Design and data source

The current study cohort was based on a nationally representative sample of community-dwelling individuals aged 65 and over from the Republic of Ireland. Data were obtained from the Irish Longitudinal Study on Ageing (Kenny, Reference Kenny2018). This large-scale, nationally representative, aging study was conducted in Ireland. The current data were derived from interviews, undertaken between January 2016 and December 2016. The response rate was 62.0%. The study was approved by the Faculty of Health Sciences Research Ethics Committee in Trinity College Dublin. Further details of sampling and study design have been described elsewhere (Whelan and Savva, Reference Whelan and Savva2013).

Participants

Eligible respondents for this study included community-dwelling individuals aged 65–85 (N = 3434). People with known or suspected dementia were not recruited. A subsample of people who did not have any information on past illnesses of any sort were removed (N = 440). Therefore, the final sample was 2994 people.

Cognitive functioning

Cognitive functioning was assessed with the Mini-Mental State Examination (MMSE) (Folstein et al., Reference Folstein1975). The MMSE is one of the most widely used research and screening measures of cognitive functioning in aging populations (Folstein et al., 1975, Reference Folstein1983). It is easy to administer, available in over 15 languages (Steis and Schrauf, 2009), and has high levels of acceptability by health professionals and researchers as a diagnostic instrument (Nieuwenhuis-Mark, 2010). The scale has 20 items to assess orientation, recall, attention, calculation, language, and visuospatial abilities. It is divided into two sections, the first of which requires vocal responses only and covers orientation, memory, and attention; the maximum score is 21. The second part tests cognitive abilities to name, follow verbal and written commands, write a sentence spontaneously, and copy a polygon; the maximum score is nine. The maximum overall score is 30, with higher scores representing better cognitive functioning (Folstein et al., Reference Folstein1983). Questions were asked and scored immediately. The test was not timed. The tester was instructed first to make the individual comfortable, to establish rapport, to praise successes, and to avoid pressing on items which the individual finds difficult.

In the current study, MMSE scores were computed, based on the original instructions (Folstein et al., Reference Folstein1975) for each of the five subscales: orientation, registration, attention and calculation, recall, and language. Total scores were computed as well. For the primary analysis, the total test scores were not dichotomized into categories of impaired cognitive functioning and intact cognitive functioning because dichotomizing leads to information lost, increased risk of a false positive and may seriously underestimate the extent of variation in outcome such that individuals close to but on opposite sides of the cut point are characterized very differently rather than similarly (Altman and Royston, Reference Altman and Royston2006). However, for robustness, when recomputing the sensitivity analysis, the total test scores were dichotomized into categories of impaired and intact cognitive functioning.

Childhood infectious diseases

The childhood infectious diseases selected for this study were chickenpox, measles, and mumps. These were chosen because they are viral diseases, infection with which are followed by an enduring immunity (Simpson, Reference Simpson1952) and were common at the time of our sample’s childhood years due to recurrent outbreaks (London and Yorke, Reference London and Yorke1973). Childhood infectious diseases were assessed self-reported by asking participant if they have had chicken pox, measles, and mumps in childhood. Responses were classified as “yes” (score of 1), “no” (score of 0), or “don’t know” (coded as missing values). A total number of childhood infectious diseases score was computed for each person as the sum of the reported diseases of chicken pox, measles, and mumps. The total score ranged from 0 (having had no diseases) to 3 (having had all three diseases).

Covariates

The following covariates were included in the analyses to adjust for potential confounding, model effect modification and to define subgroups of particular interest with possible differential effects on cognitive functioning. The covariates considered (and entered in the following order) were demographics, metabolic, and psychiatric. Demographic covariates were age at the time of data collection and sex (van der Flier and Scheltens, Reference van der Flier and Scheltens2005), highest education achieved [primary or none, secondary, third, or higher (Solfrizzi et al., Reference Solfrizzi2004; Tervo et al., Reference Tervo2004; Tyas et al., Reference Tyas2007)], and smoking status at the time of data collection [classified as smoker or non-smoker (Swan and Lessov-Schlaggar, Reference Swan and Lessov-Schlaggar2007)]. The metabolic covariate was body mass index categories [0–25, 25–30, 30–40, 40+ (Atti et al., Reference Atti2008)]. The psychiatric variable was depression (Butters et al., Reference Butters2008) as measured by the Composite International Diagnostic Interview (Robins et al., Reference Robins1988). This established instrument is widely used for assessing a clinical diagnosis of major depression in epidemiological and clinical studies. Respondents received binary scores depending on whether they had fulfilled criteria for a major depressive episode in the last 12 months or not.

Analytic approach

First, missing values were examined, and the data were imputed accordingly using MICE (Multivariate Imputation by Chained Equations; van Buuren and Groothuis-Oudshoorn, Reference van Buuren and Groothuis-Oudshoorn2010) in R package for multivariate imputation by chained equations. Second, sample characteristics were computed.

Third, the primary statistical analysis tested the association between the total MMSE scores as a function of the number of childhood infectious diseases using regression models. The assumptions of the regression models were tested. Visual inspection of residual figures was performed in order to reveal deviations from homoscedasticity or normality. An inspection for normality of error terms followed using a histogram and probability plots of the residuals. Independence of the error term was examined through a scatter plot of residuals by the predicted values to show that no discernible association existed. Next, multiple regression analysis models were computed. Regression models were computed in ascending complexity and tested without adjustment and adjusted for confounding of age at the time of data collection, sex, highest education achieved, smoking status, body mass index, and depression. Models were numbered as follows. The first model accounted for a linear effect of the number of childhood infectious diseases on MMSE scores (model 1 hereafter). The second model accounted for a linear effect of the number of childhood infectious diseases, age, sex, education level, smoking status, body mass index, and depression on MMSE scores (model 2 hereafter). The third model accounted for a quadratic effect of the number of childhood infectious diseases on MMSE scores (model 3 hereafter). The fourth model accounted for a quadratic effect of the number of childhood infectious diseases, age, sex, education level, smoking status, body mass index, and depression on MMSE scores (model 4 hereafter). The fifth model accounted for a cubic effect of the number of childhood infectious diseases on MMSE scores (model 5 hereafter). The sixth model accounted for a cubic effect of the number of childhood infectious diseases, age, sex, education level, smoking status, body mass index, and depression on MMSE scores (model 6 hereafter).

Fourth, the six models were compared for parsimony based on the Bayesian Information Criterion (BIC) for model selection (Schwarz, Reference Schwarz1978), similar to prior research (Rotstein et al., Reference Rotstein2018). Lower BIC values represent more parsimonious models and so are a better fit to the data. The best fitting model was then chosen based on the lowest BIC score and plotted using the ggplot2 library (Wickham, Reference Wickham2011). All analyses were computed in R (R Core Team, 2018).

Fifth, the robustness of the primary results was tested in 17 sensitivity analyses. The most parsimonious regression model was recomputed in subgroups with differential effects on cognitive functioning. First, the most parsimonious model was recomputed without sex as a covariate for females then males, since females are at greater risk for impaired cognitive functioning (van der Flier and Scheltens, Reference van der Flier and Scheltens2005). Second, the most parsimonious model was recomputed for persons aged 65–75 then aged 75–85, since cognitive functioning is related to age (van der Flier and Scheltens, Reference van der Flier and Scheltens2005). Third, the most parsimonious model was computed for each of the three diseases (i.e. chicken pox, measles, and mumps) separately to show their individual effect on cognitive functioning. Fourth, the most parsimonious model was computed using a saturated model in which dummy variables for one, two, or three childhood infectious diseases were entered as covariates. Fifth, the most parsimonious model was computed for each of the MMSE subscales (i.e. orientation, registration, attention and calculation, recall, and language) separately. Sixth, the most parsimonious model was computed without education to account for effects of mediation. Seventh, possible methodological confounders were considered. Although dichotomizing may be problematic (Altman and Royston, Reference Altman and Royston2006), for robustness and because dichotomous models have increased clinical lure, the most parsimonious model was recomputed dichotomized to account for intact cognitive functioning versus impaired cognitive functioning. Impaired cognitive functioning was defined using a cutoff score of 1.5 standard deviations below the mean score of the MMSE (Palmer et al., Reference Palmer2002). Additionally, the most parsimonious model was recomputed for observed data with missing values and compared to the primary analysis based on imputed data (Sterne et al., Reference Sterne2009).