Mental Illness Among Youth With Chronic Physical Conditions

We conducted a retrospective cohort study using 11 panels of the Medical Expenditure Panel Survey (MEPS) spanning from 2003 to 2014, and our final sample included 48 572 youth aged 6 to 25 years. The MEPS is a nationally representative survey of the US population conducted by the Agency for Healthcare Research and Quality.²⁹ Each year, the MEPS approaches a representative sample of US households about participating and consenting adult heads of households provide health information on each family member living within their household; college students living away from home are considered their own household. Computer-assisted interviews were used to collect data for a total of 5 interviews covering 2 calendar years. Our main inclusion criteria were (1) age 6 to 25 years (ie, old enough to reliably receive a mental health diagnosis), (2) no MHCs at baseline, and (3) no conditions likely to impair cognitive functioning in a way that might blur neurologic and mental illness (eg, developmental delay or autism). The average response rate for the included MEPS years was 57%, and the participant retention rate for our sample was 99%. The Boston Children’s Hospital Institutional Review Board considered this study exempt from review.

We used all available data within MEPS to identify youth with MHCs and CPCs. The MEPS asks respondents open-ended questions about “health problems that have actually bothered anyone in the family” and “conditions related to medical encounters that occurred during the reporting period”; professional coders then translate verbatim text responses into 3-digit International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. The MEPS also poses closed-ended questions about “priority” medical conditions (eg, asthma, attention-deficit/hyperactivity disorder, and diabetes).

We applied the Agency for Healthcare Research and Quality’s Chronic Condition Indicator to first separate youth with chronic conditions from those without.¹ The Chronic Condition Indicator is used to determine whether an ICD-9-CM code represents a chronic condition on the basis of whether the condition is expected to last at least 12 months and either (1) places limitations on self-care, independent living, and social interactions or (2) results in the need for ongoing intervention with medical products, services, or special equipment. We then dichotomized chronic conditions into MHCs or CPCs using ICD-9-CM codebook categories. Common MHCs identified in this way include major depression and generalized anxiety disorder, whereas common CPCs include asthma, migraine, and diabetes.

Our outcome of interest was the timing of incident MHCs, defined as the time from baseline to the time when the first MHC, if any, was identified. Our exposure of interest was the identification of any CPCs during the baseline reporting period. Our mediator of interest was the presence of activity limitations. These were identified by using MEPS questions regarding whether participants are limited in their ability to “work at a job, do housework, or go to school” or “participate in social, recreational, or family activities” because of an “impairment or a physical or mental health problem.” We used these questions, which are asked at 3 time points, to construct a time-varying covariate for any limitations versus none. We attempted to limit activity limitations to those associated with CPCs (rather than MHCs) by restricting our sample to youth without baseline MHCs and only observing activity limitations occurring before the onset of MHCs.

We included the following covariates a priori in all analyses: age, sex, race and/or ethnicity, family income, highest family educational achievement, insurance coverage, presence of a usual source of care, US region of residence, and metropolitan statistical area (MSA) status (urban or rural). We also adjusted for panel year in multivariate analyses after examining MHC incidence rates across panel years. Age, family income, insurance coverage, usual source of care, MSA status, and region were treated as time-varying covariates.

For all analyses, we corrected SEs due to clustering within strata and the primary sampling unit. To ensure that the sample was representative of the noninstitutionalized US population, we applied survey weights in accordance with MEPS guidelines; these account for the complex survey design and the unequal probability of selection and survey nonresponse.²⁹ We used SAS 9.4 (SAS Institute, Inc, Cary, NC) for all analyses. We characterized the clinical and sociodemographic characteristics of our study sample using descriptive statistics, looking for bivariate associations using χ² statistics.

We calculated a 2-year cumulative incidence rate of MHCs by dividing the numbers of youth who received an MHC diagnosis within the 2-year follow-up period by the total number in our 11-year study sample.

We used Cox proportional hazards models (regression models that allow the comparison of event rates, also known as “hazards”, between 2 groups while adjusting for covariates³⁰) to compare the hazard rates of MHC diagnosis within a 2-year follow-up period among youth with baseline CPCs with youth without CPCs at baseline. To measure the timing of incident MHCs, we determined the time from the beginning of the baseline reporting period until the date of participants’ first MHC, censoring participants at the date of their final interview. Cox models included CPC status at baseline, sex, race and/or ethnicity, highest family educational achievement, and panel year as fixed covariates and age, family income, insurance coverage, usual source of care, MSA status, and region as time-varying covariates. We also used Cox models to generate cumulative incidence curves for youth with and without baseline CPCs, stratified by age category (6–11, 12–18, or 19–25 years) and adjusting for covariates. We tested the proportional hazards assumption of the Cox model by visually inspecting the Schoenfeld residuals (the observed minus expected values from the model for each time point; these are expected to be centered on zero if the hazard ratio for the main effect is constant across the observation window). We also tested for an association between the Schoenfeld residuals and time, finding no significant association (P = .47), suggesting that the proportional hazards assumption was satisfied.³¹ 

We conducted a mediation analysis by evaluating how the coefficient for the exposure (baseline CPC) changed once the mediator (activity limitations) was added to the model. The magnitude of the mediation effect can be expressed via a ratio of the regression coefficients for the exposure between models with and without the mediator (the mediation ratio).³² 

We used the Sobel test to evaluate the significance of any resulting mediation effect. This is a specialized test used to determine if the indirect effect (ie, the proportion of the main effect of the exposure that operates through the mediator) is statistically significant.^33,34 

Because MHCs are heterogenous, we explored whether the effect of CPCs differed for different MHCs by constructing separate Cox models for the 3 most common types of mental health disorders: anxiety, mood, and behavior disorders (including attention-deficit/hyperactivity disorder). We also tested how 2 analytical choices could affect our findings. (1) Longer and shorter “lag” times between CPCs and MHC diagnosis could affect MHC rates because shorter lags could count MHCs as separate from CPCs when they are coincident, whereas longer lags might better capture subsequent diagnoses, so we repeated our analysis using a longer lag. (2) The relationship between CPCs and MHCs could also be affected by whether one considered the presence of CPCs only in the baseline reporting period or treated CPCs as a factor that could continue to develop across the 2-year follow-up period. We therefore repeated our analysis treating CPCs as time varying to see whether this affected our results.

In our sample of 48 572 US youth aged 6 to 25 years, 14.7% had CPCs at baseline (Table 1). Baseline CPC rates differed significantly by sex, race and/or ethnicity, family income, family educational attainment, insurance coverage, usual source of care, and region. The 2-year cumulative incidence of MHCs was 7.8% among all youth (Table 1). In general, incident MHCs were more likely to occur among older youth; the 3 diagnosis categories with the highest incidence were anxiety, mood, and behavior disorders, which had 2-year cumulative incidences of 3.2%, 2.5%, and 1.7%, respectively. Within these top 3 diagnoses, older youth were more likely to develop mood and anxiety disorders, whereas behavior disorders were highest among children aged 6 to 11 years (Fig 1A).

Among youth with CPCs at baseline, 11.5% developed incident MHCs in the subsequent 2 years compared with 7.1% in youth without baseline CPCs (P < .001) (Table 2). Youth with CPCs at baseline had a 51% greater adjusted risk (adjusted hazard ratio [aHR] 1.51; 95% confidence interval [CI] 1.30–1.74) of incident MHCs compared with youth with no baseline CPCs (Table 3; model 1). The adjusted cumulative incidence of MHCs differed by age groups, with the cumulative incidence being lower in the youngest age group (6–11 years) and higher in the older age groups (12–18 and 19–25 years) (Fig 1B).

Among all sampled youth, 1.8% reported any of the possible activity limitations (Table 2). Youth with baseline CPCs had a threefold greater odds of reporting activity limitations (adjusted odds ratio: 3.07; 95% CI 2.34–4.04; Fig 2) and youth reporting activity limitations had a 3.6-fold increased risk (aHR 3.60; 95% CI 2.83–4.57) of incident MHCs (Fig 2). After adding activity limitations to the Cox model, the association between CPCs and incident MHCs was decreased from 51% to 43% greater risk (aHR 1.43; 95% CI 1.23–1.66), yielding a mediation ratio of 13.5% (Sobel test z score = 6.38; P < .001; Fig 2), suggesting that activity limitations explained 13.5% of the relationship between CPCs and incident MHCs.

Youth with CPCs had increased MHC risk across all MHC types, with 51% (aHR 1.51; 95% CI 1.20–1.89), 70% (aHR 1.70; 95% CI 1.32–2.18), and 54% (aHR 1.54; 95% CI 1.17–2.04) greater risk of anxiety, mood, and behavior disorders, respectively, compared with youth who had no CPCs (Table 4). Our other sensitivity analyses indicated that the association between baseline CPC status and incident MHCs was robust to methodological approach. The pattern of association between sociodemographic factors and the outcome of interest varied slightly by methodological approach (Supplemental Table 5).

In this nationally representative cohort study, youth who began a 2-year period with a CPC had significantly greater risk of reporting an MHC than youth who did not begin that period with a CPC. Some youth with CPCs reported experiencing activity limitations, and these limitations, in turn, partly explained the relationship between having a CPC and the subsequent diagnosis of MHCs. Our findings suggest that having a CPC may be a risk factor for the subsequent development of MHCs in youth.

To our knowledge, the current study is the first to use nationally representative longitudinal data to estimate the risk of incident MHCs among youth with any CPCs. This work adds to the existing literature because the study design is longitudinal, data are prospectively gathered, and sampling is nationally representative. Our study is also somewhat distinctive in that it examines the potential role activity limitations may play in the development of mental illness among youth with CPCs. Despite differences in study design, our estimate of subsequent MHC risk among youth with CPCs is consistent with the meta-analysis by Secinti et al²² and the cross-sectional study by Tegethoff et al.²¹ This is especially notable given that the association observed in Secinti et al²² involved a long period of observation (ie, CPCs in childhood, MHCs in adulthood), strengthening the notion that the mental health impacts of CPCs are not transient.

Because having a pediatric-onset CPC is a risk factor for the subsequent development of MHCs, health care systems may want to consider ways to improve early detection and treatment of mental illness in this population. If CPCs indeed contribute to MHC development in youth, further investigation of the underlying biological, psychological, or social mechanisms is warranted to better understand how the development of MHCs in youth with CPCs might be interrupted. Our observation that activity limitations, although uncommon in our sample, explained >13% of the association between CPCs and MHCs is important and should be explored further to understand whether activity limitations function as a marker of CPC severity or as an etiologic mechanism in some youth. A link between activity limitations and MHC etiology is plausible because such limitations could reduce opportunities for protective exposures or lead to specific harms to mental health (eg, through decreased educational achievement,²⁶ increased social stigma,³⁵ or diminished social well-being²⁷). Such impacts may be particularly salient during adolescence and early adulthood, when gaining autonomy, forming identity, and developing positive relationships are predominant developmental tasks.^24,25 Activity limitations depend on a dynamic interplay between individuals and their environment,³ and there may be opportunities for health care systems to reduce such limitations in youth with CPCs (eg, through improvements in functional limitations or self-management skill³⁶). Policies that reduce barriers to participation in the physical or social environment may also reduce the prevalence or severity of activity limitations in these youth.^37,38 If indeed activity limitations contribute to the development of mental illness in some youth with CPCs, such interventions may be levers by which program or policy developers can reduce MHCs in this population.

We did not prespecify an examination of how sociodemographic background may independently contribute to the development or detection of CPCs and MHCs, so authors of future studies may want to focus on this matter. In the meantime, one should recognize that our sociodemographic covariates are correlated, so observed associations especially reflect available covariates. It is within this context that we observe that CPCs were more prevalent in youth with greater family education and access to care and that, holding all else constant, having greater household income or being Hispanic, African American, or Asian American race and/or ethnicity was associated with decreased risk of MHCs.

Our findings must be interpreted in light of several limitations. Although a nationally representative longitudinal study design is a strength, the association we observe may not be causal, and studies of durations >2 years are likely to be further elucidating. Also, chronic condition ascertainment relied on proxy report, which could lead to under- or overreporting of both physical health conditions and MHCs. MHCs in particular are likely to be underrecognized in youth.^39,40 Furthermore, we only had access to 3-digit ICD-9-CM codes; 5-digit ICD-9-CM codes may have provided more granular information on clinical severity.^41,42 However, condition ascertainment in the MEPS has several notable strengths, including prospective data gathering and triangulation of condition status by using open-ended questions, checklists, and questions about participants’ specific medical encounters. Finally, the MEPS lacks data on condition severity, which precluded inclusion of disease severity as a potential explanatory variable or mediator in our analysis.

Our findings indicate that youth with CPCs have increased risk of MHCs, suggesting that activity limitations may be a mechanism for this increased risk. Pediatricians are well positioned to support the mental health of youth with CPCs; primary care providers and subspecialists should include early assessment and treatment of MHCs as part of their care of youth with CPCs. Program and policy interventions to improve the care of youth with CPCs should consider the impact of CPCs on youth functioning and participation and should identify opportunities to both bolster individual strengths and to modify the physical or social environment to maximize youth ability to fully participate in activities important to their developmental progress.

aHR

adjusted hazard ratio

CI

confidence interval

CPC

chronic physical condition

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

MEPS

Medical Expenditure Panel Survey

MHC

chronic mental health condition

MSA

metropolitan statistical area