Schizophrenia is a chronic neuropsychiatric condition with a prevalence of approximately 1.1% among adults in the United States (1). Recent psychopharmacologic advances have led to treatment options, but their effectiveness is variable. While research into the etiology of schizophrenia has been ongoing for more than a century, the causes of this illness remain unknown (2).

Several risk factors have been identified, including family history/genetic predisposition, season of birth, urban birth, lower socioeconomic status, and prenatal or birth complications (2, 3). An infectious agent as the cause as well as a link between these risk factors have been proposed for more than a century and are supported by a growing body of research (2).

Much of the recent research has focused on the potential relationship between the development of schizophrenia and infections with, or antibodies to, neurotropic agents, such as Herpes simplex family viruses or Toxoplasma gondii (T. gondii) (4—9). The possible link between Toxoplasma and schizophrenia is supported by the occurrence of psychiatric symptoms in some individuals with acute infection (10) as well as by animal models documenting neurological and behavioral changes during acute infection (11, 12).

A recent review of the literature found that the preponderance of evidence obtained through 2003 supports an association between Toxoplasma and schizophrenia (13), which was confirmed by a recently published meta-analysis (14). Research designs have varied, but most studies have assessed antibody levels among acutely ill or first-onset patients (15, 16). Increased serum antibodies to T. gondii were found in patients relative to comparison subjects in several different studies (13, 15, 17).

However, notable limitations to these studies are small sample sizes and the reliance on antibody measurements from a single serum sample obtained after diagnosis. Studies of pre-onset serologic and demographic data on large numbers of affected individuals and appropriately matched comparison subjects are difficult to conduct because population-based medical and serologic data are not routinely collected. The U.S. military, however, collects data from all personnel. Serum samples are obtained at military accession and approximately every 2 years thereafter (18). Medical encounter data are electronically archived (19). This affords a unique opportunity to conduct large, population-based serological analyses on specimens obtained before the diagnosis of disease. We are conducting a series of nested case control studies to explore the hypothesized associations between psychotic disorders and selected infectious agents. In the present hypothesis-generating study, we describe findings from a panel of infectious agents that were analyzed, and we discuss an association between increased levels of antibodies to T. gondii and subsequent risk of new-onset schizophrenia among military personnel.

This project was reviewed and approved by the appropriate human protection committees. Patients were obtained from the physical disability agency databases of the Army, Navy, and Air Force and were defined as all military members who were medically discharged from military service between 1992 and 2001 for disability associated with a diagnosis of schizophrenia and who had banked serum available in the Department of Defense Serum Repository. Essentially all medical care received by military service members is through the military health care system, enabling a very complete capture. Patients and comparison subjects were selected to minimize diagnostic misclassification and maximize contrast on disease status. Prior to discharge, all patients had an extensive psychiatric evaluation that applied DSM-IV criteria with full clinical history; two psychiatrists independently reviewed and concurred on diagnosis and severity, and a third physician approved their findings. A separate physical evaluation board, including one physician, determined the fitness for duty. Onset was estimated by the first military hospitalization with a mental health diagnosis (ICD-9-CM codes 290—319). Healthy comparison subjects were selected from the population of military service members with no inpatient or outpatient diagnosis of a mental health disorder. Healthy comparison subjects were matched in a 3:1 ratio with patients on the following variables: date of birth (SD=1 year), accession date (SD=6 months), sex, race (black, white, other), branch of military service, and number of serum specimens available. A convenience cohort of potential study subjects—the first 200 subjects identified by social security number order—was selected. Patients without mental health hospitalization (N=20) were excluded because the date of onset of schizophrenia could not be estimated. The cohort of patients and matched healthy comparison subjects were used to develop hypotheses to be tested in a larger study under way of 855 patients and 1,165 comparison subjects. The 180 patients included in the present study were generally representative of the approximately 1,200 patients identified.

The Department of Defense Serum Repository provided serum specimens. The mission of the Department of Defense Serum Repository includes storage of serum that remains following mandatory human immunodeficiency virus- and operational deployment-related testing. The Department of Defense Serum Repository stores millions of specimens in a secure facility maintained by the Army Medical Surveillance Activity. All specimens are maintained at a temperature of at least —30° (19). Study sera were selected following the schematic shown in Figure 1. We attempted to obtain the same number of specimens for patients and comparison subjects, which were as follows: first available specimen during the accession medical examination (usually collected), a specimen collected during the 3- to 24-month period prior to the first mental health hospitalization, and the first specimen available after hospitalization.

The measurements of immunoglobulin G (IgG) antibody levels to T. gondii, six Herpes viruses, and influenza A and B viruses and immunoglobulin M (IgM) antibody levels to T. gondii were performed using microplate enzyme immunoassay (20—23), a method of antibody detection that is widely used in research of this nature (15, 16, 24) and found to be reliable relative to results obtained from other methods (25, 26). The enzyme-linked immunosorbent assay (ELISA) method is commonly used in large studies because of its relative speed and small volume of serum sample required. Enzyme immunoassay consists of binding serum to solid-phase antigen and subsequent reactions with enzyme-labeled antihuman IgG, or IgM, and enzyme substrate. In the present study, the amount of color generated by the enzyme substrate reaction was measured in optical density units by means of a microplate colorimeter. This method was selected because it allows for high throughput measurement of antibodies using a common platform and requires small amounts of serum sample. Standard serum samples of known amounts of antibodies were run on each microwell plate. Samples were run under code in matched groups in which neither patient nor comparison status was identified.

Matched case control serum samples were tested on the same microtiter plates; over the course of the study, samples were assayed on more than 20 different plates. To control for potential systematic error introduced by plate-to-plate variation, continuous data were normalized using the robust median normalization method, which combines the within-plate and between-plate variances (27). The manufacturer’s methods were followed in determining the T. gondii IgG status (negative, equivocal, or positive) of the serum specimens by using the optical density cutoff-value adjusted by the blank value. T. gondii IgG test results were categorized as follows: negative if the optical density cutoff-value was <0.9, positive if the optical density cutoff-value was >1.1 , and equivocal if the optical density cutoff-value was ≥0.9 or ≤1.1.

Conditional logistic regression was used to investigate the relationship between patient or comparison status and antibody levels. Failure to account for matching in logistic regression may lead to biased results, usually toward the null. During the conditional logistic-model-parameter estimation process, the model with higher (less negative) log likelihood, suggesting a better "fit" for the data, was chosen (28).

The proportional hazards model, which allows for multiple and variable numbers of comparison subjects per case and serum specimens per subject, is commonly used for conditional logistic regression modeling (28, 29). In the proportional hazards model, the estimate of association between patient or comparison status and the risk factors are reported as the hazard ratio, which is computed by exponentiating the parameter estimates. If the hazard ratio associated with a risk factor is larger than 1, an incremental increase in that factor increases the hazard rate. If the hazard ratio is less than 1, an incremental increase in the factor decreases the hazard rate.

Because scaled values with no recognized unit of measure were used to represent the antibody level, we chose the standard deviation to represent one unit of change. All results are reported as the hazard ratio for each increase of one standard deviation of the antibody level. This method also allows comparability between the effects of antibodies to other agents evaluated.

We studied the overall association as well as the temporal association between T. gondii IgG and schizophrenia, modeling T. gondii IgG alone and simultaneously with the other eight antibodies from all matched serum samples. The overall hazard ratio showed the average effect of T. gondii IgG on schizophrenia for the entire study period, and the temporal hazard ratio showed the average effect of T. gondii IgG for a given time to diagnosis. We also conducted stratified analyses to investigate heterogeneity of the T. gondii effect across demographic factors. Results from analysis of the other infectious agents will be detailed in a subsequent report.

Optical density was analyzed as both a continuous and categorical variable. When trying to detect a difference in antibody levels between patients and comparison subjects, however, the model using T. gondii IgG as a categorical variable is less informative than the continuous model. Therefore, while we report results from the categorical analysis, subsequent models include T. gondii IgG as a continuous variable only. In addition, several interaction terms were evaluated. The interaction with serum collection time to time of diagnosis shows the temporal effect of antibody levels and describes the consistency of risk across time periods. All analyses were conducted using SAS, Version 9.1.3.

Of the 1,261 potential study subjects, 99.5% had schizophrenia as the first-listed diagnosis, and more than 95% had no comorbid conditions. Among those with comorbid conditions (N=64 [<5%]), 44 subjects (3.4%) had a medical diagnosis, primarily musculoskeletal, and 21 subjects (1.6%) had an additional psychiatric diagnosis. One subject had both a medical and psychiatric comorbid condition.

A total of 180 patients and 532 comparison subjects were included in the study population. Thirty-two patients had one serum specimen available, 72 had two serum specimens, and 76 had three serum specimens. Eight patients could only be matched to two comparison subjects. Table 1 shows the demographic characteristics of patients and comparison subjects. Approximately 83% were men; more than 57% were younger than 25; 10% were older than 35 years; approximately 12% were Hispanic; and more than 56% were in the Army. Approximately 35% of patients had greater than 3 years of military service. Analysis of T. gondii IgG level-by-specimen demonstrated 28 (7%) positive serum specimens among patients, 369 negative specimens, and seven equivocal specimens. Among comparison subjects, there were 63 (5.3%) positive serum specimens, 1,110 negative specimens, and 18 equivocal specimens. Analysis of T. gondii IgG level-by-subjects showed that 15 (8%) patients and 37 (7%) comparison subjects had at least one positive specimen, while 165 (92%) patients and 491 (92.3%) comparison subjects had only negative specimens; no patients and only four comparison subjects had both equivocal and negative specimens.

We assessed the overall antibody effect for each agent (Table 2). For the analysis of T. gondii IgG alone as a continuous variable, the antibody-level effect on the risk of schizophrenia was significant (hazard ratio=1.24, p<0.01). In the model including all nine infectious agents, the T. gondii IgG effect remained significant (hazard ratio=1.26, p<0.01). When analyzed categorically by specimen, T. gondii IgG positivity was a similar, although nonsignificant, predictor of risk for schizophrenia (hazard ratio=1.31, p=0.25). Human herpesvirus (HHV)-6 (hazard ratio=1.20, p<0.01) was the only other infectious agent to have a significant positive association. Further results in this study will focus on the T. gondii IgG finding; analysis of the association between HHV-6 and risk of schizophrenia will be reported separately. None of the interactions between antibody levels and demographic factors was significant, implying that the overall estimation of effect, as shown in Table 2, is suitable.

No substantial effect modification was noted for sex (men versus women), race (black versus white), or age (<25 versus ≥25), as shown in Table 3. Because the vast majority of subjects were men (83%), the T. gondii IgG effect among men was similar to all subjects. For women, the hazard ratio was lower and not significant.

The time range from serum sample collection to the date of diagnosis varied from 11 years before to 1 year after diagnosis. To study temporal effects, the following six different time-to-diagnosis periods were defined: more than 3 years, between 2 and 3 years, between 1 and 2 years, 6 months to 1 year, within 6 months of diagnosis, and after the diagnosis date. The comparison subjects’ specimens corresponded to the same periods relative to the matched patient diagnosis date. Of note, none of the subjects are represented in more than three of the time periods. Different effects were noted across periods as shown in Table 4. The hazard ratio for T. gondii was significantly greater than 1.0 for those specimens collected within 6 months prior to diagnosis and after diagnosis, but not for other time-to-diagnosis periods.

In this study assessing relationships between antibody levels to infectious agents and risk of new-onset schizophrenia, we found a significant association between T. gondii IgG levels and schizophrenia. The addition of other infectious agents to the model did not affect this finding, nor was it substantially modified by sex, age, or race. The T. gondii IgG effect was consistent in nearly all time periods analyzed, both before and after diagnosis.

When T. gondii IgG was studied categorically (positive, equivocal, and negative), results were similar to the continuous analyses (hazard ratio=1.31) but not significant. The lack of significance may be attributable to the small proportion of seropositive specimens (7% for patients; less than 6% for comparison subjects). The hazard ratio for equivocal versus negative was 1.20 (not significant), indicating a positive tendency.

The study cohort size for female and nonwhite subjects was underpowered to assess differences in antibody-level effect. A similar association between T. gondii IgG levels and risk of schizophrenia was observed among these categories. These findings require further exploration in a larger study.

A significant positive association was also observed with HHV-6 IgG levels and schizophrenia, but not with any of the other agents tested. The Herpes family virus antibody levels were included in the model to control for any effect on the relationship between T. gondii antibodies and schizophrenia. A detailed evaluation of the relationship between these viral agents and schizophrenia will be provided in a separate study.

This study has several strengths. Since the study population had serum samples stored in a repository, we were able to retrieve and analyze samples obtained from apparently healthy individuals prior to disease diagnosis. This allowed us to document that elevated levels of Toxoplasma antibody existed prior to symptom onset, making it unlikely that the observed elevations were an artifact of disease-related exposures or a reflection of genetic predisposition to disease. We were also able to obtain multiple specimens per study subject, which will allow detailed longitudinal analysis in future studies. Furthermore, the study design allowed matching of patients and comparison subjects on demographic variables, making it unlikely that the case-control differences observed were related to confounding or other artifactual differences in the two populations.

There are, however, limitations to the current study. Ninety percent of the patients who were potentially eligible for study inclusion were hospitalized in a U.S. Department of Defense medical treatment facility with a mental illness diagnosis. Because of the lack of existing data for most of the study period, we were not able to identify any hospitalizations in non-Department of Defense facilities. According to Hoge et al. (30), less than 2% of mental health hospitalizations among military service members occur in civilian facilities for emergencies or in remote locations. Therefore, we do not feel that this is an important source of bias or diagnostic misclassification.

In addition, the data are inadequate to evaluate the impact of changes in scaled T. gondii IgG levels over time within individuals in the study population. Because of limitations on serum availability, we could not ensure that every subject was represented in each time period analyzed, resulting in varying distributions of demographic factors across time-to-diagnosis periods (data not shown). Further, we could not obtain equal numbers of specimens per subject. Although variation of the T. gondii effect by gender, race, and age was not significant, marginal differences were observed. The observed temporal effect therefore might also be because of population-specific factors in addition to differential exposure to T. gondii. The model assumed that the log (hazard ratio) is linearly dependent on T. gondii for the continuous predictor. We explored a variety of nonlinear modeling techniques but found our cohort size insufficient to fit them. We expect to find a nonlinear association with a larger cohort size in future analyses, which will increase confidence in the results and allow us to study the residuals, sensitivity, specificity, accuracy, and reliability of the model. Categorical analysis of the data was limited by the very small proportion of T. gondii-positive specimens, resulting in a nonsignificant, although positive, association between risk of schizophrenia and antibody levels.

Finally, identification of subjects for this study relied on diagnostic data obtained through electronic records rather than clinical interviews. Findings from an independent-record review of patients’ medical records, however, demonstrate a greater than 90% verification of diagnoses (31), indicating that diagnostic misclassification among patients is unlikely to be a major source of bias. Additionally, because a diagnosis of schizophrenia in this population is extremely rare—approximately 0.15 per 1,000 per year since 1990 (19)—it is likely that few, if any, of our comparison subjects were later diagnosed with schizophrenia. Any subsequent misclassification is unlikely to bias the results.

The generalizability of these results to the broader population of individuals in the United States with schizophrenia may be limited by the use of military study subjects. Military populations with schizophrenia differ somewhat from other populations in that they tend to have an older age at onset, most likely because of self-selection and medical evaluation and screening prior to entry into the military. This selection process results in few persons with overt psychosis entering the military. Additionally, the study population differed because all the individuals started out "healthy" and were drawn from a population that is younger than the general U.S. population, with a higher proportion of ethnic minorities and equal access to medical care.

Our findings are consistent with previous studies indicating increased level of antibodies to T. gondii in individuals with schizophrenia (15, 17). However, in previous studies T. gondii antibodies were measured after diagnosis, raising the possibility that the increased levels of antibodies were the result of disease-related environmental factors (32). This study indicates that increased levels of Toxoplasma antibodies can be found prior to the onset of symptoms and are thus unlikely related to symptom-associated exposures alone.

One challenge to this type of epidemiologic research, involving specimens stored for long and varying periods, is the lack of universally utilized assays and cut-points for ascertaining serum status (positive or negative) and measuring antibody levels. Various strategies have been employed, including arbitrary cutoff selection; categorization by percentile ranges; grouping results as low, intermediate, or high; and analyzing antibody levels as a continuous rather than dichotomous or categorical variable (15, 16, 24, 33). In spite of these variations, the consistent finding from most studies is an increased risk of schizophrenia among individuals with high T. gondii antibody levels, however defined (14—16, 33).

Given the high prevalence of T. gondii (approximately 20%) in the United States (34), it is clear that most infected individuals do not have schizophrenia. The reasons for potential differential reactions to Toxoplasma infection in patients with schizophrenia is not known but may be related to genetic determinants of host susceptibility (35, 36), varying degrees of pathogenicity among infecting organisms (37), or unidentified environmental factors. The timing of infection may be an important differential risk factor, since previous studies indicate that children born to mothers with increased levels of antibodies to T. gondii are at risk for the development of schizophrenia later in life (32, 38). One possible explanation of this finding is that increased antibody levels associated with schizophrenia in adults may be related to immunologic reactivation of infections acquired earlier in life.

The mechanisms by which exposure to T. gondii might lead to schizophrenia are not known. However, animals experimentally infected with T. gondii show altered behavior, including abnormal reaction to novel environmental stimuli (39). Altered novelty-seeking behavior has also been found in humans with Toxoplasma antibodies (40, 41). Since treatments are available for Toxoplasma, early detection of T. gondii infection and initiation of therapy may play a future role in schizophrenia prevention or disease modification.