Abel KM, Wicks S, Susser ES, Dalman C, Pedersen MG, Mortensen PB, Webb RT. Birth Weight, Schizophrenia, and Adult Mental DisorderIs Risk Confined to the Smallest Babies?. Arch Gen Psychiatry. 2010;67(9):923–930. doi:10.1001/archgenpsychiatry.2010.100
Studies linking birth weight and mental illness onset are inconclusive. They have primarily focused on the World Health Organization low birth weight threshold (2500 g) and schizophrenia. To our knowledge, low birth weight per se has not been conclusively linked with schizophrenia risk and specificity of the effect to birth weight below the standard threshold or to particular psychiatric diagnoses has not been demonstrated.
To examine whether (1) low birth weight (<2500 g) is associated with increased risk for adult schizophrenia; (2) risk extends into the normal weight range; and (3) risk is confined to schizophrenia or linked to other adult mental illnesses.
Population-based cohort study.
Sweden and Denmark.
Singleton live births in Sweden (1973-1984) and Denmark (1979-1986) (N = 1.49 million). Births were linked to comprehensive national registers of psychiatric treatment, with follow-up to December 31, 2002 (Sweden), or to June 30, 2005 (Denmark). There were 5445 cases of schizophrenia and 57 455 cases of any adult psychiatric disorder.
Main Outcome Measure
Crude and adjusted odds ratios for birth weight less than or more than 3500 to 3999 g in consecutive 500-g strata (from 500-1499 g to ≥4500 g) for schizophrenia, any psychiatric diagnoses, and specified psychiatric diagnoses.
Schizophrenia was associated with birth weight less than 2500 g. The association was not restricted to birth weight less than 2500 g and there was a significant linear trend of increasing odds ratios with decreasing birth weight across the birth weight range. This was mirrored for any psychiatric diagnosis and for each of the categories of psychiatric disorder.
Findings suggest there is an association between birth weight and adult mental disorder, but there is no indication this effect is specific to birth weight less than 2500 g or to schizophrenia. Future research should explore common disorder-specific mechanisms that may link birth weight to development of psychiatric disorder in adulthood.
The birth weight of an infant provides little information about the path of its growth in utero in either an adverse or favorable maternal environment. Nonetheless, low birth weight (defined by the World Health Organization as < 2500 g) has been shown to be an important marker of vulnerability to infant and child morbidity and mortality. Increasingly, studies have examined associations between low birth weight and various adult outcomes such as cardiovascular disease1 and cognition.2 Many of these studies have gone beyond the examination of low birth weight per se and have explored whether there is an inverse gradient between birth weight and risk of adult outcomes.3- 5
The exact nature of the relationship between birth weight and risk of schizophrenia in adulthood has been debated frequently in the literature but remains unresolved. Most earlier studies have explored the link between low birth weight (<2500 g) specifically and schizophrenia but have not established this relationship. Moreover, little is known about its relationship with birth weight across the normal range. If there is a gradient of risk of schizophrenia inversely related to birth weight across the normal range, it might mimic findings in physical disorders where common mechanisms are involved in poor fetal growth and the development of adult diseases such as hypertension, stroke, and cardiovascular disease.3- 5 This is important because the mechanism behind an association with schizophrenia risk confined to the smallest babies is more likely to include perinatal or intranatal events associated with smallness at birth per se. However, if the elevated risk of schizophrenia is related to poor fetal growth, rather than to low birth weight alone, then not only would this association occur most commonly in the “normal” birth weight range, but the mechanisms of risk may be very different.
In this study, we sought to reconsider the relationship between birth weight and schizophrenia in a large enough sample to detect small effects6 that may occur across the normal birth weight range. The most commonly cited meta-analysis estimated a relative risk of 1.6 for the crude association between low birth weight (<2500 g) of an infant and risk of adult schizophrenia.7 Well-known independent risk factors for low birth weight, such as parental socioeconomic data, were not generally available in these studies and the association reported in the meta-analysis was uncorrected for key confounders such as gestational age, social factors, and maternal mental illness. Most studies were restricted to a narrow definition of disorder: the largest study (Dalman et al8) had only 238 cases of schizophrenia in total and many were smaller (eg, Sacker et al,9 35 cases; Wahlbeck et al,10 114 cases; and Hultman et al,11 167 cases). There is also some suggestion of a bias toward reporting of positive findings.7,12 Furthermore, the specificity of the finding either to a birth weight of less than 2500 g or to schizophrenia itself was not addressed in the individual studies or in the meta-analysis. Since publication of the meta-analysis, the 2 largest studies assessing this association have been conducted (Gunnell et al,13 736 cases; Byrne et al,14 1039 cases). These studies were inconclusive: some association was evident between birth weight and schizophrenia in partially adjusted models but not in those fitted with a large number of covariates.13,14
In this study (total cases of schizophrenia and related disorder = 5445), we set out first to establish the relationship between low birth weight (<2500 g) and schizophrenia by combining Swedish and Danish population databases of national birth records linked with all adult psychiatric admissions. We adjusted for a range of potential confounders of the association. Second, we examined whether the increased risk extended beyond the standard low birth weight threshold or was confined to babies less than 2500 g, which required large enough numbers to be able to have sufficient disease events in each birth weight stratum. Finally, we wished to determine whether any excess risk was confined to schizophrenia or whether it was also associated with other adult mental illnesses.
We used a cohort design in which the study population consisted of all singleton live births in Sweden from 1973 to 1984 (n = 1 077 393) and in Denmark from 1979 to 1986 (n = 414 074) with information on birth weight and whether they were alive and residing in their respective country of birth on their 18th birthday. Birth weight data were 96.7% complete in the Swedish cohort and 99.6% complete in the Danish cohort. The birth cohorts were delineated through linkage between the respective medical birth registers of the 2 countries15- 18 and the Swedish National Patient Register19 and Danish Psychiatric Central Register.20
From these national registers, cases were ascertained in Sweden from all hospital discharges between January 1, 1988, through December 31, 2002, and in Denmark, from any psychiatric hospital discharges between January 1, 1994, through June 30, 2005. Cases were identified as all members of these cohorts 15 years or older who were still alive and residing in their country of birth on their 18th birthday. From Denmark, cases were from psychiatric hospital discharges and outpatient episodes, which are also included from 1995 onward in the Denmark cohort only. This resulted in 5445 people diagnosed with schizophrenia (or a related disorder including schizotypal and delusional disorders), 57 455 people on the national patient register with any psychiatric diagnosis, and approximately 1.4 million people in the general population who were not entered in the psychiatric registers.
Table 1 shows the diagnostic classifications used in the 2 countries over time. In Sweden, International Classification of Diseases, Ninth Revision was used from 1987 to 1996 and International Classification of Diseases, Tenth Revision, from 1997 to 2002; in Denmark, International Classification of Diseases, Tenth Revision was used from 1994 to June 30, 2005. We estimated effects separately for patients with any psychiatric diagnosis in the national registers (Sweden: International Classification of Diseases, Ninth Revision codes 290-319 and International Classification of Diseases, Tenth Revision codes F00-F99; Denmark: all episodes recorded in the Psychiatric Central Register as inpatient or outpatient from 1995). The diagnostic group “any psychiatric diagnosis” covers a wide range of different disorders. For the purposes of analysis, this larger group was subdivided into its main diagnostic groupings. For the age range of the cohort, these include schizophrenia, affective disorder, alcohol and drug disorders, and neurotic and stress-related disorders. Table 1 shows the diagnostic codes used in each country for each diagnosis. We studied people with schizophrenia and related disorder (schizotypal and delusional disorder) as a broad diagnostic category (see International Classification of Diseases codes in Table 1); for the remainder of the article, we refer to this category as the “schizophrenia” outcome group.
Even though they are both large longitudinal national databases, there are 2 important differences between the Swedish Patient Register and the Danish Psychiatric Central Register. First, the Swedish data only include inpatient admissions while the Danish data include both inpatient and outpatient episodes from 1995 onward. This means that the data will tend to underestimate the number of people with a diagnosis treated via outpatient care only. For example, the most serious cases of alcohol and drug disorder are likely to have an inpatient episode, but many of those with less severe illness will only be treated in outpatient care and therefore will not be included in the Swedish register data. On the other hand, Swedish data have all discharges with psychiatric diagnoses regardless of type of hospital, whereas the Danish register only includes discharges from psychiatric hospitals or outpatient clinic registers from 1995. Even though in Sweden regular hospital discharge data are routinely gathered, and the Danish data include all outpatient contacts from 1995, our cohort does not include all cases of milder illness. Although the nature of the Scandinavian databases means they tend to contain more severe cases, other aspects affecting ascertainment, such as help seeking, mean ascertainment is not exclusively related to severity.
These were conducted using Stata version 10 software (StataCorp, College Station, Texas) and SAS version 9.1 software (SAS Institute Inc, Chicago, Illinois). Two phases of analyses were conducted. The first phase entailed analysis of both national cohorts combined. To increase statistical precision, the data from both countries were merged into a large single-study data set (N = 1 491 467). The data were aggregated by 500-g birth weight strata from 500 to 1499 g and upward to 4500 g and more. The birth weight groups were stratified this broadly at both ends of the distribution so that the aggregated data files could be exported and merged together without breaching the confidentiality regulations of either country. This stratification also enabled comparison with the categories of birth weight strata used in previous studies. We estimated odds ratios (ORs) following exposure to a birth weight stratum using logistic regression models21 (always adjusted for year of birth and sex) among people registered with any psychiatric diagnosis and by specified diagnostic category.
Birth weight of 3500 to 3999 g was used as a common reference category. This was chosen to facilitate portrayal of the shape of the association between mental disorder and birth weight across its distribution. We compared the odds of developing schizophrenia and other mental disorder in other categories of birth weight with the odds in this reference category. In the first analysis phase, with data from both countries merged as 1 study cohort, the regression models were adjusted for country of birth, year of birth, and sex.
Using the combined cohort, we first addressed the question of whether low-birth-weight infants (<2500 g) were at greater risk of developing schizophrenia. We then considered whether the increased risk was confined to this group or whether individuals' birth weight was lower overall in those who developed schizophrenia compared with the general population. Finally, we asked if a similar pattern of risk applied to other mental disorder severe enough to appear in the psychiatric registers or whether this was specific to schizophrenia. To verify that the result pertained in both national cohorts, we also conducted a stratified analysis by country.
In the second analysis phase, the Swedish registers provided additional potential confounding variables of maternal psychiatric admission and parental social class. Maternal psychiatric admission was defined as biological mother admitted any time during 1973 to 2002 with a psychiatric diagnosis (F chapter in International Classification of Diseases, Tenth Revision ; codes 290-319 in International Classification of Diseases, Ninth Revision ; and codes 290-315 in International Classification of Diseases, Eighth Revision). Parental social class included parental employment status and paternal occupational social class as described in previous Swedish registry studies22; for subjects with missing paternal data, maternal data were used. Employment status (yes vs no; ±2 years from offspring birth year) was extracted from the national census at 5-year intervals from 1975. Occupational social class (blue collar, white collar, other, or self-employed, 2-6 years after birth year) was also available from the census (1980 onward). Missing data on either parental socioeconomic status or maternal mental illness were 1.1%. Using only the Swedish birth cohort, additional logistic regression models were fitted to assess the effects of these potential confounders as well as gestational age. To remove any possible effect of prematurity, we also fitted models using only term births (≥37 weeks) in the Swedish data. Gestational age determination by ultrasonography from week 17 started gradually during the 1980s, and from 1987 onward, it was rare not to undergo ultrasonography. We finally conducted a sensitivity analysis excluding all those with a maternal psychiatric admission.
In the combined Swedish and Danish national birth cohorts, we identified 57 455 people subsequently entered in the national patient register before December 31, 2002, in Sweden and before June 30, 2005, in Denmark and 1 434 012 people in the general population who were not in the psychiatric registers. Within the “any psychiatric disorder” group, there were 5445 people diagnosed with schizophrenia, 20 491 people diagnosed with drug and alcohol disorders, 11 660 people with an affective disorder, and 19 346 with neurotic, stress-related, or somatoform disorders. More than 80% of all those registered with a psychiatric disorder were diagnosed with schizophrenia, affective disorders, alcohol- and drug-related disorders, or neurotic and stress-related disorders. The remainder included behavioral disorders, personality disorder, and developmental disabilities.
In our sample, low birth weight (as defined by the World Health Organization threshold of 2500 g) was associated with increased risk of schizophrenia. The logistic regression models presented in Table 2 also indicate that overall there was a trend of increasing ORs with decreasing birth weight compared with the general population. A test for linear trend across birth weight strata generated a highly significant P value of P < .001 for schizophrenia. Although the strongest effects were consistently in the lower birth weight strata, effects were not confined to birth weights less than 2500 g but were seen within the normal-birth-weight range up to the reference group. Comparison of the crude vs adjusted ORs shown in Table 2 indicates that country of birth, year of birth, and sex had a modest combined confounding effect in attenuating the ORs somewhat, but the trends remained essentially unchanged. We found no indication of specificity of effect with any particular diagnostic category as shown in Table 2. The linear trend was also highly significant (P < .001) for each diagnostic category, without or with adjustment for year of birth, country of birth, and sex. Because the cohorts included outpatients in Denmark only, we conducted stratified analyses and assessed the results separately for the Swedish and Danish data; the effect sizes were very similar for both countries individually (data available on request).
In Table 3 and Table 4, we present ORs for the Swedish birth cohort only. Table 3 represents ORs in relation to any psychiatric disorders, with adjustment for potential confounders. This is illustrated by results for any psychiatric diagnosis in Table 3. We show results for parental social class, but adjustment for parental employment status was also performed (results available on request). Adjustment for each of these potential confounders separately made no material change to the observed point or variance estimates. We repeated the adjustments for parental social class and maternal admission in relation to schizophrenia and other diagnostic groupings, and again, the degree of attenuation in the adjusted effect size was minimal (results available on request). We also repeated analyses removing all those with maternal psychiatric admission, but again, this had little effect (results available on request).
We found an association between being small for gestational age (defined as birth weights in the population >2 SDs less than the normal birth weight for a particular gestational age) and any psychiatric diagnosis (OR, 1.35; 95% confidence interval [CI], 1.29-1.42), schizophrenia (OR, 1.34; 95% CI, 1.14-1.58), affective disorders (OR 1.16; 95% CI, 1.02-1.31), alcohol and drug disorders (OR, 1.36; 95% CI, 1.27-1.46), and neurotic, stress-related, and somatoform disorders (OR, 1.32; 95% CI, 1.20-1.46) as well as an association between low gestational age (<37 weeks) and any psychiatric diagnosis (OR, 1.25; 95% CI, 1.19-1.32), schizophrenia (OR, 1.34; 95% CI, 1.14-1.57), affective disorders (OR, 1.36; 95% CI, 1.22-1.51), alcohol and drug disorders (OR, 1.18; 95% CI, 1.11-1.27), and neurotic, stress-related, and somatoform disorders (OR, 1.40; 95% CI 1.27-1.54) (details of these analyses are available on request). Therefore, to remove any possible confounding effects of prematurity, we estimated ORs for the Swedish birth cohort in relation to any psychiatric diagnosis using only children born at term (37 weeks) or later (Table 4). The cohort size was thereby reduced to 1 025 308. The estimates adjusted for sex show attenuated effect sizes but remain significant across the birth weight range in any psychiatric diagnosis and show a similar pattern for schizophrenia and the other diagnostic groupings (results available on request).
Using 2 intercountry population cohorts to create a larger sample than has been available previously, we have been able to describe the nature of the relationship between birth weight and risk of developing severe adult psychiatric disorder ascertained in these national registries (as described earlier). First, the risk of schizophrenia is elevated in babies weighing less than 2500 g at birth. Second, although the effect is strongest for birth weight less than 2500 g, it is not confined to the smallest babies and extends into the normal birth weight range; it is clear at least to birth weights of 3000 g and the data are consistent with a graded association across the whole birth weight range. Third, this pattern of risk is not specific to schizophrenia but is mirrored in all the psychiatric diagnostic categories we considered.
Our findings are consistent with recent reports on other neurodevelopmental outcomes. Birth weight is significantly associated with global cognitive ability at ages 6, 7, and 8 years, through adolescence, and into early adulthood.2 These associations between birth weight and IQ are evident in the normal birth weight range and so are not accounted for exclusively by low birth weight.23- 25 In most, though not all, studies, comprehensive adjustment for gestational age, parity, and socioeconomic status fails to remove this effect,24,26,27 although future studies might also consider whether changes in the socioeconomic position of the parent or child across childhood affect the associations with neurodevelopmental outcomes. There is also evidence that birth weight is related to risk of cerebral palsy within the normal range from the Surveillance of Cerebral Palsy in Europe collaboration of European Cerebral Palsy Registers, which assessed birth weight and gestation in a large cohort of children compared with reference standards for the normal spread of gestation and weight for gestational age at birth.28 Parent- and teacher-reported behavioral measures in children aged 4 to 14 years in a Western Australian birth cohort showed a similar dose-response relationship between deviance from “optimal fetal growth” using a population centile estimate of optimal fetal growth.29 Finally, a recent report from the Avon Longitudinal Study of Parents and Children30 suggested that increased risk of definite psychosis-like symptoms in children aged 12 years extends into the normal birth weight range: 1 SD increase in birth weight was associated with an 18% reduction in the odds of psychosis-like symptoms.
We did not find evidence that higher-birth-weight babies are at increased risk of schizophrenia or other adult mental disorder. In this respect, our findings differ from some reports of a J-shaped pattern of risk association in which heavy babies (>4500 g) are at greater risk of some neurodevelopmental outcomes.28
Much current research attempts to understand the many causes of a birth weight less than 2500 g, its neurodevelopmental consequences, and the mechanisms linking the two,31 such as the hypoxia, periventricular hemorrhages, and white matter damage more common in low-birth-weight or very low-birth-weight infants.32,33 Our study indicates that perinatal factors that are particularly common in premature infants or those less than 2500 g are only one of a number of factors giving rise to the link between birth weight and vulnerability to abnormal neurodevelopment, such as schizophrenia. This suggests that it might be appropriate to place equal emphasis on a broader understanding of the connections between fetal growth, its control, and brain development.34- 37
At a population level, up to 7 times as many children are born with birth weights between 2500 and 3499 g than are born with low birth weight.38 Therefore, although the largest ORs were consistently found in the lower birth weight strata, the relatively small numbers in these strata mean that effects linked with having a birth weight less than 2500 g have a relatively small number of cases associated with them. By contrast, although the ORs were modestly raised for weights more than 2500 g, far more psychiatric disorder was associated with birth weight strata in the normal range. If exposure to lower birth weight (into the normal birth weight range) creates risk, future studies should also consider the timing of any change in fetal growth and the rate of fetal growth across gestation, because if growth is restricted early in pregnancy, it may be associated with quite different mechanisms and different risks compared with growth that is restricted late in pregnancy.39,40For example, the Dutch Hunger Winter produced an excess risk of schizophrenia in offspring exposed in the first trimester, whereas only those exposed in the third trimester showed an excess risk of low birth weight (<2500 g).41,42
The third finding of a lack of specificity of effects to schizophrenia makes clear the need to consider how we understand the relationship of birth weight with psychiatric disorder in general and whether the same or different mechanisms are involved in risk of schizophrenia and other disorders. As may be the case in physical illnesses such as hypertension and heart disease, birth weight could represent a common mechanism of vulnerability to psychiatric disorder5,43 or simply be a marker of risk that is common to different disorders via different mechanisms.44,45 In other physical disorders, there is growing understanding about some of the processes that might explain these associations.5,44,45 They include lifelong consequences of altered programming of growth and endocrine pathways, leading to permanent metabolic changes (eg, insulin resistance) and the consequences of compromising development of low-priority organs to protect high-priority ones with subsequent lifelong changes in growth and developmental trajectories. It is not difficult to conceive of a parallel situation occurring in neurodevelopment. Traditionally, it has been thought that the “high-priority” brain is “spared” in adverse maternal conditions. More recent evidence from structural imaging suggests that brain sparing occurs less in growth-restricted fetuses than was believed to be the case from ultrasonography measurement.46 From an alternative perspective, one could think in terms of protective aspects of development such that the better your development within a particular environment, the more protected you are from future insult. In the Scandinavian databases, however, we are limited in our interpretation of data for diagnoses other than schizophrenia because case ascertainment is generally limited to the most severe cases in the registers and therefore completeness of the record for more common, less serious disorders is not as reliable as for schizophrenia.47
An important advantage of large samples is that they allow for small effects to be detected, but this also makes them more sensitive to confounding.48 We have been able to use a sample that has more than 5 times the total number of broadly defined schizophrenia cases than the largest published study to date.14 Also, we were able to adjust for more confounders than previous studies that have reported a positive association.8- 11 However, it would be useful to have other studies where adjustment for such potential confounders as maternal smoking were possible. Heavy antenatal smokers are overrepresented in the mothers with psychiatric disorders and these mothers are more likely to have offspring who develop psychiatric disorders. However, excluding all those whose mothers were admitted with a psychiatric illness made no difference to the estimates, so this is unlikely to be an explanation. A less likely possibility is that antenatal maternal smoking itself is a cause of excess psychiatric disorder in offspring. Evidence to date suggests weak or noncausal associations.49 We could not determine the timing of maternal psychiatric admission in relation to birth of the infant, but when Jablensky et al50 examined this issue, they reported that the frequency of low birth weight was no greater in women who developed mental illness before or after the birth of the child. We did not adjust for maternal marital status, maternal age, or parity; in other studies,8,14 these adjustments did not materially change the estimates.
As noted earlier, the general population group without registered psychiatric illness is likely to have included subjects with milder disorders who were not ascertained by the psychiatric registers. These individuals may share risk factors with individuals who were included in our diagnostic groups. This would have diluted any differences in birth weights between the 2 groups, thereby attenuating our estimates. Also, our sample does not include disorders recorded in the registers before the age of 15 years, so our results cannot be generalized to childhood-onset disorders. It may well be the case that the association of birth weight with mental disorder is different for young and old.
Finally, we did not adjust for individual obstetric complications (OCs), although this is unlikely to affect the pattern of our findings. In fact, in different studies, a range of diverse OCs have been linked with schizophrenia risk,7,8 and while each is likely to have a different role in the causal pathway, those with substantial effect (eg, preeclampsia) are quite rare. In addition, for most OCs, the rarity of exposed cases means that it becomes impossible to control for them.14 One way of dealing with rarity is the use of OC scales or models that attempt to group OCs.8,14However, these make the unlikely assumption that OCs share common mechanisms of action. In the Avon Longitudinal Study of Parents and Children,30 associations between birth weight and psychosis-like symptoms in childhood were not changed when adjusting for other OCs, including resuscitation at birth as a measure of birth asphyxia.
Previous studies aiming to establish the exact nature of the relationship between low birth weight and adult schizophrenia risk have been inconclusive. In by far the largest sample to date, we have shown that low birth weight is associated with increased risk for adult schizophrenia. In contrast to those previous studies that have focused on birth weight less than 2500 g, however, we provide evidence that there is no threshold of effect for low birth weight but that risk extends into the normal birth weight range. We also report that other disorders severe enough to result in psychiatric admission or attendance at an outpatient clinic show a similar pattern of a graded association with birth weight.
Future research into the prenatal causes of adult mental disorder should place greater emphasis on the broader links between fetal growth, its control, and brain development. This should include the range of environmental, placental, and other influences on fetal growth that may work in combination with more specific causes of individual disorder such as genes, prenatal insult, and postnatal experiences to enable expression of a particular mental health outcome.37
Correspondence: Kathryn M. Abel, MD, PhD, Centre for Women's Mental Health, University of Manchester, Jean McFarlane Bldg, Manchester M13 9PL, England (firstname.lastname@example.org).
Submitted for Publication: June 22, 2009; final revision received April 3, 2010; accepted April 6, 2010.
Author Contributions: Ms Wicks, Ms Pedersen, and Dr Webb had full access to the data in the study and take responsibility for the integrity of the data and accuracy of the data analysis.
Financial Disclosure: None reported.
Funding/Support: All psychiatric epidemiological research conducted at the National Centre for Register-based Research is partially funded by the Stanley Medical Research Institute. No other external funding was sought or obtained for this study.
Role of the Sponsors: The funding sources had no role in the design or conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.