The CoLaus|PsyCoLaus cohort
CoLaus|PsyCoLaus [22, 23] is a prospective cohort study designed to study cardiovascular risk factors (CoLaus) and mental disorders (PsyCoLaus) in the community and to determine their associations. The baseline investigation was carried out between 2003 and 2006. For the allocation of the initial CoLaus sample, the subjects were randomly selected among the 35- to 75-year-old residents of the city of Lausanne (Switzerland) according to the civil register (n = 6734). The survey was conducted in the French language. Participation in the psychiatric evaluation was first confined to 35- to 67-year-olds of the CoLaus sample; 67% of subjects within this age range (n = 3720) agreed to participate in the PsyCoLaus part.
The first follow-up of CoLaus|PsyCoLaus was conducted between 2009 and 2012. Information on physical health was obtained from 5064 (75%) of the n = 6734 participants. In the PsyCoLaus subsample, 87% remained in the study, i.e., 3188 out of the 3673 participants. Participants who had missed the psychiatric part of the baseline examination (including those initially aged 67–75 years) were given the opportunity to complete it at follow-up. Therefore, the sample comprising baseline data from both interviews increased by n = 1154 and summed to n = 4874 (see overview in Fig. 1).
The Diagnostic Interview for Genetic Studies
The main survey instrument of PsyCoLaus was the French version of the semi-structured Diagnostic Interview for Genetic Studies (DIGS) [24, 25]. The DIGS collects information related to common mental disorders and returns lifetime prevalence information. It includes also data related to the course and chronology of comorbid features, somatic conditions, and a range of developmental and psychosocial risk factors [22]. A French version [24] of the DIGS [25] was used to assess diagnostic information on mental disorders (see below).
The DIGS collects information on a broad spectrum of the Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) Axis I criteria related to mental disorders and, moreover, on the course and chronology of comorbid features [25]. The brief phobia section of the DIGS was replaced by the corresponding, more extensive, sections of the Schedule for Affective Disorders and Schizophrenia—Lifetime and Anxiety Disorder Version (SADS-LA) [26]. This elicited detailed information related to the DSM-IV criteria for agoraphobia with or without panic attacks, social phobia, and specific phobias. The DIGS and the SADS-LA elicit lifetime diagnoses of common mental disorders. Successful inter-rater and test-retest reliability of the French version of the DIGS have been established for major mood and psychotic disorders [24] as well as for substance use and antisocial personality disorders [27]. Similarly, for the anxiety sections of the French version of the SADS-LA, inter-rater and test-retest reliability are good [28]. In this analysis, we grouped mental disorders into five groups: neurodevelopmental, early-onset anxiety (average age of onset up to 12), late-onset anxiety (average age of onset above 12), mood, and substance use disorders (see notes to Table 3).
The information on lifetime diagnosis of infections and somatic conditions was derived using an extended version of the medical history parts of the DIGS and an early version of the Schedule for Affective Disorders and Schizophrenia—Lifetime and Anxiety Disorder Version (SADS-LA) [26] and was based on reporting of the participants. The questions were related to ever having been diagnosed with various infections, diseases of the nervous system, and cardiovascular, respiratory, gastrointestinal, metabolic, and dermatological conditions as well as allergies and hormonal problems. The lifetime prevalence of migraine was assessed according to the criteria of the International Classification of Headache Disorders (ICHD-II) [29], using the validated French version of the Diagnostic Interview for Headache Syndromes (DIHS). For the present analysis, infections (chickenpox, measles, mumps, rubella, scarlet fever, pertussis, herpes labialis), atopic diseases (allergic asthma, other asthma, allergic eczema, other eczema, hay fever, urticaria, drug allergy), chronic inflammatory diseases (ulcer, irritable bowel syndrome, cystitis, acne, psoriasis, plus migraine), and mental disorders were selected. The age range of the sample implies that very few participants had received any measles-mumps-rubella vaccine in childhood, as measles and other vaccination schedules were only introduced by Swiss health authorities in the 1960s and 1970s and were used by a minority of children at the beginnings.
Family-related ACE were represented by the following questions (variables):
In addition, the information about traumatic experiences in childhood was extracted from the DIGS section on posttraumatic stress disorder. Potentially traumatizing events comprised accidents, physical assaults, witnessing murder, violence or death by accident, sexual abuse, and, without relevance for the present study, combat and/or war. The age before which the event must have occurred was 10 years. Socioeconomic status was assessed using the Hollingshead index [30].
Inflammatory markers and WBC counts
The biomarkers analyzed in this study comprised white blood cells (WBC) and inflammatory markers (interleukin-1β (IL-1β), interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α), and high-sensitivity C-reactive protein (hsCRP)). WBC and inflammatory markers were assessed in the baseline examination (PsyCoLaus subsample n = 4671) and the inflammatory markers also in the first follow-up (n = 4057).
Morning venous blood samples were taken from participants without current infection and allowed to clot. For the cytokine measurements, serum was preferred to plasma, as it has been shown that different anticoagulants may differentially affect absolute cytokine levels [31]. Serum samples were stored at − 80 °C before assessment and sent on dry ice to the laboratory. Cytokine levels were measured using a multiplexed particle-based flow cytometric cytokine assay [32]. Good agreement between signal and cytokine was found within the assay range (R2 ≥ 0.99). Lower detection limits (LOD) were 0.2 pg/ml. For concentrations below the LOD, a value of 0.1 pg/ml was assigned.
hsCRP was assessed during baseline and follow-up physical evaluations using immunoassay and latex HS (IMMULITE 1000-High, Diagnostic Products Corporation, LA, CA, USA), with maximum intra- and interbatch coefficients of variation of 1.3% and 4.6%, respectively [23]. Subjects with a hsCRP level higher than 10 mg/l were excluded as such an elevation is likely to be attributable to acute infection.
WBC counts were performed on an XN-2100 apparatus (Sysmex, Horgen, Switzerland) during the first follow-up of CoLaus. Data was available for 2963 participants. Participants with/without WBC counts did not differ as to sex and age; however, the latter had a slightly lower SES score (3.29 vs. 3.41, p = 0.002). Neutrophils, lymphocytes, monocytes, eosinophils, and basophils were represented as proportions of total leukocytes. The WBC variables were smoothed (square root, log)—if appropriate—in order to improve the statistical parameters of the distributions.
In order to combine values from the baseline and the follow-up measurement, the inflammatory marker values were transformed to ranks. In a preliminary step, the ranks were calculated for the participants with both measurements (by sex and separately for each measurement). In the next step, participants with only one measurement were given the ranks of their neighbors. If ranks of both measurements were available, the mean rank was used, otherwise the rank of the available measurement.
Statistical analysis
LCA has been used in comorbidity analyses both in somatic medicine [33,34,35,36,37] and psychiatric research [38,39,40]. LCA is a person-centered approach to classification, i.e., it aims to group individuals into homogeneous classes [41, 42]. The classes represent subgroups of individuals based on similar responses and characteristics.
The LCA in this study was based on variables representing infectious childhood diseases (chickenpox, measles, mumps, rubella, herpes simplex, pertussis, scarlet fever), atopic diseases (hay fever, asthma, eczema, urticaria, drug allergy), and childhood adversities (interparental violence, parental maltreatment, trauma in early childhood). LCA was conducted using Mplus version 7 [43]. To avoid problems with local maxima, the number of random starts was set to 2500 for the first step, using the 250 best solutions in the final calculation. One to six latent class models were routinely fitted to the data in order to determine the optimal number of latent classes in the final model. We considered several fit indices: the Akaike information criterion (AIC), the Bayesian information criterion (BIC), and the sample size-adjusted BIC (ABIC) and in addition the Lo-Mendell-Rubin likelihood ratio test (LMR-LRT) [44]. Typically, we prefer models with a number of classes between the number suggested by the BIC and the number suggested by the AIC. The model selection is furthermore determined by the distinction between the classes, their size, and their theoretical adequacy.
In further analyses, we applied cross-tabulations, analysis of variance (ANOVA), general linear models, Kruskal-Wallis test, and logistic regression models. Programming and this part of the analyses were carried out with SPSS Statistics (version 23). We explicitly did not perform adjustment for multiple testing for several reasons. First of all, there are two technical reasons:
-
Adjustment for multiple testing is appropriate under specific conditions including homogeneous items (as well as large Ns, precise measurements, homogeneous conditions, etc.); in contrast, the diseases/disorders we are investigating subsume heterogeneous conditions and comprise different subtypes; this applies not only to grouped disorders but similarly also to most of the specific disorders or diseases comprised in this study; one part of the analysis strategy is to prioritize the β-error over the α-error, that is, to increase the probability to account for or to detect minor subtypes
-
Another part of the analysis strategy is to focus on pattern recognition, explicitly through the LCA in this study, but also implicitly in analyses of groups of markers and disorders; pattern recognition is prioritized over analysis of single conditions; adjustment for multiple testing is uncommon in any pattern recognition analysis such as involving LCA, factor analysis, and correspondence analysis (see also [45])
Beyond these reasons, there are methodological pros and cons whether statistical precision could be helpful in crunching complex issues. We assume that complexity adapted modeling strategies are currently more urgently needed.