The Long-Term Effects of Antipsychotic Medication on Clinical Course in Schizophrenia

Psychosis can be extremely distressing to patients and their families and is associated with serious consequences, including self-injury, violence, incarceration, and disruption of social relationships. Whereas benefit for negative symptoms and cognitive deficits is not well established (10–12), the effectiveness for psychosis has been well established in randomized controlled trials; a systematic meta-analysis of 38 placebo-controlled trials of second-generation antipsychotics calculated a moderate effect size of 0.51 and a response rate of 41% with antipsychotic treatment compared with 24% with placebo (13). This magnitude of therapeutic effect compares favorably with many of the most effective treatments in nonpsychiatric fields of medicine (14). The effect size of antipsychotic efficacy compared with placebo has decreased over the past several decades, reflecting increased placebo response and possibly reflecting a decrease in drug response (15, 16), although the therapeutic effect size remains in the moderate range. Publication bias does not appear to have had a large effect on meta-analysis results (17).

An older literature described spontaneous improvement rates of approximately 20%−30% in the preantipsychotic era, when Kraepelinian diagnostic criteria were employed (18), and sustained remission in approximately 15% (19). Kraepelinian criteria, similar to contemporary diagnostic criteria, required persistence of illness prior to diagnosis. Other studies of psychosocial alternatives to antipsychotics have claimed higher rates of remission, but results are difficult to interpret due to highly selective patient samples, diagnostic heterogeneity, and high attrition rates (20, 21). While no placebo-controlled trials have been reported in first-episode psychosis patients, the National Institute of Mental Health (NIMH) Collaborative Study of Phenothiazine Treatment in Schizophrenia (22) studied acutely ill inpatients with schizophrenia, 50% of whom were in their first episode. Twenty-nine percent of placebo-treated patients were dropped prior to completing the 6-week trial due to treatment failure; of the placebo completers, 23% (or 16% of the intent-to-treat sample) were much or very much improved compared with 75% of patients treated with antipsychotics. In contrast, two recent studies of intensive individual and family psychotherapeutic approaches (23, 24) reported sufficient response in 35% and 37% of individuals with first-episode nonaffective psychosis such that elective antipsychotic medication was not administered during a 2-year follow-up. These studies did not employ epidemiologic catchment-area based sampling, and thus it is unclear whether subjects are representative. Attempts to identify clinical predictors of response in the absence of antipsychotic treatment have not produced consistent findings (25, 26), and thus currently it is unclear what proportion of individual patients experiencing a first episode of nonaffective psychosis will remit without medication, who they may be, and how long the remission will last.

From prior systematic reviews (27–29), we identified nine studies that compared antipsychotic treatment versus no pharmacotherapy in medication-naive first-episode psychosis patients with follow-up lasting at least 1 year; for studies utilizing mirror image historical controls treated in the preantipsychotic era, we required that the two treatment groups be treated within a 10-year period at the same facility. Because these studies were conducted before the practice of maintenance treatment with antipsychotics was widely adopted, in most cases patients received antipsychotics during hospitalization and subsequently were followed medication-free (28). Six of these studies utilized a mirror image design and compared longitudinal outcomes over periods of 1–5 years in a total of 7,161 first-episode psychosis patients treated in the preantipsychotic era compared with 3,510 first-episode psychosis patients treated in the antipsychotic era. Because findings from these mirror image studies are potentially confounded by factors that may have changed during the interval of time between treatment cohorts, firm conclusions cannot be drawn; however, no study reported poorer long-term outcomes in patients initially treated with antipsychotics. The three studies that focused on hospitalization rates (30–32) found increased rates of discharge and reduced rates of hospitalization associated with initial antipsychotic treatment, whereas the three studies (26, 33, 34) that measured functional outcomes reported higher rates of remission and lower rates of deterioration in the group initially treated with antipsychotics. Three additional studies provided longitudinal follow-up of medication-naive first-episode or early-course patients randomly assigned to medication compared with no medication—a design that avoids the potential limitations of nonrandomized mirror image studies using historical controls. In the first, May and colleagues (35) randomly assigned 228 first-admission patients with schizophrenia to psychotherapy, milieu treatment, antipsychotic, ECT, or antipsychotic plus psychotherapy for 1 year, after which nonresponders were openly treated with antipsychotics. Patients assigned to antipsychotic (alone or in combination) or to ECT spent less time in the hospital over the subsequent 3 years and exhibited superior social functioning; the improved outcome associated with initial antipsychotic treatment appeared to dissipate after 3 years of follow-up (36). The dissipation of the effect of initial medication status may reflect that, by the end of year 3, maintenance antipsychotic treatment was common (100% in the medication group and 85% in the psychotherapy group), consistent with clinical practice in the mid-1960s. Two other studies have been cited as evidence for negative long-term effects of initial antipsychotic treatment, but both involved sampling bias that may have confounded the results. In the first, Schooler and colleagues (37) conducted a 1-year follow-up of the 299 participants who were successfully discharged in the NIMH Collaborative Study of Phenothiazine Treatment in Schizophrenia (22) and found that individuals who received placebo during the initial 6-week trial were less likely to relapse after hospital discharge compared with individuals who received phenothiazine. Because dropout rates due to nonresponse during the initial randomized trial differed substantially between treatment groups (2% in the phenothiazine group compared with 29% in the placebo group), it is highly likely that the sample for the follow-up study of discharged patients contained a disproportionate number of poor-prognosis subjects in the phenothiazine group, since these subjects would not have achieved discharge if treated with placebo. Similarly, in a study of acutely psychotic patients, most of whom were medication naive, Rappoport and colleagues (38) reported poorer outcomes over a 3-year follow-up in patients who initially were randomly assigned to treatment with chlorpromazine compared with patients randomly assigned to placebo during the hospitalization. This design is also confounded by unequal attrition rates during the initial treatment phase—45% of placebo-treated patients dropped out compared with 26% of patients treated with chlorpromazine. After correction for the unequal loss of poor-prognosis patients in the placebo group, differences in outcomes were no longer significant between groups.

Additional information about the long-term effect of antipsychotic exposure has come from two contemporary analyses. Hegarty and colleagues (18) performed a meta-analysis comparing individuals diagnosed with schizophrenia by decade from 1895 to 1992 and found that outcomes substantially improved following the introduction of antipsychotic medication midcentury. However, outcomes became less positive following the introduction of more restrictive Kraepelinian diagnostic definitions in the 1980s, which by requiring a duration of illness of at least 6 months, decrease the likelihood of spontaneous remission. Kraepelinian diagnostic systems were similarly associated with more negative outcomes when applied in the preneuroleptic era (1895–1925). Finally, recent naturalistic studies in rural areas of China provide an opportunity to examine individuals with long durations of illness who have remained medication naive and are diagnosed using contemporary criteria. In a 14-year follow-up study comparing medicated with never-medicated individuals with schizophrenia, outcomes were poorer in the never-medicated group, including rates of partial and complete remission, homelessness, and mortality (39). Approximately 10% of individuals who had never been treated achieved remission, and approximately 8% were partially remitted compared with 30% and 37%, respectively, in individuals who regularly took medication. Because factors such as family involvement and characteristics of illness may have influenced medication status in these subjects, conclusions regarding the causal role of antipsychotics in improved long-term outcomes must be tempered. Nevertheless, this study and similar studies (40–42) do not provide evidence for a long-term negative effect of medication.

A recent meta-analysis of 33 studies found consistent although relatively modest positive correlations between a longer duration of untreated psychosis and residual positive and negative symptoms and impairments in social functioning (43). One quasi-experimental study also found that a reduction in the gap between onset of psychotic symptoms and treatment with antipsychotics was associated with improved long-term outcome (44). The relationship between the duration of untreated psychosis and outcome remains significant after controlling for variables known to influence prognosis (45), but it is not certain that the relationship between duration of untreated psychosis and outcome is causal, since no contemporary randomized placebo-controlled studies have assessed the long-term consequences of withholding antipsychotic treatment. Nevertheless, this association argues against an initial negative effect of medication and suggests that antipsychotics may have an early protective effect that persists over time.

Questions have also been raised about the value of maintenance antipsychotic treatment based on longitudinal naturalistic studies of patients with schizophrenia conducted in the United States by Harrow and colleagues (46) and in Finland by Moilanen and colleagues (47) that found superior outcomes in individuals who were previously treated but not taking medication at the time of assessment compared with individuals taking medication. However, nonmedicated patients had more favorable premorbid characteristics in the study by Harrow and colleagues (48) and had less severe illness in the study by Moilanen and colleagues (47), suggesting that in naturalistic uncontrolled studies, medication status may in part be the consequence of whether patients are doing well or poorly rather than the cause. Other naturalistic studies have found improved outcomes in individuals with schizophrenia who continued antipsychotic treatment compared with those who did not (49, 50). It is not possible to draw conclusions from these studies due to the substantial risk of confounding (51). It is clear, however, from placebo-controlled trials that maintenance treatment with antipsychotics reduces the risk of relapse. A meta-analysis of 65 randomized trials of maintenance treatment demonstrated a large reduction in relapse rates with antipsychotic maintenance compared with placebo at the 1-year follow-up (27% compared with 64%) and better quality of life (52). The effectiveness of antipsychotics in the prevention of relapse was consistent regardless of duration of illness, duration of clinical stability prior to medication discontinuation, or the use of first-generation compared with second-generation agents (52).

Several naturalistic studies have reported that a subgroup of patients, possibly as large as 20%−30%, maintain remission and function relatively well for sustained periods after discontinuing antipsychotics (6, 53–55). In a randomized trial in stabilized first-episode patients, Wunderink and colleagues (6, 56) found that a strategy of lowering antipsychotic doses and discontinuing antipsychotics when possible resulted in higher relapse rates (43% compared with 21%) over 18 months compared with treatment as usual, consistent with prior studies of targeted treatment in which medication discontinuation resulted in high rates of relapse (57). When followed naturalistically for an additional 5 years after completion of the randomized trial, rates of relapse no longer differed between experimental and control groups (61.5% compared with 68.6%) and functional remission was greater in the experimental group (46.2% compared with 19.6%), although rates of symptomatic remission did not differ (69.2% compared with 66.7%). This finding is difficult to interpret and requires replication, especially given that during the 5-year follow-up, the original randomized treatment allocation was not retained and further treatment was not controlled. Mean doses of antipsychotics (in haloperidol equivalents) were only marginally lower in the treatment discontinuation group (2.8 mg/day) compared with the standard care group (4.0 mg/day) during the final 2 years of follow-up. It is noteworthy that under the closely monitored dose reduction protocol employed by Wunderink and colleagues (56), the increase in relapse rates during the first 18 months was not associated with an increase in hospitalization or with negative outcomes, in contrast to naturalistic studies in which relapse has been associated with worsening of negative symptoms and poorer functional outcomes (58). The Wunderink study (6, 56) illustrates the difficulty of balancing the risk of relapse associated with antipsychotic dose reduction or discontinuation compared with the potential for minimizing adverse effects. Notably, one meta-analysis found that maintenance treatment with moderately low doses (50%−100% of World Health Organization-defined standard doses) was comparable in preventing relapse to standard doses, whereas very low doses (less than 50% of standard doses) increased the risk of relapse (59).

Structural abnormalities in the brains of schizophrenia patients have been well documented, particularly reductions in prefrontal cortex and temporal lobes. Brain volume is reduced in patients at ultra-high risk and is further reduced with progression to the first episode of psychosis, prior to exposure to medication (60). Unaffected twins in monozygotic and dizygotic twin pairs discordant for schizophrenia also exhibit reductions in gray matter and, based on these data, it has been estimated that most of the gray matter loss in schizophrenia is explained by genetic liability, although environmental factors also contribute (61). Whether medication plays a role in gray matter volume change, either protective or causative, remains uncertain. Antipsychotic drug dose at the time of scanning and duration of illness correlated significantly with gray matter volume loss in a meta-analysis of over 18,000 subjects with schizophrenia (62), but because medication dose may be a marker for severity of psychosis or refractoriness to treatment, correlations between medication dose and brain volume loss are difficult to interpret. Similarly, because brain regions associated with volume loss during antipsychotic treatment also are reduced in medication-naive patients, it is difficult to determine whether brain volume loss reflects progression of illness compared with a medication effect, or an interaction between the two (63). Moreover, postmortem studies prior to the use of antipsychotics showed abundant evidence of brain structural abnormalities (64), mirroring structural abnormalities found in postmortem studies of patients having received antipsychotics (65).

Along these lines, in a 5-year longitudinal study, left frontal cortical density loss correlated with the number of hospitalizations (66), consistent with a negative effect of psychosis. The duration of relapses was also found to correlate with a reduction in brain volume in a longitudinal study of first-episode patients (67), although whether associations between relapse and brain volume reduction are causal cannot be determined from naturalistic studies. In another approach to addressing this question, a study of nonrandomized medication discontinuation in stabilized first-episode psychosis patients found that gray matter volume loss over a 12-month period did not differ between medicated and nonmedicated patients, whereas both groups displayed significant brain volume loss compared with healthy controls (68), consistent with the possibility that volume changes reflect progression of illness rather than a medication effect.

Finally, it remains unclear whether brain volume changes during treatment are associated with clinical worsening. In general, gray matter volume correlates with IQ, although the pattern of relationships between brain regions and cognitive domains may be altered in schizophrenia (69). Gray matter volume loss during the course of illness has been associated with cognitive impairment and functional deficits in most studies (70, 71) but not all (72). However, studies of first-episode psychosis consistently find improvement of cognition with antipsychotic treatment, consistent with symptom reduction and practice effects on cognitive tests (73). Most evidence indicates that cognitive impairment in schizophrenia occurs prior to initiation of antipsychotic treatment and then remains stable (73); however, a subgroup of patients may exhibit cognitive decline in the latter stages of the illness (74). In addition, clozapine treatment has been associated with gray matter volume loss that occurred concurrently with clinical improvement (75, 76), suggesting a complex relationship between brain volume changes and clinical course.

The question of whether antipsychotics protect against or contribute to gray matter loss is hindered by a lack of understanding of the molecular mechanisms underlying gray matter loss in schizophrenia. Postmortem brain studies in schizophrenia patients have failed to find histopathological evidence of a neurodegenerative or neurotoxic process, including neuronal loss, ubiquitination, dystrophic neurites, astrocytosis, or microglial infiltrates (77), whereas evidence for oxidative stress is more compelling (78). Most volume changes that occur appear to be the result of reduction in neuropil related to less dendritic branching, lower spine density, and smaller cell body size (79).

A large literature has compared first- and second-generation antipsychotics on the basis of potential effects on oxidative stress, excitotoxicity, and inflammation, and these findings have generally favored second-generation agents (80). However, clinical comparisons of these two somewhat arbitrarily defined classes of antipsychotics have produced inconsistent results for both comparative efficacy and association with gray matter loss, although reduced neurologic side effects with second-generation agents is a consistent finding (81, 82). Tardive dyskinesia, which may be a marker for neurotoxicity, is more common with first-generation agents (83) and has been linked to excessive glutamate transmission and oxidative stress (84). In an early study by Lieberman and colleagues (85), haloperidol was associated with greater gray matter volume loss than olanzapine in a randomized trial in first-episode psychosis patients; the degree of gray matter loss correlated with clinical outcome. In contrast, other studies have found gray matter volume loss associated with cumulative exposure to either first- or second-generation agents, without evidence of a significant difference between the two (62, 86). One meta-analysis found that higher doses of first-generation agents were associated with greater volume loss, whereas higher doses of second-generation agents were associated with less volume loss (87); this effect may be largely attributable to clozapine (71). Because of the substantial heterogeneity among drugs within the category of second-generation antipsychotics, as well as potential effects of dosing, no definitive conclusions can be drawn about relative effects, nor are mechanisms of brain volume loss related to illness or medication clearly established.

Studies of antipsychotic exposure in animal models have drawn attention to possible neurodegenerative effects. In one study, 18 macaque monkeys were treated with vehicle, haloperidol, or olanzapine for 18–24 months administered at doses calculated to approximate typical therapeutic blood levels (88). At the end of the study, monkeys exposed to both haloperidol and olanzapine had a statistically significant 10% reduction in brain volume compared with vehicle-treated monkeys; examination of parietal cortex revealed decreased numbers of astrocytes, but not pyramidal neurons or oligodendroglia compared with monkeys in the control group (89). While of great interest, the results of this study require replication given the cross-sectional design, the small sample size, and the focus on a brain region (parietal cortex) not generally considered to be central to the disease process of schizophrenia. In a second study, rats treated with haloperidol and olanzapine demonstrated significant gray matter volume loss at 8 weeks; postmortem examination revealed a loss of neuropil but no change in the number of cells (90), suggesting that possible early effects on brain volume do not involve neuronal loss. The relevance of findings in rodents and monkeys to the treatment of psychosis in humans is uncertain, both due to species-related differences and because animals lack the pathophysiology of schizophrenia. It is possible that antipsychotics may have deleterious effects on normal brain but protective effects in the presence of schizophrenia-related neuropathology (91).

Despite evidence that antipsychotics can produce D₂ receptor sensitization and D₁ receptor desensitization in animal models, clinical studies have not provided compelling evidence that treatment with antipsychotics worsens the course of illness, increases risk of relapse, or causes cognitive deficits.