Supplemental Online Content

eMethods. eFigure 1. Study Population Flow Chart eFigure 2. Associations of Cardiovascular and Noncardiovascular Dysfunctions with Incident HF eTable 1. Systems Evaluated in Risk Models for Heart Failure eTable 2. Cardiovascular and Noncardiovascular Characteristics by Incident HF Subtypes in Women eTable 3. Cardiovascular and Noncardiovascular Characteristics by Incident HF Subtypes in Men eTable 4. Associations of Cardiovascular and Noncardiovascular Measures With Incident HF Events in Unipredictor Multivariable Models Additionally Adjusting For Income Status and Education Attainment eTable 5. Associations of Cardiovascular and Noncardiovascular Measures With Incident HF Events in Multipredictor Models eTable 6. Cox Regression Models for HF, HFpEF, and HFrEF with Multiple Imputation eTable 7. Associations of Cardiovascular and Noncardiovascular Measures With Incident HF Events in Multipredictor Models Excluding Participants With a Baseline LVEF Less Than 50% eTable 8. Exploration of Potential Interactions of Organ Function Measures with Sex for Incident HF eTable 9. Stratified Cox Regression Models for HF, HFpEF, and HFrEF by Sex Categories eReferences.

behavior; the Anxiety and Cooperation Scale ranges from 0 to 5, with a lower score indicating greater cooperation and higher score denoting greater distress, while the Frankl Scale ranges from 1-4 (1=definitely negative to 4=definitely positive) with a higher score indicating greater cooperation. Both tools have established reliability and validity. [17][18][19] Pain intensity. Child-report of pain during the dental cleaning was collected immediately after the dental encounter using the Faces Pain Scale-Revised. 20 This scale consists of six faces which express distress from no pain (0) to very much pain (10) and has strong psychometric properties within pediatric populations. 21,22 Sensory discomfort. Children reported how "bothered" they were by five different sensory stimuli (e.g., lights, sounds, smell, taste, movement) during the dental cleaning as well as the overall environment on the Dental Sensory Sensitivity Scale. 23 Each item was scored on a three-point Likert scaleno bother, a little bother, a lot of botherand summed across items, with a higher score indicating more sensory discomfort during the dental cleaning.

Data Analysis
Power Analysis. Original sample size calculations for this study determined that 165 children with two dental visits would be necessary for 80% power at a 2-sided alpha level of 0.05 to detect Cohen's d effect size differences of 0.22. Estimates for the effect sizes that would be realized ranged from 0.11-0.69 depending on the specific outcome in question, with an estimate of 0.44-0.46 for the primary outcome. The number of children with two dental visits was only 138 in this study, which contributed to low statistical power for some outcomes, while realized effect sizes were sometimes higher and sometimes lower than expected.
Primary Outcome. Electrodermal activity was scored offline using the BIOPAC AcqKnowledge program and hand-checked for accuracy and artifact exclusion, 24,25 with 25% of data double-coded to ≥85% agreement. NS-SCRs with amplitudes ≥0.05µS were converted to a rate of fluctuations per minute. The EDA variables were not normally distributed and a square root transformation was superior to a log transformation in terms of normality. 24,26 All EDA mean values reported are in their untransformed condition for ease of comparison for readers who might be familiar with these measures. Statistical models that used square root transformed values have beta estimates that are consistent with the square root transformations. The square root transformations have been noted in all applicable tables.
All the linear mixed effects regression analyses used restricted maximum likelihood, an unstructured covariance structure, and only the intercept was random. These models were adjusted for attained age and first or second clinic visit only. In crossover studies, in the absence of dropouts or missing outcome data, demographic variables which do not change over time (such as sex and race) can be effect moderators but not confounders since they are not associated with exposure; therefore, they need not be included as covariates. In fact, this property is often mentioned as a particular advantage for crossover studies. 27 Although adjusting for demographic variables in crossover studies with missing data using mixed effects regression can minimally effect results, it can have no effect on the other statistical methods that we used to validate our results.
For the paired sample t-test and Wilcoxon signed-rank tests only children with data from both dental visits were included; the RDE score was subtracted from the SADE score for each child The square root transformation was not used for Wilcoxon tests since the transformation would not change the rank, nor for the t-tests (single sample of differences between SADE and RDE) since the treatment differences were reasonably normally distributed.
Although it is reasonable to assume that children will have more favorable dental visit experiences when they are older and after having a previous clinical experience, such carry over effects would be expected to have minimal effect on a trial where treatment order is randomized and thus have not been included in analyses.
Mediation analyses. Physiological stress measures were assessed for mediation of outcomes that were significantly associated with treatment conditions. First, a linear mixed effects regression models for each of the four video-assessed behavioral distress outcomes (head movement frequency, mouth movement frequency, whimper/cry/scream frequency, whimper/cry/scream duration) assessed the effect of the treatment condition on these outcomes (adjusted for attained age and first or second clinic visit). Second, the same four models were run, but this time including square root transformed values for average skin conductance level and average NS-SCR frequency as covariates in each model. Comparison of the beta estimates for the treatment condition variable in corresponding models indicates if the association between treatment condition and outcome is attenuated by the mediators.
Moderation analyses. These analyses used linear mixed effects regression to model each mean EDA for averages and for frequencies (primary outcomes), and each of the four video-coded values for distress behavior   Note. RDE=regular dental environment; SADE=sensory adapted dental environment; FSIQ=Wechsler Abbreviated Scale of Intelligence Full Scale Intelligence Quotient (IQ score of 70 is used to determine intellectual disability); VABS= Vineland Adaptive Behavior Scales-II (raw score of 83 is equivalent to the expressive language of a four-year-old); CFSS-DS=Children's Fear Survey Schedule-Dental Subscale (score >32 indicates borderline or clinical dental fear).

eTable 1. Participant Characteristics by Number of Completed Dental Visits
a. Least squares mean (standard error) with adjustment for attained age and visit; adjusted means calculated at attained age and visit means. b. P-value for interaction of subgroup with treatment (SADE/RDE), i.e., difference in treatment effect by subgroup. c. Age at baseline d. Cut-score based on Schoen et al. 13