Longitudinal Trends in Childhood Insulin Levels and Body Mass Index and Associations With Risks of Psychosis and Depression in Young Adults

Key Points Question Are longitudinal trends in insulin levels and body mass index from childhood associated with adult depression and psychosis? Findings This cohort study of repeated-measure data from age 1 to 24 years in up to 10 463 individuals identified trajectories of fasting insulin levels and body mass index. Persistently high fasting insulin levels from age 9 years were associated with psychosis at 24 years, and puberty-onset body mass index increase was associated with depression at 24 years. Meaning This study’s findings suggest that changes in insulin sensitivity and adiposity starting from childhood may have disorder-specific associations with psychosis and depression and represent targets for prevention and treatment of cardiometabolic disorders in people with psychosis and depression.

replicating the estimation using the same seed values and comparing model parameter estimates for replication. A successfully converged model with no local solutions would have the best loglikelihood values repeated 9 . In selecting the optimum class solution, we aimed to select the solution with the lowest BIC, suitable statistical evidence (p<0.05) in VLMR-LRT and BLRT tests (suggesting the solution with n trajectories is an improvement over the solution with n-1 trajectories), high entropy values (close to 1.0), and no less than 1% of the total sample in a particular trajectory (to allow further analysis with adequate sample sizes).

Three-Step Method
After completing the first step described above, the second step is to calculate classification uncertainty, which is computed as a natural log of the average latent class probabilities for most likely class membership and the number of observations per trajectory class. These logits are used in the third step, which includes either regression on predictors of trajectory class membership (using trajectory class membership as an outcome), or regression of trajectory classes on an outcome (using trajectory class membership as a predictor). Detailed information on the statistical methodology underpinning the three-step method alongside data simulations are available elsewhere 10 . Mplus includes two methods with which to proceed with the three-step method of analysis. The first is the automatic method, in which either predictors of trajectory class membership, or outcomes, are added as auxiliary variables in the variable command, and specified as such (for example with the R3STEP option for predictors of trajectory class membership). The automatic method is suitable for simple univariable analysis, and we used this option to examine the associations of sociodemographic and lifestyle factors with trajectory class membership. The automatic method is unsuitable for analyses which include adjustment for confounders, and so the manual method must be used. In the manual method, the most likely trajectory class posterior distribution is obtained using the SAVEDATA command in MPlus, with the option SAVE=CPROB. In step 2, classification uncertainty is computed as a natural log of the average latent class probabilities for most likely class membership and the number of observations per trajectory class. These logits are provided within the Mplus output for step 1. In step 3, the desired model where the latent class variable is measured by the most likely class variable N and the measurement error for each trajectory class is fixed and prespecified to the logits computed in Step 2. In step 3, starting values are set to 0, since class membership was already determined in step 1. We used the manual method for our primary analysis examining the associations between trajectory class membership and psychiatric outcomes, since our analysis included adjusting for potential confounders. Please see an example Mplus scripts below for the manual method for our analysis of BMI trajectories.  Statistics classification system: I, II, III non-manual, III manual, IV, V). We coded a binary variable of 'lower social class' with "1" given to social class classification < III. A positive family history of cardiometabolic/cardiovascular disease was coded from self-report questionnaire data encompassing hypertension, T2DM, hypercholesterolaemia, or cardiovascular diseases. Stressful life events (SLEs) were based on self-report questionnaire data comprising a summed total of up to 42 pre-specified life events affecting the mother at 18-and 36weeks gestation, and affecting the participant at 8-weeks and 6-months postpartum. Examples included death or loss of a partner or family member, loss of employment, moving-house or financial difficulty. A full list of the 42 SLEs is reported elsewhere 11 . We compared the top tertile of summed SLE scores vs the bottom tertile. Birthweight and gestational age were coded as continuous variables derived from questionnaire data.

Time Variant Factors
Low exercise at ages 15 and 18 years was coded from self-report questionnaire data and defined as participating in any physical activity less than once per week on average in the past year. Smoking at ages 15 and 18 years was coded from self-report questionnaire data and defined as smoking on average >1 cigarette per day.

Clinical and Biochemical Phenotype of Trajectory Classes at Age 24
BMI was assessed during clinic assessment, from measures of height (m 2 ) and weight (kg). Waist circumference was measured during clinic assessment. Blood-based samples (fasting plasma glucose (FPG), fasting insulin (FI), highdensity lipoprotein, low-density lipoprotein, triglycerides and C-reactive protein) were after an 8-hour fast (water only). Samples were immediately spun, frozen and stored at -80 o C and measurements were assayed within 3 to 9 months of the samples being taken with no previous freeze-thaw cycles.

Association of Trajectory Membership with Metabolic Syndrome at Age 24
Metabolic syndrome was defined 12 as the presence of ethnicity-specific waist circumference (94cm in males and 80cm in females for caucasians; 90cm in males and 80cm in females in other ethnic groups 12 ) or raised BMI (>29.9), plus two from; elevated triglycerides (150mg/dL); reduced HDL (male <40mg/dL; female <50mg/dL); elevated systolic blood pressure (130mmHg) or elevated fasting plasma glucose (FPG) (100mg/dL). Logistic regression via the three-step method 10 was used to examine the association of trajectory membership for FI and BMI with metabolic syndrome at age 24, compared with the most common trajectory, before and after adjusting for the same confounders used in the primary analysis.