Early life mental health symptoms and objective health indicators in - - PDF document

1

Early life mental health symptoms and objective health indicators in midlife and early old age: Evidence from the 1958 British birth cohort

2

Introduction

Major depression and anxiety disorders appear in the top 10 causes of global

burden of disease(Vigo, Thornicroft, & Atun), with major depression also being the second leading cause of disability worldwide and a major contributor to the burden of suicide and ischemic heart disease ("Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013," 2015). The public health burden of these common psychological disorders is estimated to continue to increase ("Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013," 2015; Lopez & Murray, 1998). Early life mental health is known to be associated with psychological distress and other mental health phenotypes in adulthood (Clark, Rodgers, Caldwell, Power, & Stansfeld, 2007), as well as socio-economic outcomes (Colman et al., 2009) and the huge costs to society, and to the economy, of poor mental health are undisputed (Layard, 2013) as inequalities due to both social causation and selection are well documented (Goodman, Joyce, & Smith, 2011; C. Power, Stansfeld, Matthews, Manor, & Hope, 2002; Stansfeld, Clark, Rodgers, Caldwell, & Power, 2011). These findings along with those that link early life mental health with less physical activity and more alcohol use in adulthood (Maggs, Patrick, & Feinstein, 2008; PINTO PEREIRA, LI, & POWER, 2015) provide plausible mechanisms of action through which early life mental health may impact on health

• utcomes in adulthood and consequently the central hypothesis of our study was that

early life mental health is associated with objective markers of health in adulthood. Despite these plausible mechanisms of action, there is a paucity of studies documenting the prospective association of early life mental health with objectively measured markers of health in adulthood. Another limitation of the existing literature is that early life mental health is considered at a single time point (Winning, Glymour, McCormick, Gilsanz, & Kubzansky, 2015), therefore neglecting the developmental perspective in the emergence of mental health symptomatology in childhood and adolescence. The development of mental health symptoms through childhood is complex and single time point or population average estimates of symptom development over time

3

can obscure subgroups with different patterns of symptoms (Patalay, Moulton, Goodman, & Ploubidis, 2017). The absence of a single population trend for development in psychopathology across childhood and adolescence is well established and the importance and relevance of studying heterogeneous trajectories

• f symptom development has been long acknowledged (Patalay et al., 2017). In this

paper we capitalise on the availability of three assessments of mental health symptoms in childhood and adolescence and derive a longitudinal typology of early life internalising and externalising symptoms in a population based prospective birth cohort, to investigate their association with objective measures of health and disability in midlife and all – cause mortality by age 55. By doing so we were able to formally empirically test the sensitive/critical period and accumulation of risk hypotheses (Ben- Shlomo, Cooper, & Kuh, 2016) with respect to the development of internalising and externalising symptoms from childhood to adolescence

Methods

Data The National Child Development Study (NCDS) (C. Power & Elliott, 2006) is the second oldest of the British birth cohort series, with 10 major follow-ups since birth. The initial sample of 17,415 individuals – consisting of all babies born in Great Britain in a single week in 1958 – are now approaching 60 years of age (most recent follow- up at age 55), providing high quality prospective data on social, biological, physical, and psychological phenotypes at every sweep, with 9,279 study members interviewed in person in 2008. In 2002, when respondents were 44-45 years old, a biomedical survey was collected for more than 9,000 respondents. This survey collected objective measures of health, blood samples were collected from 88% of those examined, and 8018 blood samples were received from subjects who gave consent to extraction of

• DNA. In this work we make use of the biological markers obtained from this survey.

Measures Exposure – Early life mental health Externalising and Internalising symptoms in childhood were assessed using the modified version of the Rutter ‘A’ scale (Rutter, Tizard, & Whitmore, 1970). This

4

version of the scale was completed by the mothers of the participants at ages 7 and 11, but from both mother and teachers at 16. Mother and teacher reports were

• combined. Internalising symptoms include being worried, solitary or miserable, while

externalising behaviour includes symptoms such as being disobedient, destructive and/or having poor concentration. Outcomes We included objectively measured health indicators at age 43/44, these were: Fibrinogen: a marker of inflammation and cardiovascular disease (g/L); C-reactive protein: an indicator of inflammation and cardiovascular disease (g/L); Glycated haemoglobin (HbA1c): index of glucose metabolism over the previous 30−90 days, which is used as a marker of diabetes mellitus; HDL and LDL cholesterol as markers

• f cardiometabolic risk; High blood pressure: three measures of systolic and diastolic

blood pressure were taken. The mean of valid readings was used, and an individual was recorded as having high blood pressure if the average value was above 140/90; Obesity: the body mass index was calculated using information on height and weight, with obesity defined as a BMI greater than 30; Waist to Hip Ratio (WHR): waist and hip circumferences were measured and the ratio of waist over hip calculated. To assess respiratory function we used forced expiratory volume (FEV), with the highest measurement used as a valid one. FEV is a measure of how much air a person can exhale during forced breath during the first second; Disability at age 55 was assessed according to the 2010 equality act; All-cause mortality up to age 55 was recorded by NHS Digital notifications combined with information of the address database held at the Centre for Longitudinal Studies Confounders We included various confounders from ages 0 to 7 that have been previously shown to be associated both with mental health and health in adulthood (Miech, Power, & Eaton, 2007; Chris Power, Jefferis, & Manor, 2010; C. Power et al., 2002; Tabassum et al., 2008). These included birth characteristics: Birthweight; Maternal smoking during pregnancy; Maternal age; Breastfeeding; Parental characteristics: Mother working up to 5; Parents read to child; Parental interest in school; Divorce; Separation from child; Indicators of Socio Economic Position: Paternal social class at birth; Financial difficulties; Age mother stayed at school; Housing tenure; Access to

5

amenities; Housing difficulties; Characteristics of the cohort member Cognitive ability; Enuresis; Summary of objectively assessed health conditions; BMI. Statistical Modelling To derive latent summaries of externalising and internalising at 7, 11 and 16 we modelled the probability of response to the Rutter items with a 2 parameter probit unidimensional latent variable measurement model (Muthen, 1984; Rabe-Hesketh & Skrondal, 2008). The six latent summaries were entered in a latent profile analysis model (Nagin & Tremblay, 2005) that has been used in a wide range of applications (Colman, Ploubidis, Wadsworth, Jones, & Croudace, 2007) (Mavandadi, Rook, & Newsom, 2007) (Sturgis & Sullivan, 2008) in order to derive a longitudinal typology of mental health from ages 7 to 16. This approach relaxes the somewhat restrictive assumption that all individuals follow the same trajectory of externalising and internalising symptoms. The longitudinal typology was then used as a predictor of various health outcomes in midlife, disability at age 55 and all-cause mortality in multivariable models with appropriate link functions (identity. logit, Cox where appropriate). Recognising that unbiased estimates cannot obtained without properly addressing the implications of incompleteness/missing data e employed Multiple Imputation with chained equations 20 imputed datasets) using all variables in the model as well as auxiliary variables in imputation process as has been suggested in the literature (Carpenter & Kenward, 2012). Multiple imputation operates under the Missing at Random (MAR) assumption (Little & Rubin, 2002), which in this case implies that our estimates are valid if missingness is due to variables included in our models. Sensitivity analysis To further probe the assumption of no unmeasured confounders/no omitted variables, we engaged into two forms of sensitivity analysis: First, we capitalised on the availability of three assessments of mental health and estimated a lagged model, where mental health assessments at 7 and 11 (the lags) were added in the model as confounders of the association between mental health at age 16 and health outcomes in midlife. The rationale between this is that the lags block associations from unmeasured confounders with mental health at 16 therefore as shown in Directed Acyclic Graph 1, therefore protecting from unmeasured confounding/omitted variable

6

• bias. In this model we added more confounders at age 11, these were: divorce,

cognitive ability, overcrowding, illness, parental years of education, enuresis, body mass index. The second sensitivity analysis involved the use of negative controls(Lipsitch, Tchetgen, & Cohen, 2010). A negative control is a variable (outcome

• r exposure) for which there is no plausible Mechanism Of Action (MOA) other than

confounding or measurement error that links it with the actual exposure or outcome in the model of interest In other words, negative control analysis is expected to produce a null result. If an association with a negative control is observed, then this would be an indication for bias due to unmeasured confounders and/or measurement error. In this instance w employed three negative control outcomes at age 43/44: i) hair colour; ii) arm were blood was taken from by the nurse during home visit; iii) ear tested first during home visit.

Results

In both men and women a four group solution was selected based on various criteria and quality of allocation/misclassification. Unlike findings in more recently born cohorts, externalising and internalising dimensions “hang together” empirically. Their means are very similar in all waves (dots in the graph very closely aligned) and they correlate highly (>0.75) within each sweep. We preferred to include both in the model instead of a single psychological distress dimension to be consistent with the existing

• literature. In men the four latent longitudinal groups were: i) Low Stable (28%),

absence of symptoms at all ages; ii) Moderate Low (33%): Moderate experience of symptoms at 7 and 11, absence of symptoms at 16; iii) Low Childhood High Adolescence (25%): Absence of symptoms at 7 and 11, experience of symptoms at 16; v) High Stable (14%): Persistent experience of both internalising and externalising symptoms from ages 7 to 16. The four groups in women were: i) Low Stable (27%), absence of symptoms in all ages (7 to 16); ii) Moderate Low (38%): Moderate experience of symptoms at ages 7 and 11, absence of symptoms at 16; iii) Low Childhood High Adolescence (22%): Absence of symptoms at ages 7 and 11, experience of symptoms at 16; iv) Stable High (13%): Persistent experience of both internalising and externalising symptoms from ages 7 to 16. In Table 3 we present the linear regression coefficients/odds ratios and associated 95% confidence intervals for men. In all models the “Stable Low” are the reference

7

• group. Early life mental health was associated with fibrinogen, C Reactive Protein,

FEV, disability at 55 and all-cause mortality. However the association differed between

• utcomes. It was only the “Low Childhood High Adolescence” group that scored higher
• n fibrinogen and C-Reactive Protein and lower on FEV indicating that they have

worse health outcomes compared to the Low Stable group. They were also more likely to die earlier. The Stable High group were more likely to be disabled and more likely to die earlier compared to the Stable Low group. In Table 4 we present the linear regression coefficients/odds ratios and associated 95% confidence intervals for women. As in men, in all models the “Stable Low” are the reference group. In women many more associations compared to men were observed. With the exception of high blood pressure, there were associations between early life mental health and all other outcomes. All three groups, Moderate Low, Low Childhood High Adolescence and Stable High had higher fibrinogen, C – Reactive Protein, LDL cholesterol and HbA1c, as well as lower LDL cholesterol and FEV. All three groups were more likely compared to the Stable Low group to be obese, have high waist to hip ratio and be disabled at 55. With respect to all-cause mortality only the Stable High group were more likely to die earlier compared to the Low Stable group. The Low Childhood High Adolescence had the more pronounced differences with the Stable Low group, indicating that it was the least healthy compared to all other groups. In Graphs 1 and 2 we present the lagged model linear regression coefficients/odds ratios and associated 95% confidence intervals for men and women that quantify the associations between mental health at 16 and all outcomes. We see that despite the lags and added confounders added in the model we observe similar associations as in the main analysis indicating that our results are likely not due to unmeasured confounding/omitted variable bias. This was further confirmed by analysis with negative controls, where all models returned null findings (results not presented here, available from corresponding author).

8

Discussion We found that externalising and internalising symptoms in childhood/adolescence

are associated with various biomarkers at 44, all-cause mortality (up to age 55) and disability at age 55. The associations were more pronounced in women. The accumulation of risk hypothesis was confirmed in men with respect to disability and all-cause mortality, whereas for all outcomes expect blood pressure in women. The sensitive period hypothesis was confirmed for all outcomes except blood pressure in women, indicating that experience of symptoms at any age between 7 and 16 will lead to worse health and mortality outcomes in adulthood. However, in both men and women it appears that experiencing mental health symptoms only at age 16 is more detrimental with respect to adult health and mortality. From a policy perspective, it appears that mental health interventions in adolescence are more likely to make an impact on adult health and mortality, especially in women. Considering that women had lower mean externalising and similar internalising scores with men in childhood and adolescence a solely genetic explanation for the more pronounced associations between early life mental and health in adulthood seems unlikely. What seems more likely is that either societal pressures and/or gender related inequalities exacerbate the effect of early life mental health on adult health in women, or considering that depression has a stronger genetic component in women, a gene environment interaction may underlie the observed gender differences. For both men and women we found that experiencing mental health symptoms in adolescence has a particularly strong effect on adult health and mortality, whereas experiencing symptoms in childhood has a weaker (in women) or null (in men) effect. Considering that adolescence is a period marked by various physical, neurodevelopmental as well as psychosocial changes, it is difficult to speculate whether it’s the interaction of these with the experience of mental health symptoms that lead to worse outcomes in adulthood, or whether in this particular group it’s the severity and/or timing of these changes along with other environmental and social influences that cause the emergence of mental health symptoms. This group is characterised by significantly lower parental interest in school, lower cognitive ability and higher likelihood of maternal smoking during pregnancy, but is similar to other groups with respect to parental socio-economic position and health status, signifying that parenting style and interest interacting with the many changes taking place may

9

explain the emergence of mental health symptoms in adolescence and their strong link with adult health and mortality. Our findings have Implications for public health policy, especially if mental health is worse in more recently born cohorts in the UK as it has recently been shown, considering that they are expected to live longer(Ploubidis, Sullivan, Brown, & Goodman, 2017). Strengths of this study include the availability of a large population based representative prospective study, the three assessments of early life mental health and our modelling strategy that allowed us to formally test the accumulation and critical period hypotheses. However, our findings should be considered along with the limitations of our study. Our data are derived from an observational longitudinal study and bias due to unmeasured confounding/omitted variable bias cannot be ruled

• ut. Results from sensitivity analysis with lagged models and negative controls give

some reassurance that our results are nor biased, but formal causal identification was not achieved due to the nature of the data. Another limitation considers the self- reported nature of early life mental health, our latent variable modelling strategy corrects for measurement error in the form of unique variance in the mental health symptoms, but the extent to which undetected systematic error common to all items may have biased our results is unknown. The next stage of our work will involve the use of polygenic risk scores for early mental health and where available for our

• utcomes in a genetically informed design utilising one and two sample Mendelian

Randomisation (Didelez & Sheehan, 2007; Lawlor, 2016) in order to obtain formal causal identification. We also plan to extend our analysis to the more recently born 1970 British Birth Cohort for which objective measure of health will be soon available.

References

Ben-Shlomo, Y., Cooper, R., & Kuh, D. (2016). The last two decades of life course epidemiology, and its relevance for research on ageing. International Journal

• f Epidemiology, 45(4), 973-988. doi:10.1093/ije/dyw096

Carpenter, J., & Kenward, M. (2012). Multiple imputation and its application: John Wiley & Sons.

10

Clark, C., Rodgers, B., Caldwell, T., Power, C., & Stansfeld, S. (2007). Childhood and adulthood psychological ill health as predictors of midlife affective and anxiety disorders: the 1958 British Birth Cohort. Arch Gen Psychiatry, 64(6), 668-678. doi:10.1001/archpsyc.64.6.668 Colman, I., Murray, J., Abbott, R. A., Maughan, B., Kuh, D., Croudace, T. J., & Jones,

• P. B. (2009). Outcomes of conduct problems in adolescence: 40 year follow-up
• f national cohort. British Medical Journal, 338. doi:10.1136/bmj.a2981

Colman, I., Ploubidis, G. B., Wadsworth, M. E. J., Jones, P. B., & Croudace, T. J. (2007). A longitudinal typology of symptoms of depression and anxiety over the life course. Biological Psychiatry, 62(11), 1265-1271. doi:10.1016/j.biopsycli.2007.05.012 Didelez, V., & Sheehan, N. (2007). Mendelian randomization as an instrumental variable approach to causal inference. Statistical Methods in Medical Research, 16(4), 309-330. doi:10.1177/0962280206077743 Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. (2015). Lancet, 386(9995), 743-800. doi:10.1016/s0140-6736(15)60692-4 Goodman, A., Joyce, R., & Smith, J. P. (2011). The long shadow cast by childhood physical and mental problems on adult life. Proc Natl Acad Sci U S A, 108(15), 6032-6037. doi:10.1073/pnas.1016970108 Lawlor, D. A. (2016). Commentary: Two-sample Mendelian randomization:

• pportunities and challenges. International Journal of Epidemiology, 45(3),

908-915. Layard, R. (2013). Mental health: the new frontier for labour economics. IZA Journal

• f Labor Policy, 2(1), 1-16.

Lipsitch, M., Tchetgen, E. T., & Cohen, T. (2010). Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies. Epidemiology (Cambridge, Mass.), 21(3), 383-388. doi:10.1097/EDE.0b013e3181d61eeb Little, R. J. A., & Rubin, D. B. (2002). Statistical Analysis with Missing Data (Second Edition ed.). Chichester: Willey. Lopez, A. D., & Murray, C. C. (1998). The global burden of disease, 1990-2020. Nat Med, 4(11), 1241-1243. doi:10.1038/3218

11

Maggs, J. L., Patrick, M. E., & Feinstein, L. (2008). Childhood and adolescent predictors of alcohol use and problems in adolescence and adulthood in the National Child Development Study. Addiction, 103 Suppl 1, 7-22. doi:10.1111/j.1360-0443.2008.02173.x Mavandadi, S., Rook, K. S., & Newsom, J. T. (2007). Positive and negative social exchanges and disability in later life: An investigation of trajectories of change. Journals of Gerontology Series B-Psychological Sciences and Social Sciences, 62, S361-S370. Miech, R., Power, C., & Eaton, W. W. (2007). Disparities in psychological distress across education and sex: A longitudinal analysis of their persistance within a cohort

• ver

19 years. Annals

• f

Epidemiology, 17(4), 289-295. doi:10.1016/j.annepidem.2006.07.015 Muthen, B. (1984). A General Structural Equation Model with Dichotomous, Ordered Categorical, and Continuous Latent Variable Indicators. Psychometrika, 49(1), 115-132. Nagin, D. S., & Tremblay, R. E. (2005). What has been learned from group-based trajectory modeling? Examples from physical aggression and other problem

• behaviors. Annals of the American Academy of Political and Social Science,

602, 82-117. doi:10.1177/0002716205280565 Patalay, P., Moulton, V., Goodman, A., & Ploubidis, G. B. (2017). Cross-domain Symptom Development Typologies and Their Antecedents: Results From the UK Millennium Cohort Study. Journal of the American Academy of Child & Adolescent Psychiatry. doi:https://doi.org/10.1016/j.jaac.2017.06.009 PINTO PEREIRA, S. M., LI, L., & POWER, C. (2015). Early Life Factors and Adult Leisure Time Physical Inactivity Stability and Change. Medicine & Science in Sports & Exercise, 47(9), 1841-1848. doi:10.1249/mss.0000000000000610 Ploubidis, G. B., Sullivan, A., Brown, M., & Goodman, A. (2017). Psychological distress in mid-life: evidence from the 1958 and 1970 British birth cohorts. Psychol Med, 47(2), 291-303. doi:10.1017/s0033291716002464 Power, C., & Elliott, J. (2006). Cohort profile: 1958 British Birth Cohort (National Child Development Study). International Journal of Epidemiology, 35(1), 34-41. doi:10.1093/ije/dyi183 Power, C., Jefferis, B. J. M. H., & Manor, O. (2010). Childhood Cognition and Risk Factors for Cardiovascular Disease in Midadulthood: The 1958 British Birth

12

Cohort Study. American Journal of Public Health, 100(1), 129-136. doi:10.2105/AJPH.2008.155564 Power, C., Stansfeld, S. A., Matthews, S., Manor, O., & Hope, S. (2002). Childhood and adulthood risk factors for socio-economic differentials in psychological distress: evidence from the 1958 British birth cohort. Social Science & Medicine, 55(11), 1989-2004. doi:10.1016/s0277-9536(01)00325-2 Rabe-Hesketh, S., & Skrondal, A. (2008). Classical latent variable models for medical research. Statistical Methods in Medical Research, 17(1), 5-32. doi:10.1177/0962280207081236|issn 0962-2802 Rutter, M., Tizard, J., & Whitmore, K. (1970). Education, health and behaviour: Longman Publishing Group. Stansfeld, S. A., Clark, C., Rodgers, B., Caldwell, T., & Power, C. (2011). Repeated exposure to socioeconomic disadvantage and health selection as life course pathways to mid-life depressive and anxiety disorders. Soc Psychiatry Psychiatr Epidemiol, 46(7), 549-558. doi:10.1007/s00127-010-0221-3 Sturgis, P., & Sullivan, L. (2008). Exploring social mobility with latent trajectory groups. Journal of the Royal Statistical Society Series a-Statistics in Society, 171, 65- 88. Tabassum, F., Kumari, M., Rumley, A., Lowe, G., Power, C., & Strachan, D. P. (2008). Effects of socioeconomic position on inflammatory and hemostatic markers: a life-course analysis in the 1958 British birth cohort. Am J Epidemiol, 167(11), 1332-1341. doi:10.1093/aje/kwn055 Vigo, D., Thornicroft, G., & Atun, R. Estimating the true global burden of mental illness. The Lancet Psychiatry, 3(2), 171-178. doi:10.1016/S2215-0366(15)00505-2 Winning, A., Glymour, M. M., McCormick, M. C., Gilsanz, P., & Kubzansky, L. D. (2015). Psychological Distress Across the Life Course and Cardiometabolic Risk: Findings From the 1958 British Birth Cohort Study. J Am Coll Cardiol, 66(14), 1577-1586. doi:10.1016/j.jacc.2015.08.021

13 Table 1. Descriptive Statistics of outcomes Men Women Mean

• r %

SD N Mean

• r %

SD N Cardio-metabolic Risk Factors Fibrinogen (g/L) 2.88 (0.58) 3,845 3.03 (0.65) 3,838 C-Reactive Protein (g/L) 1.97 (4.40) 3,856 2.38 (4.16) 3,836 Glycated Haemoglobin 5.32 (0.76) 3,976 5.19 (0.63) 3,947 Total Cholesterol 6.07 (1.14) 3,927 5.70 (1.00) 3,897 HDL Cholesterol 1.43 (0.34) 3,914 1.69 (0.41) 3,894 LDL Cholesterol 3.57 (0.93) 3,571 3.29 (0.87) 3,820 Cholesterol Ratio (Total/HDL) 4.42 (1.19) 3,914 3.54 (1.00) 3,893 % with High Blood Pressure 16.0 4,608 5.58 4,622 BMI 27.8 (4.27) 4,585 26.9 (5.53) 4,625 % Obese 25.3 4,585 23.5 4,625 Waist to Hip Ratio 0.93 (0.06) 4,629 0.81 (0.06) 4,670 % Obese using WHR 34.0 4,629 25.1 4,670 Respiratory Functions Forced Expiratory Volume (lt.) 3.73 0.83 4,522 2.75 0.60 4,568 N in Biomedical Survey 4,665 4,712

14 Table 2. Descriptive Statistics - Confounders Men Women Mean

• r %

SD N N in the Sweep Mean

• r %

SD N N in the Sweep Early Life Socioeconomic Background Social Class at Birth - % Manual 68.2 7,227 9,004 68.1 6,887 8,411 % in Financial Hardship - Age 11 11.1 6,857 7,887 11.6 6,506 7,450 % Overcrowding - Age 11 12.0 7,065 7,887 12.1 6,727 7,450 Housing: % NO Access to 1+ (Bathroom; Indoor WC; Cooking Facilities; Hot water) - Age 11 10.82 6,989 7,887 10.9 6,640 7,450 % with Family Difficulties - Age 7 4.46 7,152 7,917 4.03 6,794 7,508 % with Divorced Parents - Age 11 4.53 7,886 7,887 4.66 7,450 7,450 % Mother in School after Minimum Age - Age 0 24.8 8,970 9,004 25.1 8,383 8,411 % Parents interested in R Education - Age 11 76.1 7,206 7,887 77.9 6,830 7,450 Early Life Health Birth Weight (ounces) 119.0 (22.0) 8,959 9,004 114.1 (21.1) 8,382 8,411 % Mother smoking when pregnant 33.2 9,004 9,004 33.2 8,411 8,411 % Out of school for 1+ months - Age 11 4.88 7,803 7,887 5.51 7,370 7,450 # times hospitalized - Age 11 0.74 (0.98) 7,089 7,887 0.57 (0.85) 6,736 7,450 % with Enuresis - Age 7 14.0 7,467 7,917 11.6 7,074 7,508 % with Enuresis - Age 11 7.37 7,069 7,887 4.73 6,723 7,450 Physical Coordination Problems - Age 11 17.2 7,028 7,887 13.5 6,624 7,450 Cognitive Ability General ability test score - Age 11 41.8 (16.3) 7,253 7,887 44.1 (15.9) 6,878 7,450

15

Table 3. Linear regression coefficients/odds ratios and 95% confidence intervals - Men

16

Table 4. Linear regression coefficients/odds ratios and 95% confidence intervals - Women

17

Directed Acyclic Graph 1. Lagged model sensitivity analysis

18

Graph 1. Lagged model sensitivity analysis results – Linear regression coefficients

19

Graph 2. Lagged model sensitivity analysis results – Odds and hazard ratios