Introduction
Emil Kraepelin (1909–1915) was the first of many researchers to suggest that men develop schizophrenia at a younger age than women. Subsequent studies addressing this gender difference in age at onset have reported findings which vary considerably; in some studies the difference is as great as 5 years (Gureje & Bamidele, Reference Gureje and Bamidele1998), and in others it is absent (Murthy et al. Reference Murthy, Janakiramaiah, Gangadhar and Subbakrishna1998; Gangadhar et al. Reference Gangadhar, Panner Selvan, Subbakrishna and Janakiramaiah2002). Thus, in a review of 53 studies, Angermeyer & Kuhn (Reference Angermeyer and Kuhn1988) reported that the majority of studies found an earlier onset in men, but in five studies, especially those from developing countries, there was no difference between the genders. More recently, three studies in India produced anomalous results. Subbakrishna et al. (Reference Subbakrishna, Murali, Gangadhar, Janakiramaiah, Subbakrishna and Kaliaperumal2001) and Murthy et al. (Reference Murthy, Janakiramaiah, Gangadhar and Subbakrishna1998) failed to find any gender difference while Gangadhar et al. (Reference Gangadhar, Panner Selvan, Subbakrishna and Janakiramaiah2002) found an earlier onset in females. These findings have led to speculation that the gender effect is absent or reversed in developing countries (Gangadhar et al. Reference Gangadhar, Panner Selvan, Subbakrishna and Janakiramaiah2002).
The variations may reflect the wide variety of definitions of age at onset. Some studies have used the first admission, some the first consultation, while others have taken the first symptom or first positive symptom. If males and females differ in their duration of untreated psychosis, or present to services at different stages of their illness, then different definitions of onset should give rise to variations in the gender difference.
A further issue is that diagnostic criteria for schizophrenia have different thresholds for length of illness. For example, the Diagnostic and Statistical Manual of Mental Disorders (DSM), Third Edition (DSM-III), Third Edition Revised (DSM-III-R) and Fourth Edition (DSM-IV) have a 6-month duration criterion while the International Classification of Diseases (ICD), Tenth Revision (ICD-10) requires only 1-month duration. Thus, briefer forms of psychosis are excluded by the DSM criteria for schizophrenia, but classified as schizophrenia in the ICD. This may be of relevance to the gender difference reported by different studies, as age at onset may vary with the severity and duration of psychosis (Castle et al. Reference Castle, Sham and Murray1998).
No recent review has systematically reviewed the published studies on gender differences in age at onset in schizophrenia, and attempted to take into account variations in the design, methodology, and the sample selected. The present meta-analysis systematically reviews the studies published between 1987 and 2009, to obtain pooled estimates of gender difference based on these studies, and to ascertain factors that may influence it.
Method
Data sources
Ovid Medline and Web of Science databases were searched between 1987 and 2009 with the keywords ‘age at onset and gender (or sex) and schizophrenia’. In addition, references were cross-checked from recent review articles related to the subject. Where relevant data were not available in the publication, the authors were contacted and the information requested.
Study selection
Only those articles or reports fulfilling the following criteria were included:
(1) Mean age at onset for schizophrenia (alone or with other psychosis) and standard deviation (s.d.) reported separately in males and females.
(2) Studies using ICD (version 9 or later) or DSM (version III or later).
(3) Studies published in English, in a peer-reviewed, indexed scientific journal.
(4) Age at onset defined as either age at first symptom, age at first positive symptom, age at first consultation or treatment, or age at first admission.
(5) Study not limited to extreme age groups such as under 18 years or over 65 years.
Exclusion criteria were as follows:
(1) Part of or the entire sample overlapped with a different publication that was already included. In such instances, the most inclusive study was used, and where this was not clear, the authors were contacted for clarification.
(2) The study was limited to childhood- and adolescent-onset psychosis or to cases with late onset (⩾45 years).
Data extraction
A total of 553 studies were identified and their abstracts screened. Where the abstract suggested that the study might contain relevant data, the paper was scrutinized. So, 46 studies were found to meet our criteria for inclusion.
Table A1 (see Appendix) shows summary data for the included studies. Where studies reported age at onset in more than one sample, each estimate was entered separately. For example, Stöber et al. (Reference Stöber, Franzek, Haubitz, Pfuhlmann and Beckmann1998) reported separate data for periodic and systemic catatonia. These data were entered as separate estimates, namely Stöber a 1998 and Stöber b 1998.
Publication bias
Possible publication bias was investigated informally by means of a funnel plot (Everitt, Reference Everitt and Everitt2003).
Meta-analysis procedure
Initial inspection of the studies revealed that the absolute age of onset varied considerably between studies, depending on the setting and inclusion criteria of the studies. However, the difference in age of onset between men and women was much more homogeneous. Therefore, the metric that we used in all our analyses was the age of onset in males minus the age at onset in females. Thus, negative numbers indicated an earlier onset in males. All analyses were conducted in Stata Version 10 for Macintosh (StataCorp LP, USA), using the metan (Bradburn et al. Reference Bradburn, Deeks and Altman1988) and metareg (Harbord & Higgins, Reference Harbord and Higgins2008) procedures. The data from the separate studies showed heterogeneity in the estimated age at onset difference for men and women (heterogeneity among all studies: χ2 = 313.53, degrees of freedom = 50, p < 0.001). Furthermore, the studies were derived from a wide range of populations, using different definitions of age of onset and schizophrenia. Consequently a random-effects model was used to obtain a combined estimate and confidence interval (CI) of the gender difference in age at onset.
We conducted separate meta-analyses to obtain pooled estimates of gender difference in age at onset for studies using the following definitions of age at onset: (1) age at first symptom; (2) age at first positive symptom; (3) age at first consultation; and (4) age at first admission.
We also combined studies with the above definitions of age at onset for schizophrenia to obtain a pooled estimate of the gender difference. We then performed a further meta-analysis, where we relaxed the diagnostic criteria to include schizo-affective disorder and other non-affective psychoses. For this analysis, we took age at first admission as the definition of age at onset, since this was the most frequently reported definition.
Meta-regression
Meta-regression was used to assess whether the diagnostic system used (DSM versus ICD) or the level of development of the host country influenced the gender differences in age at onset (any of the four above definitions used) for schizophrenia. The Human Development Index (HDI; United Nations, Reference United2001) was used to classify countries into developing (HDI 2 and 3 categories) and developed (HDI 1 category) countries.
Results
Assessment of publication bias
The funnel plot shown in Fig. A1 (see Appendix) is based on all studies reporting results for schizophrenia, regardless of the definition of age of onset. It is symmetrical about the line of estimated effect, demonstrating that publication bias is unlikely to be present.
Meta-analyses
I. Analyses for different definitions of age at onset
Age at first symptom in schizophrenia
A total of 17 estimates from 16 separate studies were included in this meta-analysis with a total sample size of 2609 males and 2016 females. Fig. 1 shows the forest plot with weighted effect size and 95% CI for each study. The weighted mean difference (WMD) in age of onset was −1.63 (95% CI −2.76 to −0.49), meaning that males had an earlier age at onset of 1.63 years.
Fig. 1. Forest plot for gender difference in age (years) at first symptom of schizophrenia. For full details of study names, see Table A1 in the Appendix. WMD, Weighted mean difference; CI, confidence interval.
Age at first positive symptom in schizophrenia
This meta-analysis included nine estimates from eight separate studies. The total sample consisted of 786 males and 573 females. Fig. 2 gives the forest plot with details of this meta-analysis. Males had an earlier onset by 1.43 years, with a WMD of −1.43 (95% CI −2.61 to −0.25).
Fig. 2. Forest plot for gender difference in age (years) at first positive symptom of schizophrenia. For full details of study names, see Table A1 in the Appendix. WMD, Weighted mean difference; CI, confidence interval.
Age at first consultation in schizophrenia
Fig. 3 gives the forest plot with weighted effect sizes and 95% CI for each individual study included in this meta-analysis. A total of 13 estimates from 12 studies were included in this meta-analysis examining the gender difference in age at first consultation in schizophrenia. The total sample size consisted of 1633 males and 1126 females. The overall WMD was −1.22, suggesting an earlier onset in males by 1.22 years. However, the 95% CI included zero (−2.49 to 0.06).
Fig. 3. Forest plot for gender difference in age (years) at first consultation for schizophrenia. For full details of study names, see Table A1 in the Appendix. WMD, Weighted mean difference; CI, confidence interval.
Age at first admission in schizophrenia
In this meta-analysis, 23 different estimates from 16 studies were included. The total sample size consisted of 17946 males and 11743 females. Fig. 4 shows the forest plot for this meta-analysis. The WMD was −1.07 (95% CI −1.93 to −0.21), indicating an earlier onset in males by 1.07 years.
Fig. 4. Forest plot for gender difference in age (years) at first admission for schizophrenia. For full details of study names, see Table A1 in the Appendix. WMD, Weighted mean difference; CI, confidence interval.
II. Analysis for any definition for age at onset in schizophrenia
As the above meta-analysis showed an earlier age at onset in males compared with females, all the studies from the above four meta-analysis were combined in a separate meta-analysis. For each study, age at first symptom was used where reported. Otherwise, age at first positive symptom was used. If neither was reported, age at first consultation was used, and if none of the above was reported, then age of first admission was used. Fig. 5 gives the forest plot for the gender difference in age at onset of schizophrenia with any of the above four definitions for age at onset considered. A total of 30 estimates from 22 studies were included. The sample size consisted of 25684 males and 16802 females. This showed an earlier onset in males, with a WMD of −1.49 years (95% CI −2.07 to −0.92).
Fig. 5. Forest plot for gender difference in age (years) at onset (any definition) of schizophrenia. For full details of study names, see Table A1 in the Appendix. WMD, Weighted mean difference; CI, confidence interval.
III. Age at first admission in non-affective psychoses
All the above meta-analyses examined gender difference in age at onset in schizophrenia. A further meta-analysis was conducted to examine the gender difference in age at onset in studies which included a broader group of disorders including schizo-affective disorders and all other non-affective psychoses. We used first admission, as this was the most frequently reported definition of onset. Fig. A2 (see Appendix) gives the forest plot for the results of this meta-analysis. A total of 30 estimates from 22 studies were included. The sample size consisted of 25684 males and 16802 females. The overall effect size was −1.37 (95% CI −2.03 to −0.71), indicating an earlier first admission in males by 1.37 years in non-affective psychoses.
Meta-regression
Meta-regression analyses were conducted to examine the effect of two variables – diagnostic system used and developed/developing country status. Meta-regression was based on the meta-analysis of studies including any definition for age at onset of schizophrenia (see Fig. 5), adjusting for definition of age at onset in the analysis.
Diagnostic system used
A total of 48 estimates from 40 studies were included in this analysis. For ICD studies alone, the gender difference was non-significant, at −0.309 (95% CI −1.229 to 0.611, p = 0.510), while for DSM studies it was −1.92 (95% CI −2.483 to −1.356, p < 0.001). The meta-regression showed that the gender difference in age of onset was significantly greater under DSM criteria than ICD, adjusted for definition of age of onset (WMD = −1.93, 95% CI −3.11 to −0.75, p = 0.001).
Developed versus developing country
The host countries of the studies were categorized according to the HDI into developed or developing countries. A total of 49 estimates from 40 studies were included in this analysis. The results showed no evidence of any difference in gender difference in age of onset between developing and developed countries (WMD adjusted for definition of onset = 1.68, 95% CI −1.41 to 1.75, p = 0.835). The gender difference in developing countries was −1.54 (95% CI −2.85 to −0.23, p = 0.021), and in developed countries was −1.43 (95% CI −2.11 to −0.77, p < 0.001).
Discussion
Our results confirm a gender difference in the age at onset of schizophrenia, with an earlier age at onset in males. When all definitions for age at onset in schizophrenia are combined, males have an earlier onset by 1.49 years. The gender difference in age at onset is larger for age at onset of first symptom or positive symptom as compared with age at first consultation or admission (1.07 years). It is known that many patients have psychotic symptoms for some years before accessing services, and it is possible that the duration of time from the first symptoms to the first access to services is related to the patients' gender (Salokangas et al. Reference Salokangas, Honkonen and Saarinen2003). We found the greatest gender difference (1.63 years) when ‘age at first symptom’ was used to define the age at onset of schizophrenia, and the smallest gender difference for age at first consultation (1.22 years) and age at first admission (1.07 years). This suggests that males may have a more insidious onset of psychosis, with longer duration of untreated psychosis, than females.
There are some potential biases to consider. First, the majority of the studies are based on hospital patients or use registers that record hospital admissions. These hospitalized cases may be more severe or disturbed, and males may be overrepresented (Thorup et al. Reference Thorup, Petersen, Jeppesen, Ohlenschlaeger, Christensen, Krarup, Jorgensen and Nordentoft2007b). Second, although we only included studies in working-age adults, there were some variations in lower and upper age limits of the studies, which may have influenced the size of the gender differences; for example, a study with an older upper limit may have included an excess of late-onset females.
The ascertainment of age at onset may not always have been accurate. For example, one study from South India (Murthy et al. Reference Murthy, Janakiramaiah, Gangadhar and Subbakrishna1998) describes the difficulties in ascertaining the age at onset in their sample in Bangalore, India. In patients who did not have date of birth records, assessment was made with information from family members using circumstantial evidence such as festivals or important national events. This highlights the difficulty in ascertaining age or age at onset in certain populations.
While we attempted to subdivide the studies according to the definition of onset used, this was not always straightforward. For example, one study defined age of onset as ‘age at first symptom’ but later stated that this was ‘the age at which symptoms such as delusions, hallucinations, excitable or aggressive behaviour, catatonia, etc began’. The definition from this study was entered as ‘age at first positive symptom’.
Furthermore, the point at which one begins to consider the presence of positive symptoms may be somewhat subjective, especially if clear cut hallucinations or delusions are not present. The demarcation of prodrome from onset of psychosis is also subjective, and may vary widely between clinicians. Gender differences in the ability to conceal symptoms may lead to earlier diagnosis in one gender than the other.
It has been demonstrated in previous studies that women have a second peak in incidence of schizophrenia in later life, albeit much smaller than the peak in the early twenties (Thorup et al. Reference Thorup, Petersen, Jeppesen, Ohlenschlaeger, Christensen, Krarup, Jorgensen and Nordentoft2007a). The estimate ‘mean age at onset’ does not take this into account and may produce skewed mean age at onset to the right for females. There is a suggestion that earlier-onset schizophrenia may be a different variant to later-onset cases (Kleinhaus et al. Reference Kleinhaus, Harlap, Perrin, Manor, Weiser, Lichtenberg and Malaspina2011). There is a need for further studies to examine gender differences in different age brackets.
In this meta-analysis, the gender difference is larger when a broader group of non-affective psychoses is included, with males being admitted 1.37 years earlier than females, compared with 1.07 years in schizophrenia. This broad group of psychoses included schizophrenia, schizo-affective disorders and other non-affective psychoses. The larger gender difference in age at onset in this group may be related to the inclusion of schizo-affective disorders. Studies of bipolar disorder have found a significant earlier onset in men, both for first depression and first mania (Viguera et al. Reference Viguera, Baldessarini and Tondo2001). A study of bipolar I disorder, in the suburbs of London, found that men had an earlier onset by 8.3 years (Viguera et al. Reference Viguera, Baldessarini and Tondo2001; Raymont et al. Reference Raymont, Bettany and Frangou2003), whilst a first-onset study in inner London over 35 years (Kennedy et al. Reference Kennedy, Boydell, Kalidindi, Fearon, Jones, van Os, Murray, Kennedy, Boydell, Kalidindi, Fearon, Jones, van Os and Murray2005) showed that men had an earlier mean onset of bipolar disorder and mania by 4.4 and 5.1 years, respectively.
The results of the meta-regression for diagnostic criteria demonstrate that the diagnostic criteria used have a large impact on the gender difference in age of onset. Indeed, studies using ICD failed to show any gender difference in age of onset. Unlike ICD, DSM-III, DSM-III-R and DSM-IV require the existence of symptoms for a 6-month period for the diagnosis of schizophrenia. Our results imply that cases of a longer duration may show a greater gender difference in age of onset. Castle and colleagues showed that the application of criteria which define a severe form of schizophrenia (DSM criteria as compared with ICD criteria) excludes more females, and results in an overall male excess, especially in those with age at onset less than 25 years (Castle et al. Reference Castle, Sham and Murray1998).
The results of the meta-regression demonstrate that the development index of the country has no effect on the gender difference in age at onset of core schizophrenia. This finding demonstrates the value of meta-analysis in addressing epidemiological questions. Many of the studies from developing countries that were included in the meta-analysis showed an earlier age of onset for males, but the fact that this difference was not significant in individual studies was sometimes interpreted by the authors as evidence for absence of gender difference, and as a result a narrative has developed in the literature suggesting that the gender difference may be absent in developing countries. In fact, the main difference between the developed and developing studies in this meta-analysis was that the latter were more heterogeneous: the estimate of between-study variance, τ2, was 2.46 in the developed countries, but 4.63 in developing countries.
Three studies from developing countries included in the meta-regression found an earlier age at onset in females than males (Thara et al. Reference Thara, Henrietta, Joseph, Rajkumar and Eaton1994; Gangadhar et al. Reference Gangadhar, Panner Selvan, Subbakrishna and Janakiramaiah2002; Kebede et al. Reference Kebede, Alem, Shibre, Negash, Fekadu, Fekadu, Deyassa, Jacobsson and Kullgren2003). However, the remaining seven studies from developing countries showed the expected earlier age at onset in males. The studies from developing countries had relatively small sample sizes. There is a need for larger studies from developing countries examining the gender difference in age at onset.
Limitations
Two databases, Ovid Medline and Web of Science, were used for literature searches. Personal communication from authors was sought for any clarification of data required. However, unpublished data or articles published in languages other than English were not included. This study did not examine gender differences in childhood-onset or late-onset schizophrenia. We were also not able to study the role of important confounders such as family history of schizophrenia, pre-morbid personality and obstetric complications as insufficient data were available. Lastly, our search covered only peer-reviewed journals. Although we cross-checked citations from individual papers, we may have missed some papers published in languages other than English and/or in non-peer-reviewed journals.
Conclusions
This meta-analysis shows a gender difference in age at onset of schizophrenia, with a small earlier age at onset in males. This difference is not specific to schizophrenia but is present in all non-affective psychosis. The gender difference was absent in studies using ICD-10 criteria but appeared confined to studies using DSM criteria, and was present to the same extent in studies from developed and developing countries. There is a need for large studies from developing countries to examine the gender difference in age at onset and factors influencing it. Further studies are required to examine the influence of confounding factors, such as family history of schizophrenia, pre-morbid personality and obstetric complications on gender differences in age at onset of schizophrenia.
Acknowledgements
This study was partly funded by the National Institute for Health Research (NIHR) Biomedical Research Centre for Mental Health at King's College London and South London and Maudsley NHS Foundation Trust.
Declaration of Interest
None.
Appendix
Fig. A1. Funnel plot with pseudo 95% confidence limits for studies using any definition of age of onset of schizophrenia (studies included in Fig. 5). s.e., Standard error; WMD, weighted mean difference.
Fig. A2. Forest plot for gender difference in age (years) at first admission for any non-affective psychoses. For full details of study names, see Table A1. WMD, Weighted mean difference; CI, confidence interval.
Table A1. Data for all studies included in the meta-analysis
