There is a wealth of data available on the prevalence of schizophrenia—a total of 1,721 estimates from 188 studies were identified in this systematic review. These estimates were drawn from 46 countries, and were based on an estimated 154,140 potentially overlapping prevalent cases.
The median prevalence estimates for persons were 4.6 per 1,000 for point prevalence, 3.3 for period prevalence, 4.0 for lifetime prevalence, and 7.2 for LMR. These estimates are congruent with an earlier narrative review of 70 studies by Torrey [8
], who reported an overall prevalence estimate of 4.6 per 1,000. Key policy documents have correctly estimated the point prevalence of schizophrenia at about four per 1,000 [2
]; however, the Diagnostic and Statistical Manual of Mental Disorders,
fourth edition (DSM-IV) [3
], reported that the lifetime prevalence of schizophrenia is “usually estimated to be between 0.5% and 1%.” This overestimate is often repeated in textbooks [226
]. As with the misunderstandings about the incidence of schizophrenia [21
], this is another example where the research community needs to review their belief systems in the face of data. It is reasonable to assume that lifetime prevalence estimates for schizophrenia would be higher than point estimates. Surprisingly, the data in this review do not support this assumption. While outside the scope of the current review, the findings raise interesting research questions about factors that may influence prevalence (e.g., recovery, suicide, or other forms of early mortality). Indeed, it is curious that the identification of the onset of psychotic disorders has received so much recent attention [227
], while we still struggle to understand the offset of schizophrenia. Point and period prevalence estimates assume that we can identify when someone has recovered from an illness. Recovery from schizophrenia clearly occurs [229
], but it is unclear whether those who are free of positive symptoms but who have mild residual disability should be counted as “active” cases or not. The definitions of recovery versus persistence are multidimensional, and future prevalence studies will benefit if these definitions can be operationalized.
The median LMR estimate was 7.2 per 1,000, which is consistent with two other narrative reviews. Fremming [232
], who reviewed 18 studies conducted in central Europe between 1926 and 1938, reported a mean LMR of 7.4 per 1,000, while Gottesman and Shields [233
] reported a mean LMR of 8.0 per 1,000 in their classic review. As predicted, LMR estimates were significantly higher than lifetime estimates, which reflects the different heritage of these two indices. It is reasonable to assume that the oft-quoted statistic that “schizophrenia affects about one in a hundred” derives from LMR data (see [234
]). However, one in a hundred is an overestimate—our systematic review agrees with two previous reviews showing that the LMR for schizophrenia is between seven and eight per 1,000. While the arithmetic mean value of 11.9 per 1,000 is more consistent with the “one in a hundred” dogma, the median is a more appropriate measure of central tendency for this skewed distribution. If we wish to provide the general public with a measure of the likelihood that individuals will develop schizophrenia during their lifetime, then a more accurate statement would be that “about seven to eight individuals per 1,000 will be affected.”
While there has been considerable debate about whether or not the incidence of schizophrenia varies between sites [21
], there is a tacit understanding that the prevalence of schizophrenia is variable. For example, in an earlier review by Eaton [5
], a 12-fold variation in point and a 10-fold variation in lifetime prevalence were noted. A recent systematic review by Goldner et al. [12
] also observed a 13-fold variation in lifetime prevalence of schizophrenia. Based on the central 80% of the estimates (10% to 90% quantiles), the present review found that the different types of prevalence estimates had from 3.4-fold (point) to 4.6-fold (period) variation. The use of the 10% and 90% quantiles to define the central segment of the distribution means that our reporting of the variability of estimates is more conservative than other commentators (i.e., we have ignored 20% of the distribution in the tails). If we had included all data points, the range of prevalence estimates would have been much higher. Regardless of whether this variability is labeled “narrow” or “prominent,” the task for the researchers is to determine how much of this variation is a function of measurement error versus “true” underlying variation. With respect to measurement error, it should be noted that this study found that quality of the study does significantly influence prevalence estimates. Future studies could explore the impact of quality on the variation in prevalence estimates.
Sex and Schizophrenia
One of the unexpected findings of this review was that there was no statistically significant difference in prevalence estimates between males and females. In our previous study of incidence of schizophrenia we found a male:female risk ratio of 1.40 [1
]. Because narrative reviews conclude that the course of the illness tends to be more severe in men than in women [235
], we assumed that this would be reflected in a higher prevalence in males than females. The lack of coherence between (a) the sex differences found in the incidence of schizophrenia, (b) the presumed difference in course of illness, and (c) the identified lack of difference in prevalence warrants closer scrutiny.
Economic Status and Schizophrenia
In keeping with our hypothesis, the prevalence of schizophrenia is lower in developing nations than in developed nations. However, we urge caution in the interpretation of these data. The use of a single economic variable is a crude way to assess a complex and multidimensional concept. Furthermore, the median prevalence estimates for emerging economies are numerically higher than those for the richest countries. While not statistically significant, the results did identify many prevalence studies from the developing world where females outnumbered males. Recently, a study from China examined whether this unexpected sex ratio was due to differential suicide rates in males with schizophrenia [24
]; however, this did not seem to explain the female excess. Our findings lend weight to the commentary by Ran and Yu-Hai Chen [25
], drawing attention to the different features of schizophrenia in the developing world. Overall, the findings suggest that factors that influence the course of illness of schizophrenia in men and women differ around the world. Regardless of the mechanisms underlying this possibility, the findings highlight the importance of using systematic techniques to identify data; 17 studies included in this review were only available in languages other than English. We speculate that the results of past narrative reviews may have been biased towards data from developed nations. From a wider perspective, the findings reinforce the importance of encouraging more research from poorer countries [236
Urbanicity and the Prevalence of Schizophrenia
In the previous systematic review of the incidence of schizophrenia, we found that urban sites had significantly higher incidence rates of schizophrenia than mixed urban/rural sites (there were too few pure rural sites to make the direct urban versus rural comparison) [1
]. Contrary to our expectations, the prevalence of schizophrenia did not differ according to urbanicity. While suggests that mixed urban/rural sites have higher prevalence estimates than pure urban and rural sites, this study found, in fact, that there was no significant difference between urban, rural, and mixed sites. Perhaps the inclusion of many sites from the developing world in this review has confounded the expected urban/rural gradient. This will be examined in more detail in future analyses.
Migrant Status and the Prevalence of Schizophrenia
As predicted, prevalence estimates for migrant groups tend to be higher than estimates for native-born populations. This finding is consistent with past systematic reviews of the incidence of schizophrenia [1
]. Migrant studies are prone to a range of methodological issues (e.g., differential pathways to care, diagnostic inaccuracies due to language and cultural practices, and uncertainty about the denominator required for the calculation of proportions). While the prevalence estimates included in this systematic review may share common biases, the increased prevalence of schizophrenia in migrant groups found in this study adds weight to the argument that migrant status is an important risk factor for schizophrenia.
Quality Scores and Other Special Groups
Reassuringly, studies that had higher overall quality scores tended to identify more cases, and thus generate higher prevalence estimates than lower quality studies. Future studies will explore whether the findings based on the overall studies persist in the subgroup of studies in the highest quality tercile.
With respect to the studies included in the category “other special groups,” the estimates are not readily comparable, but it is interesting to note that these studies reported a wide range of prevalence estimates (e.g., high in homeless populations and low in certain religious groups). Future publications will examine these groups in more detail.
Based on our experience with previous systematic reviews, we acknowledge that we may have missed studies and/or made data entry errors. We encourage readers to inform us of missing studies or errors in the data. Updated lists of relevant studies and raw data will be available from the authors. Furthermore, in the absence of clear guidelines on how to synthesize descriptive studies [26
], many of the rules we used to filter studies and extract data were necessarily ad hoc. In the future, researchers may wish to reanalyze the dataset using different criteria, and perform sensitivity analyses related to these choices.
Two of the prevalence types (LMR and inpatient-census-derived data) had distributions for persons that were higher than distributions for both males and females separately. This pattern, which is difficult to explain, was also noted in some of the previously published incidence distributions [1
]. The impact of quality scores on this pattern will be assessed in future studies.
The planned sensitivity analyses were conducted on combined data, a strategy that reduced the number of comparisons substantially (one combined analysis versus five analyses on each of point, period, lifetime, LMR, and NOS data). However, the combined prevalence estimate included studies that contributed more than one prevalence type (e.g., one study could contribute both point and period prevalence estimates). Of the 94 studies, eight contributed more than one prevalence type to the combined prevalence estimates. While the analytic technique controlled for within-study variance, the combined dataset is not based on discrete data (in contrast to the prevalence-type-specific analyses).
It was disappointing that standard errors could be allocated to so few prevalence estimates (26%). Despite this, in the future we plan to undertake a traditional meta-analysis based on this subset of estimates in order to compare the pooled estimate values with those presented in the current study.
Concerning the analyses for urbanicity, the estimates from mixed urban/rural studies are likely to be very heterogeneous. Indeed, we allocated studies to the mixed category if there was any possibility that rural sectors were included. This bias would have made any true difference between urban versus mixed urban/rural more difficult to detect. There are good reasons to review the findings for both urbanicity and sex ratio more closely when categorized by economic status. Such analyses may help generate hypotheses for future analyses, but researchers need to be extremely cautious when systematic reviews are subjected to excessive data analyses (i.e., “data torturing” [238
]). The contributing studies were not designed to test many of the hypotheses examined in this review, therefore researchers must be frugal in the use of planned sensitivity analyses, and cautious in the interpretation of the results. However, researchers are encouraged to freely explore the full data to examine additional research questions.
While there is substantial variation between sites, generally the prevalence of schizophrenia ranges from four to seven per 1,000 persons, depending on the type of prevalence estimate used. Countries from the developing world have a lower prevalence of schizophrenia. Overall, the prevalence of schizophrenia does not vary between the sexes; however, the data suggest that sex ratio of prevalence estimates may vary between sites more than previously believed. While the incidence of schizophrenia is higher in urban than rural settings, this is not reflected in the overall prevalence data. The prevalence of schizophrenia is higher in migrants than native-born individuals.
Regardless of the exact magnitude and precision of prevalence estimates, the numbers speak to a deeper, human dimension. Many people with schizophrenia have persisting symptoms, despite the best mix of interventions we can offer. This sobering reality has also emerged from research about “best buys” with respect to the cost of averting disability [239
]. For schizophrenia, with the current mix of interventions we can only reduce 13% of the burden. If we improve efficiencies within the current services, we can do somewhat better (22%). In a utopian world, even if unlimited funding were available, three-quarters of the burden of schizophrenia would remain unavoidable [240
]. This is a powerful argument for investing in applied and basic research.
As with its companion study on the incidence of schizophrenia [1
], we hope that the current review will populate the “epidemiological landscape” with data, and that this enriched environment will select the fittest (most heuristic) hypotheses [21
]. The epidemiological landscape of schizophrenia is no longer terra incognita—many of its contours have been mapped out. We can gain traction on this landscape and use the identified gradients to generate candidate risk factors for future research [241
]. Equally, these systematic reviews have brought into focus the gaps in our knowledge—parts of the map “do not fit.” Paradoxes such as these can be powerful catalysts for advancing knowledge.