[1] 15w Maternal Exposure to Respiratory Infections and Adult Schizophrenia Spectrum Disorders: A Prospective Birth Cohort Study

[1] 183w We sought to examine the relationship between maternal exposure to adult respiratory infections and schizophrenia Spectrum disorder (SSD) in the Prenatal Determinants of Schizophrenia (PDS) Study, a large birth cohort Investigation. Previous work suggests that second trimester exposure to respiratory infection may be a risk factor for SSD. We therefore examined whether this class of infection was associated with adult SSD. For this purpose, we capitalized on several design advantages of the PDS Study, including a comprehensive, prospective data base on physician-diagnosed infections and a continuous followup in which diagnoses of SSD were made, in the majority, by face-to-face interview. Second trimester exposure to respiratory infections was associated with a significantly increased risk of SSD, adjusting for maternal smoking, education, and race (rate ratio [RR] = 2.13 [1.05-435], x 2 = 436,4f = 1, p = 0.04); no associations were shown for first trimester and third trimester exposure to these respiratory infections. These findings support-and extend-previous studies suggesting that second trimester respiratory infections are risk factors for SSD. This study therefore has implications toward uncovering the etiology of schizophrenia and developing preventive strategies.

[1] 125w Demographics for the Analytic Sample. The age at delivery for the gravidas was s 19 years for 5.1 percent, 20-39 for 90.1 percent, and a 40 years for 4.8 percent. Nineteen percent of the gravidas did not complete high school, 39 percent graduated from high school but did not attend college, and 42 percent had at least some college education. The racial distribution was 61.8 percent European white, 27.3 percent black, and 10.9 percent of other ethnicity. Forty-two percent of the gravidas smoked either until the current pregnancy or during pregnancy. There was a significantly increased risk of SSD following second trimester exposure to acute respiratory infection. For these same infections, there was no increase in risk for exposure in the first and third trimesters.

[2] 86w For second trimester respiratory infection, there was a significantly increased risk of SSD, adjusting for maternal smoking, education, and race (RR = 2.13 [1.05-4.35], x 2 = 4.36, # = l,p = 0.04). There was no increased risk of SSD following first trimester (RR = 0.89 [0.28-2.84], x 2 = 0.04, 4f = 1, p = 0.84) or third trimester (RR = 0.68 [0.25-1.87], x 2 = 0.56, eff=l,p = 0.45) exposure to respiratory infection in this multivariate model. These findings are presented in figure 1.

[3] 34w The results for second trimester respiratory infection were similar when the outcome was restricted to those subjects with a diagnosis of schizophrenia (RR = 2.07 [0.80-5.36], x 2 = 2.23, df= \,p = 0.14).

[1] 102w In a prospective birth cohort study, we demonstrated that second trimester respiratory infections were associated with an increased risk of schizophrenia and SSD in adulthood. As noted earlier in this article, previous studies of maternal respiratory infection have examined only influenza in relation to adult schizophrenia, and several have reported positive associations for second trimester exposure. Thus, in the present study, we have demonstrated that second trimester exposure to respiratory infections defined more broadly was also associated with adult schizophrenia. These findings provide further evidence that-at least for this class of infections-the second trimester is a period of increased vulnerability to schizophrenia.

[2] 208w We shall now consider potential mechanisms by which maternal respiratory infections may lead to SSD. For this to be the case, we must presume that these infections affect the development of the fetal brain. The fact that many of the clinical manifestations and treatments of each infection in this group are similar suggests that they may also share one or more components of a pathogenic mechanism (or several mechanisms) that results in schizophrenia. For example, hyperthermia, a cause of neural tube defects in animal studies (Edwards 1986;Milunsky et al. 1992), is a frequent manifestation of each of these infections; and fever during pregnancy has been associated with an increased risk of schizophrenia (Jones et al. 1998). In addition, over-the-counter or prescribed cold or flu remedies may be used to treat any number of respiratory infections, and some of these medications may have teratogenic potential (Lynberg et al. 1994). Moreover, it is also conceivable that the maternal inflammatory response to these infections may have led to a disruption of fetal brain development. Notably, proinflammatory cytokines and chemokines have been shown to play important roles in brain development, and abnormalities of these immune mediators are associated with neurodevelopmental disorders, such as cerebral palsy (Dammann and Leviton 1997;Nelson et al. 1998).

[3] 59w It is also possible, however, that each of the respiratory infections may lead to schizophrenia through different mechanisms that interact with second trimester neurodevelopmental events of potential relevance to schizophrenia, such as neuronal migration. Severe respiratory infections such as empyema are associated with bacteremia (Strand and Shulman 1988). By entering the maternal bloodstream, the organism may infect the pla-

[4] 88w Table 2. Rate ratios of schizophrenia spectrum disorder following maternal exposure to respiratory Infections: Unadjusted analyses Figure 1. Adjusted rate ratio of SSD following prenatal exposure to respiratory Infection, by trimester 1 1st Trimester 2nd TrtmeB* 3rd Trtntotsf 1 Rate ratio Is adjusted for maternal smoking, education, and race. Respiratory infections include tuberculosis, Influenza, Influenza with pneumonia, bronchopneumonia, atypical pneumonia, pleurisy, empyema/Viral respiratory infections, acute bronchitis, and upper respiratory infections. Smoking is defined as smoking during or until current pregnancy. Race Is designated as white, black, or other.

[5] 113w centa and amniotic cavity, leading to fetal infection. An infection of this level of severity could also compromise maternal oxygenation, leading to fetal hypoxia, a putative risk factor for schizophrenia (Buka et al. 1993). For influenza, investigators have proposed that maternal antibodies raised against the virus cross the placenta and disrupt fetal neurodevelopment by cross-reacting with brain antigens via molecular mimicry (Laing et al. 1989;Wright et al. 1999). It is possible that the other respiratory infections shown to be associated with SSD may impair brain development through similar types of autoimmune mechanisms. Clearly, identification of the specific infectious agents associated with SSD will be necessary in order to further elaborate the pathogenic mechanisms.

[6] 264w Limitations. Several of the limitations that pertain to the overall study design are described in detail in Susser et al. (this issue). Other limitations of the study include the following: Misclassification of exposure status. Gravidas may have been misclassified for infection as either false negatives or false positives. With regard to false negatives, it is worth noting that the data on infectious exposures were abstracted from medical records on gravidas' visits to Health Plan physicians. These consisted of visits for routine obstetric care and emergent medical problems. Thus, if a gravida did not seek attention for the symptoms of infection, and the treating physician did not detect any signs of infection, the illness would not have been recorded in the records or in the data base. This would have led to an underestimate of the proportions exposed. This limitation, however, is unlikely to have biased the findings in the direction of a spurious association, which would have occurred if the likelihood of an infection being detected was greater in mothers of SSD cases as compared with noncases. The fact that all of the gravidas were enrolled in the same health plan and virtually all received adequate followup during pregnancy argues against the possibility for bias. Further analyses demonstrated that the second trimester specificity of the association between maternal respiratory infection and SSD persisted regardless of the trimester in which prenatal care was initiated (these results are not presented). Thus, the rate ratios observed are more likely to have been biased toward the null, because of nondifferential misclassification, a result of underreporting of maternal infection.

[7] 121w Table 3. Relation of individual second trimester respiratory Infections with adult schizophrenia spectrum disorder Infection Tuberculosis Influenza Influenza with pneumonia Bronchopneumonia Pneumonia, atypical Pleurisy Empyema Acute bronchitis Upper respiratory infection Any respiratory infection With regard to false positives, it should be noted that while the infections in our data base were derived from patient history and physical examination (an important strength of the study), serological assay or culture was generally not used to confirm these infections. Yet, given that the care by Health Plan physicians was generally uniform and was probably not related to risk of SSD in the offspring, it is unlikely that mothers of SSD cases are more likely to have been misdiagnosed for infection than mothers of noncases.

[8] 70w Misclassification in timing of exposure. The gestational timing of infection was also based on the diagnosis of infection, not the actual occurrence of infection. Thus, we must consider whether there may have been misclassification with respect to the gestational period of the infection. However, because the respiratory infections were nearly all acute in nature, the diagnoses are generally unlikely to have postdated the actual infection by more than 1-2 weeks.

[9] 140w Lack of detailed information on clinical status and specific exposures. First, potentially important details on the clinical status of gravidas were often not available in the data base. This occurred because the present study relied on abstracted data on maternal infection, not the actual medical records. Although the abstraction procedure was thorough and systematic, it was meant to condense the data into ICD codes. For example, as noted above, hyperthermia may have been an important intervening variable between the infection and later risk of SSD, as well as an indicator of the presence and severity of infection, but this condition was not recorded in the data base. Second, for most infections, the specific infectious agent was not identified in the data base. Thus, at present, we can only speculate as to the pathogenic organisms that may increase risk for SSD.

[1] 160w In a prospective birth cohort study with detailed prenatal data, we have demonstrated that second trimester respiratory infections are associated with an increased risk of SSD in adulthood. While independent replications that address our limitations are essential, particularly given the small number of exposed cases, these findings may have important implications for elucidating the etiopathogenesis of SSD by providing valuable clues about specific causal infectious agents. We aim to follow up on these results by testing maternal sera drawn during the pregnancies of PDS Study birth cohort members for potential infectious etiologies. For example, the findings on respiratory infection can be further explored by serological analyses for influenza and a number of viruses that can cause colds or flu-like syndromes, such as adenovirus, parainfluenza virus, and rhinovirus. Moreover, we shall examine die role of potential intervening variables such as prescribed medications, hyperthermia, pro-inflammatory cytokines, and fetal hypoxia; and assess the contribution of possible effect modifiers, including familial vulnerability to SSD.

[2] 98w This study may ultimately have implications for the prevention of SSD. At least some of the organisms that may cause the maternal respiratory infections shown to be associated with SSD in this study can be readily treated with standard antibiotics. Others, can be prevented by vaccination. In addition, our findings would predict that changes in maternal risk behaviors for these infections, such as smoking, could lead to a reduction in SSD risk among offspring. The application of these public health measures to routine obstetric, gynecologic, and general medical care therefore offers hope for prevention of this devastating illness.

[1] 187w Sample. The PDS Study (Susser et al., this issue) is a continuous followup of schizophrenia and SSD in a birth cohort of 19,044 individuals whose mothers were recruited and enrolled in the Child Health and Development Study (CHDS). The cohort members were born from 1960 to 1967 in the Oakland hospital of the Kaiser Foundation Health Plan (Health Plan), a large prepaid health care plan that provided comprehensive medical and psychiatric services to all members. Records kept by the Health Plan included psychiatric and medical care received by members; these records were computerized beginning in 1981. Using these computerized records, potential cases of SSD were ascertained. Accordingly, the PDS Study cohort was defined as the subgroup of 12,094 individuals who were Health Plan members on December 31, 1980, or thereafter. These individuals represented 96 percent of the cohort members who were in contact with the Health Plan after age 10 years. The 6,950 members who left the Health Plan before December 31, 1980, and/or were not adopted were not included in the PDS Study cohort because they could not be ascertained for psychiatric disorder by computerized records.

[1] 57w In this article we examine the relation between prospectively documented maternal exposure to respiratory infections and schizophrenia in adulthood, in the Prenatal Determinants of Schizophrenia Study (Susser et al., this issue). We focus on respiratory infection because influenza virus, a member of this class of infection, has been previously demonstrated to be associated with schizophrenia in offspring.

[2] 115w In an ecologic study in Finland, Mednick et al. (1988) were the first to demonstrate an increased risk of schizo-phrenia following second trimester exposure to the 1957 A2 influenza epidemic. Since then, a host of ecologic studies attempting to replicate this initial finding have been conducted throughout Europe and in the United States, Japan, and Australia. Positive associations between exposure to the 1957 epidemic and births of preschizophrenia subjects were reported in Great Britain (O'Callaghan et al. 1991;Adams et al. 1993;Fahy et al. 1993;Takei et al. 1993), Ireland (Cannon et al. 1996), Japan (Kunugi et al. 1995), andAustralia (McGrath et al. 1994). All but one of those studies showed second trimester specificity for the association.

[3] 101w Nevertheless, there have been several negative studies of the 1957 influenza epidemic in relation to schizophrenia risk, including studies in England (Crow et al. 1991), Holland (Susser et al. 1994), the United States (Torrey et al. 1991), andCroatia (Erlenmeyer-Kimling et al. 1994). Studies over periods of many years that related the occurrence of influenza epidemics to the risk of schizophrenia have been largely positive (Barr et al. 1990;Sham et al. 1992;Adams et al. 1993;Takei et al. 1994Takei et al. , 1996;;Wright et al. 1995), although there have been some negative studies (Grech et al. 1997;Morgan et al. 1997;Selten et al. 1998).

[4] 86w There are three major limitations of these studies. First, to define an individual as exposed, an ecologic event (an influenza epidemic or series of epidemics) was used, rather than a clinically diagnosed infection within the individual pregnancy. To date, only two previous studies (Crow et al. 1991;Cannon et al. 1996) used data on maternal infection in individuals. Although neither study found associations between maternal influenza and schizophrenia, the studies were limited by use of midwife records, retrospective collection of information on influenza, and small sample sizes.

[5] 125w The second limitation of previous studies concerns incomplete adjustment for loss to followup, which can produce bias. This occurred because either the population at risk at the time of case ascertainment was not known, prohibiting enumeration of the number of subjects leaving the population from the time of birth, or because the ascertainment of cases occurred at only one point in time, preventing specification of when subjects left the cohort during the followup interval. Complete adjustment for loss to followup, however, requires knowledge of the number of subjects in the cohort at multiple points in time, requiring frequent followup over very small intervals (Susser et al., this issue). When this information is available, potential bias can be reduced using analytic strategies such as survival analysis.

[6] 51w The third limitation of previous work relates to the outcome. Diagnoses of schizophrenia were largely clinical and were obtained from hospital registries or reviews of hospital records, rather than being based on established research protocols. These relatively crude measures of outcome may lead to diagnostic misclassification and obscure comparisons across studies.

[7] 109w With regard to other prenatal infections, there have unfortunately been few investigations of their relation to schizophrenia. In published studies that examined other infectious agents and schizophrenia, associations were demonstrated for varicella zoster (Torrey et al. 1988;O'Callaghan et al. 1994), measles (Torrey et al. 1988), and diphtheria (Watson et al. 1984). Recently, Brown et al. (in press) have reported an increased risk of nonaffective psychosis, including schizophrenia, in a birth cohort serologically documented with in utero (mostly first trimester) exposure to rubella virus. These promising findings suggest that much more effort should be devoted to the potential role of other, viral and bacterial infections in the etiology of schizophrenia.

[8] 110w In the present investigation, the research design of the PDS Study permitted us to address those limitations. First, the maternal respiratory infections were carefully and prospectively documented in individual pregnancies of this birth cohort at the time of their occurrence; virtually all diagnoses of these infections were made by obstetricians or other physicians. Second, the continuous followup afforded by the present study enabled more complete adjustment for loss to followup. Third, most diagnoses were made following direct, structured interviews in accord with research diagnostic criteria (Susser et al., this issue). Moreover, the comprehensive data set of the PDS Study permitted us to markedly expand the range of respiratory infections covered.

[9] 142w Data on all diagnoses of medical conditions were abstracted in detail from the gravidas' Health Plan charts, over the period from 6 months before the gravida's last menstrual period (LMP) to the birth of her baby. The abstraction of data was conducted by a team of professionally trained individuals, and every abstract was carefully double-checked for errors. The medical conditions were coded in accord with the 1955 revision of the International Classification of Diseases (ICD). The calendar date of diagnosis of each condition was also abstracted, and it was converted to number of days before or after the LMP. Each diagnosed condition was also qualified by the abstracter in terms of the level of confidence that the gravida had that condition. For the present study, we focused on the data pertaining to definite, probable, or possible diagnoses of second trimester respiratory infections.

[10] 179w Definition of Maternal Infections. The primary exposure variable included all second trimester (gestational days 91-180) acute respiratory infections recorded in the CHDS data set. There were two main reasons for categorizing the infections in this way. First, previous studies have demonstrated an association between second trimester influenza and risk of schizophrenia. This body of work suggests that other respiratory infections may be implicated, since they each affect the same organ system and thus share many clinical features. Consequently, they could result in schizophrenia through similar pathogenic mechanisms. Second, other respiratory infections that coincide with the usual winter occurrence of influenza may contribute to-or confound-the associations between second trimester exposure to influenza epidemics and schizophrenia. Third, influenza can predispose patients to other respiratory infections. These respiratory infections included tuberculosis; influenza; influenza with pneumonia; bronchopneumonia; pneumonia, atypical; pleurisy; empyema/viral respiratory infections; acute bronchitis; and upper respiratory infections. For the purpose of comparison, we also examined whether first trimester (gestational days 0-90) and third trimester (gestational days > 180) exposure to these respiratory infections was associated with an increased risk of SSD.

[11] 177w The procedure for ascertainment and diagnosis of cases (Susser et al., this issue) is only briefly reviewed here. SSD was defined as schizophrenia, delusional disorder, psychotic disorder not otherwise specified, schizoaffective disorder, and schizotypal personality disorder (Kendler et al. 1995;Kendler and Walsh 1995). Cases of SSD from the PDS Study birth cohort were identified by a three-step procedure consisting of (1) ascertainment of inpatient or outpatient psychiatric treatment by Health Plan registries; (2) screening to identify potential cases using first the registry data, then review of abstracted psychiatric and medical records; and (3) consensus psychiatric diagnosis following direct interview with the Diagnostic Interview for Genetic Studies (DIGS) (Numberger et al. 1994); for noninterviewed subjects, psychiatric diagnoses were made based on chart review. A total of 71 cases of SSD were diagnosed, 44 by DIGS assessment and 27 by chart review. The diagnostic breakdown of SSD cases was as follows: 43 with schizophrenia, 17 with schizoaffective disorder, 5 with schizotypal personality disorder, 1 with delusional disorder, and 5 with "other schizophrenia spectrum psychoses" (Susser et al., this issue).

[12] 181w Data Analysis. The sample selection for the data analysis of the PDS Study cohort is fully described in Susser et al. (this issue). Briefly, the sample for the present study consists of the 7,796 subjects of the 12,094 PDS Study cohort members who remained after the following two exclusionary steps. First, all noncases missing maternal interview data were excluded from the sample since the present study relies heavily on these data for information on exposure status and potential confounders. Second, the sample was reduced by including only one sibling per family. (This step was necessary because siblings represent nonindependent observations.) If a sibship included a case, the case was retained and all unaffected siblings were excluded. If a sibship did not include a case, one sibling was randomly selected for inclusion, and all others were excluded. The resulting analytic sample consists of the 71 SSD cases with maternal interview data and 7,725 noncases taken from the 12,094 PDS Study cohort members. In this analysis, we then excluded the 13 SSD cases who were missing maternal interview data, resulting in 58 cases.

[13] 130w The data were analyzed using proportional hazards regression (Kalbfleisch and Prentice 1980;Collett 1994), a method of survival analysis that takes into account different durations of followup among subjects and allows adjustment for potential confounders. The logrank test was used for unadjusted comparisons of survival distributions. Duration of followup was quantified as the time (in years) elapsed from the beginning of observation (birth) until the end of observation (the date of first treatment for SSD cases, or until either the end of Health Plan membership or the end of the study for censored observations). The population at risk is defined for each case as all subjects who were still Health Plan members and still at risk of failure (diagnosis of SSD) on the date at which the case first received treatment.

[14] 271w The primary exposure variable (second trimester acute respiratory infection) and the other trimester-specific exposure Variables were defined as binary indicator variables in these analyses. The timing of infection was based on the first diagnosis of the infection during the pregnancy, in relation to the LMP. A woman was coded as positive for acute respiratory infection during the second trimester, for example, if any of the indicated conditions for respiratory infection was recorded at least once between gestational months 4 and 6 (days from LMP were converted to months to determine the gestational month of infection). The number of infections was not considered in the analyses. This type of coding in the proportional hazards regression analyses yields estimates of the ratio of incidence rates of SSD for gravidas who experienced second trimester respiratory infection, versus gravidas who did not experience second trimester infection (assuming that the rate ratio is constant over time). The model also provides an estimate of the standard error of each rate ratio estimator, which can be used to generate test statistics and confidence intervals. Unadjusted rate ratios were obtained by fitting proportional hazards models with the exposure of interest as the only predictor. Subsequent models included potential confounding variables and yielded adjusted rate ratios. Further analyses, including assessment of first and third trimester respiratory infection, and the more restricted definition of caseness (i.e., schizophrenia only), were conducted using the same statistical approaches. An association between an exposure and SSD risk was considered statistically significant in the conventional sense when the p value was < 0.05 for a two-sided test of whether the rate ratio was equal to 1.

[15] 216w Potential confounders. Potential confounders were selected a priori in two stages. In the first stage, we considered all exposures with at least suggestive evidence from previous studies of associations with maternal infection. These included maternal age, smoking, education, race, parity, alcohol use, and marital status. We then conducted bivariate analyses, which separately examined each of these exposures for associations with second trimester respiratory infection, our main hypothesized independent variable. In these analyses, the variables were categorized as follows: maternal age (20-24 [reference category], 1-19, 25-29, 30-34, 35-39, a 40); smoking (during or until current pregnancy vs. nonsmoking [never smoked, smoked prior to current pregnancy; reference category]); education (high school graduate only [reference category], less than high school, some college, college graduate/RN); race (European white [reference category], black, other); and parity (0, 1, 2, 3, > 4). The results of the bivariate analyses are presented in table 1. As is evident from the table, we found that of all of the confounders examined, only maternal smoking had a marked association with second trimester respiratory infection. In the final model, we therefore included maternal smoking. Although maternal race and education were not associated with respiratory infection in our data set, we nonetheless included these two variables in the final model because of their well-established relation to respiratory infection.

[16] 38w Infections With SSD. Table 3 presents the breakdown of specific second trimester maternal respiratory infections by SSD case status. As can be seen, the above associations with SSD were accounted for mainly by upper respiratory infection and empyema.

[17] 165w It is conceivable that our findings may have been confounded by genetic loading for schizophrenia in the mothers, which may be associated with increased respiratory infections as a consequence of less prenatal care. To address this question, we examined whether maternal exposure to second trimester respiratory infection was related to mental conditions in the mother (including psychosis, anxiety reactions, and drug and alcohol addiction). We found no evidence of an association between maternal respiratory infection and these mental conditions. Moreover, we explored whether the timing of initiation of prenatal care (a proxy measure for level of prenatal care) may have affected the association between maternal respiratory infection and risk of SSD. We found that the addition of this covariate to the multivariate model had no effect on the strength or statistical significance of the association (RR = 2.15 [1.05-4.39], x 2 = 4.38, df= \,p = 0.04). Thus, we found no evidence that either maternal mental conditions or less prenatal care confounded the observed relationship.