors might have in altering patterns of medication-taking in contrast to routine care and to what extent they might, therefore, diminish the potential advantages of LAIs in RCTs. The substantial disparity between the large naturalistic cohort study results and those of RCTs in this context would support such concerns.

To this point, we have assumed that LAIs must be more effective than OAPs and have been trying to understand why this is not apparent in the RCTs. We should also consider the possibility that if patients are fully adherent with OAPs, the OAPs may actually be more efficacious than LAIs as suggested by 2 recent trials.^23,37,38 One interesting hypothesis is the potential supersensitivity of the dopamine D2 receptor. Long-term, continuous D2 blockade may increase the number and/or high-affinity state of dopamine receptors.^39,40 In addition, there are data suggesting that transient, rather than continuous blockade of D2 receptors by “extended dosing” (eg, every other day dosing), may be more efficacious in both animal and human studies.^41,42 Therefore, LAIs might result in more continuous D2 blockade and receptor upregulation/supersensitivity compared with OAPs. This effect would be differentially stronger when the OAPs do not consist of high-potency FGAs, as oral SGAs generally have less complete or prolonged D2 blockade.

Regarding our study methodology, 2 points require consideration. First, we utilized relapse rate preferentially based on survival analyses as a primary outcome. We acknowledge that graphically calculated relapse rates from survival curves can result in higher relapse rates compared with raw rates because the denominator decreases as dropout occurs. On the other hand, raw relapse rates do not count the potential relapse among patients who dropped out. Moreover, in case of systematically different dropout rates between the treatment arms resulting in shorter follow-up periods in one treatment compared with another, a bias is introduced that disfavors the treatment with better acceptability and greater persistence, as more time is available for relapses to occur. For these reasons, we believe that estimated rates may yield more accurate data than raw relapse rates. However, we acknowledge the problem of using estimations, rather than observed data, and the mixing of estimated and raw rates from studies not providing survival curves. To deal with this problem, we analyzed the data in several ways, utilizing raw instead of estimated relapse rates, and using raw or estimated rates exclusively. The results remained the same, independent of how relapse rates were calculated.

Second, we used each the safety/efficacy and randomized population as denominators for the analyses. While ITT analysis is the gold standard for clinical trials, ITT definitions varied, which is problematic. Some defined ITT as patients taking the medication at least once, or receiving at least one assessment (modified ITT), while others utilized the randomized population. Moreover, several studies provided survival curves and other results, which were solely based on the modified ITT sample, and 1 study²⁶ only presented the modified ITT population. It is widely accepted that the reluctance toward injections is an important obstacle for introducing LAIs.⁹ Some studies included in our analyses reported that patients dropped out when learning of their LAI allocation. We considered it important to include those patients as dropouts. However, given the fact that more studies provided data using modified ITT populations, we wanted to also examine the risk of relapse once patients are on LAIs. Therefore, we analyzed the results in both populations separately, which we considered the most conservative strategy, and there were no important differences.

Results of these analyses have to be interpreted in the context of several limitations. First, the database, though larger than in the previous meta-analysis, is still limited. For example, no long-term RCTs comparing paliperidone-LAI or aripiprazole-LAI with OAPs were available. Furthermore, relapse definitions varied. We utilized each study-defined relapse measure, and if no definition was available, we used the most-appropriate relapse-related outcome, ie, predominantly psychiatric hospitalization. The problem of heterogeneously defined relapse is not surprising because there is no universally accepted definition. On the other hand, this heterogeneity and broad-based definition could also serve to enhance the generalizability of the results. There was a 14-year gap between the last study published in 1991 and the next study published in 2005, which enabled us to examine the effect of time when the study was conducted. Because clinical and diagnostic concepts of the disease and therapeutic environments have changed over time, it is unclear, which factors in addition to the almost exclusive study of FGA-LAIs earlier and of SGA-LAIs later may have influenced the results. In addition, heterogeneity of the results was seen. Although the majority of studies targeted chronic patients with over 10 years of illness, some targeted patients with even longer illness duration,¹² while some targeted relatively early-onset patients.^43,44 One study included patients with aggressive behavior,²⁴ another included patients who were hospitalized throughout the study.²³ Such clinical characteristics and different treatment settings may have caused heterogeneity. Of note, risperidone-LAI studies were found to be most heterogeneous (τ² = 0.09, I² = 74%, Q = 30.92, df = 8, P = .0001). Moreover, the reporting of drop out due to “adverse events” is not always unambiguously restricted to physiologic/physical adverse effects. Rather, at times, discontinuation due to potential inefficacy-related outcomes, eg, anxiety, agitation, and worsening of psychosis, may be counted as a side effect-related discontinuation, which complicates the interpretation of this outcome.

Further limitations regarding the design of the analyzed studies include issues like treatment concealment (7/21 were open studies), in- vs outpatient status (6/21 included inpatients throughout the study, or presumably a significant portion of the study), study duration (6/21 were <1 year), and medication allocation (10/21 used different medications in the LAI and OAP arms). We consider the double-blind, double-dummy design favorable, as it reduces expectancy and rater biases. At the same time, however, this design itself is very different from clinical practice, and might have contributed to increased selection bias. Moreover, inpatient studies apparently ensured adherence, which may have strengthened the efficacy of the OAP arm. Because nonadherence can increase over time, one may expect also that the relative superiority of LAI compared with OAPs develops over time. Such an effect may explain the lower risk ratio in the longer term studies when we analyzed separately results from studies lasting <1 year (RR = 1.14, 95% CI: 0.82–1.58, P = .43) and those lasting ≥1 year (RR = 0.91, 95% CI: 0.77–1.07, P = .25). However, LAIs were not superior to OAPs in either study subgroup, and some recent studies with durations of ≥2 years^12,14,15 also failed to show superiority of LAIs. This calls into question that the study duration is a major determinant. Using different medications in the oral and LAI arms makes it difficult to disentangle the independent efficacy of the delivery method from potential differences in the efficacy of the 2 drugs including inappropriate dose equivalency. We attempted to examine these issues by subgroup analyses, which all yielded the same result of similar outcomes in the LAI and OAP groups.

In conclusion, while we had anticipated that LAIs (with their intrinsically better adherence) would be more effective than OAPs in preventing relapse, this was not evident in a synthesis of the available RCTs. Notably, these results are in contrast to naturalistic cohort studies showing superiority of LAIs in preventing rehospitalization. Further consideration is required to understand the reasons for this discrepancy. In order to evaluate the real-world effectivenss of LAIs compared with OAPs, large and long pragmatic trials are needed, which better resemble common clinical practice.

In our analysis, FGA-LAIs, but not SGA-LAIs, outperformed OAPs. The difference in effect was not statistically significant and in any event could be due to a cohort effect. The only way to determine if the FGA/SGA distinction is important here would be to conduct head-to-head trials of FGA-LAIs vs SGA-LAIs. It is likely that some studies included in this analysis systematically excluded patients who were expected to have poor adherence. If this is true, then the results should not be generalized to these patients and do not refute the possibility that LAIs may be superior to OAPs in this important group.

Funding

National Institute of Mental Health Advanced Center for Services and Intervention Research; the Zucker Hillside Hospital (P30MH090590).

Supplementary Material

Supplementary material is available at http:// schizophreniabulletin.oxfordjournals.org.