The literature search was carried out using the following databases: PsycINFO, Pubmed, and PSYCarticles. A search in Psycinfo and PSYCarticles for English-language articles published since 1980 using the criteria “AB = adolescent onset schizophrenia,” OR “childhood onset schizophrenia,” OR “very early onset schizophrenia,” OR “early onset schizophrenia” yielded 455 results. A search of publication titles and abstracts in PubMed based on the same terms and limitations yielded a total of 485 articles; 96 articles were chosen for further inspection. In addition, studies mentioned in previous review articles were also considered. The process of the literature search is shown in Figure 1.

Due to the interest in performing quantitative analyses based on inferential statistical tests, the following exclusion criteria were used: single case studies, studies reporting only on single or specific parameters (e.g., IQ or mortality) but no overall broad outcome measures allowing a classification into “good,” “moderate,” or “poor” outcome (see below), studies only reporting on mean outcome parameters, studies not based on internationally accepted diagnostic criteria (as reflected in the ICD and the DSM), studies with follow-up time <1 year, and studies with poor description of outcome criteria (e.g., no global functioning scores). In the case of duplicate publishing, data from the sample were included only once in the data set with the study that included the latest selected assessment. The analysis included both retrospective and prospective studies.

A mean age of ≤18 years was required. The majority of the studies only included patients aged <18 years with just a few studies also including 18 year olds 5 11 49 50 , one study including patients aged 19 years 6 , and one study 32 including a few patients aged 20 years at the time of onset; however the latter study was included because of a mean age at onset of 16.8 years. Studies reporting data on pure EOS and studies reporting combined data on EOS and other psychotic illnesses (MIX) were included in the analyses.

A total number of 21 studies were suitable for analysis 1 5 6 8 11 15 18 23 29 32 38 39 40 41 49 50 51 52 53 54 55 .

All data was collected from published material only. The studies were categorized as reporting outcome by use of either a General Functioning Scale (GFS, including Global Assessment of Functioning (GAF, 56 )), Children`s Global Assessment Scale (CGAS, 57 ), and Global Assessment Scale (GAS, 58 ) or Study-Specific Functioning (SSF) outcomes. All GFS studies used scales running from 0 to 100. A total of 10 studies used GFS scales 5 23 32 39 40 50 51 52 53 55 . The GFS studies were categorized as a “poor” outcome (score ≤ 50), “moderate” outcome (score 51–70), or “good” outcome (score >70). Out of these ten studies, five had deviating definitions of the three outcome categories. As described in Table 1, four studies used other cut-offs for “poor,” “moderate,” and “good” 5 23 40 52 . One study 51 even divided the more generally used class of “poor” outcome into deteriorated (< 30) and minimal improvement (30–50). These two groups were combined into “poor” outcome (< 50), whereas all other deviating ratings were taken directly into the analysis.

If not otherwise specified: GAF: Global Assessment of Functioning >70 = Good; 70-51 = Moderate; <51 = poor. SZ: Schizophrenia. SZ-AFF: Schizoaffective disorder. CGAS: Children Global Assesment of Functioning Scale. GAS: Global Assessment Scale. N.A. = Not assessed. PSE = Present State Examination. *Based on N at baseline.

The SSF outcomes were also rated as “poor,” “moderate,” or “good” depending on the outcome as defined in these studies and shown in Table 1. In two studies 6 29 ratings were based on outcome scales, whereas the ratings in the remaining nine studies 1 8 11 15 18 38 41 49 54 were based on categorical outcome measures. Based on the above-mentioned three categories of outcome and using the same cut-off scores, the three authors of the current study performed the ratings independently in each study. There was full consensus among the three authors in the independent ratings of 19 studies, and after two authors agreed in the remaining two studies, full consensus was also reached for these remaining two ratings.

Dropout rates were comprehensively described in 17 studies and ranged from a minimum of 0% 6 23 55 to a maximum of 59% 15 . Reasons for dropout included untraceable subjects, subjects refusing to participate, death, moving out of the area, and suicide. One study, however, included suicide as a measure of outcome 50 , but since most studies did not do so, suicides were subtracted from the data in this particular case in order to have consistent criteria for all ratings.

A total of five predicting variables were considered in the analyses as to their effect on the outcome measures: drop-out rate, type of measures of functioning, duration of follow-up, sex, and time period when patients had been diagnosed. Duration of follow-up was grouped into 1–10 and >10 years. The cut-off was chosen to obtain comparable sample sizes. If the duration of follow-up did not fit into one of these two outcome groups due to varying length of follow-up within the sample, the mean duration was used for classifying the study 5 38 49 51 .

In one study, there was no information on sex distribution 49 , and only a minority of studies reported outcomes stratified for sex 1 11 41 52 . Multiple studies noted sex differences without reporting stratified data. Time period of diagnosis considered studies including patients diagnosed before and after 1970 (<1970+) 1 6 18 23 29 32 49 and studies with all patients diagnosed in 1970 and later (≥1970) 5 8 11 15 34 36 38 40 41 52 53 55 .

Finally, diagnoses were considered by dividing the data-set into studies containing only patients with EOS and studies including both patients with schizophrenia and patients with other psychotic disorders, i.e., psychosis (MIX).

The three categories of “good”, “moderate”, and “poor” were calculated in percentages and rounded to the nearest decimal. In order to take into account the large variation in sample sizes, weighted percentages were calculated by weighting each reported rate with the size of the study group. All analyses were based on adjusted sample sizes at follow-up assessments rather than actual sample sizes after patient recruitment.

Due to consistent and significant deviation of the data from the normal distribution, non-parametric tests were used in the analyses. The effects of the four predicting variables mentioned above on the three outcome measures were analyzed using the Mann Whitney test with Bonferroni adjustments of p-values correcting for multiple testing. Considering five tests, findings were significant at the p = 0.01 level and highly significant at p = 0.002. In addition, effect sizes were calculated using the formula of rho = z/√N), where 0.1 is indicating a small effect, 0.3 a moderate effect, and 0.5 a large effect. Data analyses were performed by use of the SPSS 20 (SPSS, Chicago).

The current review is based on 21 studies containing 716 patients at follow-up. Detailed information on study characteristics and outcome findings is provided in Table 1. The sample sizes ranged from 9 to 81 patients with a mean group size of 44.4 (SD = 19.4). There were considerable differences in design, group size, methods, duration of follow-up, type of evaluation, and missing data. Diagnostic classification changed considerably over the period in which the studies were conducted given the fact that patients had been diagnosed over a wide time period ranging from 1920 to 2010. Since the 1990s, there has been an increasing reliance on DSM-IV and ICD-10 criteria. In 16 studies consisting of 592 patients, the mean age at onset was 14.9 (SD = 1.6) years; five studies reported only age ranges 6 8 11 49 51 .

The mean duration of follow-up varied between 1.5 and 42.0 years (mean = 14.4; SD = 11.4). In 20 studies based on 707 patients, a total of 394 males (56.5%) were included. Repeated follow-up assessments were based on six samples and findings were described in nine articles 8 11 29 35 41 53 54 59 60 . Unfortunately, the data from these studies are not suited for repeated measurement analysis because both the sample sizes between follow-up periods (except 11 ) and the duration of follow-up differed considerably. The total group of studies (N = 21) was divided into a group of EOS studies (N = 422) and a group of MIX studies (N = 294).

In addition to the various descriptive parameters, Table 1 contains columns reporting the outcome criteria used in the various studies, the original outcome ratings, and the outcome (in%) divided into the three categories of “good,” “moderate,” and “poor,” as calculated and rated by the us, which we based on the data in the preceding column containing the original outcome ratings.

As shown in Table 2, studies only containing EOS patients came up with a rate of 15.4% with a “good” outcome, whereas 24.5% experienced a “moderate” outcome, and 60.1% experienced a “poor” outcome. In the MIX samples, the figures were 19.6% with “good” outcome, whereas 33.6% experienced a “moderate” outcome, and 46.8% experienced a “poor” outcome. In each outcome category, though, the variation across studies proved to be remarkably high.

There were significant differences in outcome between the EOS and the MIX samples. A significantly greater proportion of the MIX samples experienced a “good” or “moderate” outcome compared to the pure EOS samples. Consequently, the percentage of patients with a poor outcome was smaller in the MIX samples than in the EOS samples. All effect sizes were moderate.

Dropout rates in 17 studies ranged between a minimum of 0% and a maximum of 59%. This distribution was dichotomized at the median, and studies were classified as having a high (>28%) or a low (<28%) dropout rate. The effect of the attrition on the three outcome parameters was assessed by Mann Whitney tests and showed highly significant differences in the “moderate” and “poor” outcome groups but not in the “good” outcome groups (see Table 3). The rate of “moderate” outcomes was significantly higher in the low attrition samples compared to the high attrition samples, whereas the opposite was the case in the “poor” outcome group with a higher rate of poor outcomes in the high attrition samples; however, the effect sizes were small. In contrast, the three studies with a dropout rate of 0% all experienced high numbers of “poor” outcome 6 23 55 , ranging from 60.5% to 79%.

In order to assess the effect of measures of functioning, studies based on GFS were compared to those using SSF measures. As shown in Table 4, there were highly significant differences in the outcomes based on these two measures of functioning in the “good” and “poor” outcome groups of the EOS samples and the MIX samples. In the latter sample, the outcome also differed significantly for the “moderate” outcome group. In the EOS samples, there were lower rates of “good” and “poor” outcomes in studies based on GFS compared to SSF outcomes. This was also true for the “moderate” outcome groups of the MIX samples. The effect sizes were small for all comparisons. In the 5 studies reporting a mean GFS in EOS patients at follow up based on a total of 199 patients 5 39 50 51 53 , the grand mean weighted for sample sizes of these studies was 47.0.

Findings that deal with the effect of duration of follow-up are presented in Table 5. In the EOS samples, the effect was highly significant for all three outcome groups. Moreover, there was a moderate effect size indicating that follow-up longer than 10 years was associated with a smaller proportion of patients with a “good” and “moderate” outcome and a larger proportion of patients with a “poor” outcome. In the MIX samples, differences for “good” and “moderate” outcomes were highly significant, and differences were significant for poor outcomes; however, the effect sizes were small. In these samples, the rate of both “good” and “poor” outcomes was increasing with longer follow-up periods, whereas the rate of “moderate” outcome was declining.

Direct calculations could only be made on the basis of five studies reporting separate results for males and females (4 with MIX and 1 with EOS patients). As Table 6 shows, highly significant differences were found for “good,” “moderate,” and “poor” outcome. These results indicate a generally less favourable outcome for males who less frequently than females experienced a “good” or “moderate” outcome and more frequently experienced a “poor” outcome. The effect sizes were small to moderate.

To further investigate the effect of sex in a larger sample, we compared 6 studies with <50% males to 14 studies with >50% males; findings are shown in Table 7. Differences were significant to highly significant on the various levels of outcome. There is a clear indication that both EOS and MIX studies containing a majority of males generally experienced a less favourable outcome. The proportion of “good” and “moderate” outcomes was lower in studies based on a male predominance, whereas the proportion was higher in the “poor” outcome groups.

The data-set allowed a dichotomization into two groups of studies, namely, those including patients diagnosed before and after 1970 and those where all patients in the sample were diagnosed in 1970 or later. Table 8 provides a comparison of the outcome of these two groups. In the EOS samples, there is a highly significant decline of “good” outcomes in all patients diagnosed in or after 1970; however, the effect is only small. In contrast, there are large effects indicating that the proportion of “moderate” outcomes increased significantly, and the proportion of “poor” outcomes decreased significantly over time. Taking all three levels into account, the overall outcome improved significantly over time.

Note: <1970+: studies including patients diagnosed before and after 1970.

≥1970: Studies containing patients diagnosed in 1970 and later.

There was only one MIX study containing patients diagnosed before 1970. On the other hand, there were clear moderate time period effects indicating highly significant improvements with increasing proportions of “good” and “moderate” outcomes and decreasing proportions of “poor” outcomes.

In the current review, we discovered that 15.4% of EOS patients experienced a “good” outcome, 24.5% experienced a “moderate” outcome, and 60.1% experienced a “poor” outcome. Clearly, these findings indicate that EOS is still a mental illness with a rather unfavourable prognosis; this conclusion is in accordance with previous reviews 5 7 12 21 31 63 64 . On the other hand, these previous reviews were based on non-aggregated data, and they did not employ rigorous data analyses as the authors do in the current review.

Furthermore, from the current analyses, it became evident that studies of patients with EOS show a worse prognosis than studies containing both patients with EOS and patients with other psychotic disorders (MIX). Unfortunately, separate analyses of the outcome of the various psychotic disorders were not feasible. In addition, differences in time points of measurement in the two samples may have been operant. Nevertheless, the different outcome in the two groups may serve as some indirect evidence that other psychotic disorders, i.e., schizoaffective disorders, schizophreniform or bipolar disorders with psychotic features, take a less serious course in terms of chronicity and functioning because all analyses based on the mixed psychotic samples showed a less severe outcome than the pure EOS samples. This finding is in accordance with similar studies in adults 2 65 .

When considering the impact of dropout rates, the general findings on outcome may be only slightly different than one would expect without any attrition in the samples. In samples with high attrition rates, patients with a “moderate” course of the disorder were less likely to be followed up, and those with a “poor” outcome were more likely to show up at follow-up assessments at the various sites, whereas there was no attrition effect on the rate of “good” outcomes. In contrast, it is unclear whether the rate of “poor” outcomes would be different. On the one hand, our analyses showed that the rate of “poor” outcomes declined significantly with low attrition rates. On the other hand, three studies without any attrition showed an increased rate of “poor” outcomes. However, one has to keep in mind that the effect sizes for attrition were only small. High dropout rates are very common in psychiatric services with estimated rates ranging from 20 to 60% 66 , which proves to be in line with the findings in the current review with dropout rates between 0 and 59% and a median of 29%.

In contrast to EOS, the outcome in studies of adult patients is generally more favourable 5 . Hegarty et al. 67 reviewed 320 adult studies from 1895 to 1992 (more than 50,000 patients in total) and found that approximately 40% improved considerably during follow-up. Jobe & Harrow 2 reviewed nine North American studies and the WHO-coordinated International Study of Schizophrenia (ISoS), all with a follow-up period of 10 years or longer, and concluded that, although adult patients with schizophrenia as a group have a worse outcome than other psychiatric patients, only a few patients show a progressive deteriorating course; depending on the strictness of the criteria used for diagnosis, 21-57% experience a “good” outcome. The ISoS compared long-term follow-up studies (10–15 years) from 14 culturally diversely treated incidence cohorts and four prevalence cohorts, totaling 1633 subjects, and found that approximately 50% experience a “good” outcome 65 .

A recent international study that examined outcome after three years of follow-up in adult outpatient schizophrenia (N = 11.078 from 37 countries) found that 66% achieved clinical remission measured with the CGI, whereas only 25.4% achieved functional remission defined as good social functioning for 6 months in terms of occupational/vocational status, independent living and active social interactions 68 . There were large regional differences in the study. Patients in Europe were less likely to achieve clinical remission but were doing better in regards to functional remission. The general outcome both in the EOS group and the mixed psychotic group in the current review clearly shows that schizophrenia and psychosis originating in childhood and adolescence on average follows a worse course than AOS. In comparison to other disorders originating in childhood or adolescence, EOS stands out by way of its particularly poor course. For instance, the outcome seen with eating disorders is much better as is shown by similar types of analyses by the senior author 69 70 .

This conclusion is also supported by a recent large cohort study from Israel with 12.071 participants. This study found that earlier onset corresponds linearly with the severity of the course of the disorder and appears to have some prognostic impact 71 . Young age at onset might have a detrimental effect on outcome because of impact at very crucial times of development and neurobiological maturation in childhood and adolescence, which prove to have more lasting effects in terms of both cognitive and psychosocial impairments 1 32 35 72 .

So far, unfortunately, there are only a small number of outcome studies based on VEOS patients only, with an over-representation of females, whereas there are more studies with a varying range of age at onset within the defined EOS age range. Furthermore, there is not a single study based exclusively on patients with adolescent onset schizophrenia; thus, there are no real solid data for a comparison of the outcome of VEOS. Clearly, more detailed analyses will be needed. Given the low prevalence rates of VEOS, only collaborative studies across several sites could arrive at sample sizes needed for a differential look at the effects of age, sex, clinical features, or treatment effects on the outcome of VEOS.

The current study is the first to make use of analyzing the impact of specific measures of functioning. Not surprisingly, the advent of global measures of functioning since the seventies also had an effect on the studies of the outcome of EOS. In contrast, a few studies continued with an older tradition to define study-specific functioning outcomes. Thus, a comparison of these two different traditions became possible. In the pure EOS studies, there were relatively small effect sizes, indicating that studies based on the more recent GFS arrived at slightly lower rates of both “good” and “poor” outcomes and no differentiation in the “moderate” outcomes than studies based on SSF outcomes. Accordingly, in EOS the overall pattern is clearly not more favourable for one of these two types of outcome. In the MIX studies, the effects were also small though more clearly showing a general pattern of less favourable outcomes based on GFS rather than on SSF assessments. Thus, the two analyses point to different findings in the two types of studies. In other words, the heterogeneity of the MIX samples favour the SFS outcomes in which the measurement might have tipped closer to the differences in the diagnostic composition of the samples.

However, this interpretation is only an assumption that needs further examination. Particularly, both the validity and the reliability of these measures need to be studied in greater detail. So far, this has been tested only in parts for some of the GFS measures in general child and adolescent psychiatry patients 73 but not specifically in patients suffering from schizophrenia. In particular, the GAF confounds symptoms and functioning with lower ratings driven by symptoms, so someone who is symptomatic but functional will receive a misleadingly low rating.

In general, there is very little information on the impact of intervention on the outcome in EOS, even though all patient samples were seen clinically and received treatment. With the exception of a single study 72 , all studies provided treatment as usual. In a recent intervention study with follow-up based on the Australian Early Psychosis Prevention and Intervention Centre (EPPIC) study, the authors found an increase in GAF score with a mean GAF score of 64 at follow-up 42 . By comparison, Oie et al. 74 found a mean GAS score of 47.7 in their EOS group containing 15 patients assessed when they were clinically stable on antipsychotic medication and followed up for 13 years. Moreover, Kao et al. 75 found a mean GAF score of 47 in 19 EOS patients after 1 year follow-up, and Gochman et al. 76 found a mean CGAS score of 43.6 after at least 8 years of follow-up. In the single intervention study included in the current review, the mean GAF score was 35 in the intervention group and 24 in the control group 55 . Only Ledda et al. 53 found a mean CGAS score of 62.1, which is quite comparable to the finding of the EPPIC study 42 .

Nevertheless, in the latter study 42 , the attrition rate was large (22/63) in the total EOS group and well explained only in a single person who committed suicide. It is unclear whether the 21 other patients that were not followed-up represent a subgroup with less favourable outcome because the authors did not provide a thorough attrition analysis. Thus, the claim of the authors that their outcome findings are superior to previous outcomes is not yet substantiated.

The current analyses revealed, with small to moderate effect sizes, that across the three levels, the outcome deteriorated with longer follow-up periods (>10 years) in the EOS samples, but the association was rather curvilinear in the MIX samples with both “good” and “poor” outcomes increasing at the expense of “moderate” outcomes with longer follow-up periods. These differences again point to the already noted different course of the other psychoses, apart from schizophrenia. Nevertheless, the present findings need to be interpreted with caution because the two follow-up periods of ≤10 and >10 years are rather broad and reflect limitations of the data not allowing a more fine-grained analysis. Furthermore, the analysis was based on a series of cross-sectional rather than longitudinal studies. Unfortunately, among the 21 outcome studies of the present analysis, there are only six samples that were assessed repeatedly for follow-up and were described in 9 articles 8 11 29 35 41 53 54 59 60 . The study by Krausz and Müller-Thomsen 11 showed an increase in the proportion of “good” outcome from the first follow-up at 5 years to the second follow-up at 11 years (19 to 31%), whereas the rate of “poor” outcomes declined (74 to 59%) with a rather constant proportion of “moderate” outcomes (7 to 10%). These findings are in contrast to the findings of the current review. Lay et al. 41 ) studied a mixed psychotic group that had been previously followed-up 35 . Unfortunately, approximately one third of the group had dropped out in between the two assessments, so it is unclear whether or not the slight shift from “moderate” to “poor” outcome (from 32 to 36%) is a valid finding. None of these longitudinal studies made use of inferential statistical tests of any significant change of the course of the disorder over time.

The current review supports the notion that male sex carries a less favourable prognosis in EOS but also in MIX samples. Nevertheless, as described in the methods section, there were profound limitations in the data for a proper analysis of sex effects. With a few exceptions 1 11 41 52 55 , the vast majority of studies did not stratify outcome by sex; thus, two rather restricted types of analyses had to be performed. The direct comparisons of a small subsample suitable for direct comparisons clearly favoured females in terms of having a better outcome. The supplementary analysis based on a larger sample compared the outcomes of samples with either less or more than half of the samples being comprised of males. The findings were in line with the previous results indicating that male sex is a negative prognostic factor.

When looking at the various studies considered in the current review, one may see that some studies reported a tendency for worse outcome for males 1 5 6 8 11 15 , but only two of these provided statistically significant differences 5 8 . One study found a specific “poor” outcome in females which proved to be not statistically significant 49 . In contrast, most of the 20 studies either reported no prognostic impact of sex 39 40 or did not specify or mention sex in relation to outcome measures 18 23 29 32 38 41 51 53 54 . One study noted that the risk of suicide was increased about 30 times in males 50 .

In the 21 studies listing the distribution of sex, the average proportion of males was only 55%, a surprising discovery given that schizophrenia usually has an earlier onset among males than among females 17 and that late onset after age 45 is more common among females 24 . Especially with regard to VEOS onset, the literature points to a male predominance 12 with a ratio of approximately 2–2.5:1 17 24 64 ; however, in the two studies of VEOS, female sex was dominant in both series of patients 23 29 . In conclusion, there is some indication of a potential sex bias in the outcome studies in terms of containing more females than expected. Potential explanations include a higher dropout rate of males from outcome assessments due to less compliance and/or a higher mortality rate.

The time span of the original diagnosis of the patients varied enormously between 1920 and 2010. During this period, major changes in the understanding of schizophrenia including the nosological classification, assessment, and intervention took place. Thus, our analysis took potential time period effects into account. The data-set was dichotomized into studies containing patients diagnosed before or after 1970 and patients all diagnosed in 1970 or later. This grouping was not ideal because it was still based on considerable heterogeneity in terms of the time when the patients were diagnosed. Nevertheless, it represented a feasible and pragmatic approach and reflected the fact that some major changes in the classification of schizophrenia both in the ICD and the DSM took place in the seventies.

The findings indicated that the overall outcome in EOS and even more clearly in MIX samples improved over time; thus, one may argue that the progress in treatment and rehabilitation of schizophrenia might have had a beneficial effect for those who were born and diagnosed later. In summary, one may also conclude that the overall relatively poor long-term outcome of EOS is, in part, due to the inclusion of studies containing patients who had been diagnosed many decades ago.

First, we decided to include only studies published after 1980, assuming that these studies would reflect a rather common international frame of understanding of the nosology of schizophrenia and psychoses. Even with this restriction, though, there was a large time span over which patients had been diagnosed. Even more importantly, there may have been general problems with the recruitments of the samples. There might have been a bias both at the time of the first clinical presentation and of follow-up assessments. The less severely affected patients with a rapid remission of symptoms may not have been included at the beginning of the studies. Furthermore, it is not fully clear which effect no attrition might have had in particular on the rate of “poor” outcomes.

In the current review, we have used the categories of “good,” “moderate,” and “poor” outcome. These categories are commonly used in the outcome literature of various mental disorders. While the three authors of the current review showed excellent convergence in the outcome ratings of the various studies regarding this classification, one may argue that the cut-offs of these three outcome groups are debatable. Nevertheless, our cut-offs (>70 = good, 51–70 = moderate, and <50 = poor) imply some face validity because they are clearly demarcating major thresholds in functioning on GFS measures. Six GFS studies were based exactly on these definitions, whereas five studies used slightly different definitions. Two studies 40 51 used a lower cut-off of >60 rather than >70 for the definition of “good” outcomes, whereas three studies 5 23 52 requested a lower cut-off of <40 for “poor” outcomes. Thus, these differences imply a less strict definition of the outcome, so our findings might have been slightly better if we had accepted these definitions. Even among these five studies, though, there is no fully congruent set of definitions. Thus, our procedure was not only plausible in terms of the construction of the various GFS measures but also served as a good compromise considering the heterogeneity of definitions of “good,” “moderate,” and “poor” outcomes.

Some of the limitations in reviews of schizophrenia, as stated by Jobe & Harrow 2 and Castle and Morgan 77 , are also relevant for the current analyses. The comparability of follow-up studies is compromised by differing criteria for diagnosis and outcome variables, sample selection (i.e., bias between inpatient and outpatient indexing), varying duration of follow-up, differences in the American and European tradition of diagnostic approaches, and prospective and retrospective designs leading to different preciseness of data acquisition. Furthermore, many studies have used different assessments for diagnosis and outcome. Various studies have lost patients due to suicide, which were counted as drop-outs; however, one could argue that suicide in terms of outcome should be listed in the “poor” outcome group, as suggested by Jarbin 50 .

Furthermore, the lack of any clear data on mortality rates in EOS and VEOS is a shortcoming of outcome studies that should be addressed in future studies. Since EOS and VEOS are very rare, patients often come from a large geographic area to the specialist research units; thus, some patients travel far to be part of the study. This might be a bias in terms of only the most affected individuals will travel this far to be part of a study, which also indicates that the patients who have the best outcome might drop out.

Finally, no firm conclusions can be made thus far as to the effects of interventions, and it is unclear whether the large variation is due to different interventions, varying clinical manifestations, or an interaction of both. As with the study of other disorders, research on the effects of intervention on course and outcome is most neglected. Further studies are clearly needed.