|
Data Extraction
We designed a data collection form to extract several key components from
relevant primary articles. These items included bibliographic information
on the article; specific study design characteristics, such as the study
population and the methods used to obtain the study sample; case-finding
techniques; and diagnostic criteria used. We extracted prevalence and incidence
data including overall, sex-specific, and age-specific rates from each
article, and when relevant, we noted the prevalence period assessed in
the study (for example, point, 1-year, or lifetime). Where serial publications
were based on the same data, only the most recent and definitive results
were extracted and included in the analysis.
Data Analysis
Qualitative Analysis
We performed a qualitative analysis of the studies’ data to summarize the
rates. We assessed the following variables to elucidate any observed differences
among rates:
1. Study population. This includes the country or region studied, the year(s)
in which assessment of prevalence or incidence was carried out, the age
range studied (the entire general population or adults only), the type
of community studied (city, town, or rural area), and the type of population
covered (the broad population or a specific ethnic group).
2. Sample characteristics. These include the sampling approach (probability
sample based on census lists, electoral roll, or household survey), the
sample size, the response rate, and the types of residence included (for
example, households, nursing homes, or hospitals).
3. Case ascertainment and diagnosis. This includes the diagnostic instrument
used, the qualifications of the person administering the instrument (for
example, lay interviewer or clinician) the information source used to ascertain
cases (interview with subject, informant, or physician, or medical records)
the means by which diagnosis was established (for example, by clinician
or by computer algorithm), the diagnostic criteria used, and the case definition
used (broad or narrow).
Estimation of Pooled Best-Estimate Rates
Each set of rates was pooled based on a Bayesian approach to metaanalysis,
using the Fastpro software program (version 1.7) by Eddy and Hasselblad.
Readers interested in a more detailed discussion of this approach should
refer to Eddy and others (6). We calculated the pooled or best-estimate
of effect values (called “median of the posterior” in this method) using
Jeffrey’s prior and a hierarchical model. We used a random- effects model
in the expectation that there would be significant heterogeneity of rates
among studies included in the pooled rate.
Heterogeneity Analysis of Pooled Rates
Because we expected significant heterogeneity of rates within the studies
pooled, we defined a systematic method for exploring why studies reported
varying rates. Each of the pooled rates was analyzed for heterogeneity
using chi-square tests according to Fleiss’ method (7). This involves grouping
the proportions according to the methodological variables that may be contributing
to differences among them. We initiated the procedure by listing the proportions
in descending order, dividing them into 2 groups by the median value, and
categorizing them according to each of the variables that were defined
a priori as being of interest (for example, country studied). Rates from
individual studies were then pooled according to variables that had the
highest magnitude of differences among groups. Figure 1 outlines the procedure
used to examine heterogeneity within the best-estimate pooled rates.
Results
Description of Studies
From the citations and abstracts generated by the initial electronic search,
we identified 26 prevalence studies and 8 incidence studies that potentially
met inclusion criteria; additionally, we identified 6 review papers. We
retrieved the full texts of these articles. To complete the initial stage
of study identification, we searched all reference lists of identified
studies and reviews, generating an additional 13 potential prevalence studies
and 45 potential incidence studies, for which we also obtained the full
texts.
Of the 39 prevalence papers for which full-text articles were reviewed,
we excluded a total of 15 articles; 13 studies did not meet eligibility
criteria, and 2 presented duplicate data. Thus, we obtained data from 24
papers that met our eligibility criteria (2,8–30). These reported the results
of 18 unique primary investigations of prevalence of schizophrenia and
related disorders. Table 1 documents our reasons for excluding those prevalence
studies fulfilling all but 1 inclusion criterion (31–42). Of the 53 papers
reporting incidence, 44 were excluded; 40 studies did not meet eligibility
criteria, and 4 reported duplicate data. Thus, we could include data from
9 publications reporting incidence of schizophrenia (2,5,43–49), describing
8 distinct primary investigations of incidence. Seven of the excluded studies
almost met inclusion criteria; Table 1 presents reasons for their exclusion
(41,50–55).
Prevalence Studies
We present findings for the 16 prevalence papers reporting 1-year or lifetime
prevalence rates only (Table 2 and 4), because relatively few studies
reported data for point prevalence (2,10,12,17,20,24,27,29) or 6-month
prevalence (13,17,20,24,30). We present findings separately for schizophrenic
disorders, schizophrenia, and schizophreniform disorder because these were
the diagnostic categories for which prevalence rates were most commonly
reported, and also because we wish to maintain important distinctions between
disorders that are occasionally conflated. We analyzed data only when 3
or more rates were reported, because this was the minimum number of values
required to produce pooled rates.
All but 1 of the studies shown in Table 2 are community surveys that examine
a series of mental disorders, using samples ranging from approximately
500 (16) to 20 000 (18) in size. One study involves a complete survey of
key informants based on an at-risk population of about 65 000 (9). For
each of these studies, the percentage confidence interval (CI) width or
error rate for estimated prevalence at a 95%CI may be calculated using
the formula provided by Kelsey and colleagues (56, p 282). For the most
part, studies used either the Diagnostic Interview Schedule (DIS) or the
Composite International Diagnostic Interview (CIDI), administered by trained
lay interviewers, and applied algorithms to derive diagnoses.
Qualitative Analysis
1-Year Prevalence. For schizophrenic disorders, 1-year prevalence rates
ranged from 0.2 per 100 in Christchurch, New Zealand (20) to 1.0 per 100
in the US ECA study (18), a 5-fold variation. For schizophrenia, 1-year
prevalence ranged from 0.2 per 100 in the Netherlands (8), in New Zealand
(20), and in rural villages in Taiwan (22) to 0.9 per 100 in the US ECA
study (18), a more than 4-fold difference. If the ECA study is excluded
as an outlier, the 1-year prevalence rates vary up to 0.42 per 100, a 2.1-fold
difference. The highest rates, observed in the US ECA study, followed by
the Swedish study (9), may be partly explained by the inclusion of institutionalized
cases. One-year prevalence rates for schizophreniform disorder ranged from
0 per 100 in New Zealand (20) and Edmonton (24) to 0.1 per 100 in the US
ECA study (18).
Lifetime Prevalence. For schizophrenic disorders, lifetime prevalence rates
ranged from 0.4 per 100 in New Zealand (21) to 2.2 per 100 in Finland (19)
and in the US National Comorbidity Study (NCS) (11), a 5.5-fold difference.
The lifetime prevalence rates for schizophrenia varied considerably, ranging
from 0.12 per 100 in Hong Kong (15) to 1.6 per 100 in Puerto Rico (26),
a difference of over 13-fold. The Puerto Rico, US ECA, and Finland studies
included institutionalized cases, which may in part explain the higher
rates from these studies. It is notable that low lifetime prevalence rates
were reported in all reviewed studies conducted in Asian countries (Hong
Kong, Taiwan, and Korea). The lifetime prevalence rates for schizophreniform
disorder ranged from 0 per 100 in Taiwan (22) to 0.2 per 100 in both the
US ECA (18) and Puerto Rico (26) studies. If the outlying rates reported
in Taiwan are excluded, all studies reported rates in the range of 0.06
to 0.2 per 100, a 3-fold variation.
Sex-Specific Lifetime Prevalence. Table 4 illustrates findings from studies
reporting sex-specific lifetime prevalence rates for schizophrenia. Male
and female subjects were found to have very similar rates across most studies.
Although the studies conducted in Taiwan and Korea both show lower lifetime
prevalence rates for female subjects, compared with male subjects, differences
are not reported to be significant.
Figure 1 Steps for examining heterogeneity in best-estimate pooled rates
Estimation and Heterogeneity Analysis of Pooled Best-Estimate Rates
Schizophrenic Disorders. The best-estimate rates for 1-year and lifetime
prevalence are 0.60 per 100 and 1.45 per 100, respectively (Table 2). Heterogeneity
analysis revealed significant differences across each set of proportions.
For both 1-year and lifetime prevalence, the variable found to have the
greatest magnitude of difference across proportions was type(s) of residence
sampled, although this variable likely does not explain much of the heterogeneity
in rates (Table 5). The pooled 1-year rate for studies that sampled both
households and institutions was approximately twice that for studies assessing
household populations only (0.85 vs 0.43 per 100, respectively).
Schizophrenia. The best-estimate rates for 1-year and lifetime prevalence
are 0.34 per 100 and 0.55 per 100, respectively (Table 2). Heterogeneity
was found across both 1-year and lifetime rates. As shown in Table 5, the
pooled 1-year and lifetime rates for studies that sampled both households
and institutions there was approximately 3 times those of the studies including
only household samples. The pooled lifetime rate for Asian studies (0.25
per 100) was almost 4 times lower than that of non-Asian studies (0.88
per 100).
Schizophreniform Disorder. The best-estimate rate for 1-year and lifetime
prevalence was found to be 0.09 per 100 and 0.11 per 100, respectively
(Table 2). Heterogeneity analysis demonstrated significant differences
across lifetime prevalence rates only. As shown in Table 5, the pooled
rate for studies conducted in Asia was found to be 6 times lower than that
of studies carried out in other parts of the world (0.03 vs 0.18 per 100,
respectively).
Incidence Studies
Table 3 presents the findings for the 8 studies for which 1-year incidence
rates of schizophrenia could be extracted. Studies reporting rates for
schizophrenia spectrum disorders (2) were not included, because this broader
category was not commonly reported.
Most studies shown in Table 3 are key-informant surveys, with at-risk populations
ranging from approximately 73 000 (43) to 5000000 (5). The only study that
involved a community-based sampling design was the US ECA study (49), with
a sample size of approximately 20 000. For the most part, the reviewed
studies used the Present State Examination (PSE) and an algorithm such
as the CATEGO system to extract diagnoses (57). Only the ECA study used
lay interviewers; all other studies used clinicians.
1 | 2 | 3 | 4 | 5
|
